]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
53b381b3 | 43 | #include <linux/blkdev.h> |
f23b5a59 | 44 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 45 | #include <linux/uio.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
31193213 | 62 | #include "qgroup.h" |
39279cc3 CM |
63 | |
64 | struct btrfs_iget_args { | |
90d3e592 | 65 | struct btrfs_key *location; |
39279cc3 CM |
66 | struct btrfs_root *root; |
67 | }; | |
68 | ||
f28a4928 FM |
69 | struct btrfs_dio_data { |
70 | u64 outstanding_extents; | |
71 | u64 reserve; | |
72 | u64 unsubmitted_oe_range_start; | |
73 | u64 unsubmitted_oe_range_end; | |
74 | }; | |
75 | ||
6e1d5dcc AD |
76 | static const struct inode_operations btrfs_dir_inode_operations; |
77 | static const struct inode_operations btrfs_symlink_inode_operations; | |
78 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
79 | static const struct inode_operations btrfs_special_inode_operations; | |
80 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
81 | static const struct address_space_operations btrfs_aops; |
82 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 83 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 84 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
85 | |
86 | static struct kmem_cache *btrfs_inode_cachep; | |
87 | struct kmem_cache *btrfs_trans_handle_cachep; | |
88 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 89 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 90 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
91 | |
92 | #define S_SHIFT 12 | |
4d4ab6d6 | 93 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
94 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
95 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
96 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
97 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
98 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
99 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
100 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
101 | }; | |
102 | ||
3972f260 | 103 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 104 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 105 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
106 | static noinline int cow_file_range(struct inode *inode, |
107 | struct page *locked_page, | |
108 | u64 start, u64 end, int *page_started, | |
109 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
110 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
111 | u64 len, u64 orig_start, | |
112 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
113 | u64 orig_block_len, u64 ram_bytes, |
114 | int type); | |
7b128766 | 115 | |
48a3b636 | 116 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 117 | |
6a3891c5 JB |
118 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
119 | void btrfs_test_inode_set_ops(struct inode *inode) | |
120 | { | |
121 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
122 | } | |
123 | #endif | |
124 | ||
f34f57a3 | 125 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
126 | struct inode *inode, struct inode *dir, |
127 | const struct qstr *qstr) | |
0279b4cd JO |
128 | { |
129 | int err; | |
130 | ||
f34f57a3 | 131 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 132 | if (!err) |
2a7dba39 | 133 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
134 | return err; |
135 | } | |
136 | ||
c8b97818 CM |
137 | /* |
138 | * this does all the hard work for inserting an inline extent into | |
139 | * the btree. The caller should have done a btrfs_drop_extents so that | |
140 | * no overlapping inline items exist in the btree | |
141 | */ | |
40f76580 | 142 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 143 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
144 | struct btrfs_root *root, struct inode *inode, |
145 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 146 | int compress_type, |
c8b97818 CM |
147 | struct page **compressed_pages) |
148 | { | |
c8b97818 CM |
149 | struct extent_buffer *leaf; |
150 | struct page *page = NULL; | |
151 | char *kaddr; | |
152 | unsigned long ptr; | |
153 | struct btrfs_file_extent_item *ei; | |
154 | int err = 0; | |
155 | int ret; | |
156 | size_t cur_size = size; | |
c8b97818 | 157 | unsigned long offset; |
c8b97818 | 158 | |
fe3f566c | 159 | if (compressed_size && compressed_pages) |
c8b97818 | 160 | cur_size = compressed_size; |
c8b97818 | 161 | |
1acae57b | 162 | inode_add_bytes(inode, size); |
c8b97818 | 163 | |
1acae57b FDBM |
164 | if (!extent_inserted) { |
165 | struct btrfs_key key; | |
166 | size_t datasize; | |
c8b97818 | 167 | |
1acae57b FDBM |
168 | key.objectid = btrfs_ino(inode); |
169 | key.offset = start; | |
962a298f | 170 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 171 | |
1acae57b FDBM |
172 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
173 | path->leave_spinning = 1; | |
174 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
175 | datasize); | |
176 | if (ret) { | |
177 | err = ret; | |
178 | goto fail; | |
179 | } | |
c8b97818 CM |
180 | } |
181 | leaf = path->nodes[0]; | |
182 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
183 | struct btrfs_file_extent_item); | |
184 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
185 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
186 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
187 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
188 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
189 | ptr = btrfs_file_extent_inline_start(ei); | |
190 | ||
261507a0 | 191 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
192 | struct page *cpage; |
193 | int i = 0; | |
d397712b | 194 | while (compressed_size > 0) { |
c8b97818 | 195 | cpage = compressed_pages[i]; |
5b050f04 | 196 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 197 | PAGE_SIZE); |
c8b97818 | 198 | |
7ac687d9 | 199 | kaddr = kmap_atomic(cpage); |
c8b97818 | 200 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 201 | kunmap_atomic(kaddr); |
c8b97818 CM |
202 | |
203 | i++; | |
204 | ptr += cur_size; | |
205 | compressed_size -= cur_size; | |
206 | } | |
207 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 208 | compress_type); |
c8b97818 CM |
209 | } else { |
210 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 211 | start >> PAGE_SHIFT); |
c8b97818 | 212 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 213 | kaddr = kmap_atomic(page); |
09cbfeaf | 214 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 215 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 216 | kunmap_atomic(kaddr); |
09cbfeaf | 217 | put_page(page); |
c8b97818 CM |
218 | } |
219 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 220 | btrfs_release_path(path); |
c8b97818 | 221 | |
c2167754 YZ |
222 | /* |
223 | * we're an inline extent, so nobody can | |
224 | * extend the file past i_size without locking | |
225 | * a page we already have locked. | |
226 | * | |
227 | * We must do any isize and inode updates | |
228 | * before we unlock the pages. Otherwise we | |
229 | * could end up racing with unlink. | |
230 | */ | |
c8b97818 | 231 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 232 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 233 | |
79787eaa | 234 | return ret; |
c8b97818 | 235 | fail: |
c8b97818 CM |
236 | return err; |
237 | } | |
238 | ||
239 | ||
240 | /* | |
241 | * conditionally insert an inline extent into the file. This | |
242 | * does the checks required to make sure the data is small enough | |
243 | * to fit as an inline extent. | |
244 | */ | |
00361589 JB |
245 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
246 | struct inode *inode, u64 start, | |
247 | u64 end, size_t compressed_size, | |
248 | int compress_type, | |
249 | struct page **compressed_pages) | |
c8b97818 | 250 | { |
00361589 | 251 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
252 | u64 isize = i_size_read(inode); |
253 | u64 actual_end = min(end + 1, isize); | |
254 | u64 inline_len = actual_end - start; | |
fda2832f | 255 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
256 | u64 data_len = inline_len; |
257 | int ret; | |
1acae57b FDBM |
258 | struct btrfs_path *path; |
259 | int extent_inserted = 0; | |
260 | u32 extent_item_size; | |
c8b97818 CM |
261 | |
262 | if (compressed_size) | |
263 | data_len = compressed_size; | |
264 | ||
265 | if (start > 0 || | |
0c29ba99 | 266 | actual_end > root->sectorsize || |
354877be | 267 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) || |
c8b97818 CM |
268 | (!compressed_size && |
269 | (actual_end & (root->sectorsize - 1)) == 0) || | |
270 | end + 1 < isize || | |
271 | data_len > root->fs_info->max_inline) { | |
272 | return 1; | |
273 | } | |
274 | ||
1acae57b FDBM |
275 | path = btrfs_alloc_path(); |
276 | if (!path) | |
277 | return -ENOMEM; | |
278 | ||
00361589 | 279 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
280 | if (IS_ERR(trans)) { |
281 | btrfs_free_path(path); | |
00361589 | 282 | return PTR_ERR(trans); |
1acae57b | 283 | } |
00361589 JB |
284 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
285 | ||
1acae57b FDBM |
286 | if (compressed_size && compressed_pages) |
287 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
288 | compressed_size); | |
289 | else | |
290 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
291 | inline_len); | |
292 | ||
293 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
294 | start, aligned_end, NULL, | |
295 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
296 | if (ret) { |
297 | btrfs_abort_transaction(trans, root, ret); | |
298 | goto out; | |
299 | } | |
c8b97818 CM |
300 | |
301 | if (isize > actual_end) | |
302 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
303 | ret = insert_inline_extent(trans, path, extent_inserted, |
304 | root, inode, start, | |
c8b97818 | 305 | inline_len, compressed_size, |
fe3f566c | 306 | compress_type, compressed_pages); |
2adcac1a | 307 | if (ret && ret != -ENOSPC) { |
79787eaa | 308 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 309 | goto out; |
2adcac1a | 310 | } else if (ret == -ENOSPC) { |
00361589 JB |
311 | ret = 1; |
312 | goto out; | |
79787eaa | 313 | } |
2adcac1a | 314 | |
bdc20e67 | 315 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 316 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 317 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 318 | out: |
94ed938a QW |
319 | /* |
320 | * Don't forget to free the reserved space, as for inlined extent | |
321 | * it won't count as data extent, free them directly here. | |
322 | * And at reserve time, it's always aligned to page size, so | |
323 | * just free one page here. | |
324 | */ | |
09cbfeaf | 325 | btrfs_qgroup_free_data(inode, 0, PAGE_SIZE); |
1acae57b | 326 | btrfs_free_path(path); |
00361589 JB |
327 | btrfs_end_transaction(trans, root); |
328 | return ret; | |
c8b97818 CM |
329 | } |
330 | ||
771ed689 CM |
331 | struct async_extent { |
332 | u64 start; | |
333 | u64 ram_size; | |
334 | u64 compressed_size; | |
335 | struct page **pages; | |
336 | unsigned long nr_pages; | |
261507a0 | 337 | int compress_type; |
771ed689 CM |
338 | struct list_head list; |
339 | }; | |
340 | ||
341 | struct async_cow { | |
342 | struct inode *inode; | |
343 | struct btrfs_root *root; | |
344 | struct page *locked_page; | |
345 | u64 start; | |
346 | u64 end; | |
347 | struct list_head extents; | |
348 | struct btrfs_work work; | |
349 | }; | |
350 | ||
351 | static noinline int add_async_extent(struct async_cow *cow, | |
352 | u64 start, u64 ram_size, | |
353 | u64 compressed_size, | |
354 | struct page **pages, | |
261507a0 LZ |
355 | unsigned long nr_pages, |
356 | int compress_type) | |
771ed689 CM |
357 | { |
358 | struct async_extent *async_extent; | |
359 | ||
360 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 361 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
362 | async_extent->start = start; |
363 | async_extent->ram_size = ram_size; | |
364 | async_extent->compressed_size = compressed_size; | |
365 | async_extent->pages = pages; | |
366 | async_extent->nr_pages = nr_pages; | |
261507a0 | 367 | async_extent->compress_type = compress_type; |
771ed689 CM |
368 | list_add_tail(&async_extent->list, &cow->extents); |
369 | return 0; | |
370 | } | |
371 | ||
f79707b0 WS |
372 | static inline int inode_need_compress(struct inode *inode) |
373 | { | |
374 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
375 | ||
376 | /* force compress */ | |
377 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
378 | return 1; | |
379 | /* bad compression ratios */ | |
380 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
381 | return 0; | |
382 | if (btrfs_test_opt(root, COMPRESS) || | |
383 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
384 | BTRFS_I(inode)->force_compress) | |
385 | return 1; | |
386 | return 0; | |
387 | } | |
388 | ||
d352ac68 | 389 | /* |
771ed689 CM |
390 | * we create compressed extents in two phases. The first |
391 | * phase compresses a range of pages that have already been | |
392 | * locked (both pages and state bits are locked). | |
c8b97818 | 393 | * |
771ed689 CM |
394 | * This is done inside an ordered work queue, and the compression |
395 | * is spread across many cpus. The actual IO submission is step | |
396 | * two, and the ordered work queue takes care of making sure that | |
397 | * happens in the same order things were put onto the queue by | |
398 | * writepages and friends. | |
c8b97818 | 399 | * |
771ed689 CM |
400 | * If this code finds it can't get good compression, it puts an |
401 | * entry onto the work queue to write the uncompressed bytes. This | |
402 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
403 | * are written in the same order that the flusher thread sent them |
404 | * down. | |
d352ac68 | 405 | */ |
c44f649e | 406 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
407 | struct page *locked_page, |
408 | u64 start, u64 end, | |
409 | struct async_cow *async_cow, | |
410 | int *num_added) | |
b888db2b CM |
411 | { |
412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 413 | u64 num_bytes; |
db94535d | 414 | u64 blocksize = root->sectorsize; |
c8b97818 | 415 | u64 actual_end; |
42dc7bab | 416 | u64 isize = i_size_read(inode); |
e6dcd2dc | 417 | int ret = 0; |
c8b97818 CM |
418 | struct page **pages = NULL; |
419 | unsigned long nr_pages; | |
420 | unsigned long nr_pages_ret = 0; | |
421 | unsigned long total_compressed = 0; | |
422 | unsigned long total_in = 0; | |
ee22184b BL |
423 | unsigned long max_compressed = SZ_128K; |
424 | unsigned long max_uncompressed = SZ_128K; | |
c8b97818 CM |
425 | int i; |
426 | int will_compress; | |
261507a0 | 427 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 428 | int redirty = 0; |
b888db2b | 429 | |
4cb13e5d | 430 | /* if this is a small write inside eof, kick off a defrag */ |
ee22184b | 431 | if ((end - start + 1) < SZ_16K && |
4cb13e5d | 432 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) |
4cb5300b CM |
433 | btrfs_add_inode_defrag(NULL, inode); |
434 | ||
42dc7bab | 435 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
436 | again: |
437 | will_compress = 0; | |
09cbfeaf KS |
438 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
439 | nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE); | |
be20aa9d | 440 | |
f03d9301 CM |
441 | /* |
442 | * we don't want to send crud past the end of i_size through | |
443 | * compression, that's just a waste of CPU time. So, if the | |
444 | * end of the file is before the start of our current | |
445 | * requested range of bytes, we bail out to the uncompressed | |
446 | * cleanup code that can deal with all of this. | |
447 | * | |
448 | * It isn't really the fastest way to fix things, but this is a | |
449 | * very uncommon corner. | |
450 | */ | |
451 | if (actual_end <= start) | |
452 | goto cleanup_and_bail_uncompressed; | |
453 | ||
c8b97818 CM |
454 | total_compressed = actual_end - start; |
455 | ||
4bcbb332 SW |
456 | /* |
457 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 458 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
459 | */ |
460 | if (total_compressed <= blocksize && | |
461 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
462 | goto cleanup_and_bail_uncompressed; | |
463 | ||
c8b97818 CM |
464 | /* we want to make sure that amount of ram required to uncompress |
465 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
466 | * of a compressed extent to 128k. This is a crucial number |
467 | * because it also controls how easily we can spread reads across | |
468 | * cpus for decompression. | |
469 | * | |
470 | * We also want to make sure the amount of IO required to do | |
471 | * a random read is reasonably small, so we limit the size of | |
472 | * a compressed extent to 128k. | |
c8b97818 CM |
473 | */ |
474 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 475 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 476 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
477 | total_in = 0; |
478 | ret = 0; | |
db94535d | 479 | |
771ed689 CM |
480 | /* |
481 | * we do compression for mount -o compress and when the | |
482 | * inode has not been flagged as nocompress. This flag can | |
483 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 484 | */ |
f79707b0 | 485 | if (inode_need_compress(inode)) { |
c8b97818 | 486 | WARN_ON(pages); |
31e818fe | 487 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
488 | if (!pages) { |
489 | /* just bail out to the uncompressed code */ | |
490 | goto cont; | |
491 | } | |
c8b97818 | 492 | |
261507a0 LZ |
493 | if (BTRFS_I(inode)->force_compress) |
494 | compress_type = BTRFS_I(inode)->force_compress; | |
495 | ||
4adaa611 CM |
496 | /* |
497 | * we need to call clear_page_dirty_for_io on each | |
498 | * page in the range. Otherwise applications with the file | |
499 | * mmap'd can wander in and change the page contents while | |
500 | * we are compressing them. | |
501 | * | |
502 | * If the compression fails for any reason, we set the pages | |
503 | * dirty again later on. | |
504 | */ | |
505 | extent_range_clear_dirty_for_io(inode, start, end); | |
506 | redirty = 1; | |
261507a0 LZ |
507 | ret = btrfs_compress_pages(compress_type, |
508 | inode->i_mapping, start, | |
509 | total_compressed, pages, | |
510 | nr_pages, &nr_pages_ret, | |
511 | &total_in, | |
512 | &total_compressed, | |
513 | max_compressed); | |
c8b97818 CM |
514 | |
515 | if (!ret) { | |
516 | unsigned long offset = total_compressed & | |
09cbfeaf | 517 | (PAGE_SIZE - 1); |
c8b97818 CM |
518 | struct page *page = pages[nr_pages_ret - 1]; |
519 | char *kaddr; | |
520 | ||
521 | /* zero the tail end of the last page, we might be | |
522 | * sending it down to disk | |
523 | */ | |
524 | if (offset) { | |
7ac687d9 | 525 | kaddr = kmap_atomic(page); |
c8b97818 | 526 | memset(kaddr + offset, 0, |
09cbfeaf | 527 | PAGE_SIZE - offset); |
7ac687d9 | 528 | kunmap_atomic(kaddr); |
c8b97818 CM |
529 | } |
530 | will_compress = 1; | |
531 | } | |
532 | } | |
560f7d75 | 533 | cont: |
c8b97818 CM |
534 | if (start == 0) { |
535 | /* lets try to make an inline extent */ | |
771ed689 | 536 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 537 | /* we didn't compress the entire range, try |
771ed689 | 538 | * to make an uncompressed inline extent. |
c8b97818 | 539 | */ |
00361589 JB |
540 | ret = cow_file_range_inline(root, inode, start, end, |
541 | 0, 0, NULL); | |
c8b97818 | 542 | } else { |
771ed689 | 543 | /* try making a compressed inline extent */ |
00361589 | 544 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
545 | total_compressed, |
546 | compress_type, pages); | |
c8b97818 | 547 | } |
79787eaa | 548 | if (ret <= 0) { |
151a41bc JB |
549 | unsigned long clear_flags = EXTENT_DELALLOC | |
550 | EXTENT_DEFRAG; | |
e6eb4314 FM |
551 | unsigned long page_error_op; |
552 | ||
151a41bc | 553 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 554 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 555 | |
771ed689 | 556 | /* |
79787eaa JM |
557 | * inline extent creation worked or returned error, |
558 | * we don't need to create any more async work items. | |
559 | * Unlock and free up our temp pages. | |
771ed689 | 560 | */ |
c2790a2e | 561 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 562 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
563 | PAGE_CLEAR_DIRTY | |
564 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 565 | page_error_op | |
c2790a2e | 566 | PAGE_END_WRITEBACK); |
c8b97818 CM |
567 | goto free_pages_out; |
568 | } | |
569 | } | |
570 | ||
571 | if (will_compress) { | |
572 | /* | |
573 | * we aren't doing an inline extent round the compressed size | |
574 | * up to a block size boundary so the allocator does sane | |
575 | * things | |
576 | */ | |
fda2832f | 577 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
578 | |
579 | /* | |
580 | * one last check to make sure the compression is really a | |
581 | * win, compare the page count read with the blocks on disk | |
582 | */ | |
09cbfeaf | 583 | total_in = ALIGN(total_in, PAGE_SIZE); |
c8b97818 CM |
584 | if (total_compressed >= total_in) { |
585 | will_compress = 0; | |
586 | } else { | |
c8b97818 CM |
587 | num_bytes = total_in; |
588 | } | |
589 | } | |
590 | if (!will_compress && pages) { | |
591 | /* | |
592 | * the compression code ran but failed to make things smaller, | |
593 | * free any pages it allocated and our page pointer array | |
594 | */ | |
595 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 596 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 597 | put_page(pages[i]); |
c8b97818 CM |
598 | } |
599 | kfree(pages); | |
600 | pages = NULL; | |
601 | total_compressed = 0; | |
602 | nr_pages_ret = 0; | |
603 | ||
604 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
605 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
606 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 607 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 608 | } |
c8b97818 | 609 | } |
771ed689 CM |
610 | if (will_compress) { |
611 | *num_added += 1; | |
c8b97818 | 612 | |
771ed689 CM |
613 | /* the async work queues will take care of doing actual |
614 | * allocation on disk for these compressed pages, | |
615 | * and will submit them to the elevator. | |
616 | */ | |
617 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
618 | total_compressed, pages, nr_pages_ret, |
619 | compress_type); | |
179e29e4 | 620 | |
24ae6365 | 621 | if (start + num_bytes < end) { |
771ed689 CM |
622 | start += num_bytes; |
623 | pages = NULL; | |
624 | cond_resched(); | |
625 | goto again; | |
626 | } | |
627 | } else { | |
f03d9301 | 628 | cleanup_and_bail_uncompressed: |
771ed689 CM |
629 | /* |
630 | * No compression, but we still need to write the pages in | |
631 | * the file we've been given so far. redirty the locked | |
632 | * page if it corresponds to our extent and set things up | |
633 | * for the async work queue to run cow_file_range to do | |
634 | * the normal delalloc dance | |
635 | */ | |
636 | if (page_offset(locked_page) >= start && | |
637 | page_offset(locked_page) <= end) { | |
638 | __set_page_dirty_nobuffers(locked_page); | |
639 | /* unlocked later on in the async handlers */ | |
640 | } | |
4adaa611 CM |
641 | if (redirty) |
642 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
643 | add_async_extent(async_cow, start, end - start + 1, |
644 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
645 | *num_added += 1; |
646 | } | |
3b951516 | 647 | |
c44f649e | 648 | return; |
771ed689 CM |
649 | |
650 | free_pages_out: | |
651 | for (i = 0; i < nr_pages_ret; i++) { | |
652 | WARN_ON(pages[i]->mapping); | |
09cbfeaf | 653 | put_page(pages[i]); |
771ed689 | 654 | } |
d397712b | 655 | kfree(pages); |
771ed689 | 656 | } |
771ed689 | 657 | |
40ae837b FM |
658 | static void free_async_extent_pages(struct async_extent *async_extent) |
659 | { | |
660 | int i; | |
661 | ||
662 | if (!async_extent->pages) | |
663 | return; | |
664 | ||
665 | for (i = 0; i < async_extent->nr_pages; i++) { | |
666 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 667 | put_page(async_extent->pages[i]); |
40ae837b FM |
668 | } |
669 | kfree(async_extent->pages); | |
670 | async_extent->nr_pages = 0; | |
671 | async_extent->pages = NULL; | |
771ed689 CM |
672 | } |
673 | ||
674 | /* | |
675 | * phase two of compressed writeback. This is the ordered portion | |
676 | * of the code, which only gets called in the order the work was | |
677 | * queued. We walk all the async extents created by compress_file_range | |
678 | * and send them down to the disk. | |
679 | */ | |
dec8f175 | 680 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
681 | struct async_cow *async_cow) |
682 | { | |
683 | struct async_extent *async_extent; | |
684 | u64 alloc_hint = 0; | |
771ed689 CM |
685 | struct btrfs_key ins; |
686 | struct extent_map *em; | |
687 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
688 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
689 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 690 | int ret = 0; |
771ed689 | 691 | |
3e04e7f1 | 692 | again: |
d397712b | 693 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
694 | async_extent = list_entry(async_cow->extents.next, |
695 | struct async_extent, list); | |
696 | list_del(&async_extent->list); | |
c8b97818 | 697 | |
771ed689 CM |
698 | io_tree = &BTRFS_I(inode)->io_tree; |
699 | ||
f5a84ee3 | 700 | retry: |
771ed689 CM |
701 | /* did the compression code fall back to uncompressed IO? */ |
702 | if (!async_extent->pages) { | |
703 | int page_started = 0; | |
704 | unsigned long nr_written = 0; | |
705 | ||
706 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 707 | async_extent->start + |
d0082371 | 708 | async_extent->ram_size - 1); |
771ed689 CM |
709 | |
710 | /* allocate blocks */ | |
f5a84ee3 JB |
711 | ret = cow_file_range(inode, async_cow->locked_page, |
712 | async_extent->start, | |
713 | async_extent->start + | |
714 | async_extent->ram_size - 1, | |
715 | &page_started, &nr_written, 0); | |
771ed689 | 716 | |
79787eaa JM |
717 | /* JDM XXX */ |
718 | ||
771ed689 CM |
719 | /* |
720 | * if page_started, cow_file_range inserted an | |
721 | * inline extent and took care of all the unlocking | |
722 | * and IO for us. Otherwise, we need to submit | |
723 | * all those pages down to the drive. | |
724 | */ | |
f5a84ee3 | 725 | if (!page_started && !ret) |
771ed689 CM |
726 | extent_write_locked_range(io_tree, |
727 | inode, async_extent->start, | |
d397712b | 728 | async_extent->start + |
771ed689 CM |
729 | async_extent->ram_size - 1, |
730 | btrfs_get_extent, | |
731 | WB_SYNC_ALL); | |
3e04e7f1 JB |
732 | else if (ret) |
733 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
734 | kfree(async_extent); |
735 | cond_resched(); | |
736 | continue; | |
737 | } | |
738 | ||
739 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 740 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 741 | |
00361589 | 742 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
743 | async_extent->compressed_size, |
744 | async_extent->compressed_size, | |
e570fd27 | 745 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 746 | if (ret) { |
40ae837b | 747 | free_async_extent_pages(async_extent); |
3e04e7f1 | 748 | |
fdf8e2ea JB |
749 | if (ret == -ENOSPC) { |
750 | unlock_extent(io_tree, async_extent->start, | |
751 | async_extent->start + | |
752 | async_extent->ram_size - 1); | |
ce62003f LB |
753 | |
754 | /* | |
755 | * we need to redirty the pages if we decide to | |
756 | * fallback to uncompressed IO, otherwise we | |
757 | * will not submit these pages down to lower | |
758 | * layers. | |
759 | */ | |
760 | extent_range_redirty_for_io(inode, | |
761 | async_extent->start, | |
762 | async_extent->start + | |
763 | async_extent->ram_size - 1); | |
764 | ||
79787eaa | 765 | goto retry; |
fdf8e2ea | 766 | } |
3e04e7f1 | 767 | goto out_free; |
f5a84ee3 | 768 | } |
c2167754 YZ |
769 | /* |
770 | * here we're doing allocation and writeback of the | |
771 | * compressed pages | |
772 | */ | |
773 | btrfs_drop_extent_cache(inode, async_extent->start, | |
774 | async_extent->start + | |
775 | async_extent->ram_size - 1, 0); | |
776 | ||
172ddd60 | 777 | em = alloc_extent_map(); |
b9aa55be LB |
778 | if (!em) { |
779 | ret = -ENOMEM; | |
3e04e7f1 | 780 | goto out_free_reserve; |
b9aa55be | 781 | } |
771ed689 CM |
782 | em->start = async_extent->start; |
783 | em->len = async_extent->ram_size; | |
445a6944 | 784 | em->orig_start = em->start; |
2ab28f32 JB |
785 | em->mod_start = em->start; |
786 | em->mod_len = em->len; | |
c8b97818 | 787 | |
771ed689 CM |
788 | em->block_start = ins.objectid; |
789 | em->block_len = ins.offset; | |
b4939680 | 790 | em->orig_block_len = ins.offset; |
cc95bef6 | 791 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 792 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 793 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
794 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
795 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 796 | em->generation = -1; |
771ed689 | 797 | |
d397712b | 798 | while (1) { |
890871be | 799 | write_lock(&em_tree->lock); |
09a2a8f9 | 800 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 801 | write_unlock(&em_tree->lock); |
771ed689 CM |
802 | if (ret != -EEXIST) { |
803 | free_extent_map(em); | |
804 | break; | |
805 | } | |
806 | btrfs_drop_extent_cache(inode, async_extent->start, | |
807 | async_extent->start + | |
808 | async_extent->ram_size - 1, 0); | |
809 | } | |
810 | ||
3e04e7f1 JB |
811 | if (ret) |
812 | goto out_free_reserve; | |
813 | ||
261507a0 LZ |
814 | ret = btrfs_add_ordered_extent_compress(inode, |
815 | async_extent->start, | |
816 | ins.objectid, | |
817 | async_extent->ram_size, | |
818 | ins.offset, | |
819 | BTRFS_ORDERED_COMPRESSED, | |
820 | async_extent->compress_type); | |
d9f85963 FM |
821 | if (ret) { |
822 | btrfs_drop_extent_cache(inode, async_extent->start, | |
823 | async_extent->start + | |
824 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 825 | goto out_free_reserve; |
d9f85963 | 826 | } |
9cfa3e34 | 827 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
771ed689 | 828 | |
771ed689 CM |
829 | /* |
830 | * clear dirty, set writeback and unlock the pages. | |
831 | */ | |
c2790a2e | 832 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
833 | async_extent->start + |
834 | async_extent->ram_size - 1, | |
151a41bc JB |
835 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
836 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 837 | PAGE_SET_WRITEBACK); |
771ed689 | 838 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
839 | async_extent->start, |
840 | async_extent->ram_size, | |
841 | ins.objectid, | |
842 | ins.offset, async_extent->pages, | |
843 | async_extent->nr_pages); | |
fce2a4e6 FM |
844 | if (ret) { |
845 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
846 | struct page *p = async_extent->pages[0]; | |
847 | const u64 start = async_extent->start; | |
848 | const u64 end = start + async_extent->ram_size - 1; | |
849 | ||
850 | p->mapping = inode->i_mapping; | |
851 | tree->ops->writepage_end_io_hook(p, start, end, | |
852 | NULL, 0); | |
853 | p->mapping = NULL; | |
854 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
855 | PAGE_END_WRITEBACK | | |
856 | PAGE_SET_ERROR); | |
40ae837b | 857 | free_async_extent_pages(async_extent); |
fce2a4e6 | 858 | } |
771ed689 CM |
859 | alloc_hint = ins.objectid + ins.offset; |
860 | kfree(async_extent); | |
861 | cond_resched(); | |
862 | } | |
dec8f175 | 863 | return; |
3e04e7f1 | 864 | out_free_reserve: |
9cfa3e34 | 865 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
e570fd27 | 866 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 867 | out_free: |
c2790a2e | 868 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
869 | async_extent->start + |
870 | async_extent->ram_size - 1, | |
c2790a2e | 871 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
872 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
873 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
874 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
875 | PAGE_SET_ERROR); | |
40ae837b | 876 | free_async_extent_pages(async_extent); |
79787eaa | 877 | kfree(async_extent); |
3e04e7f1 | 878 | goto again; |
771ed689 CM |
879 | } |
880 | ||
4b46fce2 JB |
881 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
882 | u64 num_bytes) | |
883 | { | |
884 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
885 | struct extent_map *em; | |
886 | u64 alloc_hint = 0; | |
887 | ||
888 | read_lock(&em_tree->lock); | |
889 | em = search_extent_mapping(em_tree, start, num_bytes); | |
890 | if (em) { | |
891 | /* | |
892 | * if block start isn't an actual block number then find the | |
893 | * first block in this inode and use that as a hint. If that | |
894 | * block is also bogus then just don't worry about it. | |
895 | */ | |
896 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
897 | free_extent_map(em); | |
898 | em = search_extent_mapping(em_tree, 0, 0); | |
899 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
900 | alloc_hint = em->block_start; | |
901 | if (em) | |
902 | free_extent_map(em); | |
903 | } else { | |
904 | alloc_hint = em->block_start; | |
905 | free_extent_map(em); | |
906 | } | |
907 | } | |
908 | read_unlock(&em_tree->lock); | |
909 | ||
910 | return alloc_hint; | |
911 | } | |
912 | ||
771ed689 CM |
913 | /* |
914 | * when extent_io.c finds a delayed allocation range in the file, | |
915 | * the call backs end up in this code. The basic idea is to | |
916 | * allocate extents on disk for the range, and create ordered data structs | |
917 | * in ram to track those extents. | |
918 | * | |
919 | * locked_page is the page that writepage had locked already. We use | |
920 | * it to make sure we don't do extra locks or unlocks. | |
921 | * | |
922 | * *page_started is set to one if we unlock locked_page and do everything | |
923 | * required to start IO on it. It may be clean and already done with | |
924 | * IO when we return. | |
925 | */ | |
00361589 JB |
926 | static noinline int cow_file_range(struct inode *inode, |
927 | struct page *locked_page, | |
928 | u64 start, u64 end, int *page_started, | |
929 | unsigned long *nr_written, | |
930 | int unlock) | |
771ed689 | 931 | { |
00361589 | 932 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
933 | u64 alloc_hint = 0; |
934 | u64 num_bytes; | |
935 | unsigned long ram_size; | |
936 | u64 disk_num_bytes; | |
937 | u64 cur_alloc_size; | |
938 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
939 | struct btrfs_key ins; |
940 | struct extent_map *em; | |
941 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
942 | int ret = 0; | |
943 | ||
02ecd2c2 JB |
944 | if (btrfs_is_free_space_inode(inode)) { |
945 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
946 | ret = -EINVAL; |
947 | goto out_unlock; | |
02ecd2c2 | 948 | } |
771ed689 | 949 | |
fda2832f | 950 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
951 | num_bytes = max(blocksize, num_bytes); |
952 | disk_num_bytes = num_bytes; | |
771ed689 | 953 | |
4cb5300b | 954 | /* if this is a small write inside eof, kick off defrag */ |
ee22184b | 955 | if (num_bytes < SZ_64K && |
4cb13e5d | 956 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) |
00361589 | 957 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 958 | |
771ed689 CM |
959 | if (start == 0) { |
960 | /* lets try to make an inline extent */ | |
00361589 JB |
961 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
962 | NULL); | |
771ed689 | 963 | if (ret == 0) { |
c2790a2e JB |
964 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
965 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 966 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
967 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
968 | PAGE_END_WRITEBACK); | |
c2167754 | 969 | |
771ed689 | 970 | *nr_written = *nr_written + |
09cbfeaf | 971 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 972 | *page_started = 1; |
771ed689 | 973 | goto out; |
79787eaa | 974 | } else if (ret < 0) { |
79787eaa | 975 | goto out_unlock; |
771ed689 CM |
976 | } |
977 | } | |
978 | ||
979 | BUG_ON(disk_num_bytes > | |
6c41761f | 980 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 981 | |
4b46fce2 | 982 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
983 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
984 | ||
d397712b | 985 | while (disk_num_bytes > 0) { |
a791e35e CM |
986 | unsigned long op; |
987 | ||
287a0ab9 | 988 | cur_alloc_size = disk_num_bytes; |
00361589 | 989 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 990 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 991 | &ins, 1, 1); |
00361589 | 992 | if (ret < 0) |
79787eaa | 993 | goto out_unlock; |
d397712b | 994 | |
172ddd60 | 995 | em = alloc_extent_map(); |
b9aa55be LB |
996 | if (!em) { |
997 | ret = -ENOMEM; | |
ace68bac | 998 | goto out_reserve; |
b9aa55be | 999 | } |
e6dcd2dc | 1000 | em->start = start; |
445a6944 | 1001 | em->orig_start = em->start; |
771ed689 CM |
1002 | ram_size = ins.offset; |
1003 | em->len = ins.offset; | |
2ab28f32 JB |
1004 | em->mod_start = em->start; |
1005 | em->mod_len = em->len; | |
c8b97818 | 1006 | |
e6dcd2dc | 1007 | em->block_start = ins.objectid; |
c8b97818 | 1008 | em->block_len = ins.offset; |
b4939680 | 1009 | em->orig_block_len = ins.offset; |
cc95bef6 | 1010 | em->ram_bytes = ram_size; |
e6dcd2dc | 1011 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 1012 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 1013 | em->generation = -1; |
c8b97818 | 1014 | |
d397712b | 1015 | while (1) { |
890871be | 1016 | write_lock(&em_tree->lock); |
09a2a8f9 | 1017 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1018 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
1019 | if (ret != -EEXIST) { |
1020 | free_extent_map(em); | |
1021 | break; | |
1022 | } | |
1023 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1024 | start + ram_size - 1, 0); |
e6dcd2dc | 1025 | } |
ace68bac LB |
1026 | if (ret) |
1027 | goto out_reserve; | |
e6dcd2dc | 1028 | |
98d20f67 | 1029 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1030 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1031 | ram_size, cur_alloc_size, 0); |
ace68bac | 1032 | if (ret) |
d9f85963 | 1033 | goto out_drop_extent_cache; |
c8b97818 | 1034 | |
17d217fe YZ |
1035 | if (root->root_key.objectid == |
1036 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1037 | ret = btrfs_reloc_clone_csums(inode, start, | |
1038 | cur_alloc_size); | |
00361589 | 1039 | if (ret) |
d9f85963 | 1040 | goto out_drop_extent_cache; |
17d217fe YZ |
1041 | } |
1042 | ||
9cfa3e34 FM |
1043 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
1044 | ||
d397712b | 1045 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1046 | break; |
d397712b | 1047 | |
c8b97818 CM |
1048 | /* we're not doing compressed IO, don't unlock the first |
1049 | * page (which the caller expects to stay locked), don't | |
1050 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1051 | * |
1052 | * Do set the Private2 bit so we know this page was properly | |
1053 | * setup for writepage | |
c8b97818 | 1054 | */ |
c2790a2e JB |
1055 | op = unlock ? PAGE_UNLOCK : 0; |
1056 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1057 | |
c2790a2e JB |
1058 | extent_clear_unlock_delalloc(inode, start, |
1059 | start + ram_size - 1, locked_page, | |
1060 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1061 | op); | |
c8b97818 | 1062 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1063 | num_bytes -= cur_alloc_size; |
1064 | alloc_hint = ins.objectid + ins.offset; | |
1065 | start += cur_alloc_size; | |
b888db2b | 1066 | } |
79787eaa | 1067 | out: |
be20aa9d | 1068 | return ret; |
b7d5b0a8 | 1069 | |
d9f85963 FM |
1070 | out_drop_extent_cache: |
1071 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1072 | out_reserve: |
9cfa3e34 | 1073 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
e570fd27 | 1074 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1075 | out_unlock: |
c2790a2e | 1076 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1077 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1078 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1079 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1080 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1081 | goto out; |
771ed689 | 1082 | } |
c8b97818 | 1083 | |
771ed689 CM |
1084 | /* |
1085 | * work queue call back to started compression on a file and pages | |
1086 | */ | |
1087 | static noinline void async_cow_start(struct btrfs_work *work) | |
1088 | { | |
1089 | struct async_cow *async_cow; | |
1090 | int num_added = 0; | |
1091 | async_cow = container_of(work, struct async_cow, work); | |
1092 | ||
1093 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1094 | async_cow->start, async_cow->end, async_cow, | |
1095 | &num_added); | |
8180ef88 | 1096 | if (num_added == 0) { |
cb77fcd8 | 1097 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1098 | async_cow->inode = NULL; |
8180ef88 | 1099 | } |
771ed689 CM |
1100 | } |
1101 | ||
1102 | /* | |
1103 | * work queue call back to submit previously compressed pages | |
1104 | */ | |
1105 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1106 | { | |
1107 | struct async_cow *async_cow; | |
1108 | struct btrfs_root *root; | |
1109 | unsigned long nr_pages; | |
1110 | ||
1111 | async_cow = container_of(work, struct async_cow, work); | |
1112 | ||
1113 | root = async_cow->root; | |
09cbfeaf KS |
1114 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1115 | PAGE_SHIFT; | |
771ed689 | 1116 | |
ee863954 DS |
1117 | /* |
1118 | * atomic_sub_return implies a barrier for waitqueue_active | |
1119 | */ | |
66657b31 | 1120 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
ee22184b | 1121 | 5 * SZ_1M && |
771ed689 CM |
1122 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1123 | wake_up(&root->fs_info->async_submit_wait); | |
1124 | ||
d397712b | 1125 | if (async_cow->inode) |
771ed689 | 1126 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1127 | } |
c8b97818 | 1128 | |
771ed689 CM |
1129 | static noinline void async_cow_free(struct btrfs_work *work) |
1130 | { | |
1131 | struct async_cow *async_cow; | |
1132 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1133 | if (async_cow->inode) |
cb77fcd8 | 1134 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1135 | kfree(async_cow); |
1136 | } | |
1137 | ||
1138 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1139 | u64 start, u64 end, int *page_started, | |
1140 | unsigned long *nr_written) | |
1141 | { | |
1142 | struct async_cow *async_cow; | |
1143 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1144 | unsigned long nr_pages; | |
1145 | u64 cur_end; | |
ee22184b | 1146 | int limit = 10 * SZ_1M; |
771ed689 | 1147 | |
a3429ab7 CM |
1148 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1149 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1150 | while (start < end) { |
771ed689 | 1151 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1152 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1153 | async_cow->inode = igrab(inode); |
771ed689 CM |
1154 | async_cow->root = root; |
1155 | async_cow->locked_page = locked_page; | |
1156 | async_cow->start = start; | |
1157 | ||
f79707b0 WS |
1158 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1159 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1160 | cur_end = end; |
1161 | else | |
ee22184b | 1162 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1163 | |
1164 | async_cow->end = cur_end; | |
1165 | INIT_LIST_HEAD(&async_cow->extents); | |
1166 | ||
9e0af237 LB |
1167 | btrfs_init_work(&async_cow->work, |
1168 | btrfs_delalloc_helper, | |
1169 | async_cow_start, async_cow_submit, | |
1170 | async_cow_free); | |
771ed689 | 1171 | |
09cbfeaf KS |
1172 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1173 | PAGE_SHIFT; | |
771ed689 CM |
1174 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); |
1175 | ||
afe3d242 QW |
1176 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1177 | &async_cow->work); | |
771ed689 CM |
1178 | |
1179 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1180 | wait_event(root->fs_info->async_submit_wait, | |
1181 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1182 | limit)); | |
1183 | } | |
1184 | ||
d397712b | 1185 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1186 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1187 | wait_event(root->fs_info->async_submit_wait, | |
1188 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1189 | 0)); | |
1190 | } | |
1191 | ||
1192 | *nr_written += nr_pages; | |
1193 | start = cur_end + 1; | |
1194 | } | |
1195 | *page_started = 1; | |
1196 | return 0; | |
be20aa9d CM |
1197 | } |
1198 | ||
d397712b | 1199 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1200 | u64 bytenr, u64 num_bytes) |
1201 | { | |
1202 | int ret; | |
1203 | struct btrfs_ordered_sum *sums; | |
1204 | LIST_HEAD(list); | |
1205 | ||
07d400a6 | 1206 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1207 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1208 | if (ret == 0 && list_empty(&list)) |
1209 | return 0; | |
1210 | ||
1211 | while (!list_empty(&list)) { | |
1212 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1213 | list_del(&sums->list); | |
1214 | kfree(sums); | |
1215 | } | |
1216 | return 1; | |
1217 | } | |
1218 | ||
d352ac68 CM |
1219 | /* |
1220 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1221 | * of the extents that exist in the file, and COWs the file as required. | |
1222 | * | |
1223 | * If no cow copies or snapshots exist, we write directly to the existing | |
1224 | * blocks on disk | |
1225 | */ | |
7f366cfe CM |
1226 | static noinline int run_delalloc_nocow(struct inode *inode, |
1227 | struct page *locked_page, | |
771ed689 CM |
1228 | u64 start, u64 end, int *page_started, int force, |
1229 | unsigned long *nr_written) | |
be20aa9d | 1230 | { |
be20aa9d | 1231 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1232 | struct btrfs_trans_handle *trans; |
be20aa9d | 1233 | struct extent_buffer *leaf; |
be20aa9d | 1234 | struct btrfs_path *path; |
80ff3856 | 1235 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1236 | struct btrfs_key found_key; |
80ff3856 YZ |
1237 | u64 cow_start; |
1238 | u64 cur_offset; | |
1239 | u64 extent_end; | |
5d4f98a2 | 1240 | u64 extent_offset; |
80ff3856 YZ |
1241 | u64 disk_bytenr; |
1242 | u64 num_bytes; | |
b4939680 | 1243 | u64 disk_num_bytes; |
cc95bef6 | 1244 | u64 ram_bytes; |
80ff3856 | 1245 | int extent_type; |
79787eaa | 1246 | int ret, err; |
d899e052 | 1247 | int type; |
80ff3856 YZ |
1248 | int nocow; |
1249 | int check_prev = 1; | |
82d5902d | 1250 | bool nolock; |
33345d01 | 1251 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1252 | |
1253 | path = btrfs_alloc_path(); | |
17ca04af | 1254 | if (!path) { |
c2790a2e JB |
1255 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1256 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1257 | EXTENT_DO_ACCOUNTING | |
1258 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1259 | PAGE_CLEAR_DIRTY | |
1260 | PAGE_SET_WRITEBACK | | |
1261 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1262 | return -ENOMEM; |
17ca04af | 1263 | } |
82d5902d | 1264 | |
83eea1f1 | 1265 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1266 | |
1267 | if (nolock) | |
7a7eaa40 | 1268 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1269 | else |
7a7eaa40 | 1270 | trans = btrfs_join_transaction(root); |
ff5714cc | 1271 | |
79787eaa | 1272 | if (IS_ERR(trans)) { |
c2790a2e JB |
1273 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1274 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1275 | EXTENT_DO_ACCOUNTING | |
1276 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1277 | PAGE_CLEAR_DIRTY | |
1278 | PAGE_SET_WRITEBACK | | |
1279 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1280 | btrfs_free_path(path); |
1281 | return PTR_ERR(trans); | |
1282 | } | |
1283 | ||
74b21075 | 1284 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1285 | |
80ff3856 YZ |
1286 | cow_start = (u64)-1; |
1287 | cur_offset = start; | |
1288 | while (1) { | |
33345d01 | 1289 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1290 | cur_offset, 0); |
d788a349 | 1291 | if (ret < 0) |
79787eaa | 1292 | goto error; |
80ff3856 YZ |
1293 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1294 | leaf = path->nodes[0]; | |
1295 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1296 | path->slots[0] - 1); | |
33345d01 | 1297 | if (found_key.objectid == ino && |
80ff3856 YZ |
1298 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1299 | path->slots[0]--; | |
1300 | } | |
1301 | check_prev = 0; | |
1302 | next_slot: | |
1303 | leaf = path->nodes[0]; | |
1304 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1305 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1306 | if (ret < 0) |
79787eaa | 1307 | goto error; |
80ff3856 YZ |
1308 | if (ret > 0) |
1309 | break; | |
1310 | leaf = path->nodes[0]; | |
1311 | } | |
be20aa9d | 1312 | |
80ff3856 YZ |
1313 | nocow = 0; |
1314 | disk_bytenr = 0; | |
17d217fe | 1315 | num_bytes = 0; |
80ff3856 YZ |
1316 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1317 | ||
1d512cb7 FM |
1318 | if (found_key.objectid > ino) |
1319 | break; | |
1320 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1321 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1322 | path->slots[0]++; | |
1323 | goto next_slot; | |
1324 | } | |
1325 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1326 | found_key.offset > end) |
1327 | break; | |
1328 | ||
1329 | if (found_key.offset > cur_offset) { | |
1330 | extent_end = found_key.offset; | |
e9061e21 | 1331 | extent_type = 0; |
80ff3856 YZ |
1332 | goto out_check; |
1333 | } | |
1334 | ||
1335 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1336 | struct btrfs_file_extent_item); | |
1337 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1338 | ||
cc95bef6 | 1339 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1340 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1341 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1342 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1343 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1344 | extent_end = found_key.offset + |
1345 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1346 | disk_num_bytes = |
1347 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1348 | if (extent_end <= start) { |
1349 | path->slots[0]++; | |
1350 | goto next_slot; | |
1351 | } | |
17d217fe YZ |
1352 | if (disk_bytenr == 0) |
1353 | goto out_check; | |
80ff3856 YZ |
1354 | if (btrfs_file_extent_compression(leaf, fi) || |
1355 | btrfs_file_extent_encryption(leaf, fi) || | |
1356 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1357 | goto out_check; | |
d899e052 YZ |
1358 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1359 | goto out_check; | |
d2fb3437 | 1360 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1361 | goto out_check; |
33345d01 | 1362 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1363 | found_key.offset - |
1364 | extent_offset, disk_bytenr)) | |
17d217fe | 1365 | goto out_check; |
5d4f98a2 | 1366 | disk_bytenr += extent_offset; |
17d217fe YZ |
1367 | disk_bytenr += cur_offset - found_key.offset; |
1368 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1369 | /* |
1370 | * if there are pending snapshots for this root, | |
1371 | * we fall into common COW way. | |
1372 | */ | |
1373 | if (!nolock) { | |
9ea24bbe | 1374 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1375 | if (!err) |
1376 | goto out_check; | |
1377 | } | |
17d217fe YZ |
1378 | /* |
1379 | * force cow if csum exists in the range. | |
1380 | * this ensure that csum for a given extent are | |
1381 | * either valid or do not exist. | |
1382 | */ | |
1383 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1384 | goto out_check; | |
f78c436c FM |
1385 | if (!btrfs_inc_nocow_writers(root->fs_info, |
1386 | disk_bytenr)) | |
1387 | goto out_check; | |
80ff3856 YZ |
1388 | nocow = 1; |
1389 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1390 | extent_end = found_key.offset + | |
514ac8ad CM |
1391 | btrfs_file_extent_inline_len(leaf, |
1392 | path->slots[0], fi); | |
80ff3856 YZ |
1393 | extent_end = ALIGN(extent_end, root->sectorsize); |
1394 | } else { | |
1395 | BUG_ON(1); | |
1396 | } | |
1397 | out_check: | |
1398 | if (extent_end <= start) { | |
1399 | path->slots[0]++; | |
e9894fd3 | 1400 | if (!nolock && nocow) |
9ea24bbe | 1401 | btrfs_end_write_no_snapshoting(root); |
f78c436c FM |
1402 | if (nocow) |
1403 | btrfs_dec_nocow_writers(root->fs_info, | |
1404 | disk_bytenr); | |
80ff3856 YZ |
1405 | goto next_slot; |
1406 | } | |
1407 | if (!nocow) { | |
1408 | if (cow_start == (u64)-1) | |
1409 | cow_start = cur_offset; | |
1410 | cur_offset = extent_end; | |
1411 | if (cur_offset > end) | |
1412 | break; | |
1413 | path->slots[0]++; | |
1414 | goto next_slot; | |
7ea394f1 YZ |
1415 | } |
1416 | ||
b3b4aa74 | 1417 | btrfs_release_path(path); |
80ff3856 | 1418 | if (cow_start != (u64)-1) { |
00361589 JB |
1419 | ret = cow_file_range(inode, locked_page, |
1420 | cow_start, found_key.offset - 1, | |
1421 | page_started, nr_written, 1); | |
e9894fd3 WS |
1422 | if (ret) { |
1423 | if (!nolock && nocow) | |
9ea24bbe | 1424 | btrfs_end_write_no_snapshoting(root); |
f78c436c FM |
1425 | if (nocow) |
1426 | btrfs_dec_nocow_writers(root->fs_info, | |
1427 | disk_bytenr); | |
79787eaa | 1428 | goto error; |
e9894fd3 | 1429 | } |
80ff3856 | 1430 | cow_start = (u64)-1; |
7ea394f1 | 1431 | } |
80ff3856 | 1432 | |
d899e052 YZ |
1433 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1434 | struct extent_map *em; | |
1435 | struct extent_map_tree *em_tree; | |
1436 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1437 | em = alloc_extent_map(); |
79787eaa | 1438 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1439 | em->start = cur_offset; |
70c8a91c | 1440 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1441 | em->len = num_bytes; |
1442 | em->block_len = num_bytes; | |
1443 | em->block_start = disk_bytenr; | |
b4939680 | 1444 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1445 | em->ram_bytes = ram_bytes; |
d899e052 | 1446 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1447 | em->mod_start = em->start; |
1448 | em->mod_len = em->len; | |
d899e052 | 1449 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1450 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1451 | em->generation = -1; |
d899e052 | 1452 | while (1) { |
890871be | 1453 | write_lock(&em_tree->lock); |
09a2a8f9 | 1454 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1455 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1456 | if (ret != -EEXIST) { |
1457 | free_extent_map(em); | |
1458 | break; | |
1459 | } | |
1460 | btrfs_drop_extent_cache(inode, em->start, | |
1461 | em->start + em->len - 1, 0); | |
1462 | } | |
1463 | type = BTRFS_ORDERED_PREALLOC; | |
1464 | } else { | |
1465 | type = BTRFS_ORDERED_NOCOW; | |
1466 | } | |
80ff3856 YZ |
1467 | |
1468 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1469 | num_bytes, num_bytes, type); |
f78c436c FM |
1470 | if (nocow) |
1471 | btrfs_dec_nocow_writers(root->fs_info, disk_bytenr); | |
79787eaa | 1472 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1473 | |
efa56464 YZ |
1474 | if (root->root_key.objectid == |
1475 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1476 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1477 | num_bytes); | |
e9894fd3 WS |
1478 | if (ret) { |
1479 | if (!nolock && nocow) | |
9ea24bbe | 1480 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1481 | goto error; |
e9894fd3 | 1482 | } |
efa56464 YZ |
1483 | } |
1484 | ||
c2790a2e JB |
1485 | extent_clear_unlock_delalloc(inode, cur_offset, |
1486 | cur_offset + num_bytes - 1, | |
1487 | locked_page, EXTENT_LOCKED | | |
1488 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1489 | PAGE_SET_PRIVATE2); | |
e9894fd3 | 1490 | if (!nolock && nocow) |
9ea24bbe | 1491 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1492 | cur_offset = extent_end; |
1493 | if (cur_offset > end) | |
1494 | break; | |
be20aa9d | 1495 | } |
b3b4aa74 | 1496 | btrfs_release_path(path); |
80ff3856 | 1497 | |
17ca04af | 1498 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1499 | cow_start = cur_offset; |
17ca04af JB |
1500 | cur_offset = end; |
1501 | } | |
1502 | ||
80ff3856 | 1503 | if (cow_start != (u64)-1) { |
00361589 JB |
1504 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1505 | page_started, nr_written, 1); | |
d788a349 | 1506 | if (ret) |
79787eaa | 1507 | goto error; |
80ff3856 YZ |
1508 | } |
1509 | ||
79787eaa | 1510 | error: |
a698d075 | 1511 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1512 | if (!ret) |
1513 | ret = err; | |
1514 | ||
17ca04af | 1515 | if (ret && cur_offset < end) |
c2790a2e JB |
1516 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1517 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1518 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1519 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1520 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1521 | PAGE_SET_WRITEBACK | |
1522 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1523 | btrfs_free_path(path); |
79787eaa | 1524 | return ret; |
be20aa9d CM |
1525 | } |
1526 | ||
47059d93 WS |
1527 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1528 | { | |
1529 | ||
1530 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1531 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1532 | return 0; | |
1533 | ||
1534 | /* | |
1535 | * @defrag_bytes is a hint value, no spinlock held here, | |
1536 | * if is not zero, it means the file is defragging. | |
1537 | * Force cow if given extent needs to be defragged. | |
1538 | */ | |
1539 | if (BTRFS_I(inode)->defrag_bytes && | |
1540 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1541 | EXTENT_DEFRAG, 0, NULL)) | |
1542 | return 1; | |
1543 | ||
1544 | return 0; | |
1545 | } | |
1546 | ||
d352ac68 CM |
1547 | /* |
1548 | * extent_io.c call back to do delayed allocation processing | |
1549 | */ | |
c8b97818 | 1550 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1551 | u64 start, u64 end, int *page_started, |
1552 | unsigned long *nr_written) | |
be20aa9d | 1553 | { |
be20aa9d | 1554 | int ret; |
47059d93 | 1555 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1556 | |
47059d93 | 1557 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1558 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1559 | page_started, 1, nr_written); |
47059d93 | 1560 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1561 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1562 | page_started, 0, nr_written); |
7816030e | 1563 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1564 | ret = cow_file_range(inode, locked_page, start, end, |
1565 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1566 | } else { |
1567 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1568 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1569 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1570 | page_started, nr_written); |
7ddf5a42 | 1571 | } |
b888db2b CM |
1572 | return ret; |
1573 | } | |
1574 | ||
1bf85046 JM |
1575 | static void btrfs_split_extent_hook(struct inode *inode, |
1576 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1577 | { |
dcab6a3b JB |
1578 | u64 size; |
1579 | ||
0ca1f7ce | 1580 | /* not delalloc, ignore it */ |
9ed74f2d | 1581 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1582 | return; |
9ed74f2d | 1583 | |
dcab6a3b JB |
1584 | size = orig->end - orig->start + 1; |
1585 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
1586 | u64 num_extents; | |
1587 | u64 new_size; | |
1588 | ||
1589 | /* | |
ba117213 JB |
1590 | * See the explanation in btrfs_merge_extent_hook, the same |
1591 | * applies here, just in reverse. | |
dcab6a3b JB |
1592 | */ |
1593 | new_size = orig->end - split + 1; | |
ba117213 | 1594 | num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
dcab6a3b | 1595 | BTRFS_MAX_EXTENT_SIZE); |
ba117213 JB |
1596 | new_size = split - orig->start; |
1597 | num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1598 | BTRFS_MAX_EXTENT_SIZE); | |
1599 | if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1600 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) | |
dcab6a3b JB |
1601 | return; |
1602 | } | |
1603 | ||
9e0baf60 JB |
1604 | spin_lock(&BTRFS_I(inode)->lock); |
1605 | BTRFS_I(inode)->outstanding_extents++; | |
1606 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1607 | } |
1608 | ||
1609 | /* | |
1610 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1611 | * extents so we can keep track of new extents that are just merged onto old | |
1612 | * extents, such as when we are doing sequential writes, so we can properly | |
1613 | * account for the metadata space we'll need. | |
1614 | */ | |
1bf85046 JM |
1615 | static void btrfs_merge_extent_hook(struct inode *inode, |
1616 | struct extent_state *new, | |
1617 | struct extent_state *other) | |
9ed74f2d | 1618 | { |
dcab6a3b JB |
1619 | u64 new_size, old_size; |
1620 | u64 num_extents; | |
1621 | ||
9ed74f2d JB |
1622 | /* not delalloc, ignore it */ |
1623 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1624 | return; |
9ed74f2d | 1625 | |
8461a3de JB |
1626 | if (new->start > other->start) |
1627 | new_size = new->end - other->start + 1; | |
1628 | else | |
1629 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1630 | |
1631 | /* we're not bigger than the max, unreserve the space and go */ | |
1632 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1633 | spin_lock(&BTRFS_I(inode)->lock); | |
1634 | BTRFS_I(inode)->outstanding_extents--; | |
1635 | spin_unlock(&BTRFS_I(inode)->lock); | |
1636 | return; | |
1637 | } | |
1638 | ||
1639 | /* | |
ba117213 JB |
1640 | * We have to add up either side to figure out how many extents were |
1641 | * accounted for before we merged into one big extent. If the number of | |
1642 | * extents we accounted for is <= the amount we need for the new range | |
1643 | * then we can return, otherwise drop. Think of it like this | |
1644 | * | |
1645 | * [ 4k][MAX_SIZE] | |
1646 | * | |
1647 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1648 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1649 | * we have 1 so they are == and we can return. But in this case | |
1650 | * | |
1651 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1652 | * | |
1653 | * Each range on their own accounts for 2 extents, but merged together | |
1654 | * they are only 3 extents worth of accounting, so we need to drop in | |
1655 | * this case. | |
dcab6a3b | 1656 | */ |
ba117213 | 1657 | old_size = other->end - other->start + 1; |
dcab6a3b JB |
1658 | num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, |
1659 | BTRFS_MAX_EXTENT_SIZE); | |
ba117213 JB |
1660 | old_size = new->end - new->start + 1; |
1661 | num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1662 | BTRFS_MAX_EXTENT_SIZE); | |
1663 | ||
dcab6a3b | 1664 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
ba117213 | 1665 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) |
dcab6a3b JB |
1666 | return; |
1667 | ||
9e0baf60 JB |
1668 | spin_lock(&BTRFS_I(inode)->lock); |
1669 | BTRFS_I(inode)->outstanding_extents--; | |
1670 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1671 | } |
1672 | ||
eb73c1b7 MX |
1673 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1674 | struct inode *inode) | |
1675 | { | |
1676 | spin_lock(&root->delalloc_lock); | |
1677 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1678 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1679 | &root->delalloc_inodes); | |
1680 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1681 | &BTRFS_I(inode)->runtime_flags); | |
1682 | root->nr_delalloc_inodes++; | |
1683 | if (root->nr_delalloc_inodes == 1) { | |
1684 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1685 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1686 | list_add_tail(&root->delalloc_root, | |
1687 | &root->fs_info->delalloc_roots); | |
1688 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1689 | } | |
1690 | } | |
1691 | spin_unlock(&root->delalloc_lock); | |
1692 | } | |
1693 | ||
1694 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1695 | struct inode *inode) | |
1696 | { | |
1697 | spin_lock(&root->delalloc_lock); | |
1698 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1699 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1700 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1701 | &BTRFS_I(inode)->runtime_flags); | |
1702 | root->nr_delalloc_inodes--; | |
1703 | if (!root->nr_delalloc_inodes) { | |
1704 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1705 | BUG_ON(list_empty(&root->delalloc_root)); | |
1706 | list_del_init(&root->delalloc_root); | |
1707 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1708 | } | |
1709 | } | |
1710 | spin_unlock(&root->delalloc_lock); | |
1711 | } | |
1712 | ||
d352ac68 CM |
1713 | /* |
1714 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1715 | * bytes in this file, and to maintain the list of inodes that | |
1716 | * have pending delalloc work to be done. | |
1717 | */ | |
1bf85046 | 1718 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1719 | struct extent_state *state, unsigned *bits) |
291d673e | 1720 | { |
9ed74f2d | 1721 | |
47059d93 WS |
1722 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1723 | WARN_ON(1); | |
75eff68e CM |
1724 | /* |
1725 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1726 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1727 | * bit, which is only set or cleared with irqs on |
1728 | */ | |
0ca1f7ce | 1729 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1730 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1731 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1732 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1733 | |
9e0baf60 | 1734 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1735 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1736 | } else { |
1737 | spin_lock(&BTRFS_I(inode)->lock); | |
1738 | BTRFS_I(inode)->outstanding_extents++; | |
1739 | spin_unlock(&BTRFS_I(inode)->lock); | |
1740 | } | |
287a0ab9 | 1741 | |
6a3891c5 JB |
1742 | /* For sanity tests */ |
1743 | if (btrfs_test_is_dummy_root(root)) | |
1744 | return; | |
1745 | ||
963d678b MX |
1746 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1747 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1748 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1749 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1750 | if (*bits & EXTENT_DEFRAG) |
1751 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1752 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1753 | &BTRFS_I(inode)->runtime_flags)) |
1754 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1755 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1756 | } |
291d673e CM |
1757 | } |
1758 | ||
d352ac68 CM |
1759 | /* |
1760 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1761 | */ | |
1bf85046 | 1762 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1763 | struct extent_state *state, |
9ee49a04 | 1764 | unsigned *bits) |
291d673e | 1765 | { |
47059d93 | 1766 | u64 len = state->end + 1 - state->start; |
dcab6a3b JB |
1767 | u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1, |
1768 | BTRFS_MAX_EXTENT_SIZE); | |
47059d93 WS |
1769 | |
1770 | spin_lock(&BTRFS_I(inode)->lock); | |
1771 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1772 | BTRFS_I(inode)->defrag_bytes -= len; | |
1773 | spin_unlock(&BTRFS_I(inode)->lock); | |
1774 | ||
75eff68e CM |
1775 | /* |
1776 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1777 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1778 | * bit, which is only set or cleared with irqs on |
1779 | */ | |
0ca1f7ce | 1780 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1781 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1782 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1783 | |
9e0baf60 | 1784 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1785 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1786 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1787 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1788 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1789 | spin_unlock(&BTRFS_I(inode)->lock); |
1790 | } | |
0ca1f7ce | 1791 | |
b6d08f06 JB |
1792 | /* |
1793 | * We don't reserve metadata space for space cache inodes so we | |
1794 | * don't need to call dellalloc_release_metadata if there is an | |
1795 | * error. | |
1796 | */ | |
1797 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1798 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1799 | btrfs_delalloc_release_metadata(inode, len); |
1800 | ||
6a3891c5 JB |
1801 | /* For sanity tests. */ |
1802 | if (btrfs_test_is_dummy_root(root)) | |
1803 | return; | |
1804 | ||
0cb59c99 | 1805 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1806 | && do_list && !(state->state & EXTENT_NORESERVE)) |
51773bec QW |
1807 | btrfs_free_reserved_data_space_noquota(inode, |
1808 | state->start, len); | |
9ed74f2d | 1809 | |
963d678b MX |
1810 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1811 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1812 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1813 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1814 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1815 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1816 | &BTRFS_I(inode)->runtime_flags)) |
1817 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1818 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1819 | } |
291d673e CM |
1820 | } |
1821 | ||
d352ac68 CM |
1822 | /* |
1823 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1824 | * we don't create bios that span stripes or chunks | |
1825 | */ | |
64a16701 | 1826 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1827 | size_t size, struct bio *bio, |
1828 | unsigned long bio_flags) | |
239b14b3 CM |
1829 | { |
1830 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1831 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1832 | u64 length = 0; |
1833 | u64 map_length; | |
239b14b3 CM |
1834 | int ret; |
1835 | ||
771ed689 CM |
1836 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1837 | return 0; | |
1838 | ||
4f024f37 | 1839 | length = bio->bi_iter.bi_size; |
239b14b3 | 1840 | map_length = length; |
64a16701 | 1841 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1842 | &map_length, NULL, 0); |
3ec706c8 | 1843 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1844 | BUG_ON(ret < 0); |
d397712b | 1845 | if (map_length < length + size) |
239b14b3 | 1846 | return 1; |
3444a972 | 1847 | return 0; |
239b14b3 CM |
1848 | } |
1849 | ||
d352ac68 CM |
1850 | /* |
1851 | * in order to insert checksums into the metadata in large chunks, | |
1852 | * we wait until bio submission time. All the pages in the bio are | |
1853 | * checksummed and sums are attached onto the ordered extent record. | |
1854 | * | |
1855 | * At IO completion time the cums attached on the ordered extent record | |
1856 | * are inserted into the btree | |
1857 | */ | |
d397712b CM |
1858 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1859 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1860 | unsigned long bio_flags, |
1861 | u64 bio_offset) | |
065631f6 | 1862 | { |
065631f6 | 1863 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1864 | int ret = 0; |
e015640f | 1865 | |
d20f7043 | 1866 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1867 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1868 | return 0; |
1869 | } | |
e015640f | 1870 | |
4a69a410 CM |
1871 | /* |
1872 | * in order to insert checksums into the metadata in large chunks, | |
1873 | * we wait until bio submission time. All the pages in the bio are | |
1874 | * checksummed and sums are attached onto the ordered extent record. | |
1875 | * | |
1876 | * At IO completion time the cums attached on the ordered extent record | |
1877 | * are inserted into the btree | |
1878 | */ | |
b2950863 | 1879 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1880 | int mirror_num, unsigned long bio_flags, |
1881 | u64 bio_offset) | |
4a69a410 CM |
1882 | { |
1883 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1884 | int ret; |
1885 | ||
1886 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
4246a0b6 CH |
1887 | if (ret) { |
1888 | bio->bi_error = ret; | |
1889 | bio_endio(bio); | |
1890 | } | |
61891923 | 1891 | return ret; |
44b8bd7e CM |
1892 | } |
1893 | ||
d352ac68 | 1894 | /* |
cad321ad CM |
1895 | * extent_io.c submission hook. This does the right thing for csum calculation |
1896 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1897 | */ |
b2950863 | 1898 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1899 | int mirror_num, unsigned long bio_flags, |
1900 | u64 bio_offset) | |
44b8bd7e CM |
1901 | { |
1902 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
0d51e28a | 1903 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1904 | int ret = 0; |
19b9bdb0 | 1905 | int skip_sum; |
b812ce28 | 1906 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1907 | |
6cbff00f | 1908 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1909 | |
83eea1f1 | 1910 | if (btrfs_is_free_space_inode(inode)) |
0d51e28a | 1911 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1912 | |
7b6d91da | 1913 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1914 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1915 | if (ret) | |
61891923 | 1916 | goto out; |
5fd02043 | 1917 | |
d20f7043 | 1918 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1919 | ret = btrfs_submit_compressed_read(inode, bio, |
1920 | mirror_num, | |
1921 | bio_flags); | |
1922 | goto out; | |
c2db1073 TI |
1923 | } else if (!skip_sum) { |
1924 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1925 | if (ret) | |
61891923 | 1926 | goto out; |
c2db1073 | 1927 | } |
4d1b5fb4 | 1928 | goto mapit; |
b812ce28 | 1929 | } else if (async && !skip_sum) { |
17d217fe YZ |
1930 | /* csum items have already been cloned */ |
1931 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1932 | goto mapit; | |
19b9bdb0 | 1933 | /* we're doing a write, do the async checksumming */ |
61891923 | 1934 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1935 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1936 | bio_flags, bio_offset, |
1937 | __btrfs_submit_bio_start, | |
4a69a410 | 1938 | __btrfs_submit_bio_done); |
61891923 | 1939 | goto out; |
b812ce28 JB |
1940 | } else if (!skip_sum) { |
1941 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1942 | if (ret) | |
1943 | goto out; | |
19b9bdb0 CM |
1944 | } |
1945 | ||
0b86a832 | 1946 | mapit: |
61891923 SB |
1947 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1948 | ||
1949 | out: | |
4246a0b6 CH |
1950 | if (ret < 0) { |
1951 | bio->bi_error = ret; | |
1952 | bio_endio(bio); | |
1953 | } | |
61891923 | 1954 | return ret; |
065631f6 | 1955 | } |
6885f308 | 1956 | |
d352ac68 CM |
1957 | /* |
1958 | * given a list of ordered sums record them in the inode. This happens | |
1959 | * at IO completion time based on sums calculated at bio submission time. | |
1960 | */ | |
ba1da2f4 | 1961 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1962 | struct inode *inode, u64 file_offset, |
1963 | struct list_head *list) | |
1964 | { | |
e6dcd2dc CM |
1965 | struct btrfs_ordered_sum *sum; |
1966 | ||
c6e30871 | 1967 | list_for_each_entry(sum, list, list) { |
39847c4d | 1968 | trans->adding_csums = 1; |
d20f7043 CM |
1969 | btrfs_csum_file_blocks(trans, |
1970 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1971 | trans->adding_csums = 0; |
e6dcd2dc CM |
1972 | } |
1973 | return 0; | |
1974 | } | |
1975 | ||
2ac55d41 JB |
1976 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1977 | struct extent_state **cached_state) | |
ea8c2819 | 1978 | { |
09cbfeaf | 1979 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 1980 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
7cd8c752 | 1981 | cached_state); |
ea8c2819 CM |
1982 | } |
1983 | ||
d352ac68 | 1984 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1985 | struct btrfs_writepage_fixup { |
1986 | struct page *page; | |
1987 | struct btrfs_work work; | |
1988 | }; | |
1989 | ||
b2950863 | 1990 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1991 | { |
1992 | struct btrfs_writepage_fixup *fixup; | |
1993 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1994 | struct extent_state *cached_state = NULL; |
247e743c CM |
1995 | struct page *page; |
1996 | struct inode *inode; | |
1997 | u64 page_start; | |
1998 | u64 page_end; | |
87826df0 | 1999 | int ret; |
247e743c CM |
2000 | |
2001 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2002 | page = fixup->page; | |
4a096752 | 2003 | again: |
247e743c CM |
2004 | lock_page(page); |
2005 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2006 | ClearPageChecked(page); | |
2007 | goto out_page; | |
2008 | } | |
2009 | ||
2010 | inode = page->mapping->host; | |
2011 | page_start = page_offset(page); | |
09cbfeaf | 2012 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2013 | |
ff13db41 | 2014 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2015 | &cached_state); |
4a096752 CM |
2016 | |
2017 | /* already ordered? We're done */ | |
8b62b72b | 2018 | if (PagePrivate2(page)) |
247e743c | 2019 | goto out; |
4a096752 | 2020 | |
dbfdb6d1 | 2021 | ordered = btrfs_lookup_ordered_range(inode, page_start, |
09cbfeaf | 2022 | PAGE_SIZE); |
4a096752 | 2023 | if (ordered) { |
2ac55d41 JB |
2024 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2025 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2026 | unlock_page(page); |
2027 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2028 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2029 | goto again; |
2030 | } | |
247e743c | 2031 | |
7cf5b976 | 2032 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
09cbfeaf | 2033 | PAGE_SIZE); |
87826df0 JM |
2034 | if (ret) { |
2035 | mapping_set_error(page->mapping, ret); | |
2036 | end_extent_writepage(page, ret, page_start, page_end); | |
2037 | ClearPageChecked(page); | |
2038 | goto out; | |
2039 | } | |
2040 | ||
2ac55d41 | 2041 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 2042 | ClearPageChecked(page); |
87826df0 | 2043 | set_page_dirty(page); |
247e743c | 2044 | out: |
2ac55d41 JB |
2045 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2046 | &cached_state, GFP_NOFS); | |
247e743c CM |
2047 | out_page: |
2048 | unlock_page(page); | |
09cbfeaf | 2049 | put_page(page); |
b897abec | 2050 | kfree(fixup); |
247e743c CM |
2051 | } |
2052 | ||
2053 | /* | |
2054 | * There are a few paths in the higher layers of the kernel that directly | |
2055 | * set the page dirty bit without asking the filesystem if it is a | |
2056 | * good idea. This causes problems because we want to make sure COW | |
2057 | * properly happens and the data=ordered rules are followed. | |
2058 | * | |
c8b97818 | 2059 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2060 | * hasn't been properly setup for IO. We kick off an async process |
2061 | * to fix it up. The async helper will wait for ordered extents, set | |
2062 | * the delalloc bit and make it safe to write the page. | |
2063 | */ | |
b2950863 | 2064 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2065 | { |
2066 | struct inode *inode = page->mapping->host; | |
2067 | struct btrfs_writepage_fixup *fixup; | |
2068 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 2069 | |
8b62b72b CM |
2070 | /* this page is properly in the ordered list */ |
2071 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2072 | return 0; |
2073 | ||
2074 | if (PageChecked(page)) | |
2075 | return -EAGAIN; | |
2076 | ||
2077 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2078 | if (!fixup) | |
2079 | return -EAGAIN; | |
f421950f | 2080 | |
247e743c | 2081 | SetPageChecked(page); |
09cbfeaf | 2082 | get_page(page); |
9e0af237 LB |
2083 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2084 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2085 | fixup->page = page; |
dc6e3209 | 2086 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 2087 | return -EBUSY; |
247e743c CM |
2088 | } |
2089 | ||
d899e052 YZ |
2090 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2091 | struct inode *inode, u64 file_pos, | |
2092 | u64 disk_bytenr, u64 disk_num_bytes, | |
2093 | u64 num_bytes, u64 ram_bytes, | |
2094 | u8 compression, u8 encryption, | |
2095 | u16 other_encoding, int extent_type) | |
2096 | { | |
2097 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2098 | struct btrfs_file_extent_item *fi; | |
2099 | struct btrfs_path *path; | |
2100 | struct extent_buffer *leaf; | |
2101 | struct btrfs_key ins; | |
1acae57b | 2102 | int extent_inserted = 0; |
d899e052 YZ |
2103 | int ret; |
2104 | ||
2105 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2106 | if (!path) |
2107 | return -ENOMEM; | |
d899e052 | 2108 | |
a1ed835e CM |
2109 | /* |
2110 | * we may be replacing one extent in the tree with another. | |
2111 | * The new extent is pinned in the extent map, and we don't want | |
2112 | * to drop it from the cache until it is completely in the btree. | |
2113 | * | |
2114 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2115 | * the caller is expected to unpin it and allow it to be merged | |
2116 | * with the others. | |
2117 | */ | |
1acae57b FDBM |
2118 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2119 | file_pos + num_bytes, NULL, 0, | |
2120 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2121 | if (ret) |
2122 | goto out; | |
d899e052 | 2123 | |
1acae57b FDBM |
2124 | if (!extent_inserted) { |
2125 | ins.objectid = btrfs_ino(inode); | |
2126 | ins.offset = file_pos; | |
2127 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2128 | ||
2129 | path->leave_spinning = 1; | |
2130 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2131 | sizeof(*fi)); | |
2132 | if (ret) | |
2133 | goto out; | |
2134 | } | |
d899e052 YZ |
2135 | leaf = path->nodes[0]; |
2136 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2137 | struct btrfs_file_extent_item); | |
2138 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2139 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2140 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2141 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2142 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2143 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2144 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2145 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2146 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2147 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2148 | |
d899e052 | 2149 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2150 | btrfs_release_path(path); |
d899e052 YZ |
2151 | |
2152 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2153 | |
2154 | ins.objectid = disk_bytenr; | |
2155 | ins.offset = disk_num_bytes; | |
2156 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2157 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2158 | root->root_key.objectid, | |
5846a3c2 QW |
2159 | btrfs_ino(inode), file_pos, |
2160 | ram_bytes, &ins); | |
297d750b | 2161 | /* |
5846a3c2 QW |
2162 | * Release the reserved range from inode dirty range map, as it is |
2163 | * already moved into delayed_ref_head | |
297d750b QW |
2164 | */ |
2165 | btrfs_qgroup_release_data(inode, file_pos, ram_bytes); | |
79787eaa | 2166 | out: |
d899e052 | 2167 | btrfs_free_path(path); |
b9473439 | 2168 | |
79787eaa | 2169 | return ret; |
d899e052 YZ |
2170 | } |
2171 | ||
38c227d8 LB |
2172 | /* snapshot-aware defrag */ |
2173 | struct sa_defrag_extent_backref { | |
2174 | struct rb_node node; | |
2175 | struct old_sa_defrag_extent *old; | |
2176 | u64 root_id; | |
2177 | u64 inum; | |
2178 | u64 file_pos; | |
2179 | u64 extent_offset; | |
2180 | u64 num_bytes; | |
2181 | u64 generation; | |
2182 | }; | |
2183 | ||
2184 | struct old_sa_defrag_extent { | |
2185 | struct list_head list; | |
2186 | struct new_sa_defrag_extent *new; | |
2187 | ||
2188 | u64 extent_offset; | |
2189 | u64 bytenr; | |
2190 | u64 offset; | |
2191 | u64 len; | |
2192 | int count; | |
2193 | }; | |
2194 | ||
2195 | struct new_sa_defrag_extent { | |
2196 | struct rb_root root; | |
2197 | struct list_head head; | |
2198 | struct btrfs_path *path; | |
2199 | struct inode *inode; | |
2200 | u64 file_pos; | |
2201 | u64 len; | |
2202 | u64 bytenr; | |
2203 | u64 disk_len; | |
2204 | u8 compress_type; | |
2205 | }; | |
2206 | ||
2207 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2208 | struct sa_defrag_extent_backref *b2) | |
2209 | { | |
2210 | if (b1->root_id < b2->root_id) | |
2211 | return -1; | |
2212 | else if (b1->root_id > b2->root_id) | |
2213 | return 1; | |
2214 | ||
2215 | if (b1->inum < b2->inum) | |
2216 | return -1; | |
2217 | else if (b1->inum > b2->inum) | |
2218 | return 1; | |
2219 | ||
2220 | if (b1->file_pos < b2->file_pos) | |
2221 | return -1; | |
2222 | else if (b1->file_pos > b2->file_pos) | |
2223 | return 1; | |
2224 | ||
2225 | /* | |
2226 | * [------------------------------] ===> (a range of space) | |
2227 | * |<--->| |<---->| =============> (fs/file tree A) | |
2228 | * |<---------------------------->| ===> (fs/file tree B) | |
2229 | * | |
2230 | * A range of space can refer to two file extents in one tree while | |
2231 | * refer to only one file extent in another tree. | |
2232 | * | |
2233 | * So we may process a disk offset more than one time(two extents in A) | |
2234 | * and locate at the same extent(one extent in B), then insert two same | |
2235 | * backrefs(both refer to the extent in B). | |
2236 | */ | |
2237 | return 0; | |
2238 | } | |
2239 | ||
2240 | static void backref_insert(struct rb_root *root, | |
2241 | struct sa_defrag_extent_backref *backref) | |
2242 | { | |
2243 | struct rb_node **p = &root->rb_node; | |
2244 | struct rb_node *parent = NULL; | |
2245 | struct sa_defrag_extent_backref *entry; | |
2246 | int ret; | |
2247 | ||
2248 | while (*p) { | |
2249 | parent = *p; | |
2250 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2251 | ||
2252 | ret = backref_comp(backref, entry); | |
2253 | if (ret < 0) | |
2254 | p = &(*p)->rb_left; | |
2255 | else | |
2256 | p = &(*p)->rb_right; | |
2257 | } | |
2258 | ||
2259 | rb_link_node(&backref->node, parent, p); | |
2260 | rb_insert_color(&backref->node, root); | |
2261 | } | |
2262 | ||
2263 | /* | |
2264 | * Note the backref might has changed, and in this case we just return 0. | |
2265 | */ | |
2266 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2267 | void *ctx) | |
2268 | { | |
2269 | struct btrfs_file_extent_item *extent; | |
2270 | struct btrfs_fs_info *fs_info; | |
2271 | struct old_sa_defrag_extent *old = ctx; | |
2272 | struct new_sa_defrag_extent *new = old->new; | |
2273 | struct btrfs_path *path = new->path; | |
2274 | struct btrfs_key key; | |
2275 | struct btrfs_root *root; | |
2276 | struct sa_defrag_extent_backref *backref; | |
2277 | struct extent_buffer *leaf; | |
2278 | struct inode *inode = new->inode; | |
2279 | int slot; | |
2280 | int ret; | |
2281 | u64 extent_offset; | |
2282 | u64 num_bytes; | |
2283 | ||
2284 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2285 | inum == btrfs_ino(inode)) | |
2286 | return 0; | |
2287 | ||
2288 | key.objectid = root_id; | |
2289 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2290 | key.offset = (u64)-1; | |
2291 | ||
2292 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2293 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2294 | if (IS_ERR(root)) { | |
2295 | if (PTR_ERR(root) == -ENOENT) | |
2296 | return 0; | |
2297 | WARN_ON(1); | |
2298 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2299 | inum, offset, root_id); | |
2300 | return PTR_ERR(root); | |
2301 | } | |
2302 | ||
2303 | key.objectid = inum; | |
2304 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2305 | if (offset > (u64)-1 << 32) | |
2306 | key.offset = 0; | |
2307 | else | |
2308 | key.offset = offset; | |
2309 | ||
2310 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2311 | if (WARN_ON(ret < 0)) |
38c227d8 | 2312 | return ret; |
50f1319c | 2313 | ret = 0; |
38c227d8 LB |
2314 | |
2315 | while (1) { | |
2316 | cond_resched(); | |
2317 | ||
2318 | leaf = path->nodes[0]; | |
2319 | slot = path->slots[0]; | |
2320 | ||
2321 | if (slot >= btrfs_header_nritems(leaf)) { | |
2322 | ret = btrfs_next_leaf(root, path); | |
2323 | if (ret < 0) { | |
2324 | goto out; | |
2325 | } else if (ret > 0) { | |
2326 | ret = 0; | |
2327 | goto out; | |
2328 | } | |
2329 | continue; | |
2330 | } | |
2331 | ||
2332 | path->slots[0]++; | |
2333 | ||
2334 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2335 | ||
2336 | if (key.objectid > inum) | |
2337 | goto out; | |
2338 | ||
2339 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2340 | continue; | |
2341 | ||
2342 | extent = btrfs_item_ptr(leaf, slot, | |
2343 | struct btrfs_file_extent_item); | |
2344 | ||
2345 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2346 | continue; | |
2347 | ||
e68afa49 LB |
2348 | /* |
2349 | * 'offset' refers to the exact key.offset, | |
2350 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2351 | * (key.offset - extent_offset). | |
2352 | */ | |
2353 | if (key.offset != offset) | |
38c227d8 LB |
2354 | continue; |
2355 | ||
e68afa49 | 2356 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2357 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2358 | |
38c227d8 LB |
2359 | if (extent_offset >= old->extent_offset + old->offset + |
2360 | old->len || extent_offset + num_bytes <= | |
2361 | old->extent_offset + old->offset) | |
2362 | continue; | |
38c227d8 LB |
2363 | break; |
2364 | } | |
2365 | ||
2366 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2367 | if (!backref) { | |
2368 | ret = -ENOENT; | |
2369 | goto out; | |
2370 | } | |
2371 | ||
2372 | backref->root_id = root_id; | |
2373 | backref->inum = inum; | |
e68afa49 | 2374 | backref->file_pos = offset; |
38c227d8 LB |
2375 | backref->num_bytes = num_bytes; |
2376 | backref->extent_offset = extent_offset; | |
2377 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2378 | backref->old = old; | |
2379 | backref_insert(&new->root, backref); | |
2380 | old->count++; | |
2381 | out: | |
2382 | btrfs_release_path(path); | |
2383 | WARN_ON(ret); | |
2384 | return ret; | |
2385 | } | |
2386 | ||
2387 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2388 | struct new_sa_defrag_extent *new) | |
2389 | { | |
2390 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2391 | struct old_sa_defrag_extent *old, *tmp; | |
2392 | int ret; | |
2393 | ||
2394 | new->path = path; | |
2395 | ||
2396 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2397 | ret = iterate_inodes_from_logical(old->bytenr + |
2398 | old->extent_offset, fs_info, | |
38c227d8 LB |
2399 | path, record_one_backref, |
2400 | old); | |
4724b106 JB |
2401 | if (ret < 0 && ret != -ENOENT) |
2402 | return false; | |
38c227d8 LB |
2403 | |
2404 | /* no backref to be processed for this extent */ | |
2405 | if (!old->count) { | |
2406 | list_del(&old->list); | |
2407 | kfree(old); | |
2408 | } | |
2409 | } | |
2410 | ||
2411 | if (list_empty(&new->head)) | |
2412 | return false; | |
2413 | ||
2414 | return true; | |
2415 | } | |
2416 | ||
2417 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2418 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2419 | struct new_sa_defrag_extent *new) |
38c227d8 | 2420 | { |
116e0024 | 2421 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2422 | return 0; |
2423 | ||
2424 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2425 | return 0; | |
2426 | ||
116e0024 LB |
2427 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2428 | return 0; | |
2429 | ||
2430 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2431 | btrfs_file_extent_other_encoding(leaf, fi)) |
2432 | return 0; | |
2433 | ||
2434 | return 1; | |
2435 | } | |
2436 | ||
2437 | /* | |
2438 | * Note the backref might has changed, and in this case we just return 0. | |
2439 | */ | |
2440 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2441 | struct sa_defrag_extent_backref *prev, | |
2442 | struct sa_defrag_extent_backref *backref) | |
2443 | { | |
2444 | struct btrfs_file_extent_item *extent; | |
2445 | struct btrfs_file_extent_item *item; | |
2446 | struct btrfs_ordered_extent *ordered; | |
2447 | struct btrfs_trans_handle *trans; | |
2448 | struct btrfs_fs_info *fs_info; | |
2449 | struct btrfs_root *root; | |
2450 | struct btrfs_key key; | |
2451 | struct extent_buffer *leaf; | |
2452 | struct old_sa_defrag_extent *old = backref->old; | |
2453 | struct new_sa_defrag_extent *new = old->new; | |
2454 | struct inode *src_inode = new->inode; | |
2455 | struct inode *inode; | |
2456 | struct extent_state *cached = NULL; | |
2457 | int ret = 0; | |
2458 | u64 start; | |
2459 | u64 len; | |
2460 | u64 lock_start; | |
2461 | u64 lock_end; | |
2462 | bool merge = false; | |
2463 | int index; | |
2464 | ||
2465 | if (prev && prev->root_id == backref->root_id && | |
2466 | prev->inum == backref->inum && | |
2467 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2468 | merge = true; | |
2469 | ||
2470 | /* step 1: get root */ | |
2471 | key.objectid = backref->root_id; | |
2472 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2473 | key.offset = (u64)-1; | |
2474 | ||
2475 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2476 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2477 | ||
2478 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2479 | if (IS_ERR(root)) { | |
2480 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2481 | if (PTR_ERR(root) == -ENOENT) | |
2482 | return 0; | |
2483 | return PTR_ERR(root); | |
2484 | } | |
38c227d8 | 2485 | |
bcbba5e6 WS |
2486 | if (btrfs_root_readonly(root)) { |
2487 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2488 | return 0; | |
2489 | } | |
2490 | ||
38c227d8 LB |
2491 | /* step 2: get inode */ |
2492 | key.objectid = backref->inum; | |
2493 | key.type = BTRFS_INODE_ITEM_KEY; | |
2494 | key.offset = 0; | |
2495 | ||
2496 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2497 | if (IS_ERR(inode)) { | |
2498 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2499 | return 0; | |
2500 | } | |
2501 | ||
2502 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2503 | ||
2504 | /* step 3: relink backref */ | |
2505 | lock_start = backref->file_pos; | |
2506 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2507 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2508 | &cached); |
38c227d8 LB |
2509 | |
2510 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2511 | if (ordered) { | |
2512 | btrfs_put_ordered_extent(ordered); | |
2513 | goto out_unlock; | |
2514 | } | |
2515 | ||
2516 | trans = btrfs_join_transaction(root); | |
2517 | if (IS_ERR(trans)) { | |
2518 | ret = PTR_ERR(trans); | |
2519 | goto out_unlock; | |
2520 | } | |
2521 | ||
2522 | key.objectid = backref->inum; | |
2523 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2524 | key.offset = backref->file_pos; | |
2525 | ||
2526 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2527 | if (ret < 0) { | |
2528 | goto out_free_path; | |
2529 | } else if (ret > 0) { | |
2530 | ret = 0; | |
2531 | goto out_free_path; | |
2532 | } | |
2533 | ||
2534 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2535 | struct btrfs_file_extent_item); | |
2536 | ||
2537 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2538 | backref->generation) | |
2539 | goto out_free_path; | |
2540 | ||
2541 | btrfs_release_path(path); | |
2542 | ||
2543 | start = backref->file_pos; | |
2544 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2545 | start += old->extent_offset + old->offset - | |
2546 | backref->extent_offset; | |
2547 | ||
2548 | len = min(backref->extent_offset + backref->num_bytes, | |
2549 | old->extent_offset + old->offset + old->len); | |
2550 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2551 | ||
2552 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2553 | start + len, 1); | |
2554 | if (ret) | |
2555 | goto out_free_path; | |
2556 | again: | |
2557 | key.objectid = btrfs_ino(inode); | |
2558 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2559 | key.offset = start; | |
2560 | ||
a09a0a70 | 2561 | path->leave_spinning = 1; |
38c227d8 LB |
2562 | if (merge) { |
2563 | struct btrfs_file_extent_item *fi; | |
2564 | u64 extent_len; | |
2565 | struct btrfs_key found_key; | |
2566 | ||
3c9665df | 2567 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2568 | if (ret < 0) |
2569 | goto out_free_path; | |
2570 | ||
2571 | path->slots[0]--; | |
2572 | leaf = path->nodes[0]; | |
2573 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2574 | ||
2575 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2576 | struct btrfs_file_extent_item); | |
2577 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2578 | ||
116e0024 LB |
2579 | if (extent_len + found_key.offset == start && |
2580 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2581 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2582 | extent_len + len); | |
2583 | btrfs_mark_buffer_dirty(leaf); | |
2584 | inode_add_bytes(inode, len); | |
2585 | ||
2586 | ret = 1; | |
2587 | goto out_free_path; | |
2588 | } else { | |
2589 | merge = false; | |
2590 | btrfs_release_path(path); | |
2591 | goto again; | |
2592 | } | |
2593 | } | |
2594 | ||
2595 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2596 | sizeof(*extent)); | |
2597 | if (ret) { | |
2598 | btrfs_abort_transaction(trans, root, ret); | |
2599 | goto out_free_path; | |
2600 | } | |
2601 | ||
2602 | leaf = path->nodes[0]; | |
2603 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2604 | struct btrfs_file_extent_item); | |
2605 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2606 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2607 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2608 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2609 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2610 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2611 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2612 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2613 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2614 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2615 | ||
2616 | btrfs_mark_buffer_dirty(leaf); | |
2617 | inode_add_bytes(inode, len); | |
a09a0a70 | 2618 | btrfs_release_path(path); |
38c227d8 LB |
2619 | |
2620 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2621 | new->disk_len, 0, | |
2622 | backref->root_id, backref->inum, | |
b06c4bf5 | 2623 | new->file_pos); /* start - extent_offset */ |
38c227d8 LB |
2624 | if (ret) { |
2625 | btrfs_abort_transaction(trans, root, ret); | |
2626 | goto out_free_path; | |
2627 | } | |
2628 | ||
2629 | ret = 1; | |
2630 | out_free_path: | |
2631 | btrfs_release_path(path); | |
a09a0a70 | 2632 | path->leave_spinning = 0; |
38c227d8 LB |
2633 | btrfs_end_transaction(trans, root); |
2634 | out_unlock: | |
2635 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2636 | &cached, GFP_NOFS); | |
2637 | iput(inode); | |
2638 | return ret; | |
2639 | } | |
2640 | ||
6f519564 LB |
2641 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2642 | { | |
2643 | struct old_sa_defrag_extent *old, *tmp; | |
2644 | ||
2645 | if (!new) | |
2646 | return; | |
2647 | ||
2648 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2649 | kfree(old); |
2650 | } | |
2651 | kfree(new); | |
2652 | } | |
2653 | ||
38c227d8 LB |
2654 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2655 | { | |
2656 | struct btrfs_path *path; | |
38c227d8 LB |
2657 | struct sa_defrag_extent_backref *backref; |
2658 | struct sa_defrag_extent_backref *prev = NULL; | |
2659 | struct inode *inode; | |
2660 | struct btrfs_root *root; | |
2661 | struct rb_node *node; | |
2662 | int ret; | |
2663 | ||
2664 | inode = new->inode; | |
2665 | root = BTRFS_I(inode)->root; | |
2666 | ||
2667 | path = btrfs_alloc_path(); | |
2668 | if (!path) | |
2669 | return; | |
2670 | ||
2671 | if (!record_extent_backrefs(path, new)) { | |
2672 | btrfs_free_path(path); | |
2673 | goto out; | |
2674 | } | |
2675 | btrfs_release_path(path); | |
2676 | ||
2677 | while (1) { | |
2678 | node = rb_first(&new->root); | |
2679 | if (!node) | |
2680 | break; | |
2681 | rb_erase(node, &new->root); | |
2682 | ||
2683 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2684 | ||
2685 | ret = relink_extent_backref(path, prev, backref); | |
2686 | WARN_ON(ret < 0); | |
2687 | ||
2688 | kfree(prev); | |
2689 | ||
2690 | if (ret == 1) | |
2691 | prev = backref; | |
2692 | else | |
2693 | prev = NULL; | |
2694 | cond_resched(); | |
2695 | } | |
2696 | kfree(prev); | |
2697 | ||
2698 | btrfs_free_path(path); | |
38c227d8 | 2699 | out: |
6f519564 LB |
2700 | free_sa_defrag_extent(new); |
2701 | ||
38c227d8 LB |
2702 | atomic_dec(&root->fs_info->defrag_running); |
2703 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2704 | } |
2705 | ||
2706 | static struct new_sa_defrag_extent * | |
2707 | record_old_file_extents(struct inode *inode, | |
2708 | struct btrfs_ordered_extent *ordered) | |
2709 | { | |
2710 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2711 | struct btrfs_path *path; | |
2712 | struct btrfs_key key; | |
6f519564 | 2713 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2714 | struct new_sa_defrag_extent *new; |
2715 | int ret; | |
2716 | ||
2717 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2718 | if (!new) | |
2719 | return NULL; | |
2720 | ||
2721 | new->inode = inode; | |
2722 | new->file_pos = ordered->file_offset; | |
2723 | new->len = ordered->len; | |
2724 | new->bytenr = ordered->start; | |
2725 | new->disk_len = ordered->disk_len; | |
2726 | new->compress_type = ordered->compress_type; | |
2727 | new->root = RB_ROOT; | |
2728 | INIT_LIST_HEAD(&new->head); | |
2729 | ||
2730 | path = btrfs_alloc_path(); | |
2731 | if (!path) | |
2732 | goto out_kfree; | |
2733 | ||
2734 | key.objectid = btrfs_ino(inode); | |
2735 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2736 | key.offset = new->file_pos; | |
2737 | ||
2738 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2739 | if (ret < 0) | |
2740 | goto out_free_path; | |
2741 | if (ret > 0 && path->slots[0] > 0) | |
2742 | path->slots[0]--; | |
2743 | ||
2744 | /* find out all the old extents for the file range */ | |
2745 | while (1) { | |
2746 | struct btrfs_file_extent_item *extent; | |
2747 | struct extent_buffer *l; | |
2748 | int slot; | |
2749 | u64 num_bytes; | |
2750 | u64 offset; | |
2751 | u64 end; | |
2752 | u64 disk_bytenr; | |
2753 | u64 extent_offset; | |
2754 | ||
2755 | l = path->nodes[0]; | |
2756 | slot = path->slots[0]; | |
2757 | ||
2758 | if (slot >= btrfs_header_nritems(l)) { | |
2759 | ret = btrfs_next_leaf(root, path); | |
2760 | if (ret < 0) | |
6f519564 | 2761 | goto out_free_path; |
38c227d8 LB |
2762 | else if (ret > 0) |
2763 | break; | |
2764 | continue; | |
2765 | } | |
2766 | ||
2767 | btrfs_item_key_to_cpu(l, &key, slot); | |
2768 | ||
2769 | if (key.objectid != btrfs_ino(inode)) | |
2770 | break; | |
2771 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2772 | break; | |
2773 | if (key.offset >= new->file_pos + new->len) | |
2774 | break; | |
2775 | ||
2776 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2777 | ||
2778 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2779 | if (key.offset + num_bytes < new->file_pos) | |
2780 | goto next; | |
2781 | ||
2782 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2783 | if (!disk_bytenr) | |
2784 | goto next; | |
2785 | ||
2786 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2787 | ||
2788 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2789 | if (!old) | |
6f519564 | 2790 | goto out_free_path; |
38c227d8 LB |
2791 | |
2792 | offset = max(new->file_pos, key.offset); | |
2793 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2794 | ||
2795 | old->bytenr = disk_bytenr; | |
2796 | old->extent_offset = extent_offset; | |
2797 | old->offset = offset - key.offset; | |
2798 | old->len = end - offset; | |
2799 | old->new = new; | |
2800 | old->count = 0; | |
2801 | list_add_tail(&old->list, &new->head); | |
2802 | next: | |
2803 | path->slots[0]++; | |
2804 | cond_resched(); | |
2805 | } | |
2806 | ||
2807 | btrfs_free_path(path); | |
2808 | atomic_inc(&root->fs_info->defrag_running); | |
2809 | ||
2810 | return new; | |
2811 | ||
38c227d8 LB |
2812 | out_free_path: |
2813 | btrfs_free_path(path); | |
2814 | out_kfree: | |
6f519564 | 2815 | free_sa_defrag_extent(new); |
38c227d8 LB |
2816 | return NULL; |
2817 | } | |
2818 | ||
e570fd27 MX |
2819 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2820 | u64 start, u64 len) | |
2821 | { | |
2822 | struct btrfs_block_group_cache *cache; | |
2823 | ||
2824 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2825 | ASSERT(cache); | |
2826 | ||
2827 | spin_lock(&cache->lock); | |
2828 | cache->delalloc_bytes -= len; | |
2829 | spin_unlock(&cache->lock); | |
2830 | ||
2831 | btrfs_put_block_group(cache); | |
2832 | } | |
2833 | ||
d352ac68 CM |
2834 | /* as ordered data IO finishes, this gets called so we can finish |
2835 | * an ordered extent if the range of bytes in the file it covers are | |
2836 | * fully written. | |
2837 | */ | |
5fd02043 | 2838 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2839 | { |
5fd02043 | 2840 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2841 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2842 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2843 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2844 | struct extent_state *cached_state = NULL; |
38c227d8 | 2845 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2846 | int compress_type = 0; |
77cef2ec JB |
2847 | int ret = 0; |
2848 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2849 | bool nolock; |
77cef2ec | 2850 | bool truncated = false; |
e6dcd2dc | 2851 | |
83eea1f1 | 2852 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2853 | |
5fd02043 JB |
2854 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2855 | ret = -EIO; | |
2856 | goto out; | |
2857 | } | |
2858 | ||
f612496b MX |
2859 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2860 | ordered_extent->file_offset + | |
2861 | ordered_extent->len - 1); | |
2862 | ||
77cef2ec JB |
2863 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2864 | truncated = true; | |
2865 | logical_len = ordered_extent->truncated_len; | |
2866 | /* Truncated the entire extent, don't bother adding */ | |
2867 | if (!logical_len) | |
2868 | goto out; | |
2869 | } | |
2870 | ||
c2167754 | 2871 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2872 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2873 | |
2874 | /* | |
2875 | * For mwrite(mmap + memset to write) case, we still reserve | |
2876 | * space for NOCOW range. | |
2877 | * As NOCOW won't cause a new delayed ref, just free the space | |
2878 | */ | |
2879 | btrfs_qgroup_free_data(inode, ordered_extent->file_offset, | |
2880 | ordered_extent->len); | |
6c760c07 JB |
2881 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2882 | if (nolock) | |
2883 | trans = btrfs_join_transaction_nolock(root); | |
2884 | else | |
2885 | trans = btrfs_join_transaction(root); | |
2886 | if (IS_ERR(trans)) { | |
2887 | ret = PTR_ERR(trans); | |
2888 | trans = NULL; | |
2889 | goto out; | |
c2167754 | 2890 | } |
6c760c07 JB |
2891 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2892 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2893 | if (ret) /* -ENOMEM or corruption */ | |
2894 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2895 | goto out; |
2896 | } | |
e6dcd2dc | 2897 | |
2ac55d41 JB |
2898 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2899 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2900 | &cached_state); |
e6dcd2dc | 2901 | |
38c227d8 LB |
2902 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2903 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2904 | EXTENT_DEFRAG, 1, cached_state); | |
2905 | if (ret) { | |
2906 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2907 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2908 | /* the inode is shared */ |
2909 | new = record_old_file_extents(inode, ordered_extent); | |
2910 | ||
2911 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2912 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2913 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2914 | } | |
2915 | ||
0cb59c99 | 2916 | if (nolock) |
7a7eaa40 | 2917 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2918 | else |
7a7eaa40 | 2919 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2920 | if (IS_ERR(trans)) { |
2921 | ret = PTR_ERR(trans); | |
2922 | trans = NULL; | |
2923 | goto out_unlock; | |
2924 | } | |
a79b7d4b | 2925 | |
0ca1f7ce | 2926 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2927 | |
c8b97818 | 2928 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2929 | compress_type = ordered_extent->compress_type; |
d899e052 | 2930 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2931 | BUG_ON(compress_type); |
920bbbfb | 2932 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2933 | ordered_extent->file_offset, |
2934 | ordered_extent->file_offset + | |
77cef2ec | 2935 | logical_len); |
d899e052 | 2936 | } else { |
0af3d00b | 2937 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2938 | ret = insert_reserved_file_extent(trans, inode, |
2939 | ordered_extent->file_offset, | |
2940 | ordered_extent->start, | |
2941 | ordered_extent->disk_len, | |
77cef2ec | 2942 | logical_len, logical_len, |
261507a0 | 2943 | compress_type, 0, 0, |
d899e052 | 2944 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2945 | if (!ret) |
2946 | btrfs_release_delalloc_bytes(root, | |
2947 | ordered_extent->start, | |
2948 | ordered_extent->disk_len); | |
d899e052 | 2949 | } |
5dc562c5 JB |
2950 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2951 | ordered_extent->file_offset, ordered_extent->len, | |
2952 | trans->transid); | |
79787eaa JM |
2953 | if (ret < 0) { |
2954 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2955 | goto out_unlock; |
79787eaa | 2956 | } |
2ac55d41 | 2957 | |
e6dcd2dc CM |
2958 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2959 | &ordered_extent->list); | |
2960 | ||
6c760c07 JB |
2961 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2962 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2963 | if (ret) { /* -ENOMEM or corruption */ | |
2964 | btrfs_abort_transaction(trans, root, ret); | |
2965 | goto out_unlock; | |
1ef30be1 JB |
2966 | } |
2967 | ret = 0; | |
5fd02043 JB |
2968 | out_unlock: |
2969 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2970 | ordered_extent->file_offset + | |
2971 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2972 | out: |
5b0e95bf | 2973 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2974 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2975 | if (trans) |
2976 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2977 | |
77cef2ec JB |
2978 | if (ret || truncated) { |
2979 | u64 start, end; | |
2980 | ||
2981 | if (truncated) | |
2982 | start = ordered_extent->file_offset + logical_len; | |
2983 | else | |
2984 | start = ordered_extent->file_offset; | |
2985 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2986 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2987 | ||
2988 | /* Drop the cache for the part of the extent we didn't write. */ | |
2989 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2990 | |
0bec9ef5 JB |
2991 | /* |
2992 | * If the ordered extent had an IOERR or something else went | |
2993 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2994 | * back to the allocator. We only free the extent in the |
2995 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2996 | */ |
77cef2ec JB |
2997 | if ((ret || !logical_len) && |
2998 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2999 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
3000 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 3001 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3002 | } |
3003 | ||
3004 | ||
5fd02043 | 3005 | /* |
8bad3c02 LB |
3006 | * This needs to be done to make sure anybody waiting knows we are done |
3007 | * updating everything for this ordered extent. | |
5fd02043 JB |
3008 | */ |
3009 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3010 | ||
38c227d8 | 3011 | /* for snapshot-aware defrag */ |
6f519564 LB |
3012 | if (new) { |
3013 | if (ret) { | |
3014 | free_sa_defrag_extent(new); | |
3015 | atomic_dec(&root->fs_info->defrag_running); | |
3016 | } else { | |
3017 | relink_file_extents(new); | |
3018 | } | |
3019 | } | |
38c227d8 | 3020 | |
e6dcd2dc CM |
3021 | /* once for us */ |
3022 | btrfs_put_ordered_extent(ordered_extent); | |
3023 | /* once for the tree */ | |
3024 | btrfs_put_ordered_extent(ordered_extent); | |
3025 | ||
5fd02043 JB |
3026 | return ret; |
3027 | } | |
3028 | ||
3029 | static void finish_ordered_fn(struct btrfs_work *work) | |
3030 | { | |
3031 | struct btrfs_ordered_extent *ordered_extent; | |
3032 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3033 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3034 | } |
3035 | ||
b2950863 | 3036 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3037 | struct extent_state *state, int uptodate) |
3038 | { | |
5fd02043 JB |
3039 | struct inode *inode = page->mapping->host; |
3040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3041 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
3042 | struct btrfs_workqueue *wq; |
3043 | btrfs_work_func_t func; | |
5fd02043 | 3044 | |
1abe9b8a | 3045 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3046 | ||
8b62b72b | 3047 | ClearPagePrivate2(page); |
5fd02043 JB |
3048 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3049 | end - start + 1, uptodate)) | |
3050 | return 0; | |
3051 | ||
9e0af237 LB |
3052 | if (btrfs_is_free_space_inode(inode)) { |
3053 | wq = root->fs_info->endio_freespace_worker; | |
3054 | func = btrfs_freespace_write_helper; | |
3055 | } else { | |
3056 | wq = root->fs_info->endio_write_workers; | |
3057 | func = btrfs_endio_write_helper; | |
3058 | } | |
5fd02043 | 3059 | |
9e0af237 LB |
3060 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3061 | NULL); | |
3062 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
3063 | |
3064 | return 0; | |
211f90e6 CM |
3065 | } |
3066 | ||
dc380aea MX |
3067 | static int __readpage_endio_check(struct inode *inode, |
3068 | struct btrfs_io_bio *io_bio, | |
3069 | int icsum, struct page *page, | |
3070 | int pgoff, u64 start, size_t len) | |
3071 | { | |
3072 | char *kaddr; | |
3073 | u32 csum_expected; | |
3074 | u32 csum = ~(u32)0; | |
dc380aea MX |
3075 | |
3076 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3077 | ||
3078 | kaddr = kmap_atomic(page); | |
3079 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
3080 | btrfs_csum_final(csum, (char *)&csum); | |
3081 | if (csum != csum_expected) | |
3082 | goto zeroit; | |
3083 | ||
3084 | kunmap_atomic(kaddr); | |
3085 | return 0; | |
3086 | zeroit: | |
94647322 DS |
3087 | btrfs_warn_rl(BTRFS_I(inode)->root->fs_info, |
3088 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
dc380aea MX |
3089 | btrfs_ino(inode), start, csum, csum_expected); |
3090 | memset(kaddr + pgoff, 1, len); | |
3091 | flush_dcache_page(page); | |
3092 | kunmap_atomic(kaddr); | |
3093 | if (csum_expected == 0) | |
3094 | return 0; | |
3095 | return -EIO; | |
3096 | } | |
3097 | ||
d352ac68 CM |
3098 | /* |
3099 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3100 | * if there's a match, we allow the bio to finish. If not, the code in |
3101 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3102 | */ |
facc8a22 MX |
3103 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3104 | u64 phy_offset, struct page *page, | |
3105 | u64 start, u64 end, int mirror) | |
07157aac | 3106 | { |
4eee4fa4 | 3107 | size_t offset = start - page_offset(page); |
07157aac | 3108 | struct inode *inode = page->mapping->host; |
d1310b2e | 3109 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3110 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3111 | |
d20f7043 CM |
3112 | if (PageChecked(page)) { |
3113 | ClearPageChecked(page); | |
dc380aea | 3114 | return 0; |
d20f7043 | 3115 | } |
6cbff00f CH |
3116 | |
3117 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3118 | return 0; |
17d217fe YZ |
3119 | |
3120 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3121 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3122 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3123 | return 0; |
17d217fe | 3124 | } |
d20f7043 | 3125 | |
facc8a22 | 3126 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3127 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3128 | start, (size_t)(end - start + 1)); | |
07157aac | 3129 | } |
b888db2b | 3130 | |
24bbcf04 YZ |
3131 | void btrfs_add_delayed_iput(struct inode *inode) |
3132 | { | |
3133 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
8089fe62 | 3134 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3135 | |
3136 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3137 | return; | |
3138 | ||
24bbcf04 | 3139 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3140 | if (binode->delayed_iput_count == 0) { |
3141 | ASSERT(list_empty(&binode->delayed_iput)); | |
3142 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3143 | } else { | |
3144 | binode->delayed_iput_count++; | |
3145 | } | |
24bbcf04 YZ |
3146 | spin_unlock(&fs_info->delayed_iput_lock); |
3147 | } | |
3148 | ||
3149 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3150 | { | |
24bbcf04 | 3151 | struct btrfs_fs_info *fs_info = root->fs_info; |
24bbcf04 | 3152 | |
24bbcf04 | 3153 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3154 | while (!list_empty(&fs_info->delayed_iputs)) { |
3155 | struct btrfs_inode *inode; | |
3156 | ||
3157 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3158 | struct btrfs_inode, delayed_iput); | |
3159 | if (inode->delayed_iput_count) { | |
3160 | inode->delayed_iput_count--; | |
3161 | list_move_tail(&inode->delayed_iput, | |
3162 | &fs_info->delayed_iputs); | |
3163 | } else { | |
3164 | list_del_init(&inode->delayed_iput); | |
3165 | } | |
3166 | spin_unlock(&fs_info->delayed_iput_lock); | |
3167 | iput(&inode->vfs_inode); | |
3168 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3169 | } |
8089fe62 | 3170 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3171 | } |
3172 | ||
d68fc57b | 3173 | /* |
42b2aa86 | 3174 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3175 | * files in the subvolume, it removes orphan item and frees block_rsv |
3176 | * structure. | |
3177 | */ | |
3178 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3179 | struct btrfs_root *root) | |
3180 | { | |
90290e19 | 3181 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3182 | int ret; |
3183 | ||
8a35d95f | 3184 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3185 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3186 | return; | |
3187 | ||
90290e19 | 3188 | spin_lock(&root->orphan_lock); |
8a35d95f | 3189 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3190 | spin_unlock(&root->orphan_lock); |
3191 | return; | |
3192 | } | |
3193 | ||
3194 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3195 | spin_unlock(&root->orphan_lock); | |
3196 | return; | |
3197 | } | |
3198 | ||
3199 | block_rsv = root->orphan_block_rsv; | |
3200 | root->orphan_block_rsv = NULL; | |
3201 | spin_unlock(&root->orphan_lock); | |
3202 | ||
27cdeb70 | 3203 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3204 | btrfs_root_refs(&root->root_item) > 0) { |
3205 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3206 | root->root_key.objectid); | |
4ef31a45 JB |
3207 | if (ret) |
3208 | btrfs_abort_transaction(trans, root, ret); | |
3209 | else | |
27cdeb70 MX |
3210 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3211 | &root->state); | |
d68fc57b YZ |
3212 | } |
3213 | ||
90290e19 JB |
3214 | if (block_rsv) { |
3215 | WARN_ON(block_rsv->size > 0); | |
3216 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3217 | } |
3218 | } | |
3219 | ||
7b128766 JB |
3220 | /* |
3221 | * This creates an orphan entry for the given inode in case something goes | |
3222 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3223 | * |
3224 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3225 | * this function. | |
7b128766 JB |
3226 | */ |
3227 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3228 | { | |
3229 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3230 | struct btrfs_block_rsv *block_rsv = NULL; |
3231 | int reserve = 0; | |
3232 | int insert = 0; | |
3233 | int ret; | |
7b128766 | 3234 | |
d68fc57b | 3235 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3236 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3237 | if (!block_rsv) |
3238 | return -ENOMEM; | |
d68fc57b | 3239 | } |
7b128766 | 3240 | |
d68fc57b YZ |
3241 | spin_lock(&root->orphan_lock); |
3242 | if (!root->orphan_block_rsv) { | |
3243 | root->orphan_block_rsv = block_rsv; | |
3244 | } else if (block_rsv) { | |
3245 | btrfs_free_block_rsv(root, block_rsv); | |
3246 | block_rsv = NULL; | |
7b128766 | 3247 | } |
7b128766 | 3248 | |
8a35d95f JB |
3249 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3250 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3251 | #if 0 |
3252 | /* | |
3253 | * For proper ENOSPC handling, we should do orphan | |
3254 | * cleanup when mounting. But this introduces backward | |
3255 | * compatibility issue. | |
3256 | */ | |
3257 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3258 | insert = 2; | |
3259 | else | |
3260 | insert = 1; | |
3261 | #endif | |
3262 | insert = 1; | |
321f0e70 | 3263 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3264 | } |
3265 | ||
72ac3c0d JB |
3266 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3267 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3268 | reserve = 1; |
d68fc57b | 3269 | spin_unlock(&root->orphan_lock); |
7b128766 | 3270 | |
d68fc57b YZ |
3271 | /* grab metadata reservation from transaction handle */ |
3272 | if (reserve) { | |
3273 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3274 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3275 | } |
7b128766 | 3276 | |
d68fc57b YZ |
3277 | /* insert an orphan item to track this unlinked/truncated file */ |
3278 | if (insert >= 1) { | |
33345d01 | 3279 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3280 | if (ret) { |
703c88e0 | 3281 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3282 | if (reserve) { |
3283 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3284 | &BTRFS_I(inode)->runtime_flags); | |
3285 | btrfs_orphan_release_metadata(inode); | |
3286 | } | |
3287 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3288 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3289 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3290 | btrfs_abort_transaction(trans, root, ret); |
3291 | return ret; | |
3292 | } | |
79787eaa JM |
3293 | } |
3294 | ret = 0; | |
d68fc57b YZ |
3295 | } |
3296 | ||
3297 | /* insert an orphan item to track subvolume contains orphan files */ | |
3298 | if (insert >= 2) { | |
3299 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3300 | root->root_key.objectid); | |
79787eaa JM |
3301 | if (ret && ret != -EEXIST) { |
3302 | btrfs_abort_transaction(trans, root, ret); | |
3303 | return ret; | |
3304 | } | |
d68fc57b YZ |
3305 | } |
3306 | return 0; | |
7b128766 JB |
3307 | } |
3308 | ||
3309 | /* | |
3310 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3311 | * item for this particular inode. | |
3312 | */ | |
48a3b636 ES |
3313 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3314 | struct inode *inode) | |
7b128766 JB |
3315 | { |
3316 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3317 | int delete_item = 0; |
3318 | int release_rsv = 0; | |
7b128766 JB |
3319 | int ret = 0; |
3320 | ||
d68fc57b | 3321 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3322 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3323 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3324 | delete_item = 1; |
7b128766 | 3325 | |
72ac3c0d JB |
3326 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3327 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3328 | release_rsv = 1; |
d68fc57b | 3329 | spin_unlock(&root->orphan_lock); |
7b128766 | 3330 | |
703c88e0 | 3331 | if (delete_item) { |
8a35d95f | 3332 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3333 | if (trans) |
3334 | ret = btrfs_del_orphan_item(trans, root, | |
3335 | btrfs_ino(inode)); | |
8a35d95f | 3336 | } |
7b128766 | 3337 | |
703c88e0 FDBM |
3338 | if (release_rsv) |
3339 | btrfs_orphan_release_metadata(inode); | |
3340 | ||
4ef31a45 | 3341 | return ret; |
7b128766 JB |
3342 | } |
3343 | ||
3344 | /* | |
3345 | * this cleans up any orphans that may be left on the list from the last use | |
3346 | * of this root. | |
3347 | */ | |
66b4ffd1 | 3348 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3349 | { |
3350 | struct btrfs_path *path; | |
3351 | struct extent_buffer *leaf; | |
7b128766 JB |
3352 | struct btrfs_key key, found_key; |
3353 | struct btrfs_trans_handle *trans; | |
3354 | struct inode *inode; | |
8f6d7f4f | 3355 | u64 last_objectid = 0; |
7b128766 JB |
3356 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3357 | ||
d68fc57b | 3358 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3359 | return 0; |
c71bf099 YZ |
3360 | |
3361 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3362 | if (!path) { |
3363 | ret = -ENOMEM; | |
3364 | goto out; | |
3365 | } | |
e4058b54 | 3366 | path->reada = READA_BACK; |
7b128766 JB |
3367 | |
3368 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3369 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3370 | key.offset = (u64)-1; |
3371 | ||
7b128766 JB |
3372 | while (1) { |
3373 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3374 | if (ret < 0) |
3375 | goto out; | |
7b128766 JB |
3376 | |
3377 | /* | |
3378 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3379 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3380 | * find the key and see if we have stuff that matches |
3381 | */ | |
3382 | if (ret > 0) { | |
66b4ffd1 | 3383 | ret = 0; |
7b128766 JB |
3384 | if (path->slots[0] == 0) |
3385 | break; | |
3386 | path->slots[0]--; | |
3387 | } | |
3388 | ||
3389 | /* pull out the item */ | |
3390 | leaf = path->nodes[0]; | |
7b128766 JB |
3391 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3392 | ||
3393 | /* make sure the item matches what we want */ | |
3394 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3395 | break; | |
962a298f | 3396 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3397 | break; |
3398 | ||
3399 | /* release the path since we're done with it */ | |
b3b4aa74 | 3400 | btrfs_release_path(path); |
7b128766 JB |
3401 | |
3402 | /* | |
3403 | * this is where we are basically btrfs_lookup, without the | |
3404 | * crossing root thing. we store the inode number in the | |
3405 | * offset of the orphan item. | |
3406 | */ | |
8f6d7f4f JB |
3407 | |
3408 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3409 | btrfs_err(root->fs_info, |
3410 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3411 | ret = -EINVAL; |
3412 | goto out; | |
3413 | } | |
3414 | ||
3415 | last_objectid = found_key.offset; | |
3416 | ||
5d4f98a2 YZ |
3417 | found_key.objectid = found_key.offset; |
3418 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3419 | found_key.offset = 0; | |
73f73415 | 3420 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3421 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3422 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3423 | goto out; |
7b128766 | 3424 | |
f8e9e0b0 AJ |
3425 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3426 | struct btrfs_root *dead_root; | |
3427 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3428 | int is_dead_root = 0; | |
3429 | ||
3430 | /* | |
3431 | * this is an orphan in the tree root. Currently these | |
3432 | * could come from 2 sources: | |
3433 | * a) a snapshot deletion in progress | |
3434 | * b) a free space cache inode | |
3435 | * We need to distinguish those two, as the snapshot | |
3436 | * orphan must not get deleted. | |
3437 | * find_dead_roots already ran before us, so if this | |
3438 | * is a snapshot deletion, we should find the root | |
3439 | * in the dead_roots list | |
3440 | */ | |
3441 | spin_lock(&fs_info->trans_lock); | |
3442 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3443 | root_list) { | |
3444 | if (dead_root->root_key.objectid == | |
3445 | found_key.objectid) { | |
3446 | is_dead_root = 1; | |
3447 | break; | |
3448 | } | |
3449 | } | |
3450 | spin_unlock(&fs_info->trans_lock); | |
3451 | if (is_dead_root) { | |
3452 | /* prevent this orphan from being found again */ | |
3453 | key.offset = found_key.objectid - 1; | |
3454 | continue; | |
3455 | } | |
3456 | } | |
7b128766 | 3457 | /* |
a8c9e576 JB |
3458 | * Inode is already gone but the orphan item is still there, |
3459 | * kill the orphan item. | |
7b128766 | 3460 | */ |
a8c9e576 JB |
3461 | if (ret == -ESTALE) { |
3462 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3463 | if (IS_ERR(trans)) { |
3464 | ret = PTR_ERR(trans); | |
3465 | goto out; | |
3466 | } | |
c2cf52eb SK |
3467 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3468 | found_key.objectid); | |
a8c9e576 JB |
3469 | ret = btrfs_del_orphan_item(trans, root, |
3470 | found_key.objectid); | |
5b21f2ed | 3471 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3472 | if (ret) |
3473 | goto out; | |
7b128766 JB |
3474 | continue; |
3475 | } | |
3476 | ||
a8c9e576 JB |
3477 | /* |
3478 | * add this inode to the orphan list so btrfs_orphan_del does | |
3479 | * the proper thing when we hit it | |
3480 | */ | |
8a35d95f JB |
3481 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3482 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3483 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3484 | |
7b128766 JB |
3485 | /* if we have links, this was a truncate, lets do that */ |
3486 | if (inode->i_nlink) { | |
fae7f21c | 3487 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3488 | iput(inode); |
3489 | continue; | |
3490 | } | |
7b128766 | 3491 | nr_truncate++; |
f3fe820c JB |
3492 | |
3493 | /* 1 for the orphan item deletion. */ | |
3494 | trans = btrfs_start_transaction(root, 1); | |
3495 | if (IS_ERR(trans)) { | |
c69b26b0 | 3496 | iput(inode); |
f3fe820c JB |
3497 | ret = PTR_ERR(trans); |
3498 | goto out; | |
3499 | } | |
3500 | ret = btrfs_orphan_add(trans, inode); | |
3501 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3502 | if (ret) { |
3503 | iput(inode); | |
f3fe820c | 3504 | goto out; |
c69b26b0 | 3505 | } |
f3fe820c | 3506 | |
66b4ffd1 | 3507 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3508 | if (ret) |
3509 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3510 | } else { |
3511 | nr_unlink++; | |
3512 | } | |
3513 | ||
3514 | /* this will do delete_inode and everything for us */ | |
3515 | iput(inode); | |
66b4ffd1 JB |
3516 | if (ret) |
3517 | goto out; | |
7b128766 | 3518 | } |
3254c876 MX |
3519 | /* release the path since we're done with it */ |
3520 | btrfs_release_path(path); | |
3521 | ||
d68fc57b YZ |
3522 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3523 | ||
3524 | if (root->orphan_block_rsv) | |
3525 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3526 | (u64)-1); | |
3527 | ||
27cdeb70 MX |
3528 | if (root->orphan_block_rsv || |
3529 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3530 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3531 | if (!IS_ERR(trans)) |
3532 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3533 | } |
7b128766 JB |
3534 | |
3535 | if (nr_unlink) | |
4884b476 | 3536 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3537 | if (nr_truncate) |
4884b476 | 3538 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3539 | |
3540 | out: | |
3541 | if (ret) | |
68b663d1 | 3542 | btrfs_err(root->fs_info, |
c2cf52eb | 3543 | "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3544 | btrfs_free_path(path); |
3545 | return ret; | |
7b128766 JB |
3546 | } |
3547 | ||
46a53cca CM |
3548 | /* |
3549 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3550 | * don't find any xattrs, we know there can't be any acls. | |
3551 | * | |
3552 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3553 | */ | |
3554 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3555 | int slot, u64 objectid, |
3556 | int *first_xattr_slot) | |
46a53cca CM |
3557 | { |
3558 | u32 nritems = btrfs_header_nritems(leaf); | |
3559 | struct btrfs_key found_key; | |
f23b5a59 JB |
3560 | static u64 xattr_access = 0; |
3561 | static u64 xattr_default = 0; | |
46a53cca CM |
3562 | int scanned = 0; |
3563 | ||
f23b5a59 | 3564 | if (!xattr_access) { |
97d79299 AG |
3565 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3566 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3567 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3568 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3569 | } |
3570 | ||
46a53cca | 3571 | slot++; |
63541927 | 3572 | *first_xattr_slot = -1; |
46a53cca CM |
3573 | while (slot < nritems) { |
3574 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3575 | ||
3576 | /* we found a different objectid, there must not be acls */ | |
3577 | if (found_key.objectid != objectid) | |
3578 | return 0; | |
3579 | ||
3580 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3581 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3582 | if (*first_xattr_slot == -1) |
3583 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3584 | if (found_key.offset == xattr_access || |
3585 | found_key.offset == xattr_default) | |
3586 | return 1; | |
3587 | } | |
46a53cca CM |
3588 | |
3589 | /* | |
3590 | * we found a key greater than an xattr key, there can't | |
3591 | * be any acls later on | |
3592 | */ | |
3593 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3594 | return 0; | |
3595 | ||
3596 | slot++; | |
3597 | scanned++; | |
3598 | ||
3599 | /* | |
3600 | * it goes inode, inode backrefs, xattrs, extents, | |
3601 | * so if there are a ton of hard links to an inode there can | |
3602 | * be a lot of backrefs. Don't waste time searching too hard, | |
3603 | * this is just an optimization | |
3604 | */ | |
3605 | if (scanned >= 8) | |
3606 | break; | |
3607 | } | |
3608 | /* we hit the end of the leaf before we found an xattr or | |
3609 | * something larger than an xattr. We have to assume the inode | |
3610 | * has acls | |
3611 | */ | |
63541927 FDBM |
3612 | if (*first_xattr_slot == -1) |
3613 | *first_xattr_slot = slot; | |
46a53cca CM |
3614 | return 1; |
3615 | } | |
3616 | ||
d352ac68 CM |
3617 | /* |
3618 | * read an inode from the btree into the in-memory inode | |
3619 | */ | |
5d4f98a2 | 3620 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3621 | { |
3622 | struct btrfs_path *path; | |
5f39d397 | 3623 | struct extent_buffer *leaf; |
39279cc3 CM |
3624 | struct btrfs_inode_item *inode_item; |
3625 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3626 | struct btrfs_key location; | |
67de1176 | 3627 | unsigned long ptr; |
46a53cca | 3628 | int maybe_acls; |
618e21d5 | 3629 | u32 rdev; |
39279cc3 | 3630 | int ret; |
2f7e33d4 | 3631 | bool filled = false; |
63541927 | 3632 | int first_xattr_slot; |
2f7e33d4 MX |
3633 | |
3634 | ret = btrfs_fill_inode(inode, &rdev); | |
3635 | if (!ret) | |
3636 | filled = true; | |
39279cc3 CM |
3637 | |
3638 | path = btrfs_alloc_path(); | |
1748f843 MF |
3639 | if (!path) |
3640 | goto make_bad; | |
3641 | ||
39279cc3 | 3642 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3643 | |
39279cc3 | 3644 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3645 | if (ret) |
39279cc3 | 3646 | goto make_bad; |
39279cc3 | 3647 | |
5f39d397 | 3648 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3649 | |
3650 | if (filled) | |
67de1176 | 3651 | goto cache_index; |
2f7e33d4 | 3652 | |
5f39d397 CM |
3653 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3654 | struct btrfs_inode_item); | |
5f39d397 | 3655 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3656 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3657 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3658 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3659 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3660 | |
a937b979 DS |
3661 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3662 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3663 | |
a937b979 DS |
3664 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3665 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3666 | |
a937b979 DS |
3667 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3668 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3669 | |
9cc97d64 | 3670 | BTRFS_I(inode)->i_otime.tv_sec = |
3671 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3672 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3673 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3674 | |
a76a3cd4 | 3675 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3676 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3677 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3678 | ||
6e17d30b YD |
3679 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3680 | inode->i_generation = BTRFS_I(inode)->generation; | |
3681 | inode->i_rdev = 0; | |
3682 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3683 | ||
3684 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3685 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3686 | ||
3687 | cache_index: | |
5dc562c5 JB |
3688 | /* |
3689 | * If we were modified in the current generation and evicted from memory | |
3690 | * and then re-read we need to do a full sync since we don't have any | |
3691 | * idea about which extents were modified before we were evicted from | |
3692 | * cache. | |
6e17d30b YD |
3693 | * |
3694 | * This is required for both inode re-read from disk and delayed inode | |
3695 | * in delayed_nodes_tree. | |
5dc562c5 JB |
3696 | */ |
3697 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3698 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3699 | &BTRFS_I(inode)->runtime_flags); | |
3700 | ||
bde6c242 FM |
3701 | /* |
3702 | * We don't persist the id of the transaction where an unlink operation | |
3703 | * against the inode was last made. So here we assume the inode might | |
3704 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3705 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3706 | * between the inode and its parent if the inode is fsync'ed and the log | |
3707 | * replayed. For example, in the scenario: | |
3708 | * | |
3709 | * touch mydir/foo | |
3710 | * ln mydir/foo mydir/bar | |
3711 | * sync | |
3712 | * unlink mydir/bar | |
3713 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3714 | * xfs_io -c fsync mydir/foo | |
3715 | * <power failure> | |
3716 | * mount fs, triggers fsync log replay | |
3717 | * | |
3718 | * We must make sure that when we fsync our inode foo we also log its | |
3719 | * parent inode, otherwise after log replay the parent still has the | |
3720 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3721 | * and doesn't have an inode ref with the name "bar" anymore. | |
3722 | * | |
3723 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3724 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3725 | * transaction commits on fsync if our inode is a directory, or if our |
3726 | * inode is not a directory, logging its parent unnecessarily. | |
3727 | */ | |
3728 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3729 | ||
67de1176 MX |
3730 | path->slots[0]++; |
3731 | if (inode->i_nlink != 1 || | |
3732 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3733 | goto cache_acl; | |
3734 | ||
3735 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3736 | if (location.objectid != btrfs_ino(inode)) | |
3737 | goto cache_acl; | |
3738 | ||
3739 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3740 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3741 | struct btrfs_inode_ref *ref; | |
3742 | ||
3743 | ref = (struct btrfs_inode_ref *)ptr; | |
3744 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3745 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3746 | struct btrfs_inode_extref *extref; | |
3747 | ||
3748 | extref = (struct btrfs_inode_extref *)ptr; | |
3749 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3750 | extref); | |
3751 | } | |
2f7e33d4 | 3752 | cache_acl: |
46a53cca CM |
3753 | /* |
3754 | * try to precache a NULL acl entry for files that don't have | |
3755 | * any xattrs or acls | |
3756 | */ | |
33345d01 | 3757 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3758 | btrfs_ino(inode), &first_xattr_slot); |
3759 | if (first_xattr_slot != -1) { | |
3760 | path->slots[0] = first_xattr_slot; | |
3761 | ret = btrfs_load_inode_props(inode, path); | |
3762 | if (ret) | |
3763 | btrfs_err(root->fs_info, | |
351fd353 | 3764 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3765 | btrfs_ino(inode), |
3766 | root->root_key.objectid, ret); | |
3767 | } | |
3768 | btrfs_free_path(path); | |
3769 | ||
72c04902 AV |
3770 | if (!maybe_acls) |
3771 | cache_no_acl(inode); | |
46a53cca | 3772 | |
39279cc3 | 3773 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3774 | case S_IFREG: |
3775 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3776 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3777 | inode->i_fop = &btrfs_file_operations; |
3778 | inode->i_op = &btrfs_file_inode_operations; | |
3779 | break; | |
3780 | case S_IFDIR: | |
3781 | inode->i_fop = &btrfs_dir_file_operations; | |
3782 | if (root == root->fs_info->tree_root) | |
3783 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3784 | else | |
3785 | inode->i_op = &btrfs_dir_inode_operations; | |
3786 | break; | |
3787 | case S_IFLNK: | |
3788 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3789 | inode_nohighmem(inode); |
39279cc3 CM |
3790 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3791 | break; | |
618e21d5 | 3792 | default: |
0279b4cd | 3793 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3794 | init_special_inode(inode, inode->i_mode, rdev); |
3795 | break; | |
39279cc3 | 3796 | } |
6cbff00f CH |
3797 | |
3798 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3799 | return; |
3800 | ||
3801 | make_bad: | |
39279cc3 | 3802 | btrfs_free_path(path); |
39279cc3 CM |
3803 | make_bad_inode(inode); |
3804 | } | |
3805 | ||
d352ac68 CM |
3806 | /* |
3807 | * given a leaf and an inode, copy the inode fields into the leaf | |
3808 | */ | |
e02119d5 CM |
3809 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3810 | struct extent_buffer *leaf, | |
5f39d397 | 3811 | struct btrfs_inode_item *item, |
39279cc3 CM |
3812 | struct inode *inode) |
3813 | { | |
51fab693 LB |
3814 | struct btrfs_map_token token; |
3815 | ||
3816 | btrfs_init_map_token(&token); | |
5f39d397 | 3817 | |
51fab693 LB |
3818 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3819 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3820 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3821 | &token); | |
3822 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3823 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3824 | |
a937b979 | 3825 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3826 | inode->i_atime.tv_sec, &token); |
a937b979 | 3827 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3828 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3829 | |
a937b979 | 3830 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3831 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3832 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3833 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3834 | |
a937b979 | 3835 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3836 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3837 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3838 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3839 | |
9cc97d64 | 3840 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3841 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3842 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3843 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3844 | ||
51fab693 LB |
3845 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3846 | &token); | |
3847 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3848 | &token); | |
3849 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3850 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3851 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3852 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3853 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3854 | } |
3855 | ||
d352ac68 CM |
3856 | /* |
3857 | * copy everything in the in-memory inode into the btree. | |
3858 | */ | |
2115133f | 3859 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3860 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3861 | { |
3862 | struct btrfs_inode_item *inode_item; | |
3863 | struct btrfs_path *path; | |
5f39d397 | 3864 | struct extent_buffer *leaf; |
39279cc3 CM |
3865 | int ret; |
3866 | ||
3867 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3868 | if (!path) |
3869 | return -ENOMEM; | |
3870 | ||
b9473439 | 3871 | path->leave_spinning = 1; |
16cdcec7 MX |
3872 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3873 | 1); | |
39279cc3 CM |
3874 | if (ret) { |
3875 | if (ret > 0) | |
3876 | ret = -ENOENT; | |
3877 | goto failed; | |
3878 | } | |
3879 | ||
5f39d397 CM |
3880 | leaf = path->nodes[0]; |
3881 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3882 | struct btrfs_inode_item); |
39279cc3 | 3883 | |
e02119d5 | 3884 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3885 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3886 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3887 | ret = 0; |
3888 | failed: | |
39279cc3 CM |
3889 | btrfs_free_path(path); |
3890 | return ret; | |
3891 | } | |
3892 | ||
2115133f CM |
3893 | /* |
3894 | * copy everything in the in-memory inode into the btree. | |
3895 | */ | |
3896 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3897 | struct btrfs_root *root, struct inode *inode) | |
3898 | { | |
3899 | int ret; | |
3900 | ||
3901 | /* | |
3902 | * If the inode is a free space inode, we can deadlock during commit | |
3903 | * if we put it into the delayed code. | |
3904 | * | |
3905 | * The data relocation inode should also be directly updated | |
3906 | * without delay | |
3907 | */ | |
83eea1f1 | 3908 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3909 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3910 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3911 | btrfs_update_root_times(trans, root); |
3912 | ||
2115133f CM |
3913 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3914 | if (!ret) | |
3915 | btrfs_set_inode_last_trans(trans, inode); | |
3916 | return ret; | |
3917 | } | |
3918 | ||
3919 | return btrfs_update_inode_item(trans, root, inode); | |
3920 | } | |
3921 | ||
be6aef60 JB |
3922 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3923 | struct btrfs_root *root, | |
3924 | struct inode *inode) | |
2115133f CM |
3925 | { |
3926 | int ret; | |
3927 | ||
3928 | ret = btrfs_update_inode(trans, root, inode); | |
3929 | if (ret == -ENOSPC) | |
3930 | return btrfs_update_inode_item(trans, root, inode); | |
3931 | return ret; | |
3932 | } | |
3933 | ||
d352ac68 CM |
3934 | /* |
3935 | * unlink helper that gets used here in inode.c and in the tree logging | |
3936 | * recovery code. It remove a link in a directory with a given name, and | |
3937 | * also drops the back refs in the inode to the directory | |
3938 | */ | |
92986796 AV |
3939 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3940 | struct btrfs_root *root, | |
3941 | struct inode *dir, struct inode *inode, | |
3942 | const char *name, int name_len) | |
39279cc3 CM |
3943 | { |
3944 | struct btrfs_path *path; | |
39279cc3 | 3945 | int ret = 0; |
5f39d397 | 3946 | struct extent_buffer *leaf; |
39279cc3 | 3947 | struct btrfs_dir_item *di; |
5f39d397 | 3948 | struct btrfs_key key; |
aec7477b | 3949 | u64 index; |
33345d01 LZ |
3950 | u64 ino = btrfs_ino(inode); |
3951 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3952 | |
3953 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3954 | if (!path) { |
3955 | ret = -ENOMEM; | |
554233a6 | 3956 | goto out; |
54aa1f4d CM |
3957 | } |
3958 | ||
b9473439 | 3959 | path->leave_spinning = 1; |
33345d01 | 3960 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3961 | name, name_len, -1); |
3962 | if (IS_ERR(di)) { | |
3963 | ret = PTR_ERR(di); | |
3964 | goto err; | |
3965 | } | |
3966 | if (!di) { | |
3967 | ret = -ENOENT; | |
3968 | goto err; | |
3969 | } | |
5f39d397 CM |
3970 | leaf = path->nodes[0]; |
3971 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3972 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3973 | if (ret) |
3974 | goto err; | |
b3b4aa74 | 3975 | btrfs_release_path(path); |
39279cc3 | 3976 | |
67de1176 MX |
3977 | /* |
3978 | * If we don't have dir index, we have to get it by looking up | |
3979 | * the inode ref, since we get the inode ref, remove it directly, | |
3980 | * it is unnecessary to do delayed deletion. | |
3981 | * | |
3982 | * But if we have dir index, needn't search inode ref to get it. | |
3983 | * Since the inode ref is close to the inode item, it is better | |
3984 | * that we delay to delete it, and just do this deletion when | |
3985 | * we update the inode item. | |
3986 | */ | |
3987 | if (BTRFS_I(inode)->dir_index) { | |
3988 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3989 | if (!ret) { | |
3990 | index = BTRFS_I(inode)->dir_index; | |
3991 | goto skip_backref; | |
3992 | } | |
3993 | } | |
3994 | ||
33345d01 LZ |
3995 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3996 | dir_ino, &index); | |
aec7477b | 3997 | if (ret) { |
c2cf52eb SK |
3998 | btrfs_info(root->fs_info, |
3999 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 4000 | name_len, name, ino, dir_ino); |
79787eaa | 4001 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
4002 | goto err; |
4003 | } | |
67de1176 | 4004 | skip_backref: |
16cdcec7 | 4005 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4006 | if (ret) { |
4007 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4008 | goto err; |
79787eaa | 4009 | } |
39279cc3 | 4010 | |
e02119d5 | 4011 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 4012 | inode, dir_ino); |
79787eaa JM |
4013 | if (ret != 0 && ret != -ENOENT) { |
4014 | btrfs_abort_transaction(trans, root, ret); | |
4015 | goto err; | |
4016 | } | |
e02119d5 CM |
4017 | |
4018 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
4019 | dir, index); | |
6418c961 CM |
4020 | if (ret == -ENOENT) |
4021 | ret = 0; | |
d4e3991b ZB |
4022 | else if (ret) |
4023 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
4024 | err: |
4025 | btrfs_free_path(path); | |
e02119d5 CM |
4026 | if (ret) |
4027 | goto out; | |
4028 | ||
4029 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
4030 | inode_inc_iversion(inode); |
4031 | inode_inc_iversion(dir); | |
04b285f3 DD |
4032 | inode->i_ctime = dir->i_mtime = |
4033 | dir->i_ctime = current_fs_time(inode->i_sb); | |
b9959295 | 4034 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4035 | out: |
39279cc3 CM |
4036 | return ret; |
4037 | } | |
4038 | ||
92986796 AV |
4039 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4040 | struct btrfs_root *root, | |
4041 | struct inode *dir, struct inode *inode, | |
4042 | const char *name, int name_len) | |
4043 | { | |
4044 | int ret; | |
4045 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4046 | if (!ret) { | |
8b558c5f | 4047 | drop_nlink(inode); |
92986796 AV |
4048 | ret = btrfs_update_inode(trans, root, inode); |
4049 | } | |
4050 | return ret; | |
4051 | } | |
39279cc3 | 4052 | |
a22285a6 YZ |
4053 | /* |
4054 | * helper to start transaction for unlink and rmdir. | |
4055 | * | |
d52be818 JB |
4056 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4057 | * if we cannot make our reservations the normal way try and see if there is | |
4058 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4059 | * allow the unlink to occur. | |
a22285a6 | 4060 | */ |
d52be818 | 4061 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4062 | { |
a22285a6 | 4063 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4064 | |
e70bea5f JB |
4065 | /* |
4066 | * 1 for the possible orphan item | |
4067 | * 1 for the dir item | |
4068 | * 1 for the dir index | |
4069 | * 1 for the inode ref | |
e70bea5f JB |
4070 | * 1 for the inode |
4071 | */ | |
8eab77ff | 4072 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4073 | } |
4074 | ||
4075 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4076 | { | |
4077 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4078 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4079 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4080 | int ret; |
a22285a6 | 4081 | |
d52be818 | 4082 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4083 | if (IS_ERR(trans)) |
4084 | return PTR_ERR(trans); | |
5f39d397 | 4085 | |
2b0143b5 | 4086 | btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0); |
12fcfd22 | 4087 | |
2b0143b5 | 4088 | ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4089 | dentry->d_name.name, dentry->d_name.len); |
b532402e TI |
4090 | if (ret) |
4091 | goto out; | |
7b128766 | 4092 | |
a22285a6 | 4093 | if (inode->i_nlink == 0) { |
7b128766 | 4094 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4095 | if (ret) |
4096 | goto out; | |
a22285a6 | 4097 | } |
7b128766 | 4098 | |
b532402e | 4099 | out: |
d52be818 | 4100 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4101 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
4102 | return ret; |
4103 | } | |
4104 | ||
4df27c4d YZ |
4105 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4106 | struct btrfs_root *root, | |
4107 | struct inode *dir, u64 objectid, | |
4108 | const char *name, int name_len) | |
4109 | { | |
4110 | struct btrfs_path *path; | |
4111 | struct extent_buffer *leaf; | |
4112 | struct btrfs_dir_item *di; | |
4113 | struct btrfs_key key; | |
4114 | u64 index; | |
4115 | int ret; | |
33345d01 | 4116 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
4117 | |
4118 | path = btrfs_alloc_path(); | |
4119 | if (!path) | |
4120 | return -ENOMEM; | |
4121 | ||
33345d01 | 4122 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4123 | name, name_len, -1); |
79787eaa JM |
4124 | if (IS_ERR_OR_NULL(di)) { |
4125 | if (!di) | |
4126 | ret = -ENOENT; | |
4127 | else | |
4128 | ret = PTR_ERR(di); | |
4129 | goto out; | |
4130 | } | |
4df27c4d YZ |
4131 | |
4132 | leaf = path->nodes[0]; | |
4133 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4134 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4135 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
4136 | if (ret) { |
4137 | btrfs_abort_transaction(trans, root, ret); | |
4138 | goto out; | |
4139 | } | |
b3b4aa74 | 4140 | btrfs_release_path(path); |
4df27c4d YZ |
4141 | |
4142 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4143 | objectid, root->root_key.objectid, | |
33345d01 | 4144 | dir_ino, &index, name, name_len); |
4df27c4d | 4145 | if (ret < 0) { |
79787eaa JM |
4146 | if (ret != -ENOENT) { |
4147 | btrfs_abort_transaction(trans, root, ret); | |
4148 | goto out; | |
4149 | } | |
33345d01 | 4150 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4151 | name, name_len); |
79787eaa JM |
4152 | if (IS_ERR_OR_NULL(di)) { |
4153 | if (!di) | |
4154 | ret = -ENOENT; | |
4155 | else | |
4156 | ret = PTR_ERR(di); | |
4157 | btrfs_abort_transaction(trans, root, ret); | |
4158 | goto out; | |
4159 | } | |
4df27c4d YZ |
4160 | |
4161 | leaf = path->nodes[0]; | |
4162 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4163 | btrfs_release_path(path); |
4df27c4d YZ |
4164 | index = key.offset; |
4165 | } | |
945d8962 | 4166 | btrfs_release_path(path); |
4df27c4d | 4167 | |
16cdcec7 | 4168 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4169 | if (ret) { |
4170 | btrfs_abort_transaction(trans, root, ret); | |
4171 | goto out; | |
4172 | } | |
4df27c4d YZ |
4173 | |
4174 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4175 | inode_inc_iversion(dir); |
04b285f3 | 4176 | dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); |
5a24e84c | 4177 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4178 | if (ret) |
4179 | btrfs_abort_transaction(trans, root, ret); | |
4180 | out: | |
71d7aed0 | 4181 | btrfs_free_path(path); |
79787eaa | 4182 | return ret; |
4df27c4d YZ |
4183 | } |
4184 | ||
39279cc3 CM |
4185 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4186 | { | |
2b0143b5 | 4187 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4188 | int err = 0; |
39279cc3 | 4189 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4190 | struct btrfs_trans_handle *trans; |
39279cc3 | 4191 | |
b3ae244e | 4192 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4193 | return -ENOTEMPTY; |
b3ae244e DS |
4194 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4195 | return -EPERM; | |
134d4512 | 4196 | |
d52be818 | 4197 | trans = __unlink_start_trans(dir); |
a22285a6 | 4198 | if (IS_ERR(trans)) |
5df6a9f6 | 4199 | return PTR_ERR(trans); |
5df6a9f6 | 4200 | |
33345d01 | 4201 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4202 | err = btrfs_unlink_subvol(trans, root, dir, |
4203 | BTRFS_I(inode)->location.objectid, | |
4204 | dentry->d_name.name, | |
4205 | dentry->d_name.len); | |
4206 | goto out; | |
4207 | } | |
4208 | ||
7b128766 JB |
4209 | err = btrfs_orphan_add(trans, inode); |
4210 | if (err) | |
4df27c4d | 4211 | goto out; |
7b128766 | 4212 | |
39279cc3 | 4213 | /* now the directory is empty */ |
2b0143b5 | 4214 | err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4215 | dentry->d_name.name, dentry->d_name.len); |
d397712b | 4216 | if (!err) |
dbe674a9 | 4217 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4218 | out: |
d52be818 | 4219 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4220 | btrfs_btree_balance_dirty(root); |
3954401f | 4221 | |
39279cc3 CM |
4222 | return err; |
4223 | } | |
4224 | ||
28f75a0e CM |
4225 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4226 | struct btrfs_root *root, | |
4227 | u64 bytes_deleted) | |
4228 | { | |
4229 | int ret; | |
4230 | ||
dc95f7bf JB |
4231 | /* |
4232 | * This is only used to apply pressure to the enospc system, we don't | |
4233 | * intend to use this reservation at all. | |
4234 | */ | |
28f75a0e | 4235 | bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted); |
dc95f7bf | 4236 | bytes_deleted *= root->nodesize; |
28f75a0e CM |
4237 | ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv, |
4238 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); | |
dc95f7bf JB |
4239 | if (!ret) { |
4240 | trace_btrfs_space_reservation(root->fs_info, "transaction", | |
4241 | trans->transid, | |
4242 | bytes_deleted, 1); | |
28f75a0e | 4243 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4244 | } |
28f75a0e CM |
4245 | return ret; |
4246 | ||
4247 | } | |
4248 | ||
0305cd5f FM |
4249 | static int truncate_inline_extent(struct inode *inode, |
4250 | struct btrfs_path *path, | |
4251 | struct btrfs_key *found_key, | |
4252 | const u64 item_end, | |
4253 | const u64 new_size) | |
4254 | { | |
4255 | struct extent_buffer *leaf = path->nodes[0]; | |
4256 | int slot = path->slots[0]; | |
4257 | struct btrfs_file_extent_item *fi; | |
4258 | u32 size = (u32)(new_size - found_key->offset); | |
4259 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4260 | ||
4261 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4262 | ||
4263 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4264 | loff_t offset = new_size; | |
09cbfeaf | 4265 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4266 | |
4267 | /* | |
4268 | * Zero out the remaining of the last page of our inline extent, | |
4269 | * instead of directly truncating our inline extent here - that | |
4270 | * would be much more complex (decompressing all the data, then | |
4271 | * compressing the truncated data, which might be bigger than | |
4272 | * the size of the inline extent, resize the extent, etc). | |
4273 | * We release the path because to get the page we might need to | |
4274 | * read the extent item from disk (data not in the page cache). | |
4275 | */ | |
4276 | btrfs_release_path(path); | |
9703fefe CR |
4277 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4278 | 0); | |
0305cd5f FM |
4279 | } |
4280 | ||
4281 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4282 | size = btrfs_file_extent_calc_inline_size(size); | |
4283 | btrfs_truncate_item(root, path, size, 1); | |
4284 | ||
4285 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4286 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4287 | ||
4288 | return 0; | |
4289 | } | |
4290 | ||
39279cc3 CM |
4291 | /* |
4292 | * this can truncate away extent items, csum items and directory items. | |
4293 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4294 | * any higher than new_size |
39279cc3 CM |
4295 | * |
4296 | * csum items that cross the new i_size are truncated to the new size | |
4297 | * as well. | |
7b128766 JB |
4298 | * |
4299 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4300 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4301 | */ |
8082510e YZ |
4302 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4303 | struct btrfs_root *root, | |
4304 | struct inode *inode, | |
4305 | u64 new_size, u32 min_type) | |
39279cc3 | 4306 | { |
39279cc3 | 4307 | struct btrfs_path *path; |
5f39d397 | 4308 | struct extent_buffer *leaf; |
39279cc3 | 4309 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4310 | struct btrfs_key key; |
4311 | struct btrfs_key found_key; | |
39279cc3 | 4312 | u64 extent_start = 0; |
db94535d | 4313 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4314 | u64 extent_offset = 0; |
39279cc3 | 4315 | u64 item_end = 0; |
c1aa4575 | 4316 | u64 last_size = new_size; |
8082510e | 4317 | u32 found_type = (u8)-1; |
39279cc3 CM |
4318 | int found_extent; |
4319 | int del_item; | |
85e21bac CM |
4320 | int pending_del_nr = 0; |
4321 | int pending_del_slot = 0; | |
179e29e4 | 4322 | int extent_type = -1; |
8082510e YZ |
4323 | int ret; |
4324 | int err = 0; | |
33345d01 | 4325 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4326 | u64 bytes_deleted = 0; |
1262133b JB |
4327 | bool be_nice = 0; |
4328 | bool should_throttle = 0; | |
28f75a0e | 4329 | bool should_end = 0; |
8082510e YZ |
4330 | |
4331 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4332 | |
28ed1345 CM |
4333 | /* |
4334 | * for non-free space inodes and ref cows, we want to back off from | |
4335 | * time to time | |
4336 | */ | |
4337 | if (!btrfs_is_free_space_inode(inode) && | |
4338 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4339 | be_nice = 1; | |
4340 | ||
0eb0e19c MF |
4341 | path = btrfs_alloc_path(); |
4342 | if (!path) | |
4343 | return -ENOMEM; | |
e4058b54 | 4344 | path->reada = READA_BACK; |
0eb0e19c | 4345 | |
5dc562c5 JB |
4346 | /* |
4347 | * We want to drop from the next block forward in case this new size is | |
4348 | * not block aligned since we will be keeping the last block of the | |
4349 | * extent just the way it is. | |
4350 | */ | |
27cdeb70 MX |
4351 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4352 | root == root->fs_info->tree_root) | |
fda2832f QW |
4353 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4354 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4355 | |
16cdcec7 MX |
4356 | /* |
4357 | * This function is also used to drop the items in the log tree before | |
4358 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4359 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4360 | * items. | |
4361 | */ | |
4362 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4363 | btrfs_kill_delayed_inode_items(inode); | |
4364 | ||
33345d01 | 4365 | key.objectid = ino; |
39279cc3 | 4366 | key.offset = (u64)-1; |
5f39d397 CM |
4367 | key.type = (u8)-1; |
4368 | ||
85e21bac | 4369 | search_again: |
28ed1345 CM |
4370 | /* |
4371 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4372 | * up a huge file in a single leaf. Most of the time that | |
4373 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4374 | */ | |
ee22184b | 4375 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4376 | if (btrfs_should_end_transaction(trans, root)) { |
4377 | err = -EAGAIN; | |
4378 | goto error; | |
4379 | } | |
4380 | } | |
4381 | ||
4382 | ||
b9473439 | 4383 | path->leave_spinning = 1; |
85e21bac | 4384 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4385 | if (ret < 0) { |
4386 | err = ret; | |
4387 | goto out; | |
4388 | } | |
d397712b | 4389 | |
85e21bac | 4390 | if (ret > 0) { |
e02119d5 CM |
4391 | /* there are no items in the tree for us to truncate, we're |
4392 | * done | |
4393 | */ | |
8082510e YZ |
4394 | if (path->slots[0] == 0) |
4395 | goto out; | |
85e21bac CM |
4396 | path->slots[0]--; |
4397 | } | |
4398 | ||
d397712b | 4399 | while (1) { |
39279cc3 | 4400 | fi = NULL; |
5f39d397 CM |
4401 | leaf = path->nodes[0]; |
4402 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4403 | found_type = found_key.type; |
39279cc3 | 4404 | |
33345d01 | 4405 | if (found_key.objectid != ino) |
39279cc3 | 4406 | break; |
5f39d397 | 4407 | |
85e21bac | 4408 | if (found_type < min_type) |
39279cc3 CM |
4409 | break; |
4410 | ||
5f39d397 | 4411 | item_end = found_key.offset; |
39279cc3 | 4412 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4413 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4414 | struct btrfs_file_extent_item); |
179e29e4 CM |
4415 | extent_type = btrfs_file_extent_type(leaf, fi); |
4416 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4417 | item_end += |
db94535d | 4418 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4419 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4420 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4421 | path->slots[0], fi); |
39279cc3 | 4422 | } |
008630c1 | 4423 | item_end--; |
39279cc3 | 4424 | } |
8082510e YZ |
4425 | if (found_type > min_type) { |
4426 | del_item = 1; | |
4427 | } else { | |
4428 | if (item_end < new_size) | |
b888db2b | 4429 | break; |
8082510e YZ |
4430 | if (found_key.offset >= new_size) |
4431 | del_item = 1; | |
4432 | else | |
4433 | del_item = 0; | |
39279cc3 | 4434 | } |
39279cc3 | 4435 | found_extent = 0; |
39279cc3 | 4436 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4437 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4438 | goto delete; | |
4439 | ||
7f4f6e0a JB |
4440 | if (del_item) |
4441 | last_size = found_key.offset; | |
4442 | else | |
4443 | last_size = new_size; | |
4444 | ||
179e29e4 | 4445 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4446 | u64 num_dec; |
db94535d | 4447 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4448 | if (!del_item) { |
db94535d CM |
4449 | u64 orig_num_bytes = |
4450 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4451 | extent_num_bytes = ALIGN(new_size - |
4452 | found_key.offset, | |
4453 | root->sectorsize); | |
db94535d CM |
4454 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4455 | extent_num_bytes); | |
4456 | num_dec = (orig_num_bytes - | |
9069218d | 4457 | extent_num_bytes); |
27cdeb70 MX |
4458 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4459 | &root->state) && | |
4460 | extent_start != 0) | |
a76a3cd4 | 4461 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4462 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4463 | } else { |
db94535d CM |
4464 | extent_num_bytes = |
4465 | btrfs_file_extent_disk_num_bytes(leaf, | |
4466 | fi); | |
5d4f98a2 YZ |
4467 | extent_offset = found_key.offset - |
4468 | btrfs_file_extent_offset(leaf, fi); | |
4469 | ||
39279cc3 | 4470 | /* FIXME blocksize != 4096 */ |
9069218d | 4471 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4472 | if (extent_start != 0) { |
4473 | found_extent = 1; | |
27cdeb70 MX |
4474 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4475 | &root->state)) | |
a76a3cd4 | 4476 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4477 | } |
39279cc3 | 4478 | } |
9069218d | 4479 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4480 | /* |
4481 | * we can't truncate inline items that have had | |
4482 | * special encodings | |
4483 | */ | |
4484 | if (!del_item && | |
c8b97818 CM |
4485 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4486 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4487 | |
4488 | /* | |
0305cd5f FM |
4489 | * Need to release path in order to truncate a |
4490 | * compressed extent. So delete any accumulated | |
4491 | * extent items so far. | |
514ac8ad | 4492 | */ |
0305cd5f FM |
4493 | if (btrfs_file_extent_compression(leaf, fi) != |
4494 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4495 | err = btrfs_del_items(trans, root, path, | |
4496 | pending_del_slot, | |
4497 | pending_del_nr); | |
4498 | if (err) { | |
4499 | btrfs_abort_transaction(trans, | |
4500 | root, | |
4501 | err); | |
4502 | goto error; | |
4503 | } | |
4504 | pending_del_nr = 0; | |
4505 | } | |
4506 | ||
4507 | err = truncate_inline_extent(inode, path, | |
4508 | &found_key, | |
4509 | item_end, | |
4510 | new_size); | |
4511 | if (err) { | |
4512 | btrfs_abort_transaction(trans, | |
4513 | root, err); | |
4514 | goto error; | |
4515 | } | |
27cdeb70 MX |
4516 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4517 | &root->state)) { | |
0305cd5f | 4518 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4519 | } |
39279cc3 | 4520 | } |
179e29e4 | 4521 | delete: |
39279cc3 | 4522 | if (del_item) { |
85e21bac CM |
4523 | if (!pending_del_nr) { |
4524 | /* no pending yet, add ourselves */ | |
4525 | pending_del_slot = path->slots[0]; | |
4526 | pending_del_nr = 1; | |
4527 | } else if (pending_del_nr && | |
4528 | path->slots[0] + 1 == pending_del_slot) { | |
4529 | /* hop on the pending chunk */ | |
4530 | pending_del_nr++; | |
4531 | pending_del_slot = path->slots[0]; | |
4532 | } else { | |
d397712b | 4533 | BUG(); |
85e21bac | 4534 | } |
39279cc3 CM |
4535 | } else { |
4536 | break; | |
4537 | } | |
28f75a0e CM |
4538 | should_throttle = 0; |
4539 | ||
27cdeb70 MX |
4540 | if (found_extent && |
4541 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4542 | root == root->fs_info->tree_root)) { | |
b9473439 | 4543 | btrfs_set_path_blocking(path); |
28ed1345 | 4544 | bytes_deleted += extent_num_bytes; |
39279cc3 | 4545 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4546 | extent_num_bytes, 0, |
4547 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4548 | ino, extent_offset); |
39279cc3 | 4549 | BUG_ON(ret); |
1262133b | 4550 | if (btrfs_should_throttle_delayed_refs(trans, root)) |
28ed1345 CM |
4551 | btrfs_async_run_delayed_refs(root, |
4552 | trans->delayed_ref_updates * 2, 0); | |
28f75a0e CM |
4553 | if (be_nice) { |
4554 | if (truncate_space_check(trans, root, | |
4555 | extent_num_bytes)) { | |
4556 | should_end = 1; | |
4557 | } | |
4558 | if (btrfs_should_throttle_delayed_refs(trans, | |
4559 | root)) { | |
4560 | should_throttle = 1; | |
4561 | } | |
4562 | } | |
39279cc3 | 4563 | } |
85e21bac | 4564 | |
8082510e YZ |
4565 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4566 | break; | |
4567 | ||
4568 | if (path->slots[0] == 0 || | |
1262133b | 4569 | path->slots[0] != pending_del_slot || |
28f75a0e | 4570 | should_throttle || should_end) { |
8082510e YZ |
4571 | if (pending_del_nr) { |
4572 | ret = btrfs_del_items(trans, root, path, | |
4573 | pending_del_slot, | |
4574 | pending_del_nr); | |
79787eaa JM |
4575 | if (ret) { |
4576 | btrfs_abort_transaction(trans, | |
4577 | root, ret); | |
4578 | goto error; | |
4579 | } | |
8082510e YZ |
4580 | pending_del_nr = 0; |
4581 | } | |
b3b4aa74 | 4582 | btrfs_release_path(path); |
28f75a0e | 4583 | if (should_throttle) { |
1262133b JB |
4584 | unsigned long updates = trans->delayed_ref_updates; |
4585 | if (updates) { | |
4586 | trans->delayed_ref_updates = 0; | |
4587 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4588 | if (ret && !err) | |
4589 | err = ret; | |
4590 | } | |
4591 | } | |
28f75a0e CM |
4592 | /* |
4593 | * if we failed to refill our space rsv, bail out | |
4594 | * and let the transaction restart | |
4595 | */ | |
4596 | if (should_end) { | |
4597 | err = -EAGAIN; | |
4598 | goto error; | |
4599 | } | |
85e21bac | 4600 | goto search_again; |
8082510e YZ |
4601 | } else { |
4602 | path->slots[0]--; | |
85e21bac | 4603 | } |
39279cc3 | 4604 | } |
8082510e | 4605 | out: |
85e21bac CM |
4606 | if (pending_del_nr) { |
4607 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4608 | pending_del_nr); | |
79787eaa JM |
4609 | if (ret) |
4610 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4611 | } |
79787eaa | 4612 | error: |
c1aa4575 | 4613 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7f4f6e0a | 4614 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
28ed1345 | 4615 | |
39279cc3 | 4616 | btrfs_free_path(path); |
28ed1345 | 4617 | |
ee22184b | 4618 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4619 | unsigned long updates = trans->delayed_ref_updates; |
4620 | if (updates) { | |
4621 | trans->delayed_ref_updates = 0; | |
4622 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4623 | if (ret && !err) | |
4624 | err = ret; | |
4625 | } | |
4626 | } | |
8082510e | 4627 | return err; |
39279cc3 CM |
4628 | } |
4629 | ||
4630 | /* | |
9703fefe | 4631 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4632 | * @inode - inode that we're zeroing |
4633 | * @from - the offset to start zeroing | |
4634 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4635 | * offset | |
4636 | * @front - zero up to the offset instead of from the offset on | |
4637 | * | |
9703fefe | 4638 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4639 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4640 | */ |
9703fefe | 4641 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4642 | int front) |
39279cc3 | 4643 | { |
2aaa6655 | 4644 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4645 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4646 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4647 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4648 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4649 | char *kaddr; |
db94535d | 4650 | u32 blocksize = root->sectorsize; |
09cbfeaf | 4651 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4652 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4653 | struct page *page; |
3b16a4e3 | 4654 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4655 | int ret = 0; |
9703fefe CR |
4656 | u64 block_start; |
4657 | u64 block_end; | |
39279cc3 | 4658 | |
2aaa6655 JB |
4659 | if ((offset & (blocksize - 1)) == 0 && |
4660 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4661 | goto out; |
9703fefe | 4662 | |
7cf5b976 | 4663 | ret = btrfs_delalloc_reserve_space(inode, |
9703fefe | 4664 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4665 | if (ret) |
4666 | goto out; | |
39279cc3 | 4667 | |
211c17f5 | 4668 | again: |
3b16a4e3 | 4669 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4670 | if (!page) { |
7cf5b976 | 4671 | btrfs_delalloc_release_space(inode, |
9703fefe CR |
4672 | round_down(from, blocksize), |
4673 | blocksize); | |
ac6a2b36 | 4674 | ret = -ENOMEM; |
39279cc3 | 4675 | goto out; |
5d5e103a | 4676 | } |
e6dcd2dc | 4677 | |
9703fefe CR |
4678 | block_start = round_down(from, blocksize); |
4679 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4680 | |
39279cc3 | 4681 | if (!PageUptodate(page)) { |
9ebefb18 | 4682 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4683 | lock_page(page); |
211c17f5 CM |
4684 | if (page->mapping != mapping) { |
4685 | unlock_page(page); | |
09cbfeaf | 4686 | put_page(page); |
211c17f5 CM |
4687 | goto again; |
4688 | } | |
39279cc3 CM |
4689 | if (!PageUptodate(page)) { |
4690 | ret = -EIO; | |
89642229 | 4691 | goto out_unlock; |
39279cc3 CM |
4692 | } |
4693 | } | |
211c17f5 | 4694 | wait_on_page_writeback(page); |
e6dcd2dc | 4695 | |
9703fefe | 4696 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4697 | set_page_extent_mapped(page); |
4698 | ||
9703fefe | 4699 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4700 | if (ordered) { |
9703fefe | 4701 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4702 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4703 | unlock_page(page); |
09cbfeaf | 4704 | put_page(page); |
eb84ae03 | 4705 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4706 | btrfs_put_ordered_extent(ordered); |
4707 | goto again; | |
4708 | } | |
4709 | ||
9703fefe | 4710 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4711 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4712 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4713 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4714 | |
9703fefe | 4715 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
2ac55d41 | 4716 | &cached_state); |
9ed74f2d | 4717 | if (ret) { |
9703fefe | 4718 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4719 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4720 | goto out_unlock; |
4721 | } | |
4722 | ||
9703fefe | 4723 | if (offset != blocksize) { |
2aaa6655 | 4724 | if (!len) |
9703fefe | 4725 | len = blocksize - offset; |
e6dcd2dc | 4726 | kaddr = kmap(page); |
2aaa6655 | 4727 | if (front) |
9703fefe CR |
4728 | memset(kaddr + (block_start - page_offset(page)), |
4729 | 0, offset); | |
2aaa6655 | 4730 | else |
9703fefe CR |
4731 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4732 | 0, len); | |
e6dcd2dc CM |
4733 | flush_dcache_page(page); |
4734 | kunmap(page); | |
4735 | } | |
247e743c | 4736 | ClearPageChecked(page); |
e6dcd2dc | 4737 | set_page_dirty(page); |
9703fefe | 4738 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4739 | GFP_NOFS); |
39279cc3 | 4740 | |
89642229 | 4741 | out_unlock: |
5d5e103a | 4742 | if (ret) |
9703fefe CR |
4743 | btrfs_delalloc_release_space(inode, block_start, |
4744 | blocksize); | |
39279cc3 | 4745 | unlock_page(page); |
09cbfeaf | 4746 | put_page(page); |
39279cc3 CM |
4747 | out: |
4748 | return ret; | |
4749 | } | |
4750 | ||
16e7549f JB |
4751 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4752 | u64 offset, u64 len) | |
4753 | { | |
4754 | struct btrfs_trans_handle *trans; | |
4755 | int ret; | |
4756 | ||
4757 | /* | |
4758 | * Still need to make sure the inode looks like it's been updated so | |
4759 | * that any holes get logged if we fsync. | |
4760 | */ | |
4761 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4762 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4763 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4764 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4765 | return 0; | |
4766 | } | |
4767 | ||
4768 | /* | |
4769 | * 1 - for the one we're dropping | |
4770 | * 1 - for the one we're adding | |
4771 | * 1 - for updating the inode. | |
4772 | */ | |
4773 | trans = btrfs_start_transaction(root, 3); | |
4774 | if (IS_ERR(trans)) | |
4775 | return PTR_ERR(trans); | |
4776 | ||
4777 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4778 | if (ret) { | |
4779 | btrfs_abort_transaction(trans, root, ret); | |
4780 | btrfs_end_transaction(trans, root); | |
4781 | return ret; | |
4782 | } | |
4783 | ||
4784 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4785 | 0, 0, len, 0, len, 0, 0, 0); | |
4786 | if (ret) | |
4787 | btrfs_abort_transaction(trans, root, ret); | |
4788 | else | |
4789 | btrfs_update_inode(trans, root, inode); | |
4790 | btrfs_end_transaction(trans, root); | |
4791 | return ret; | |
4792 | } | |
4793 | ||
695a0d0d JB |
4794 | /* |
4795 | * This function puts in dummy file extents for the area we're creating a hole | |
4796 | * for. So if we are truncating this file to a larger size we need to insert | |
4797 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4798 | * the range between oldsize and size | |
4799 | */ | |
a41ad394 | 4800 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4801 | { |
9036c102 YZ |
4802 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4803 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4804 | struct extent_map *em = NULL; |
2ac55d41 | 4805 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4806 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4807 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4808 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4809 | u64 last_byte; |
4810 | u64 cur_offset; | |
4811 | u64 hole_size; | |
9ed74f2d | 4812 | int err = 0; |
39279cc3 | 4813 | |
a71754fc | 4814 | /* |
9703fefe CR |
4815 | * If our size started in the middle of a block we need to zero out the |
4816 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4817 | * expose stale data. |
4818 | */ | |
9703fefe | 4819 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4820 | if (err) |
4821 | return err; | |
4822 | ||
9036c102 YZ |
4823 | if (size <= hole_start) |
4824 | return 0; | |
4825 | ||
9036c102 YZ |
4826 | while (1) { |
4827 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4828 | |
ff13db41 | 4829 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4830 | &cached_state); |
fa7c1494 MX |
4831 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4832 | block_end - hole_start); | |
9036c102 YZ |
4833 | if (!ordered) |
4834 | break; | |
2ac55d41 JB |
4835 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4836 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4837 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4838 | btrfs_put_ordered_extent(ordered); |
4839 | } | |
39279cc3 | 4840 | |
9036c102 YZ |
4841 | cur_offset = hole_start; |
4842 | while (1) { | |
4843 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4844 | block_end - cur_offset, 0); | |
79787eaa JM |
4845 | if (IS_ERR(em)) { |
4846 | err = PTR_ERR(em); | |
f2767956 | 4847 | em = NULL; |
79787eaa JM |
4848 | break; |
4849 | } | |
9036c102 | 4850 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4851 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4852 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4853 | struct extent_map *hole_em; |
9036c102 | 4854 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4855 | |
16e7549f JB |
4856 | err = maybe_insert_hole(root, inode, cur_offset, |
4857 | hole_size); | |
4858 | if (err) | |
3893e33b | 4859 | break; |
5dc562c5 JB |
4860 | btrfs_drop_extent_cache(inode, cur_offset, |
4861 | cur_offset + hole_size - 1, 0); | |
4862 | hole_em = alloc_extent_map(); | |
4863 | if (!hole_em) { | |
4864 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4865 | &BTRFS_I(inode)->runtime_flags); | |
4866 | goto next; | |
4867 | } | |
4868 | hole_em->start = cur_offset; | |
4869 | hole_em->len = hole_size; | |
4870 | hole_em->orig_start = cur_offset; | |
8082510e | 4871 | |
5dc562c5 JB |
4872 | hole_em->block_start = EXTENT_MAP_HOLE; |
4873 | hole_em->block_len = 0; | |
b4939680 | 4874 | hole_em->orig_block_len = 0; |
cc95bef6 | 4875 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4876 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4877 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4878 | hole_em->generation = root->fs_info->generation; |
8082510e | 4879 | |
5dc562c5 JB |
4880 | while (1) { |
4881 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4882 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4883 | write_unlock(&em_tree->lock); |
4884 | if (err != -EEXIST) | |
4885 | break; | |
4886 | btrfs_drop_extent_cache(inode, cur_offset, | |
4887 | cur_offset + | |
4888 | hole_size - 1, 0); | |
4889 | } | |
4890 | free_extent_map(hole_em); | |
9036c102 | 4891 | } |
16e7549f | 4892 | next: |
9036c102 | 4893 | free_extent_map(em); |
a22285a6 | 4894 | em = NULL; |
9036c102 | 4895 | cur_offset = last_byte; |
8082510e | 4896 | if (cur_offset >= block_end) |
9036c102 YZ |
4897 | break; |
4898 | } | |
a22285a6 | 4899 | free_extent_map(em); |
2ac55d41 JB |
4900 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4901 | GFP_NOFS); | |
9036c102 YZ |
4902 | return err; |
4903 | } | |
39279cc3 | 4904 | |
3972f260 | 4905 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4906 | { |
f4a2f4c5 MX |
4907 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4908 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4909 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4910 | loff_t newsize = attr->ia_size; |
4911 | int mask = attr->ia_valid; | |
8082510e YZ |
4912 | int ret; |
4913 | ||
3972f260 ES |
4914 | /* |
4915 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4916 | * special case where we need to update the times despite not having | |
4917 | * these flags set. For all other operations the VFS set these flags | |
4918 | * explicitly if it wants a timestamp update. | |
4919 | */ | |
dff6efc3 CH |
4920 | if (newsize != oldsize) { |
4921 | inode_inc_iversion(inode); | |
4922 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4923 | inode->i_ctime = inode->i_mtime = | |
4924 | current_fs_time(inode->i_sb); | |
4925 | } | |
3972f260 | 4926 | |
a41ad394 | 4927 | if (newsize > oldsize) { |
9ea24bbe FM |
4928 | /* |
4929 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4930 | * This is to ensure the snapshot captures a fully consistent | |
4931 | * state of this file - if the snapshot captures this expanding | |
4932 | * truncation, it must capture all writes that happened before | |
4933 | * this truncation. | |
4934 | */ | |
0bc19f90 | 4935 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 4936 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4937 | if (ret) { |
4938 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4939 | return ret; |
9ea24bbe | 4940 | } |
8082510e | 4941 | |
f4a2f4c5 | 4942 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4943 | if (IS_ERR(trans)) { |
4944 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4945 | return PTR_ERR(trans); |
9ea24bbe | 4946 | } |
f4a2f4c5 MX |
4947 | |
4948 | i_size_write(inode, newsize); | |
4949 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 4950 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 4951 | ret = btrfs_update_inode(trans, root, inode); |
9ea24bbe | 4952 | btrfs_end_write_no_snapshoting(root); |
7ad85bb7 | 4953 | btrfs_end_transaction(trans, root); |
a41ad394 | 4954 | } else { |
8082510e | 4955 | |
a41ad394 JB |
4956 | /* |
4957 | * We're truncating a file that used to have good data down to | |
4958 | * zero. Make sure it gets into the ordered flush list so that | |
4959 | * any new writes get down to disk quickly. | |
4960 | */ | |
4961 | if (newsize == 0) | |
72ac3c0d JB |
4962 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4963 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4964 | |
f3fe820c JB |
4965 | /* |
4966 | * 1 for the orphan item we're going to add | |
4967 | * 1 for the orphan item deletion. | |
4968 | */ | |
4969 | trans = btrfs_start_transaction(root, 2); | |
4970 | if (IS_ERR(trans)) | |
4971 | return PTR_ERR(trans); | |
4972 | ||
4973 | /* | |
4974 | * We need to do this in case we fail at _any_ point during the | |
4975 | * actual truncate. Once we do the truncate_setsize we could | |
4976 | * invalidate pages which forces any outstanding ordered io to | |
4977 | * be instantly completed which will give us extents that need | |
4978 | * to be truncated. If we fail to get an orphan inode down we | |
4979 | * could have left over extents that were never meant to live, | |
01327610 | 4980 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
4981 | * will be consistent. |
4982 | */ | |
4983 | ret = btrfs_orphan_add(trans, inode); | |
4984 | btrfs_end_transaction(trans, root); | |
4985 | if (ret) | |
4986 | return ret; | |
4987 | ||
a41ad394 JB |
4988 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4989 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4990 | |
4991 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4992 | btrfs_inode_block_unlocked_dio(inode); | |
4993 | inode_dio_wait(inode); | |
4994 | btrfs_inode_resume_unlocked_dio(inode); | |
4995 | ||
a41ad394 | 4996 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4997 | if (ret && inode->i_nlink) { |
4998 | int err; | |
4999 | ||
5000 | /* | |
5001 | * failed to truncate, disk_i_size is only adjusted down | |
5002 | * as we remove extents, so it should represent the true | |
5003 | * size of the inode, so reset the in memory size and | |
5004 | * delete our orphan entry. | |
5005 | */ | |
5006 | trans = btrfs_join_transaction(root); | |
5007 | if (IS_ERR(trans)) { | |
5008 | btrfs_orphan_del(NULL, inode); | |
5009 | return ret; | |
5010 | } | |
5011 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
5012 | err = btrfs_orphan_del(trans, inode); | |
5013 | if (err) | |
5014 | btrfs_abort_transaction(trans, root, err); | |
5015 | btrfs_end_transaction(trans, root); | |
5016 | } | |
8082510e YZ |
5017 | } |
5018 | ||
a41ad394 | 5019 | return ret; |
8082510e YZ |
5020 | } |
5021 | ||
9036c102 YZ |
5022 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5023 | { | |
2b0143b5 | 5024 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5025 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5026 | int err; |
39279cc3 | 5027 | |
b83cc969 LZ |
5028 | if (btrfs_root_readonly(root)) |
5029 | return -EROFS; | |
5030 | ||
9036c102 YZ |
5031 | err = inode_change_ok(inode, attr); |
5032 | if (err) | |
5033 | return err; | |
2bf5a725 | 5034 | |
5a3f23d5 | 5035 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5036 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5037 | if (err) |
5038 | return err; | |
39279cc3 | 5039 | } |
9036c102 | 5040 | |
1025774c CH |
5041 | if (attr->ia_valid) { |
5042 | setattr_copy(inode, attr); | |
0c4d2d95 | 5043 | inode_inc_iversion(inode); |
22c44fe6 | 5044 | err = btrfs_dirty_inode(inode); |
1025774c | 5045 | |
22c44fe6 | 5046 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5047 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5048 | } |
33268eaf | 5049 | |
39279cc3 CM |
5050 | return err; |
5051 | } | |
61295eb8 | 5052 | |
131e404a FDBM |
5053 | /* |
5054 | * While truncating the inode pages during eviction, we get the VFS calling | |
5055 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5056 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5057 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5058 | * extent_state structures over and over, wasting lots of time. | |
5059 | * | |
5060 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5061 | * those expensive operations on a per page basis and do only the ordered io | |
5062 | * finishing, while we release here the extent_map and extent_state structures, | |
5063 | * without the excessive merging and splitting. | |
5064 | */ | |
5065 | static void evict_inode_truncate_pages(struct inode *inode) | |
5066 | { | |
5067 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5068 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5069 | struct rb_node *node; | |
5070 | ||
5071 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5072 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5073 | |
5074 | write_lock(&map_tree->lock); | |
5075 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5076 | struct extent_map *em; | |
5077 | ||
5078 | node = rb_first(&map_tree->map); | |
5079 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5080 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5081 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5082 | remove_extent_mapping(map_tree, em); |
5083 | free_extent_map(em); | |
7064dd5c FM |
5084 | if (need_resched()) { |
5085 | write_unlock(&map_tree->lock); | |
5086 | cond_resched(); | |
5087 | write_lock(&map_tree->lock); | |
5088 | } | |
131e404a FDBM |
5089 | } |
5090 | write_unlock(&map_tree->lock); | |
5091 | ||
6ca07097 FM |
5092 | /* |
5093 | * Keep looping until we have no more ranges in the io tree. | |
5094 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5095 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5096 | * still in progress (unlocked the pages in the bio but did not yet | |
5097 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5098 | * ranges can still be locked and eviction started because before |
5099 | * submitting those bios, which are executed by a separate task (work | |
5100 | * queue kthread), inode references (inode->i_count) were not taken | |
5101 | * (which would be dropped in the end io callback of each bio). | |
5102 | * Therefore here we effectively end up waiting for those bios and | |
5103 | * anyone else holding locked ranges without having bumped the inode's | |
5104 | * reference count - if we don't do it, when they access the inode's | |
5105 | * io_tree to unlock a range it may be too late, leading to an | |
5106 | * use-after-free issue. | |
5107 | */ | |
131e404a FDBM |
5108 | spin_lock(&io_tree->lock); |
5109 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5110 | struct extent_state *state; | |
5111 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5112 | u64 start; |
5113 | u64 end; | |
131e404a FDBM |
5114 | |
5115 | node = rb_first(&io_tree->state); | |
5116 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5117 | start = state->start; |
5118 | end = state->end; | |
131e404a FDBM |
5119 | spin_unlock(&io_tree->lock); |
5120 | ||
ff13db41 | 5121 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5122 | |
5123 | /* | |
5124 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5125 | * and its reserved space won't be freed by delayed_ref. | |
5126 | * So we need to free its reserved space here. | |
5127 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5128 | * | |
5129 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5130 | */ | |
5131 | if (state->state & EXTENT_DELALLOC) | |
5132 | btrfs_qgroup_free_data(inode, start, end - start + 1); | |
5133 | ||
6ca07097 | 5134 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5135 | EXTENT_LOCKED | EXTENT_DIRTY | |
5136 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5137 | EXTENT_DEFRAG, 1, 1, | |
5138 | &cached_state, GFP_NOFS); | |
131e404a | 5139 | |
7064dd5c | 5140 | cond_resched(); |
131e404a FDBM |
5141 | spin_lock(&io_tree->lock); |
5142 | } | |
5143 | spin_unlock(&io_tree->lock); | |
5144 | } | |
5145 | ||
bd555975 | 5146 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
5147 | { |
5148 | struct btrfs_trans_handle *trans; | |
5149 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5150 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5151 | int steal_from_global = 0; |
07127184 | 5152 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
5153 | int ret; |
5154 | ||
1abe9b8a | 5155 | trace_btrfs_inode_evict(inode); |
5156 | ||
131e404a FDBM |
5157 | evict_inode_truncate_pages(inode); |
5158 | ||
69e9c6c6 SB |
5159 | if (inode->i_nlink && |
5160 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5161 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
5162 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
5163 | goto no_delete; |
5164 | ||
39279cc3 | 5165 | if (is_bad_inode(inode)) { |
7b128766 | 5166 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
5167 | goto no_delete; |
5168 | } | |
bd555975 | 5169 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5170 | if (!special_file(inode->i_mode)) |
5171 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5172 | |
f612496b MX |
5173 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
5174 | ||
c71bf099 | 5175 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 5176 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5177 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5178 | goto no_delete; |
5179 | } | |
5180 | ||
76dda93c | 5181 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5182 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5183 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5184 | goto no_delete; |
5185 | } | |
5186 | ||
0e8c36a9 MX |
5187 | ret = btrfs_commit_inode_delayed_inode(inode); |
5188 | if (ret) { | |
5189 | btrfs_orphan_del(NULL, inode); | |
5190 | goto no_delete; | |
5191 | } | |
5192 | ||
66d8f3dd | 5193 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
5194 | if (!rsv) { |
5195 | btrfs_orphan_del(NULL, inode); | |
5196 | goto no_delete; | |
5197 | } | |
4a338542 | 5198 | rsv->size = min_size; |
ca7e70f5 | 5199 | rsv->failfast = 1; |
726c35fa | 5200 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 5201 | |
dbe674a9 | 5202 | btrfs_i_size_write(inode, 0); |
5f39d397 | 5203 | |
4289a667 | 5204 | /* |
8407aa46 MX |
5205 | * This is a bit simpler than btrfs_truncate since we've already |
5206 | * reserved our space for our orphan item in the unlink, so we just | |
5207 | * need to reserve some slack space in case we add bytes and update | |
5208 | * inode item when doing the truncate. | |
4289a667 | 5209 | */ |
8082510e | 5210 | while (1) { |
08e007d2 MX |
5211 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5212 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5213 | |
5214 | /* | |
5215 | * Try and steal from the global reserve since we will | |
5216 | * likely not use this space anyway, we want to try as | |
5217 | * hard as possible to get this to work. | |
5218 | */ | |
5219 | if (ret) | |
3bce876f JB |
5220 | steal_from_global++; |
5221 | else | |
5222 | steal_from_global = 0; | |
5223 | ret = 0; | |
d68fc57b | 5224 | |
3bce876f JB |
5225 | /* |
5226 | * steal_from_global == 0: we reserved stuff, hooray! | |
5227 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5228 | * steal_from_global == 2: we've committed, still not a lot of | |
5229 | * room but maybe we'll have room in the global reserve this | |
5230 | * time. | |
5231 | * steal_from_global == 3: abandon all hope! | |
5232 | */ | |
5233 | if (steal_from_global > 2) { | |
c2cf52eb SK |
5234 | btrfs_warn(root->fs_info, |
5235 | "Could not get space for a delete, will truncate on mount %d", | |
5236 | ret); | |
4289a667 JB |
5237 | btrfs_orphan_del(NULL, inode); |
5238 | btrfs_free_block_rsv(root, rsv); | |
5239 | goto no_delete; | |
d68fc57b | 5240 | } |
7b128766 | 5241 | |
0e8c36a9 | 5242 | trans = btrfs_join_transaction(root); |
4289a667 JB |
5243 | if (IS_ERR(trans)) { |
5244 | btrfs_orphan_del(NULL, inode); | |
5245 | btrfs_free_block_rsv(root, rsv); | |
5246 | goto no_delete; | |
d68fc57b | 5247 | } |
7b128766 | 5248 | |
3bce876f | 5249 | /* |
01327610 | 5250 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5251 | * sure there is room to do it, if not we need to commit and try |
5252 | * again. | |
5253 | */ | |
5254 | if (steal_from_global) { | |
5255 | if (!btrfs_check_space_for_delayed_refs(trans, root)) | |
5256 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, | |
5257 | min_size); | |
5258 | else | |
5259 | ret = -ENOSPC; | |
5260 | } | |
5261 | ||
5262 | /* | |
5263 | * Couldn't steal from the global reserve, we have too much | |
5264 | * pending stuff built up, commit the transaction and try it | |
5265 | * again. | |
5266 | */ | |
5267 | if (ret) { | |
5268 | ret = btrfs_commit_transaction(trans, root); | |
5269 | if (ret) { | |
5270 | btrfs_orphan_del(NULL, inode); | |
5271 | btrfs_free_block_rsv(root, rsv); | |
5272 | goto no_delete; | |
5273 | } | |
5274 | continue; | |
5275 | } else { | |
5276 | steal_from_global = 0; | |
5277 | } | |
5278 | ||
4289a667 JB |
5279 | trans->block_rsv = rsv; |
5280 | ||
d68fc57b | 5281 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5282 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5283 | break; |
85e21bac | 5284 | |
8407aa46 | 5285 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
5286 | btrfs_end_transaction(trans, root); |
5287 | trans = NULL; | |
b53d3f5d | 5288 | btrfs_btree_balance_dirty(root); |
8082510e | 5289 | } |
5f39d397 | 5290 | |
4289a667 JB |
5291 | btrfs_free_block_rsv(root, rsv); |
5292 | ||
4ef31a45 JB |
5293 | /* |
5294 | * Errors here aren't a big deal, it just means we leave orphan items | |
5295 | * in the tree. They will be cleaned up on the next mount. | |
5296 | */ | |
8082510e | 5297 | if (ret == 0) { |
4289a667 | 5298 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5299 | btrfs_orphan_del(trans, inode); |
5300 | } else { | |
5301 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5302 | } |
54aa1f4d | 5303 | |
4289a667 | 5304 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
5305 | if (!(root == root->fs_info->tree_root || |
5306 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 5307 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 5308 | |
54aa1f4d | 5309 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5310 | btrfs_btree_balance_dirty(root); |
39279cc3 | 5311 | no_delete: |
89042e5a | 5312 | btrfs_remove_delayed_node(inode); |
dbd5768f | 5313 | clear_inode(inode); |
39279cc3 CM |
5314 | } |
5315 | ||
5316 | /* | |
5317 | * this returns the key found in the dir entry in the location pointer. | |
5318 | * If no dir entries were found, location->objectid is 0. | |
5319 | */ | |
5320 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5321 | struct btrfs_key *location) | |
5322 | { | |
5323 | const char *name = dentry->d_name.name; | |
5324 | int namelen = dentry->d_name.len; | |
5325 | struct btrfs_dir_item *di; | |
5326 | struct btrfs_path *path; | |
5327 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5328 | int ret = 0; |
39279cc3 CM |
5329 | |
5330 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5331 | if (!path) |
5332 | return -ENOMEM; | |
3954401f | 5333 | |
33345d01 | 5334 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 5335 | namelen, 0); |
0d9f7f3e Y |
5336 | if (IS_ERR(di)) |
5337 | ret = PTR_ERR(di); | |
d397712b | 5338 | |
c704005d | 5339 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5340 | goto out_err; |
d397712b | 5341 | |
5f39d397 | 5342 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5343 | out: |
39279cc3 CM |
5344 | btrfs_free_path(path); |
5345 | return ret; | |
3954401f CM |
5346 | out_err: |
5347 | location->objectid = 0; | |
5348 | goto out; | |
39279cc3 CM |
5349 | } |
5350 | ||
5351 | /* | |
5352 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5353 | * needs to be changed to reflect the root directory of the tree root. This | |
5354 | * is kind of like crossing a mount point. | |
5355 | */ | |
5356 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
5357 | struct inode *dir, |
5358 | struct dentry *dentry, | |
5359 | struct btrfs_key *location, | |
5360 | struct btrfs_root **sub_root) | |
39279cc3 | 5361 | { |
4df27c4d YZ |
5362 | struct btrfs_path *path; |
5363 | struct btrfs_root *new_root; | |
5364 | struct btrfs_root_ref *ref; | |
5365 | struct extent_buffer *leaf; | |
1d4c08e0 | 5366 | struct btrfs_key key; |
4df27c4d YZ |
5367 | int ret; |
5368 | int err = 0; | |
39279cc3 | 5369 | |
4df27c4d YZ |
5370 | path = btrfs_alloc_path(); |
5371 | if (!path) { | |
5372 | err = -ENOMEM; | |
5373 | goto out; | |
5374 | } | |
39279cc3 | 5375 | |
4df27c4d | 5376 | err = -ENOENT; |
1d4c08e0 DS |
5377 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5378 | key.type = BTRFS_ROOT_REF_KEY; | |
5379 | key.offset = location->objectid; | |
5380 | ||
5381 | ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path, | |
5382 | 0, 0); | |
4df27c4d YZ |
5383 | if (ret) { |
5384 | if (ret < 0) | |
5385 | err = ret; | |
5386 | goto out; | |
5387 | } | |
39279cc3 | 5388 | |
4df27c4d YZ |
5389 | leaf = path->nodes[0]; |
5390 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 5391 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
5392 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5393 | goto out; | |
39279cc3 | 5394 | |
4df27c4d YZ |
5395 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5396 | (unsigned long)(ref + 1), | |
5397 | dentry->d_name.len); | |
5398 | if (ret) | |
5399 | goto out; | |
5400 | ||
b3b4aa74 | 5401 | btrfs_release_path(path); |
4df27c4d YZ |
5402 | |
5403 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5404 | if (IS_ERR(new_root)) { | |
5405 | err = PTR_ERR(new_root); | |
5406 | goto out; | |
5407 | } | |
5408 | ||
4df27c4d YZ |
5409 | *sub_root = new_root; |
5410 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5411 | location->type = BTRFS_INODE_ITEM_KEY; | |
5412 | location->offset = 0; | |
5413 | err = 0; | |
5414 | out: | |
5415 | btrfs_free_path(path); | |
5416 | return err; | |
39279cc3 CM |
5417 | } |
5418 | ||
5d4f98a2 YZ |
5419 | static void inode_tree_add(struct inode *inode) |
5420 | { | |
5421 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5422 | struct btrfs_inode *entry; | |
03e860bd FNP |
5423 | struct rb_node **p; |
5424 | struct rb_node *parent; | |
cef21937 | 5425 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5426 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5427 | |
1d3382cb | 5428 | if (inode_unhashed(inode)) |
76dda93c | 5429 | return; |
e1409cef | 5430 | parent = NULL; |
5d4f98a2 | 5431 | spin_lock(&root->inode_lock); |
e1409cef | 5432 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5433 | while (*p) { |
5434 | parent = *p; | |
5435 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5436 | ||
33345d01 | 5437 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5438 | p = &parent->rb_left; |
33345d01 | 5439 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5440 | p = &parent->rb_right; |
5d4f98a2 YZ |
5441 | else { |
5442 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5443 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5444 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5445 | RB_CLEAR_NODE(parent); |
5446 | spin_unlock(&root->inode_lock); | |
cef21937 | 5447 | return; |
5d4f98a2 YZ |
5448 | } |
5449 | } | |
cef21937 FDBM |
5450 | rb_link_node(new, parent, p); |
5451 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5452 | spin_unlock(&root->inode_lock); |
5453 | } | |
5454 | ||
5455 | static void inode_tree_del(struct inode *inode) | |
5456 | { | |
5457 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5458 | int empty = 0; |
5d4f98a2 | 5459 | |
03e860bd | 5460 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5461 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5462 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5463 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5464 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5465 | } |
03e860bd | 5466 | spin_unlock(&root->inode_lock); |
76dda93c | 5467 | |
69e9c6c6 | 5468 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5469 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5470 | spin_lock(&root->inode_lock); | |
5471 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5472 | spin_unlock(&root->inode_lock); | |
5473 | if (empty) | |
5474 | btrfs_add_dead_root(root); | |
5475 | } | |
5476 | } | |
5477 | ||
143bede5 | 5478 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5479 | { |
5480 | struct rb_node *node; | |
5481 | struct rb_node *prev; | |
5482 | struct btrfs_inode *entry; | |
5483 | struct inode *inode; | |
5484 | u64 objectid = 0; | |
5485 | ||
7813b3db LB |
5486 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5487 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5488 | |
5489 | spin_lock(&root->inode_lock); | |
5490 | again: | |
5491 | node = root->inode_tree.rb_node; | |
5492 | prev = NULL; | |
5493 | while (node) { | |
5494 | prev = node; | |
5495 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5496 | ||
33345d01 | 5497 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5498 | node = node->rb_left; |
33345d01 | 5499 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5500 | node = node->rb_right; |
5501 | else | |
5502 | break; | |
5503 | } | |
5504 | if (!node) { | |
5505 | while (prev) { | |
5506 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5507 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5508 | node = prev; |
5509 | break; | |
5510 | } | |
5511 | prev = rb_next(prev); | |
5512 | } | |
5513 | } | |
5514 | while (node) { | |
5515 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5516 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5517 | inode = igrab(&entry->vfs_inode); |
5518 | if (inode) { | |
5519 | spin_unlock(&root->inode_lock); | |
5520 | if (atomic_read(&inode->i_count) > 1) | |
5521 | d_prune_aliases(inode); | |
5522 | /* | |
45321ac5 | 5523 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5524 | * the inode cache when its usage count |
5525 | * hits zero. | |
5526 | */ | |
5527 | iput(inode); | |
5528 | cond_resched(); | |
5529 | spin_lock(&root->inode_lock); | |
5530 | goto again; | |
5531 | } | |
5532 | ||
5533 | if (cond_resched_lock(&root->inode_lock)) | |
5534 | goto again; | |
5535 | ||
5536 | node = rb_next(node); | |
5537 | } | |
5538 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5539 | } |
5540 | ||
e02119d5 CM |
5541 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5542 | { | |
5543 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5544 | inode->i_ino = args->location->objectid; |
5545 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5546 | sizeof(*args->location)); | |
e02119d5 | 5547 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5548 | return 0; |
5549 | } | |
5550 | ||
5551 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5552 | { | |
5553 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5554 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5555 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5556 | } |
5557 | ||
5d4f98a2 | 5558 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5559 | struct btrfs_key *location, |
5d4f98a2 | 5560 | struct btrfs_root *root) |
39279cc3 CM |
5561 | { |
5562 | struct inode *inode; | |
5563 | struct btrfs_iget_args args; | |
90d3e592 | 5564 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5565 | |
90d3e592 | 5566 | args.location = location; |
39279cc3 CM |
5567 | args.root = root; |
5568 | ||
778ba82b | 5569 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5570 | btrfs_init_locked_inode, |
5571 | (void *)&args); | |
5572 | return inode; | |
5573 | } | |
5574 | ||
1a54ef8c BR |
5575 | /* Get an inode object given its location and corresponding root. |
5576 | * Returns in *is_new if the inode was read from disk | |
5577 | */ | |
5578 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5579 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5580 | { |
5581 | struct inode *inode; | |
5582 | ||
90d3e592 | 5583 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5584 | if (!inode) |
5d4f98a2 | 5585 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5586 | |
5587 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5588 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5589 | if (!is_bad_inode(inode)) { |
5590 | inode_tree_add(inode); | |
5591 | unlock_new_inode(inode); | |
5592 | if (new) | |
5593 | *new = 1; | |
5594 | } else { | |
e0b6d65b ST |
5595 | unlock_new_inode(inode); |
5596 | iput(inode); | |
5597 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5598 | } |
5599 | } | |
5600 | ||
1a54ef8c BR |
5601 | return inode; |
5602 | } | |
5603 | ||
4df27c4d YZ |
5604 | static struct inode *new_simple_dir(struct super_block *s, |
5605 | struct btrfs_key *key, | |
5606 | struct btrfs_root *root) | |
5607 | { | |
5608 | struct inode *inode = new_inode(s); | |
5609 | ||
5610 | if (!inode) | |
5611 | return ERR_PTR(-ENOMEM); | |
5612 | ||
4df27c4d YZ |
5613 | BTRFS_I(inode)->root = root; |
5614 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5615 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5616 | |
5617 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5618 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5619 | inode->i_fop = &simple_dir_operations; |
5620 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
04b285f3 | 5621 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 5622 | inode->i_atime = inode->i_mtime; |
5623 | inode->i_ctime = inode->i_mtime; | |
5624 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5625 | |
5626 | return inode; | |
5627 | } | |
5628 | ||
3de4586c | 5629 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5630 | { |
d397712b | 5631 | struct inode *inode; |
4df27c4d | 5632 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5633 | struct btrfs_root *sub_root = root; |
5634 | struct btrfs_key location; | |
76dda93c | 5635 | int index; |
b4aff1f8 | 5636 | int ret = 0; |
39279cc3 CM |
5637 | |
5638 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5639 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5640 | |
39e3c955 | 5641 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5642 | if (ret < 0) |
5643 | return ERR_PTR(ret); | |
5f39d397 | 5644 | |
4df27c4d | 5645 | if (location.objectid == 0) |
5662344b | 5646 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5647 | |
5648 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5649 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5650 | return inode; |
5651 | } | |
5652 | ||
5653 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5654 | ||
76dda93c | 5655 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5656 | ret = fixup_tree_root_location(root, dir, dentry, |
5657 | &location, &sub_root); | |
5658 | if (ret < 0) { | |
5659 | if (ret != -ENOENT) | |
5660 | inode = ERR_PTR(ret); | |
5661 | else | |
5662 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5663 | } else { | |
73f73415 | 5664 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5665 | } |
76dda93c YZ |
5666 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5667 | ||
34d19bad | 5668 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5669 | down_read(&root->fs_info->cleanup_work_sem); |
5670 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5671 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5672 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5673 | if (ret) { |
5674 | iput(inode); | |
66b4ffd1 | 5675 | inode = ERR_PTR(ret); |
01cd3367 | 5676 | } |
c71bf099 YZ |
5677 | } |
5678 | ||
3de4586c CM |
5679 | return inode; |
5680 | } | |
5681 | ||
fe15ce44 | 5682 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5683 | { |
5684 | struct btrfs_root *root; | |
2b0143b5 | 5685 | struct inode *inode = d_inode(dentry); |
76dda93c | 5686 | |
848cce0d | 5687 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5688 | inode = d_inode(dentry->d_parent); |
76dda93c | 5689 | |
848cce0d LZ |
5690 | if (inode) { |
5691 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5692 | if (btrfs_root_refs(&root->root_item) == 0) |
5693 | return 1; | |
848cce0d LZ |
5694 | |
5695 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5696 | return 1; | |
efefb143 | 5697 | } |
76dda93c YZ |
5698 | return 0; |
5699 | } | |
5700 | ||
b4aff1f8 JB |
5701 | static void btrfs_dentry_release(struct dentry *dentry) |
5702 | { | |
944a4515 | 5703 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5704 | } |
5705 | ||
3de4586c | 5706 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5707 | unsigned int flags) |
3de4586c | 5708 | { |
5662344b | 5709 | struct inode *inode; |
a66e7cc6 | 5710 | |
5662344b TI |
5711 | inode = btrfs_lookup_dentry(dir, dentry); |
5712 | if (IS_ERR(inode)) { | |
5713 | if (PTR_ERR(inode) == -ENOENT) | |
5714 | inode = NULL; | |
5715 | else | |
5716 | return ERR_CAST(inode); | |
5717 | } | |
5718 | ||
41d28bca | 5719 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5720 | } |
5721 | ||
16cdcec7 | 5722 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5723 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5724 | }; | |
5725 | ||
9cdda8d3 | 5726 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5727 | { |
9cdda8d3 | 5728 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5729 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5730 | struct btrfs_item *item; | |
5731 | struct btrfs_dir_item *di; | |
5732 | struct btrfs_key key; | |
5f39d397 | 5733 | struct btrfs_key found_key; |
39279cc3 | 5734 | struct btrfs_path *path; |
16cdcec7 MX |
5735 | struct list_head ins_list; |
5736 | struct list_head del_list; | |
39279cc3 | 5737 | int ret; |
5f39d397 | 5738 | struct extent_buffer *leaf; |
39279cc3 | 5739 | int slot; |
39279cc3 CM |
5740 | unsigned char d_type; |
5741 | int over = 0; | |
5742 | u32 di_cur; | |
5743 | u32 di_total; | |
5744 | u32 di_len; | |
5745 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5746 | char tmp_name[32]; |
5747 | char *name_ptr; | |
5748 | int name_len; | |
9cdda8d3 | 5749 | int is_curr = 0; /* ctx->pos points to the current index? */ |
bc4ef759 | 5750 | bool emitted; |
39279cc3 CM |
5751 | |
5752 | /* FIXME, use a real flag for deciding about the key type */ | |
5753 | if (root->fs_info->tree_root == root) | |
5754 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5755 | |
9cdda8d3 AV |
5756 | if (!dir_emit_dots(file, ctx)) |
5757 | return 0; | |
5758 | ||
49593bfa | 5759 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5760 | if (!path) |
5761 | return -ENOMEM; | |
ff5714cc | 5762 | |
e4058b54 | 5763 | path->reada = READA_FORWARD; |
49593bfa | 5764 | |
16cdcec7 MX |
5765 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5766 | INIT_LIST_HEAD(&ins_list); | |
5767 | INIT_LIST_HEAD(&del_list); | |
5768 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5769 | } | |
5770 | ||
962a298f | 5771 | key.type = key_type; |
9cdda8d3 | 5772 | key.offset = ctx->pos; |
33345d01 | 5773 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5774 | |
39279cc3 CM |
5775 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5776 | if (ret < 0) | |
5777 | goto err; | |
49593bfa | 5778 | |
bc4ef759 | 5779 | emitted = false; |
49593bfa | 5780 | while (1) { |
5f39d397 | 5781 | leaf = path->nodes[0]; |
39279cc3 | 5782 | slot = path->slots[0]; |
b9e03af0 LZ |
5783 | if (slot >= btrfs_header_nritems(leaf)) { |
5784 | ret = btrfs_next_leaf(root, path); | |
5785 | if (ret < 0) | |
5786 | goto err; | |
5787 | else if (ret > 0) | |
5788 | break; | |
5789 | continue; | |
39279cc3 | 5790 | } |
3de4586c | 5791 | |
dd3cc16b | 5792 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5793 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5794 | ||
5795 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5796 | break; |
962a298f | 5797 | if (found_key.type != key_type) |
39279cc3 | 5798 | break; |
9cdda8d3 | 5799 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5800 | goto next; |
16cdcec7 MX |
5801 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5802 | btrfs_should_delete_dir_index(&del_list, | |
5803 | found_key.offset)) | |
5804 | goto next; | |
5f39d397 | 5805 | |
9cdda8d3 | 5806 | ctx->pos = found_key.offset; |
16cdcec7 | 5807 | is_curr = 1; |
49593bfa | 5808 | |
39279cc3 CM |
5809 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5810 | di_cur = 0; | |
5f39d397 | 5811 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5812 | |
5813 | while (di_cur < di_total) { | |
5f39d397 CM |
5814 | struct btrfs_key location; |
5815 | ||
22a94d44 JB |
5816 | if (verify_dir_item(root, leaf, di)) |
5817 | break; | |
5818 | ||
5f39d397 | 5819 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5820 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5821 | name_ptr = tmp_name; |
5822 | } else { | |
49e350a4 | 5823 | name_ptr = kmalloc(name_len, GFP_KERNEL); |
49593bfa DW |
5824 | if (!name_ptr) { |
5825 | ret = -ENOMEM; | |
5826 | goto err; | |
5827 | } | |
5f39d397 CM |
5828 | } |
5829 | read_extent_buffer(leaf, name_ptr, | |
5830 | (unsigned long)(di + 1), name_len); | |
5831 | ||
5832 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5833 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5834 | |
fede766f | 5835 | |
3de4586c | 5836 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5837 | * skip it. |
5838 | * | |
5839 | * In contrast to old kernels, we insert the snapshot's | |
5840 | * dir item and dir index after it has been created, so | |
5841 | * we won't find a reference to our own snapshot. We | |
5842 | * still keep the following code for backward | |
5843 | * compatibility. | |
3de4586c CM |
5844 | */ |
5845 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5846 | location.objectid == root->root_key.objectid) { | |
5847 | over = 0; | |
5848 | goto skip; | |
5849 | } | |
9cdda8d3 AV |
5850 | over = !dir_emit(ctx, name_ptr, name_len, |
5851 | location.objectid, d_type); | |
5f39d397 | 5852 | |
3de4586c | 5853 | skip: |
5f39d397 CM |
5854 | if (name_ptr != tmp_name) |
5855 | kfree(name_ptr); | |
5856 | ||
39279cc3 CM |
5857 | if (over) |
5858 | goto nopos; | |
bc4ef759 | 5859 | emitted = true; |
5103e947 | 5860 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5861 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5862 | di_cur += di_len; |
5863 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5864 | } | |
b9e03af0 LZ |
5865 | next: |
5866 | path->slots[0]++; | |
39279cc3 | 5867 | } |
49593bfa | 5868 | |
16cdcec7 MX |
5869 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5870 | if (is_curr) | |
9cdda8d3 | 5871 | ctx->pos++; |
bc4ef759 | 5872 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted); |
16cdcec7 MX |
5873 | if (ret) |
5874 | goto nopos; | |
5875 | } | |
5876 | ||
bc4ef759 DS |
5877 | /* |
5878 | * If we haven't emitted any dir entry, we must not touch ctx->pos as | |
5879 | * it was was set to the termination value in previous call. We assume | |
5880 | * that "." and ".." were emitted if we reach this point and set the | |
5881 | * termination value as well for an empty directory. | |
5882 | */ | |
5883 | if (ctx->pos > 2 && !emitted) | |
5884 | goto nopos; | |
5885 | ||
49593bfa | 5886 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5887 | ctx->pos++; |
5888 | ||
5889 | /* | |
5890 | * Stop new entries from being returned after we return the last | |
5891 | * entry. | |
5892 | * | |
5893 | * New directory entries are assigned a strictly increasing | |
5894 | * offset. This means that new entries created during readdir | |
5895 | * are *guaranteed* to be seen in the future by that readdir. | |
5896 | * This has broken buggy programs which operate on names as | |
5897 | * they're returned by readdir. Until we re-use freed offsets | |
5898 | * we have this hack to stop new entries from being returned | |
5899 | * under the assumption that they'll never reach this huge | |
5900 | * offset. | |
5901 | * | |
5902 | * This is being careful not to overflow 32bit loff_t unless the | |
5903 | * last entry requires it because doing so has broken 32bit apps | |
5904 | * in the past. | |
5905 | */ | |
5906 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5907 | if (ctx->pos >= INT_MAX) | |
5908 | ctx->pos = LLONG_MAX; | |
5909 | else | |
5910 | ctx->pos = INT_MAX; | |
5911 | } | |
39279cc3 CM |
5912 | nopos: |
5913 | ret = 0; | |
5914 | err: | |
16cdcec7 MX |
5915 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5916 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5917 | btrfs_free_path(path); |
39279cc3 CM |
5918 | return ret; |
5919 | } | |
5920 | ||
a9185b41 | 5921 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5922 | { |
5923 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5924 | struct btrfs_trans_handle *trans; | |
5925 | int ret = 0; | |
0af3d00b | 5926 | bool nolock = false; |
39279cc3 | 5927 | |
72ac3c0d | 5928 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5929 | return 0; |
5930 | ||
83eea1f1 | 5931 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5932 | nolock = true; |
0af3d00b | 5933 | |
a9185b41 | 5934 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5935 | if (nolock) |
7a7eaa40 | 5936 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5937 | else |
7a7eaa40 | 5938 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5939 | if (IS_ERR(trans)) |
5940 | return PTR_ERR(trans); | |
a698d075 | 5941 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5942 | } |
5943 | return ret; | |
5944 | } | |
5945 | ||
5946 | /* | |
54aa1f4d | 5947 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5948 | * inode changes. But, it is most likely to find the inode in cache. |
5949 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5950 | * to keep or drop this code. | |
5951 | */ | |
48a3b636 | 5952 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5953 | { |
5954 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5955 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5956 | int ret; |
5957 | ||
72ac3c0d | 5958 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5959 | return 0; |
39279cc3 | 5960 | |
7a7eaa40 | 5961 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5962 | if (IS_ERR(trans)) |
5963 | return PTR_ERR(trans); | |
8929ecfa YZ |
5964 | |
5965 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5966 | if (ret && ret == -ENOSPC) { |
5967 | /* whoops, lets try again with the full transaction */ | |
5968 | btrfs_end_transaction(trans, root); | |
5969 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5970 | if (IS_ERR(trans)) |
5971 | return PTR_ERR(trans); | |
8929ecfa | 5972 | |
94b60442 | 5973 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5974 | } |
39279cc3 | 5975 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5976 | if (BTRFS_I(inode)->delayed_node) |
5977 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5978 | |
5979 | return ret; | |
5980 | } | |
5981 | ||
5982 | /* | |
5983 | * This is a copy of file_update_time. We need this so we can return error on | |
5984 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5985 | */ | |
e41f941a JB |
5986 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5987 | int flags) | |
22c44fe6 | 5988 | { |
2bc55652 AB |
5989 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5990 | ||
5991 | if (btrfs_root_readonly(root)) | |
5992 | return -EROFS; | |
5993 | ||
e41f941a | 5994 | if (flags & S_VERSION) |
22c44fe6 | 5995 | inode_inc_iversion(inode); |
e41f941a JB |
5996 | if (flags & S_CTIME) |
5997 | inode->i_ctime = *now; | |
5998 | if (flags & S_MTIME) | |
5999 | inode->i_mtime = *now; | |
6000 | if (flags & S_ATIME) | |
6001 | inode->i_atime = *now; | |
6002 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6003 | } |
6004 | ||
d352ac68 CM |
6005 | /* |
6006 | * find the highest existing sequence number in a directory | |
6007 | * and then set the in-memory index_cnt variable to reflect | |
6008 | * free sequence numbers | |
6009 | */ | |
aec7477b JB |
6010 | static int btrfs_set_inode_index_count(struct inode *inode) |
6011 | { | |
6012 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6013 | struct btrfs_key key, found_key; | |
6014 | struct btrfs_path *path; | |
6015 | struct extent_buffer *leaf; | |
6016 | int ret; | |
6017 | ||
33345d01 | 6018 | key.objectid = btrfs_ino(inode); |
962a298f | 6019 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6020 | key.offset = (u64)-1; |
6021 | ||
6022 | path = btrfs_alloc_path(); | |
6023 | if (!path) | |
6024 | return -ENOMEM; | |
6025 | ||
6026 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6027 | if (ret < 0) | |
6028 | goto out; | |
6029 | /* FIXME: we should be able to handle this */ | |
6030 | if (ret == 0) | |
6031 | goto out; | |
6032 | ret = 0; | |
6033 | ||
6034 | /* | |
6035 | * MAGIC NUMBER EXPLANATION: | |
6036 | * since we search a directory based on f_pos we have to start at 2 | |
6037 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6038 | * else has to start at 2 | |
6039 | */ | |
6040 | if (path->slots[0] == 0) { | |
6041 | BTRFS_I(inode)->index_cnt = 2; | |
6042 | goto out; | |
6043 | } | |
6044 | ||
6045 | path->slots[0]--; | |
6046 | ||
6047 | leaf = path->nodes[0]; | |
6048 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6049 | ||
33345d01 | 6050 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6051 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
6052 | BTRFS_I(inode)->index_cnt = 2; |
6053 | goto out; | |
6054 | } | |
6055 | ||
6056 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
6057 | out: | |
6058 | btrfs_free_path(path); | |
6059 | return ret; | |
6060 | } | |
6061 | ||
d352ac68 CM |
6062 | /* |
6063 | * helper to find a free sequence number in a given directory. This current | |
6064 | * code is very simple, later versions will do smarter things in the btree | |
6065 | */ | |
3de4586c | 6066 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
6067 | { |
6068 | int ret = 0; | |
6069 | ||
6070 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
6071 | ret = btrfs_inode_delayed_dir_index_count(dir); |
6072 | if (ret) { | |
6073 | ret = btrfs_set_inode_index_count(dir); | |
6074 | if (ret) | |
6075 | return ret; | |
6076 | } | |
aec7477b JB |
6077 | } |
6078 | ||
00e4e6b3 | 6079 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
6080 | BTRFS_I(dir)->index_cnt++; |
6081 | ||
6082 | return ret; | |
6083 | } | |
6084 | ||
b0d5d10f CM |
6085 | static int btrfs_insert_inode_locked(struct inode *inode) |
6086 | { | |
6087 | struct btrfs_iget_args args; | |
6088 | args.location = &BTRFS_I(inode)->location; | |
6089 | args.root = BTRFS_I(inode)->root; | |
6090 | ||
6091 | return insert_inode_locked4(inode, | |
6092 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6093 | btrfs_find_actor, &args); | |
6094 | } | |
6095 | ||
39279cc3 CM |
6096 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6097 | struct btrfs_root *root, | |
aec7477b | 6098 | struct inode *dir, |
9c58309d | 6099 | const char *name, int name_len, |
175a4eb7 AV |
6100 | u64 ref_objectid, u64 objectid, |
6101 | umode_t mode, u64 *index) | |
39279cc3 CM |
6102 | { |
6103 | struct inode *inode; | |
5f39d397 | 6104 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6105 | struct btrfs_key *location; |
5f39d397 | 6106 | struct btrfs_path *path; |
9c58309d CM |
6107 | struct btrfs_inode_ref *ref; |
6108 | struct btrfs_key key[2]; | |
6109 | u32 sizes[2]; | |
ef3b9af5 | 6110 | int nitems = name ? 2 : 1; |
9c58309d | 6111 | unsigned long ptr; |
39279cc3 | 6112 | int ret; |
39279cc3 | 6113 | |
5f39d397 | 6114 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6115 | if (!path) |
6116 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6117 | |
39279cc3 | 6118 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
6119 | if (!inode) { |
6120 | btrfs_free_path(path); | |
39279cc3 | 6121 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6122 | } |
39279cc3 | 6123 | |
5762b5c9 FM |
6124 | /* |
6125 | * O_TMPFILE, set link count to 0, so that after this point, | |
6126 | * we fill in an inode item with the correct link count. | |
6127 | */ | |
6128 | if (!name) | |
6129 | set_nlink(inode, 0); | |
6130 | ||
581bb050 LZ |
6131 | /* |
6132 | * we have to initialize this early, so we can reclaim the inode | |
6133 | * number if we fail afterwards in this function. | |
6134 | */ | |
6135 | inode->i_ino = objectid; | |
6136 | ||
ef3b9af5 | 6137 | if (dir && name) { |
1abe9b8a | 6138 | trace_btrfs_inode_request(dir); |
6139 | ||
3de4586c | 6140 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 6141 | if (ret) { |
8fb27640 | 6142 | btrfs_free_path(path); |
09771430 | 6143 | iput(inode); |
aec7477b | 6144 | return ERR_PTR(ret); |
09771430 | 6145 | } |
ef3b9af5 FM |
6146 | } else if (dir) { |
6147 | *index = 0; | |
aec7477b JB |
6148 | } |
6149 | /* | |
6150 | * index_cnt is ignored for everything but a dir, | |
6151 | * btrfs_get_inode_index_count has an explanation for the magic | |
6152 | * number | |
6153 | */ | |
6154 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6155 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6156 | BTRFS_I(inode)->root = root; |
e02119d5 | 6157 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6158 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6159 | |
5dc562c5 JB |
6160 | /* |
6161 | * We could have gotten an inode number from somebody who was fsynced | |
6162 | * and then removed in this same transaction, so let's just set full | |
6163 | * sync since it will be a full sync anyway and this will blow away the | |
6164 | * old info in the log. | |
6165 | */ | |
6166 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6167 | ||
9c58309d | 6168 | key[0].objectid = objectid; |
962a298f | 6169 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6170 | key[0].offset = 0; |
6171 | ||
9c58309d | 6172 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6173 | |
6174 | if (name) { | |
6175 | /* | |
6176 | * Start new inodes with an inode_ref. This is slightly more | |
6177 | * efficient for small numbers of hard links since they will | |
6178 | * be packed into one item. Extended refs will kick in if we | |
6179 | * add more hard links than can fit in the ref item. | |
6180 | */ | |
6181 | key[1].objectid = objectid; | |
962a298f | 6182 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6183 | key[1].offset = ref_objectid; |
6184 | ||
6185 | sizes[1] = name_len + sizeof(*ref); | |
6186 | } | |
9c58309d | 6187 | |
b0d5d10f CM |
6188 | location = &BTRFS_I(inode)->location; |
6189 | location->objectid = objectid; | |
6190 | location->offset = 0; | |
962a298f | 6191 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6192 | |
6193 | ret = btrfs_insert_inode_locked(inode); | |
6194 | if (ret < 0) | |
6195 | goto fail; | |
6196 | ||
b9473439 | 6197 | path->leave_spinning = 1; |
ef3b9af5 | 6198 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6199 | if (ret != 0) |
b0d5d10f | 6200 | goto fail_unlock; |
5f39d397 | 6201 | |
ecc11fab | 6202 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6203 | inode_set_bytes(inode, 0); |
9cc97d64 | 6204 | |
04b285f3 | 6205 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 6206 | inode->i_atime = inode->i_mtime; |
6207 | inode->i_ctime = inode->i_mtime; | |
6208 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6209 | ||
5f39d397 CM |
6210 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6211 | struct btrfs_inode_item); | |
293f7e07 LZ |
6212 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
6213 | sizeof(*inode_item)); | |
e02119d5 | 6214 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6215 | |
ef3b9af5 FM |
6216 | if (name) { |
6217 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6218 | struct btrfs_inode_ref); | |
6219 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6220 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6221 | ptr = (unsigned long)(ref + 1); | |
6222 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6223 | } | |
9c58309d | 6224 | |
5f39d397 CM |
6225 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6226 | btrfs_free_path(path); | |
6227 | ||
6cbff00f CH |
6228 | btrfs_inherit_iflags(inode, dir); |
6229 | ||
569254b0 | 6230 | if (S_ISREG(mode)) { |
94272164 CM |
6231 | if (btrfs_test_opt(root, NODATASUM)) |
6232 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 6233 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
6234 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6235 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6236 | } |
6237 | ||
5d4f98a2 | 6238 | inode_tree_add(inode); |
1abe9b8a | 6239 | |
6240 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6241 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6242 | |
8ea05e3a AB |
6243 | btrfs_update_root_times(trans, root); |
6244 | ||
63541927 FDBM |
6245 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6246 | if (ret) | |
6247 | btrfs_err(root->fs_info, | |
6248 | "error inheriting props for ino %llu (root %llu): %d", | |
6249 | btrfs_ino(inode), root->root_key.objectid, ret); | |
6250 | ||
39279cc3 | 6251 | return inode; |
b0d5d10f CM |
6252 | |
6253 | fail_unlock: | |
6254 | unlock_new_inode(inode); | |
5f39d397 | 6255 | fail: |
ef3b9af5 | 6256 | if (dir && name) |
aec7477b | 6257 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6258 | btrfs_free_path(path); |
09771430 | 6259 | iput(inode); |
5f39d397 | 6260 | return ERR_PTR(ret); |
39279cc3 CM |
6261 | } |
6262 | ||
6263 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6264 | { | |
6265 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6266 | } | |
6267 | ||
d352ac68 CM |
6268 | /* |
6269 | * utility function to add 'inode' into 'parent_inode' with | |
6270 | * a give name and a given sequence number. | |
6271 | * if 'add_backref' is true, also insert a backref from the | |
6272 | * inode to the parent directory. | |
6273 | */ | |
e02119d5 CM |
6274 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
6275 | struct inode *parent_inode, struct inode *inode, | |
6276 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 6277 | { |
4df27c4d | 6278 | int ret = 0; |
39279cc3 | 6279 | struct btrfs_key key; |
e02119d5 | 6280 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
6281 | u64 ino = btrfs_ino(inode); |
6282 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6283 | |
33345d01 | 6284 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6285 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
6286 | } else { | |
33345d01 | 6287 | key.objectid = ino; |
962a298f | 6288 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6289 | key.offset = 0; |
6290 | } | |
6291 | ||
33345d01 | 6292 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6293 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
6294 | key.objectid, root->root_key.objectid, | |
33345d01 | 6295 | parent_ino, index, name, name_len); |
4df27c4d | 6296 | } else if (add_backref) { |
33345d01 LZ |
6297 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6298 | parent_ino, index); | |
4df27c4d | 6299 | } |
39279cc3 | 6300 | |
79787eaa JM |
6301 | /* Nothing to clean up yet */ |
6302 | if (ret) | |
6303 | return ret; | |
4df27c4d | 6304 | |
79787eaa JM |
6305 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6306 | parent_inode, &key, | |
6307 | btrfs_inode_type(inode), index); | |
9c52057c | 6308 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6309 | goto fail_dir_item; |
6310 | else if (ret) { | |
6311 | btrfs_abort_transaction(trans, root, ret); | |
6312 | return ret; | |
39279cc3 | 6313 | } |
79787eaa JM |
6314 | |
6315 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6316 | name_len * 2); | |
0c4d2d95 | 6317 | inode_inc_iversion(parent_inode); |
04b285f3 DD |
6318 | parent_inode->i_mtime = parent_inode->i_ctime = |
6319 | current_fs_time(parent_inode->i_sb); | |
79787eaa JM |
6320 | ret = btrfs_update_inode(trans, root, parent_inode); |
6321 | if (ret) | |
6322 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 6323 | return ret; |
fe66a05a CM |
6324 | |
6325 | fail_dir_item: | |
6326 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6327 | u64 local_index; | |
6328 | int err; | |
6329 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
6330 | key.objectid, root->root_key.objectid, | |
6331 | parent_ino, &local_index, name, name_len); | |
6332 | ||
6333 | } else if (add_backref) { | |
6334 | u64 local_index; | |
6335 | int err; | |
6336 | ||
6337 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6338 | ino, parent_ino, &local_index); | |
6339 | } | |
6340 | return ret; | |
39279cc3 CM |
6341 | } |
6342 | ||
6343 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6344 | struct inode *dir, struct dentry *dentry, |
6345 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6346 | { |
a1b075d2 JB |
6347 | int err = btrfs_add_link(trans, dir, inode, |
6348 | dentry->d_name.name, dentry->d_name.len, | |
6349 | backref, index); | |
39279cc3 CM |
6350 | if (err > 0) |
6351 | err = -EEXIST; | |
6352 | return err; | |
6353 | } | |
6354 | ||
618e21d5 | 6355 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6356 | umode_t mode, dev_t rdev) |
618e21d5 JB |
6357 | { |
6358 | struct btrfs_trans_handle *trans; | |
6359 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6360 | struct inode *inode = NULL; |
618e21d5 JB |
6361 | int err; |
6362 | int drop_inode = 0; | |
6363 | u64 objectid; | |
00e4e6b3 | 6364 | u64 index = 0; |
618e21d5 | 6365 | |
9ed74f2d JB |
6366 | /* |
6367 | * 2 for inode item and ref | |
6368 | * 2 for dir items | |
6369 | * 1 for xattr if selinux is on | |
6370 | */ | |
a22285a6 YZ |
6371 | trans = btrfs_start_transaction(root, 5); |
6372 | if (IS_ERR(trans)) | |
6373 | return PTR_ERR(trans); | |
1832a6d5 | 6374 | |
581bb050 LZ |
6375 | err = btrfs_find_free_ino(root, &objectid); |
6376 | if (err) | |
6377 | goto out_unlock; | |
6378 | ||
aec7477b | 6379 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6380 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6381 | mode, &index); |
7cf96da3 TI |
6382 | if (IS_ERR(inode)) { |
6383 | err = PTR_ERR(inode); | |
618e21d5 | 6384 | goto out_unlock; |
7cf96da3 | 6385 | } |
618e21d5 | 6386 | |
ad19db71 CS |
6387 | /* |
6388 | * If the active LSM wants to access the inode during | |
6389 | * d_instantiate it needs these. Smack checks to see | |
6390 | * if the filesystem supports xattrs by looking at the | |
6391 | * ops vector. | |
6392 | */ | |
ad19db71 | 6393 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6394 | init_special_inode(inode, inode->i_mode, rdev); |
6395 | ||
6396 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6397 | if (err) |
b0d5d10f CM |
6398 | goto out_unlock_inode; |
6399 | ||
6400 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6401 | if (err) { | |
6402 | goto out_unlock_inode; | |
6403 | } else { | |
1b4ab1bb | 6404 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6405 | unlock_new_inode(inode); |
08c422c2 | 6406 | d_instantiate(dentry, inode); |
618e21d5 | 6407 | } |
b0d5d10f | 6408 | |
618e21d5 | 6409 | out_unlock: |
7ad85bb7 | 6410 | btrfs_end_transaction(trans, root); |
c581afc8 | 6411 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6412 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6413 | if (drop_inode) { |
6414 | inode_dec_link_count(inode); | |
6415 | iput(inode); | |
6416 | } | |
618e21d5 | 6417 | return err; |
b0d5d10f CM |
6418 | |
6419 | out_unlock_inode: | |
6420 | drop_inode = 1; | |
6421 | unlock_new_inode(inode); | |
6422 | goto out_unlock; | |
6423 | ||
618e21d5 JB |
6424 | } |
6425 | ||
39279cc3 | 6426 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6427 | umode_t mode, bool excl) |
39279cc3 CM |
6428 | { |
6429 | struct btrfs_trans_handle *trans; | |
6430 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6431 | struct inode *inode = NULL; |
43baa579 | 6432 | int drop_inode_on_err = 0; |
a22285a6 | 6433 | int err; |
39279cc3 | 6434 | u64 objectid; |
00e4e6b3 | 6435 | u64 index = 0; |
39279cc3 | 6436 | |
9ed74f2d JB |
6437 | /* |
6438 | * 2 for inode item and ref | |
6439 | * 2 for dir items | |
6440 | * 1 for xattr if selinux is on | |
6441 | */ | |
a22285a6 YZ |
6442 | trans = btrfs_start_transaction(root, 5); |
6443 | if (IS_ERR(trans)) | |
6444 | return PTR_ERR(trans); | |
9ed74f2d | 6445 | |
581bb050 LZ |
6446 | err = btrfs_find_free_ino(root, &objectid); |
6447 | if (err) | |
6448 | goto out_unlock; | |
6449 | ||
aec7477b | 6450 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6451 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6452 | mode, &index); |
7cf96da3 TI |
6453 | if (IS_ERR(inode)) { |
6454 | err = PTR_ERR(inode); | |
39279cc3 | 6455 | goto out_unlock; |
7cf96da3 | 6456 | } |
43baa579 | 6457 | drop_inode_on_err = 1; |
ad19db71 CS |
6458 | /* |
6459 | * If the active LSM wants to access the inode during | |
6460 | * d_instantiate it needs these. Smack checks to see | |
6461 | * if the filesystem supports xattrs by looking at the | |
6462 | * ops vector. | |
6463 | */ | |
6464 | inode->i_fop = &btrfs_file_operations; | |
6465 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6466 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6467 | |
6468 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6469 | if (err) | |
6470 | goto out_unlock_inode; | |
6471 | ||
6472 | err = btrfs_update_inode(trans, root, inode); | |
6473 | if (err) | |
6474 | goto out_unlock_inode; | |
ad19db71 | 6475 | |
a1b075d2 | 6476 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6477 | if (err) |
b0d5d10f | 6478 | goto out_unlock_inode; |
43baa579 | 6479 | |
43baa579 | 6480 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6481 | unlock_new_inode(inode); |
43baa579 FB |
6482 | d_instantiate(dentry, inode); |
6483 | ||
39279cc3 | 6484 | out_unlock: |
7ad85bb7 | 6485 | btrfs_end_transaction(trans, root); |
43baa579 | 6486 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6487 | inode_dec_link_count(inode); |
6488 | iput(inode); | |
6489 | } | |
c581afc8 | 6490 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6491 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6492 | return err; |
b0d5d10f CM |
6493 | |
6494 | out_unlock_inode: | |
6495 | unlock_new_inode(inode); | |
6496 | goto out_unlock; | |
6497 | ||
39279cc3 CM |
6498 | } |
6499 | ||
6500 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6501 | struct dentry *dentry) | |
6502 | { | |
271dba45 | 6503 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6504 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6505 | struct inode *inode = d_inode(old_dentry); |
00e4e6b3 | 6506 | u64 index; |
39279cc3 CM |
6507 | int err; |
6508 | int drop_inode = 0; | |
6509 | ||
4a8be425 TH |
6510 | /* do not allow sys_link's with other subvols of the same device */ |
6511 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6512 | return -EXDEV; |
4a8be425 | 6513 | |
f186373f | 6514 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6515 | return -EMLINK; |
4a8be425 | 6516 | |
3de4586c | 6517 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6518 | if (err) |
6519 | goto fail; | |
6520 | ||
a22285a6 | 6521 | /* |
7e6b6465 | 6522 | * 2 items for inode and inode ref |
a22285a6 | 6523 | * 2 items for dir items |
7e6b6465 | 6524 | * 1 item for parent inode |
a22285a6 | 6525 | */ |
7e6b6465 | 6526 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6527 | if (IS_ERR(trans)) { |
6528 | err = PTR_ERR(trans); | |
271dba45 | 6529 | trans = NULL; |
a22285a6 YZ |
6530 | goto fail; |
6531 | } | |
5f39d397 | 6532 | |
67de1176 MX |
6533 | /* There are several dir indexes for this inode, clear the cache. */ |
6534 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6535 | inc_nlink(inode); |
0c4d2d95 | 6536 | inode_inc_iversion(inode); |
04b285f3 | 6537 | inode->i_ctime = current_fs_time(inode->i_sb); |
7de9c6ee | 6538 | ihold(inode); |
e9976151 | 6539 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6540 | |
a1b075d2 | 6541 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6542 | |
a5719521 | 6543 | if (err) { |
54aa1f4d | 6544 | drop_inode = 1; |
a5719521 | 6545 | } else { |
10d9f309 | 6546 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6547 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6548 | if (err) |
6549 | goto fail; | |
ef3b9af5 FM |
6550 | if (inode->i_nlink == 1) { |
6551 | /* | |
6552 | * If new hard link count is 1, it's a file created | |
6553 | * with open(2) O_TMPFILE flag. | |
6554 | */ | |
6555 | err = btrfs_orphan_del(trans, inode); | |
6556 | if (err) | |
6557 | goto fail; | |
6558 | } | |
08c422c2 | 6559 | d_instantiate(dentry, inode); |
6a912213 | 6560 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6561 | } |
39279cc3 | 6562 | |
c581afc8 | 6563 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6564 | fail: |
271dba45 FM |
6565 | if (trans) |
6566 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
6567 | if (drop_inode) { |
6568 | inode_dec_link_count(inode); | |
6569 | iput(inode); | |
6570 | } | |
b53d3f5d | 6571 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6572 | return err; |
6573 | } | |
6574 | ||
18bb1db3 | 6575 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6576 | { |
b9d86667 | 6577 | struct inode *inode = NULL; |
39279cc3 CM |
6578 | struct btrfs_trans_handle *trans; |
6579 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6580 | int err = 0; | |
6581 | int drop_on_err = 0; | |
b9d86667 | 6582 | u64 objectid = 0; |
00e4e6b3 | 6583 | u64 index = 0; |
39279cc3 | 6584 | |
9ed74f2d JB |
6585 | /* |
6586 | * 2 items for inode and ref | |
6587 | * 2 items for dir items | |
6588 | * 1 for xattr if selinux is on | |
6589 | */ | |
a22285a6 YZ |
6590 | trans = btrfs_start_transaction(root, 5); |
6591 | if (IS_ERR(trans)) | |
6592 | return PTR_ERR(trans); | |
39279cc3 | 6593 | |
581bb050 LZ |
6594 | err = btrfs_find_free_ino(root, &objectid); |
6595 | if (err) | |
6596 | goto out_fail; | |
6597 | ||
aec7477b | 6598 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6599 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6600 | S_IFDIR | mode, &index); |
39279cc3 CM |
6601 | if (IS_ERR(inode)) { |
6602 | err = PTR_ERR(inode); | |
6603 | goto out_fail; | |
6604 | } | |
5f39d397 | 6605 | |
39279cc3 | 6606 | drop_on_err = 1; |
b0d5d10f CM |
6607 | /* these must be set before we unlock the inode */ |
6608 | inode->i_op = &btrfs_dir_inode_operations; | |
6609 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6610 | |
2a7dba39 | 6611 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6612 | if (err) |
b0d5d10f | 6613 | goto out_fail_inode; |
39279cc3 | 6614 | |
dbe674a9 | 6615 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6616 | err = btrfs_update_inode(trans, root, inode); |
6617 | if (err) | |
b0d5d10f | 6618 | goto out_fail_inode; |
5f39d397 | 6619 | |
a1b075d2 JB |
6620 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6621 | dentry->d_name.len, 0, index); | |
39279cc3 | 6622 | if (err) |
b0d5d10f | 6623 | goto out_fail_inode; |
5f39d397 | 6624 | |
39279cc3 | 6625 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6626 | /* |
6627 | * mkdir is special. We're unlocking after we call d_instantiate | |
6628 | * to avoid a race with nfsd calling d_instantiate. | |
6629 | */ | |
6630 | unlock_new_inode(inode); | |
39279cc3 | 6631 | drop_on_err = 0; |
39279cc3 CM |
6632 | |
6633 | out_fail: | |
7ad85bb7 | 6634 | btrfs_end_transaction(trans, root); |
c7cfb8a5 WS |
6635 | if (drop_on_err) { |
6636 | inode_dec_link_count(inode); | |
39279cc3 | 6637 | iput(inode); |
c7cfb8a5 | 6638 | } |
c581afc8 | 6639 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6640 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6641 | return err; |
b0d5d10f CM |
6642 | |
6643 | out_fail_inode: | |
6644 | unlock_new_inode(inode); | |
6645 | goto out_fail; | |
39279cc3 CM |
6646 | } |
6647 | ||
e6c4efd8 QW |
6648 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6649 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6650 | { | |
6651 | struct rb_node *next; | |
6652 | ||
6653 | next = rb_next(&em->rb_node); | |
6654 | if (!next) | |
6655 | return NULL; | |
6656 | return container_of(next, struct extent_map, rb_node); | |
6657 | } | |
6658 | ||
6659 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6660 | { | |
6661 | struct rb_node *prev; | |
6662 | ||
6663 | prev = rb_prev(&em->rb_node); | |
6664 | if (!prev) | |
6665 | return NULL; | |
6666 | return container_of(prev, struct extent_map, rb_node); | |
6667 | } | |
6668 | ||
d352ac68 | 6669 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6670 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6671 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6672 | * the best fitted new extent into the tree. |
d352ac68 | 6673 | */ |
3b951516 CM |
6674 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6675 | struct extent_map *existing, | |
e6dcd2dc | 6676 | struct extent_map *em, |
51f395ad | 6677 | u64 map_start) |
3b951516 | 6678 | { |
e6c4efd8 QW |
6679 | struct extent_map *prev; |
6680 | struct extent_map *next; | |
6681 | u64 start; | |
6682 | u64 end; | |
3b951516 | 6683 | u64 start_diff; |
3b951516 | 6684 | |
e6dcd2dc | 6685 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6686 | |
6687 | if (existing->start > map_start) { | |
6688 | next = existing; | |
6689 | prev = prev_extent_map(next); | |
6690 | } else { | |
6691 | prev = existing; | |
6692 | next = next_extent_map(prev); | |
6693 | } | |
6694 | ||
6695 | start = prev ? extent_map_end(prev) : em->start; | |
6696 | start = max_t(u64, start, em->start); | |
6697 | end = next ? next->start : extent_map_end(em); | |
6698 | end = min_t(u64, end, extent_map_end(em)); | |
6699 | start_diff = start - em->start; | |
6700 | em->start = start; | |
6701 | em->len = end - start; | |
c8b97818 CM |
6702 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6703 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6704 | em->block_start += start_diff; |
c8b97818 CM |
6705 | em->block_len -= start_diff; |
6706 | } | |
09a2a8f9 | 6707 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6708 | } |
6709 | ||
c8b97818 | 6710 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6711 | struct page *page, |
c8b97818 CM |
6712 | size_t pg_offset, u64 extent_offset, |
6713 | struct btrfs_file_extent_item *item) | |
6714 | { | |
6715 | int ret; | |
6716 | struct extent_buffer *leaf = path->nodes[0]; | |
6717 | char *tmp; | |
6718 | size_t max_size; | |
6719 | unsigned long inline_size; | |
6720 | unsigned long ptr; | |
261507a0 | 6721 | int compress_type; |
c8b97818 CM |
6722 | |
6723 | WARN_ON(pg_offset != 0); | |
261507a0 | 6724 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6725 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6726 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6727 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6728 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6729 | if (!tmp) |
6730 | return -ENOMEM; | |
c8b97818 CM |
6731 | ptr = btrfs_file_extent_inline_start(item); |
6732 | ||
6733 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6734 | ||
09cbfeaf | 6735 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6736 | ret = btrfs_decompress(compress_type, tmp, page, |
6737 | extent_offset, inline_size, max_size); | |
c8b97818 | 6738 | kfree(tmp); |
166ae5a4 | 6739 | return ret; |
c8b97818 CM |
6740 | } |
6741 | ||
d352ac68 CM |
6742 | /* |
6743 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6744 | * the ugly parts come from merging extents from the disk with the in-ram |
6745 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6746 | * where the in-ram extents might be locked pending data=ordered completion. |
6747 | * | |
6748 | * This also copies inline extents directly into the page. | |
6749 | */ | |
d397712b | 6750 | |
a52d9a80 | 6751 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6752 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6753 | int create) |
6754 | { | |
6755 | int ret; | |
6756 | int err = 0; | |
a52d9a80 CM |
6757 | u64 extent_start = 0; |
6758 | u64 extent_end = 0; | |
33345d01 | 6759 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6760 | u32 found_type; |
f421950f | 6761 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6762 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6763 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6764 | struct extent_buffer *leaf; |
6765 | struct btrfs_key found_key; | |
a52d9a80 CM |
6766 | struct extent_map *em = NULL; |
6767 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6768 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6769 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6770 | const bool new_inline = !page || create; |
a52d9a80 | 6771 | |
a52d9a80 | 6772 | again: |
890871be | 6773 | read_lock(&em_tree->lock); |
d1310b2e | 6774 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6775 | if (em) |
6776 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6777 | read_unlock(&em_tree->lock); |
d1310b2e | 6778 | |
a52d9a80 | 6779 | if (em) { |
e1c4b745 CM |
6780 | if (em->start > start || em->start + em->len <= start) |
6781 | free_extent_map(em); | |
6782 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6783 | free_extent_map(em); |
6784 | else | |
6785 | goto out; | |
a52d9a80 | 6786 | } |
172ddd60 | 6787 | em = alloc_extent_map(); |
a52d9a80 | 6788 | if (!em) { |
d1310b2e CM |
6789 | err = -ENOMEM; |
6790 | goto out; | |
a52d9a80 | 6791 | } |
e6dcd2dc | 6792 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6793 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6794 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6795 | em->len = (u64)-1; |
c8b97818 | 6796 | em->block_len = (u64)-1; |
f421950f CM |
6797 | |
6798 | if (!path) { | |
6799 | path = btrfs_alloc_path(); | |
026fd317 JB |
6800 | if (!path) { |
6801 | err = -ENOMEM; | |
6802 | goto out; | |
6803 | } | |
6804 | /* | |
6805 | * Chances are we'll be called again, so go ahead and do | |
6806 | * readahead | |
6807 | */ | |
e4058b54 | 6808 | path->reada = READA_FORWARD; |
f421950f CM |
6809 | } |
6810 | ||
179e29e4 CM |
6811 | ret = btrfs_lookup_file_extent(trans, root, path, |
6812 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6813 | if (ret < 0) { |
6814 | err = ret; | |
6815 | goto out; | |
6816 | } | |
6817 | ||
6818 | if (ret != 0) { | |
6819 | if (path->slots[0] == 0) | |
6820 | goto not_found; | |
6821 | path->slots[0]--; | |
6822 | } | |
6823 | ||
5f39d397 CM |
6824 | leaf = path->nodes[0]; |
6825 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6826 | struct btrfs_file_extent_item); |
a52d9a80 | 6827 | /* are we inside the extent that was found? */ |
5f39d397 | 6828 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6829 | found_type = found_key.type; |
5f39d397 | 6830 | if (found_key.objectid != objectid || |
a52d9a80 | 6831 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6832 | /* |
6833 | * If we backup past the first extent we want to move forward | |
6834 | * and see if there is an extent in front of us, otherwise we'll | |
6835 | * say there is a hole for our whole search range which can | |
6836 | * cause problems. | |
6837 | */ | |
6838 | extent_end = start; | |
6839 | goto next; | |
a52d9a80 CM |
6840 | } |
6841 | ||
5f39d397 CM |
6842 | found_type = btrfs_file_extent_type(leaf, item); |
6843 | extent_start = found_key.offset; | |
d899e052 YZ |
6844 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6845 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6846 | extent_end = extent_start + |
db94535d | 6847 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6848 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6849 | size_t size; | |
514ac8ad | 6850 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6851 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6852 | } |
25a50341 | 6853 | next: |
9036c102 YZ |
6854 | if (start >= extent_end) { |
6855 | path->slots[0]++; | |
6856 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6857 | ret = btrfs_next_leaf(root, path); | |
6858 | if (ret < 0) { | |
6859 | err = ret; | |
6860 | goto out; | |
a52d9a80 | 6861 | } |
9036c102 YZ |
6862 | if (ret > 0) |
6863 | goto not_found; | |
6864 | leaf = path->nodes[0]; | |
a52d9a80 | 6865 | } |
9036c102 YZ |
6866 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6867 | if (found_key.objectid != objectid || | |
6868 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6869 | goto not_found; | |
6870 | if (start + len <= found_key.offset) | |
6871 | goto not_found; | |
e2eca69d WS |
6872 | if (start > found_key.offset) |
6873 | goto next; | |
9036c102 | 6874 | em->start = start; |
70c8a91c | 6875 | em->orig_start = start; |
9036c102 YZ |
6876 | em->len = found_key.offset - start; |
6877 | goto not_found_em; | |
6878 | } | |
6879 | ||
7ffbb598 FM |
6880 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6881 | ||
d899e052 YZ |
6882 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6883 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6884 | goto insert; |
6885 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6886 | unsigned long ptr; |
a52d9a80 | 6887 | char *map; |
3326d1b0 CM |
6888 | size_t size; |
6889 | size_t extent_offset; | |
6890 | size_t copy_size; | |
a52d9a80 | 6891 | |
7ffbb598 | 6892 | if (new_inline) |
689f9346 | 6893 | goto out; |
5f39d397 | 6894 | |
514ac8ad | 6895 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6896 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6897 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6898 | size - extent_offset); | |
3326d1b0 | 6899 | em->start = extent_start + extent_offset; |
fda2832f | 6900 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6901 | em->orig_block_len = em->len; |
70c8a91c | 6902 | em->orig_start = em->start; |
689f9346 | 6903 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6904 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6905 | if (btrfs_file_extent_compression(leaf, item) != |
6906 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6907 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6908 | extent_offset, item); |
166ae5a4 ZB |
6909 | if (ret) { |
6910 | err = ret; | |
6911 | goto out; | |
6912 | } | |
c8b97818 CM |
6913 | } else { |
6914 | map = kmap(page); | |
6915 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6916 | copy_size); | |
09cbfeaf | 6917 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6918 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6919 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6920 | copy_size); |
6921 | } | |
c8b97818 CM |
6922 | kunmap(page); |
6923 | } | |
179e29e4 CM |
6924 | flush_dcache_page(page); |
6925 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6926 | BUG(); |
179e29e4 CM |
6927 | if (!trans) { |
6928 | kunmap(page); | |
6929 | free_extent_map(em); | |
6930 | em = NULL; | |
ff5714cc | 6931 | |
b3b4aa74 | 6932 | btrfs_release_path(path); |
7a7eaa40 | 6933 | trans = btrfs_join_transaction(root); |
ff5714cc | 6934 | |
3612b495 TI |
6935 | if (IS_ERR(trans)) |
6936 | return ERR_CAST(trans); | |
179e29e4 CM |
6937 | goto again; |
6938 | } | |
c8b97818 | 6939 | map = kmap(page); |
70dec807 | 6940 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6941 | copy_size); |
c8b97818 | 6942 | kunmap(page); |
179e29e4 | 6943 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6944 | } |
d1310b2e | 6945 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6946 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6947 | goto insert; |
a52d9a80 CM |
6948 | } |
6949 | not_found: | |
6950 | em->start = start; | |
70c8a91c | 6951 | em->orig_start = start; |
d1310b2e | 6952 | em->len = len; |
a52d9a80 | 6953 | not_found_em: |
5f39d397 | 6954 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6955 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6956 | insert: |
b3b4aa74 | 6957 | btrfs_release_path(path); |
d1310b2e | 6958 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6959 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6960 | em->start, em->len, start, len); |
a52d9a80 CM |
6961 | err = -EIO; |
6962 | goto out; | |
6963 | } | |
d1310b2e CM |
6964 | |
6965 | err = 0; | |
890871be | 6966 | write_lock(&em_tree->lock); |
09a2a8f9 | 6967 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6968 | /* it is possible that someone inserted the extent into the tree |
6969 | * while we had the lock dropped. It is also possible that | |
6970 | * an overlapping map exists in the tree | |
6971 | */ | |
a52d9a80 | 6972 | if (ret == -EEXIST) { |
3b951516 | 6973 | struct extent_map *existing; |
e6dcd2dc CM |
6974 | |
6975 | ret = 0; | |
6976 | ||
e6c4efd8 QW |
6977 | existing = search_extent_mapping(em_tree, start, len); |
6978 | /* | |
6979 | * existing will always be non-NULL, since there must be | |
6980 | * extent causing the -EEXIST. | |
6981 | */ | |
8dff9c85 CM |
6982 | if (existing->start == em->start && |
6983 | extent_map_end(existing) == extent_map_end(em) && | |
6984 | em->block_start == existing->block_start) { | |
6985 | /* | |
6986 | * these two extents are the same, it happens | |
6987 | * with inlines especially | |
6988 | */ | |
6989 | free_extent_map(em); | |
6990 | em = existing; | |
6991 | err = 0; | |
6992 | ||
6993 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 6994 | start <= existing->start) { |
e6c4efd8 QW |
6995 | /* |
6996 | * The existing extent map is the one nearest to | |
6997 | * the [start, start + len) range which overlaps | |
6998 | */ | |
6999 | err = merge_extent_mapping(em_tree, existing, | |
7000 | em, start); | |
e1c4b745 | 7001 | free_extent_map(existing); |
e6c4efd8 | 7002 | if (err) { |
3b951516 CM |
7003 | free_extent_map(em); |
7004 | em = NULL; | |
7005 | } | |
7006 | } else { | |
7007 | free_extent_map(em); | |
7008 | em = existing; | |
e6dcd2dc | 7009 | err = 0; |
a52d9a80 | 7010 | } |
a52d9a80 | 7011 | } |
890871be | 7012 | write_unlock(&em_tree->lock); |
a52d9a80 | 7013 | out: |
1abe9b8a | 7014 | |
4cd8587c | 7015 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 7016 | |
527afb44 | 7017 | btrfs_free_path(path); |
a52d9a80 CM |
7018 | if (trans) { |
7019 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 7020 | if (!err) |
a52d9a80 CM |
7021 | err = ret; |
7022 | } | |
a52d9a80 CM |
7023 | if (err) { |
7024 | free_extent_map(em); | |
a52d9a80 CM |
7025 | return ERR_PTR(err); |
7026 | } | |
79787eaa | 7027 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7028 | return em; |
7029 | } | |
7030 | ||
ec29ed5b CM |
7031 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
7032 | size_t pg_offset, u64 start, u64 len, | |
7033 | int create) | |
7034 | { | |
7035 | struct extent_map *em; | |
7036 | struct extent_map *hole_em = NULL; | |
7037 | u64 range_start = start; | |
7038 | u64 end; | |
7039 | u64 found; | |
7040 | u64 found_end; | |
7041 | int err = 0; | |
7042 | ||
7043 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7044 | if (IS_ERR(em)) | |
7045 | return em; | |
7046 | if (em) { | |
7047 | /* | |
f9e4fb53 LB |
7048 | * if our em maps to |
7049 | * - a hole or | |
7050 | * - a pre-alloc extent, | |
7051 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7052 | */ |
f9e4fb53 LB |
7053 | if (em->block_start != EXTENT_MAP_HOLE && |
7054 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7055 | return em; |
7056 | else | |
7057 | hole_em = em; | |
7058 | } | |
7059 | ||
7060 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7061 | end = start + len; | |
7062 | if (end < start) | |
7063 | end = (u64)-1; | |
7064 | else | |
7065 | end -= 1; | |
7066 | ||
7067 | em = NULL; | |
7068 | ||
7069 | /* ok, we didn't find anything, lets look for delalloc */ | |
7070 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
7071 | end, len, EXTENT_DELALLOC, 1); | |
7072 | found_end = range_start + found; | |
7073 | if (found_end < range_start) | |
7074 | found_end = (u64)-1; | |
7075 | ||
7076 | /* | |
7077 | * we didn't find anything useful, return | |
7078 | * the original results from get_extent() | |
7079 | */ | |
7080 | if (range_start > end || found_end <= start) { | |
7081 | em = hole_em; | |
7082 | hole_em = NULL; | |
7083 | goto out; | |
7084 | } | |
7085 | ||
7086 | /* adjust the range_start to make sure it doesn't | |
7087 | * go backwards from the start they passed in | |
7088 | */ | |
67871254 | 7089 | range_start = max(start, range_start); |
ec29ed5b CM |
7090 | found = found_end - range_start; |
7091 | ||
7092 | if (found > 0) { | |
7093 | u64 hole_start = start; | |
7094 | u64 hole_len = len; | |
7095 | ||
172ddd60 | 7096 | em = alloc_extent_map(); |
ec29ed5b CM |
7097 | if (!em) { |
7098 | err = -ENOMEM; | |
7099 | goto out; | |
7100 | } | |
7101 | /* | |
7102 | * when btrfs_get_extent can't find anything it | |
7103 | * returns one huge hole | |
7104 | * | |
7105 | * make sure what it found really fits our range, and | |
7106 | * adjust to make sure it is based on the start from | |
7107 | * the caller | |
7108 | */ | |
7109 | if (hole_em) { | |
7110 | u64 calc_end = extent_map_end(hole_em); | |
7111 | ||
7112 | if (calc_end <= start || (hole_em->start > end)) { | |
7113 | free_extent_map(hole_em); | |
7114 | hole_em = NULL; | |
7115 | } else { | |
7116 | hole_start = max(hole_em->start, start); | |
7117 | hole_len = calc_end - hole_start; | |
7118 | } | |
7119 | } | |
7120 | em->bdev = NULL; | |
7121 | if (hole_em && range_start > hole_start) { | |
7122 | /* our hole starts before our delalloc, so we | |
7123 | * have to return just the parts of the hole | |
7124 | * that go until the delalloc starts | |
7125 | */ | |
7126 | em->len = min(hole_len, | |
7127 | range_start - hole_start); | |
7128 | em->start = hole_start; | |
7129 | em->orig_start = hole_start; | |
7130 | /* | |
7131 | * don't adjust block start at all, | |
7132 | * it is fixed at EXTENT_MAP_HOLE | |
7133 | */ | |
7134 | em->block_start = hole_em->block_start; | |
7135 | em->block_len = hole_len; | |
f9e4fb53 LB |
7136 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7137 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7138 | } else { |
7139 | em->start = range_start; | |
7140 | em->len = found; | |
7141 | em->orig_start = range_start; | |
7142 | em->block_start = EXTENT_MAP_DELALLOC; | |
7143 | em->block_len = found; | |
7144 | } | |
7145 | } else if (hole_em) { | |
7146 | return hole_em; | |
7147 | } | |
7148 | out: | |
7149 | ||
7150 | free_extent_map(hole_em); | |
7151 | if (err) { | |
7152 | free_extent_map(em); | |
7153 | return ERR_PTR(err); | |
7154 | } | |
7155 | return em; | |
7156 | } | |
7157 | ||
5f9a8a51 FM |
7158 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7159 | const u64 start, | |
7160 | const u64 len, | |
7161 | const u64 orig_start, | |
7162 | const u64 block_start, | |
7163 | const u64 block_len, | |
7164 | const u64 orig_block_len, | |
7165 | const u64 ram_bytes, | |
7166 | const int type) | |
7167 | { | |
7168 | struct extent_map *em = NULL; | |
7169 | int ret; | |
7170 | ||
7171 | down_read(&BTRFS_I(inode)->dio_sem); | |
7172 | if (type != BTRFS_ORDERED_NOCOW) { | |
7173 | em = create_pinned_em(inode, start, len, orig_start, | |
7174 | block_start, block_len, orig_block_len, | |
7175 | ram_bytes, type); | |
7176 | if (IS_ERR(em)) | |
7177 | goto out; | |
7178 | } | |
7179 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7180 | len, block_len, type); | |
7181 | if (ret) { | |
7182 | if (em) { | |
7183 | free_extent_map(em); | |
7184 | btrfs_drop_extent_cache(inode, start, | |
7185 | start + len - 1, 0); | |
7186 | } | |
7187 | em = ERR_PTR(ret); | |
7188 | } | |
7189 | out: | |
7190 | up_read(&BTRFS_I(inode)->dio_sem); | |
7191 | ||
7192 | return em; | |
7193 | } | |
7194 | ||
4b46fce2 JB |
7195 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7196 | u64 start, u64 len) | |
7197 | { | |
7198 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 7199 | struct extent_map *em; |
4b46fce2 JB |
7200 | struct btrfs_key ins; |
7201 | u64 alloc_hint; | |
7202 | int ret; | |
4b46fce2 | 7203 | |
4b46fce2 | 7204 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 7205 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 7206 | alloc_hint, &ins, 1, 1); |
00361589 JB |
7207 | if (ret) |
7208 | return ERR_PTR(ret); | |
4b46fce2 | 7209 | |
5f9a8a51 FM |
7210 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7211 | ins.objectid, ins.offset, ins.offset, | |
7212 | ins.offset, 0); | |
9cfa3e34 | 7213 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
5f9a8a51 | 7214 | if (IS_ERR(em)) |
e570fd27 | 7215 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
de0ee0ed | 7216 | |
4b46fce2 JB |
7217 | return em; |
7218 | } | |
7219 | ||
46bfbb5c CM |
7220 | /* |
7221 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7222 | * block must be cow'd | |
7223 | */ | |
00361589 | 7224 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7225 | u64 *orig_start, u64 *orig_block_len, |
7226 | u64 *ram_bytes) | |
46bfbb5c | 7227 | { |
00361589 | 7228 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
7229 | struct btrfs_path *path; |
7230 | int ret; | |
7231 | struct extent_buffer *leaf; | |
7232 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7233 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7234 | struct btrfs_file_extent_item *fi; |
7235 | struct btrfs_key key; | |
7236 | u64 disk_bytenr; | |
7237 | u64 backref_offset; | |
7238 | u64 extent_end; | |
7239 | u64 num_bytes; | |
7240 | int slot; | |
7241 | int found_type; | |
7ee9e440 | 7242 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7243 | |
46bfbb5c CM |
7244 | path = btrfs_alloc_path(); |
7245 | if (!path) | |
7246 | return -ENOMEM; | |
7247 | ||
00361589 | 7248 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
7249 | offset, 0); |
7250 | if (ret < 0) | |
7251 | goto out; | |
7252 | ||
7253 | slot = path->slots[0]; | |
7254 | if (ret == 1) { | |
7255 | if (slot == 0) { | |
7256 | /* can't find the item, must cow */ | |
7257 | ret = 0; | |
7258 | goto out; | |
7259 | } | |
7260 | slot--; | |
7261 | } | |
7262 | ret = 0; | |
7263 | leaf = path->nodes[0]; | |
7264 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 7265 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
7266 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7267 | /* not our file or wrong item type, must cow */ | |
7268 | goto out; | |
7269 | } | |
7270 | ||
7271 | if (key.offset > offset) { | |
7272 | /* Wrong offset, must cow */ | |
7273 | goto out; | |
7274 | } | |
7275 | ||
7276 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7277 | found_type = btrfs_file_extent_type(leaf, fi); | |
7278 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7279 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7280 | /* not a regular extent, must cow */ | |
7281 | goto out; | |
7282 | } | |
7ee9e440 JB |
7283 | |
7284 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7285 | goto out; | |
7286 | ||
e77751aa MX |
7287 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7288 | if (extent_end <= offset) | |
7289 | goto out; | |
7290 | ||
46bfbb5c | 7291 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7292 | if (disk_bytenr == 0) |
7293 | goto out; | |
7294 | ||
7295 | if (btrfs_file_extent_compression(leaf, fi) || | |
7296 | btrfs_file_extent_encryption(leaf, fi) || | |
7297 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7298 | goto out; | |
7299 | ||
46bfbb5c CM |
7300 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7301 | ||
7ee9e440 JB |
7302 | if (orig_start) { |
7303 | *orig_start = key.offset - backref_offset; | |
7304 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7305 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7306 | } | |
eb384b55 | 7307 | |
46bfbb5c CM |
7308 | if (btrfs_extent_readonly(root, disk_bytenr)) |
7309 | goto out; | |
7b2b7085 MX |
7310 | |
7311 | num_bytes = min(offset + *len, extent_end) - offset; | |
7312 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7313 | u64 range_end; | |
7314 | ||
7315 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
7316 | ret = test_range_bit(io_tree, offset, range_end, | |
7317 | EXTENT_DELALLOC, 0, NULL); | |
7318 | if (ret) { | |
7319 | ret = -EAGAIN; | |
7320 | goto out; | |
7321 | } | |
7322 | } | |
7323 | ||
1bda19eb | 7324 | btrfs_release_path(path); |
46bfbb5c CM |
7325 | |
7326 | /* | |
7327 | * look for other files referencing this extent, if we | |
7328 | * find any we must cow | |
7329 | */ | |
00361589 JB |
7330 | trans = btrfs_join_transaction(root); |
7331 | if (IS_ERR(trans)) { | |
7332 | ret = 0; | |
46bfbb5c | 7333 | goto out; |
00361589 JB |
7334 | } |
7335 | ||
7336 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
7337 | key.offset - backref_offset, disk_bytenr); | |
7338 | btrfs_end_transaction(trans, root); | |
7339 | if (ret) { | |
7340 | ret = 0; | |
7341 | goto out; | |
7342 | } | |
46bfbb5c CM |
7343 | |
7344 | /* | |
7345 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7346 | * in this extent we are about to write. If there | |
7347 | * are any csums in that range we have to cow in order | |
7348 | * to keep the csums correct | |
7349 | */ | |
7350 | disk_bytenr += backref_offset; | |
7351 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
7352 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
7353 | goto out; | |
7354 | /* | |
7355 | * all of the above have passed, it is safe to overwrite this extent | |
7356 | * without cow | |
7357 | */ | |
eb384b55 | 7358 | *len = num_bytes; |
46bfbb5c CM |
7359 | ret = 1; |
7360 | out: | |
7361 | btrfs_free_path(path); | |
7362 | return ret; | |
7363 | } | |
7364 | ||
fc4adbff AG |
7365 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7366 | { | |
7367 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7368 | int found = false; | |
7369 | void **pagep = NULL; | |
7370 | struct page *page = NULL; | |
7371 | int start_idx; | |
7372 | int end_idx; | |
7373 | ||
09cbfeaf | 7374 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7375 | |
7376 | /* | |
7377 | * end is the last byte in the last page. end == start is legal | |
7378 | */ | |
09cbfeaf | 7379 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7380 | |
7381 | rcu_read_lock(); | |
7382 | ||
7383 | /* Most of the code in this while loop is lifted from | |
7384 | * find_get_page. It's been modified to begin searching from a | |
7385 | * page and return just the first page found in that range. If the | |
7386 | * found idx is less than or equal to the end idx then we know that | |
7387 | * a page exists. If no pages are found or if those pages are | |
7388 | * outside of the range then we're fine (yay!) */ | |
7389 | while (page == NULL && | |
7390 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7391 | page = radix_tree_deref_slot(pagep); | |
7392 | if (unlikely(!page)) | |
7393 | break; | |
7394 | ||
7395 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7396 | if (radix_tree_deref_retry(page)) { |
7397 | page = NULL; | |
fc4adbff | 7398 | continue; |
809f9016 | 7399 | } |
fc4adbff AG |
7400 | /* |
7401 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7402 | * here as an exceptional entry: so return it without | |
7403 | * attempting to raise page count. | |
7404 | */ | |
6fdef6d4 | 7405 | page = NULL; |
fc4adbff AG |
7406 | break; /* TODO: Is this relevant for this use case? */ |
7407 | } | |
7408 | ||
91405151 FM |
7409 | if (!page_cache_get_speculative(page)) { |
7410 | page = NULL; | |
fc4adbff | 7411 | continue; |
91405151 | 7412 | } |
fc4adbff AG |
7413 | |
7414 | /* | |
7415 | * Has the page moved? | |
7416 | * This is part of the lockless pagecache protocol. See | |
7417 | * include/linux/pagemap.h for details. | |
7418 | */ | |
7419 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7420 | put_page(page); |
fc4adbff AG |
7421 | page = NULL; |
7422 | } | |
7423 | } | |
7424 | ||
7425 | if (page) { | |
7426 | if (page->index <= end_idx) | |
7427 | found = true; | |
09cbfeaf | 7428 | put_page(page); |
fc4adbff AG |
7429 | } |
7430 | ||
7431 | rcu_read_unlock(); | |
7432 | return found; | |
7433 | } | |
7434 | ||
eb838e73 JB |
7435 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7436 | struct extent_state **cached_state, int writing) | |
7437 | { | |
7438 | struct btrfs_ordered_extent *ordered; | |
7439 | int ret = 0; | |
7440 | ||
7441 | while (1) { | |
7442 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7443 | cached_state); |
eb838e73 JB |
7444 | /* |
7445 | * We're concerned with the entire range that we're going to be | |
01327610 | 7446 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7447 | * extents in this range. |
7448 | */ | |
7449 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7450 | lockend - lockstart + 1); | |
7451 | ||
7452 | /* | |
7453 | * We need to make sure there are no buffered pages in this | |
7454 | * range either, we could have raced between the invalidate in | |
7455 | * generic_file_direct_write and locking the extent. The | |
7456 | * invalidate needs to happen so that reads after a write do not | |
7457 | * get stale data. | |
7458 | */ | |
fc4adbff AG |
7459 | if (!ordered && |
7460 | (!writing || | |
7461 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7462 | break; |
7463 | ||
7464 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7465 | cached_state, GFP_NOFS); | |
7466 | ||
7467 | if (ordered) { | |
ade77029 FM |
7468 | /* |
7469 | * If we are doing a DIO read and the ordered extent we | |
7470 | * found is for a buffered write, we can not wait for it | |
7471 | * to complete and retry, because if we do so we can | |
7472 | * deadlock with concurrent buffered writes on page | |
7473 | * locks. This happens only if our DIO read covers more | |
7474 | * than one extent map, if at this point has already | |
7475 | * created an ordered extent for a previous extent map | |
7476 | * and locked its range in the inode's io tree, and a | |
7477 | * concurrent write against that previous extent map's | |
7478 | * range and this range started (we unlock the ranges | |
7479 | * in the io tree only when the bios complete and | |
7480 | * buffered writes always lock pages before attempting | |
7481 | * to lock range in the io tree). | |
7482 | */ | |
7483 | if (writing || | |
7484 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7485 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7486 | else | |
7487 | ret = -ENOTBLK; | |
eb838e73 JB |
7488 | btrfs_put_ordered_extent(ordered); |
7489 | } else { | |
eb838e73 | 7490 | /* |
b850ae14 FM |
7491 | * We could trigger writeback for this range (and wait |
7492 | * for it to complete) and then invalidate the pages for | |
7493 | * this range (through invalidate_inode_pages2_range()), | |
7494 | * but that can lead us to a deadlock with a concurrent | |
7495 | * call to readpages() (a buffered read or a defrag call | |
7496 | * triggered a readahead) on a page lock due to an | |
7497 | * ordered dio extent we created before but did not have | |
7498 | * yet a corresponding bio submitted (whence it can not | |
7499 | * complete), which makes readpages() wait for that | |
7500 | * ordered extent to complete while holding a lock on | |
7501 | * that page. | |
eb838e73 | 7502 | */ |
b850ae14 | 7503 | ret = -ENOTBLK; |
eb838e73 JB |
7504 | } |
7505 | ||
ade77029 FM |
7506 | if (ret) |
7507 | break; | |
7508 | ||
eb838e73 JB |
7509 | cond_resched(); |
7510 | } | |
7511 | ||
7512 | return ret; | |
7513 | } | |
7514 | ||
69ffb543 JB |
7515 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7516 | u64 len, u64 orig_start, | |
7517 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7518 | u64 orig_block_len, u64 ram_bytes, |
7519 | int type) | |
69ffb543 JB |
7520 | { |
7521 | struct extent_map_tree *em_tree; | |
7522 | struct extent_map *em; | |
7523 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7524 | int ret; | |
7525 | ||
7526 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7527 | em = alloc_extent_map(); | |
7528 | if (!em) | |
7529 | return ERR_PTR(-ENOMEM); | |
7530 | ||
7531 | em->start = start; | |
7532 | em->orig_start = orig_start; | |
2ab28f32 JB |
7533 | em->mod_start = start; |
7534 | em->mod_len = len; | |
69ffb543 JB |
7535 | em->len = len; |
7536 | em->block_len = block_len; | |
7537 | em->block_start = block_start; | |
7538 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7539 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7540 | em->ram_bytes = ram_bytes; |
70c8a91c | 7541 | em->generation = -1; |
69ffb543 JB |
7542 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7543 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7544 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7545 | |
7546 | do { | |
7547 | btrfs_drop_extent_cache(inode, em->start, | |
7548 | em->start + em->len - 1, 0); | |
7549 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7550 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7551 | write_unlock(&em_tree->lock); |
7552 | } while (ret == -EEXIST); | |
7553 | ||
7554 | if (ret) { | |
7555 | free_extent_map(em); | |
7556 | return ERR_PTR(ret); | |
7557 | } | |
7558 | ||
7559 | return em; | |
7560 | } | |
7561 | ||
9c9464cc FM |
7562 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7563 | struct btrfs_dio_data *dio_data, | |
7564 | const u64 len) | |
7565 | { | |
7566 | unsigned num_extents; | |
7567 | ||
7568 | num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1, | |
7569 | BTRFS_MAX_EXTENT_SIZE); | |
7570 | /* | |
7571 | * If we have an outstanding_extents count still set then we're | |
7572 | * within our reservation, otherwise we need to adjust our inode | |
7573 | * counter appropriately. | |
7574 | */ | |
7575 | if (dio_data->outstanding_extents) { | |
7576 | dio_data->outstanding_extents -= num_extents; | |
7577 | } else { | |
7578 | spin_lock(&BTRFS_I(inode)->lock); | |
7579 | BTRFS_I(inode)->outstanding_extents += num_extents; | |
7580 | spin_unlock(&BTRFS_I(inode)->lock); | |
7581 | } | |
7582 | } | |
7583 | ||
4b46fce2 JB |
7584 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7585 | struct buffer_head *bh_result, int create) | |
7586 | { | |
7587 | struct extent_map *em; | |
7588 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7589 | struct extent_state *cached_state = NULL; |
50745b0a | 7590 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7591 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7592 | u64 lockstart, lockend; |
4b46fce2 | 7593 | u64 len = bh_result->b_size; |
eb838e73 | 7594 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7595 | int ret = 0; |
eb838e73 | 7596 | |
172a5049 | 7597 | if (create) |
3266789f | 7598 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7599 | else |
c329861d | 7600 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7601 | |
c329861d JB |
7602 | lockstart = start; |
7603 | lockend = start + len - 1; | |
7604 | ||
e1cbbfa5 JB |
7605 | if (current->journal_info) { |
7606 | /* | |
7607 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7608 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7609 | * confused. |
7610 | */ | |
50745b0a | 7611 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7612 | current->journal_info = NULL; |
7613 | } | |
7614 | ||
eb838e73 JB |
7615 | /* |
7616 | * If this errors out it's because we couldn't invalidate pagecache for | |
7617 | * this range and we need to fallback to buffered. | |
7618 | */ | |
9c9464cc FM |
7619 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7620 | create)) { | |
7621 | ret = -ENOTBLK; | |
7622 | goto err; | |
7623 | } | |
eb838e73 | 7624 | |
4b46fce2 | 7625 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7626 | if (IS_ERR(em)) { |
7627 | ret = PTR_ERR(em); | |
7628 | goto unlock_err; | |
7629 | } | |
4b46fce2 JB |
7630 | |
7631 | /* | |
7632 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7633 | * io. INLINE is special, and we could probably kludge it in here, but | |
7634 | * it's still buffered so for safety lets just fall back to the generic | |
7635 | * buffered path. | |
7636 | * | |
7637 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7638 | * decompress it, so there will be buffering required no matter what we | |
7639 | * do, so go ahead and fallback to buffered. | |
7640 | * | |
01327610 | 7641 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7642 | * to buffered IO. Don't blame me, this is the price we pay for using |
7643 | * the generic code. | |
7644 | */ | |
7645 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7646 | em->block_start == EXTENT_MAP_INLINE) { | |
7647 | free_extent_map(em); | |
eb838e73 JB |
7648 | ret = -ENOTBLK; |
7649 | goto unlock_err; | |
4b46fce2 JB |
7650 | } |
7651 | ||
7652 | /* Just a good old fashioned hole, return */ | |
7653 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7654 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7655 | free_extent_map(em); | |
eb838e73 | 7656 | goto unlock_err; |
4b46fce2 JB |
7657 | } |
7658 | ||
7659 | /* | |
7660 | * We don't allocate a new extent in the following cases | |
7661 | * | |
7662 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7663 | * existing extent. | |
7664 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7665 | * just use the extent. | |
7666 | * | |
7667 | */ | |
46bfbb5c | 7668 | if (!create) { |
eb838e73 JB |
7669 | len = min(len, em->len - (start - em->start)); |
7670 | lockstart = start + len; | |
7671 | goto unlock; | |
46bfbb5c | 7672 | } |
4b46fce2 JB |
7673 | |
7674 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7675 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7676 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7677 | int type; |
eb384b55 | 7678 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7679 | |
7680 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7681 | type = BTRFS_ORDERED_PREALLOC; | |
7682 | else | |
7683 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7684 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7685 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7686 | |
00361589 | 7687 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c FM |
7688 | &orig_block_len, &ram_bytes) == 1 && |
7689 | btrfs_inc_nocow_writers(root->fs_info, block_start)) { | |
5f9a8a51 | 7690 | struct extent_map *em2; |
0b901916 | 7691 | |
5f9a8a51 FM |
7692 | em2 = btrfs_create_dio_extent(inode, start, len, |
7693 | orig_start, block_start, | |
7694 | len, orig_block_len, | |
7695 | ram_bytes, type); | |
f78c436c | 7696 | btrfs_dec_nocow_writers(root->fs_info, block_start); |
69ffb543 JB |
7697 | if (type == BTRFS_ORDERED_PREALLOC) { |
7698 | free_extent_map(em); | |
5f9a8a51 | 7699 | em = em2; |
69ffb543 | 7700 | } |
5f9a8a51 FM |
7701 | if (em2 && IS_ERR(em2)) { |
7702 | ret = PTR_ERR(em2); | |
eb838e73 | 7703 | goto unlock_err; |
46bfbb5c CM |
7704 | } |
7705 | goto unlock; | |
4b46fce2 | 7706 | } |
4b46fce2 | 7707 | } |
00361589 | 7708 | |
46bfbb5c CM |
7709 | /* |
7710 | * this will cow the extent, reset the len in case we changed | |
7711 | * it above | |
7712 | */ | |
7713 | len = bh_result->b_size; | |
70c8a91c JB |
7714 | free_extent_map(em); |
7715 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7716 | if (IS_ERR(em)) { |
7717 | ret = PTR_ERR(em); | |
7718 | goto unlock_err; | |
7719 | } | |
46bfbb5c CM |
7720 | len = min(len, em->len - (start - em->start)); |
7721 | unlock: | |
4b46fce2 JB |
7722 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7723 | inode->i_blkbits; | |
46bfbb5c | 7724 | bh_result->b_size = len; |
4b46fce2 JB |
7725 | bh_result->b_bdev = em->bdev; |
7726 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7727 | if (create) { |
7728 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7729 | set_buffer_new(bh_result); | |
7730 | ||
7731 | /* | |
7732 | * Need to update the i_size under the extent lock so buffered | |
7733 | * readers will get the updated i_size when we unlock. | |
7734 | */ | |
7735 | if (start + len > i_size_read(inode)) | |
7736 | i_size_write(inode, start + len); | |
0934856d | 7737 | |
9c9464cc | 7738 | adjust_dio_outstanding_extents(inode, dio_data, len); |
7cf5b976 | 7739 | btrfs_free_reserved_data_space(inode, start, len); |
50745b0a | 7740 | WARN_ON(dio_data->reserve < len); |
7741 | dio_data->reserve -= len; | |
f28a4928 | 7742 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7743 | current->journal_info = dio_data; |
c3473e83 | 7744 | } |
4b46fce2 | 7745 | |
eb838e73 JB |
7746 | /* |
7747 | * In the case of write we need to clear and unlock the entire range, | |
7748 | * in the case of read we need to unlock only the end area that we | |
7749 | * aren't using if there is any left over space. | |
7750 | */ | |
24c03fa5 | 7751 | if (lockstart < lockend) { |
0934856d MX |
7752 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7753 | lockend, unlock_bits, 1, 0, | |
7754 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7755 | } else { |
eb838e73 | 7756 | free_extent_state(cached_state); |
24c03fa5 | 7757 | } |
eb838e73 | 7758 | |
4b46fce2 JB |
7759 | free_extent_map(em); |
7760 | ||
7761 | return 0; | |
eb838e73 JB |
7762 | |
7763 | unlock_err: | |
eb838e73 JB |
7764 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7765 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7766 | err: |
50745b0a | 7767 | if (dio_data) |
7768 | current->journal_info = dio_data; | |
9c9464cc FM |
7769 | /* |
7770 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
7771 | * write less data then expected, so that we don't underflow our inode's | |
7772 | * outstanding extents counter. | |
7773 | */ | |
7774 | if (create && dio_data) | |
7775 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
7776 | ||
eb838e73 | 7777 | return ret; |
4b46fce2 JB |
7778 | } |
7779 | ||
8b110e39 MX |
7780 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7781 | int rw, int mirror_num) | |
7782 | { | |
7783 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7784 | int ret; | |
7785 | ||
7786 | BUG_ON(rw & REQ_WRITE); | |
7787 | ||
7788 | bio_get(bio); | |
7789 | ||
7790 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7791 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7792 | if (ret) | |
7793 | goto err; | |
7794 | ||
7795 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7796 | err: | |
7797 | bio_put(bio); | |
7798 | return ret; | |
7799 | } | |
7800 | ||
7801 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7802 | struct bio *failed_bio, | |
7803 | struct io_failure_record *failrec, | |
7804 | int failed_mirror) | |
7805 | { | |
7806 | int num_copies; | |
7807 | ||
7808 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7809 | failrec->logical, failrec->len); | |
7810 | if (num_copies == 1) { | |
7811 | /* | |
7812 | * we only have a single copy of the data, so don't bother with | |
7813 | * all the retry and error correction code that follows. no | |
7814 | * matter what the error is, it is very likely to persist. | |
7815 | */ | |
7816 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7817 | num_copies, failrec->this_mirror, failed_mirror); | |
7818 | return 0; | |
7819 | } | |
7820 | ||
7821 | failrec->failed_mirror = failed_mirror; | |
7822 | failrec->this_mirror++; | |
7823 | if (failrec->this_mirror == failed_mirror) | |
7824 | failrec->this_mirror++; | |
7825 | ||
7826 | if (failrec->this_mirror > num_copies) { | |
7827 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7828 | num_copies, failrec->this_mirror, failed_mirror); | |
7829 | return 0; | |
7830 | } | |
7831 | ||
7832 | return 1; | |
7833 | } | |
7834 | ||
7835 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
2dabb324 CR |
7836 | struct page *page, unsigned int pgoff, |
7837 | u64 start, u64 end, int failed_mirror, | |
7838 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7839 | { |
7840 | struct io_failure_record *failrec; | |
7841 | struct bio *bio; | |
7842 | int isector; | |
7843 | int read_mode; | |
7844 | int ret; | |
7845 | ||
7846 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7847 | ||
7848 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7849 | if (ret) | |
7850 | return ret; | |
7851 | ||
7852 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7853 | failed_mirror); | |
7854 | if (!ret) { | |
7855 | free_io_failure(inode, failrec); | |
7856 | return -EIO; | |
7857 | } | |
7858 | ||
2dabb324 CR |
7859 | if ((failed_bio->bi_vcnt > 1) |
7860 | || (failed_bio->bi_io_vec->bv_len | |
7861 | > BTRFS_I(inode)->root->sectorsize)) | |
8b110e39 MX |
7862 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; |
7863 | else | |
7864 | read_mode = READ_SYNC; | |
7865 | ||
7866 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7867 | isector >>= inode->i_sb->s_blocksize_bits; | |
7868 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7869 | pgoff, isector, repair_endio, repair_arg); |
8b110e39 MX |
7870 | if (!bio) { |
7871 | free_io_failure(inode, failrec); | |
7872 | return -EIO; | |
7873 | } | |
7874 | ||
7875 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7876 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7877 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7878 | ||
7879 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7880 | failrec->this_mirror); | |
7881 | if (ret) { | |
7882 | free_io_failure(inode, failrec); | |
7883 | bio_put(bio); | |
7884 | } | |
7885 | ||
7886 | return ret; | |
7887 | } | |
7888 | ||
7889 | struct btrfs_retry_complete { | |
7890 | struct completion done; | |
7891 | struct inode *inode; | |
7892 | u64 start; | |
7893 | int uptodate; | |
7894 | }; | |
7895 | ||
4246a0b6 | 7896 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7897 | { |
7898 | struct btrfs_retry_complete *done = bio->bi_private; | |
2dabb324 | 7899 | struct inode *inode; |
8b110e39 MX |
7900 | struct bio_vec *bvec; |
7901 | int i; | |
7902 | ||
4246a0b6 | 7903 | if (bio->bi_error) |
8b110e39 MX |
7904 | goto end; |
7905 | ||
2dabb324 CR |
7906 | ASSERT(bio->bi_vcnt == 1); |
7907 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
7908 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
7909 | ||
8b110e39 MX |
7910 | done->uptodate = 1; |
7911 | bio_for_each_segment_all(bvec, bio, i) | |
7912 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7913 | end: | |
7914 | complete(&done->done); | |
7915 | bio_put(bio); | |
7916 | } | |
7917 | ||
7918 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7919 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7920 | { |
2dabb324 | 7921 | struct btrfs_fs_info *fs_info; |
2c30c71b | 7922 | struct bio_vec *bvec; |
8b110e39 | 7923 | struct btrfs_retry_complete done; |
4b46fce2 | 7924 | u64 start; |
2dabb324 CR |
7925 | unsigned int pgoff; |
7926 | u32 sectorsize; | |
7927 | int nr_sectors; | |
2c30c71b | 7928 | int i; |
c1dc0896 | 7929 | int ret; |
4b46fce2 | 7930 | |
2dabb324 CR |
7931 | fs_info = BTRFS_I(inode)->root->fs_info; |
7932 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
7933 | ||
8b110e39 MX |
7934 | start = io_bio->logical; |
7935 | done.inode = inode; | |
7936 | ||
7937 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
2dabb324 CR |
7938 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
7939 | pgoff = bvec->bv_offset; | |
7940 | ||
7941 | next_block_or_try_again: | |
8b110e39 MX |
7942 | done.uptodate = 0; |
7943 | done.start = start; | |
7944 | init_completion(&done.done); | |
7945 | ||
2dabb324 CR |
7946 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
7947 | pgoff, start, start + sectorsize - 1, | |
7948 | io_bio->mirror_num, | |
7949 | btrfs_retry_endio_nocsum, &done); | |
8b110e39 MX |
7950 | if (ret) |
7951 | return ret; | |
7952 | ||
7953 | wait_for_completion(&done.done); | |
7954 | ||
7955 | if (!done.uptodate) { | |
7956 | /* We might have another mirror, so try again */ | |
2dabb324 | 7957 | goto next_block_or_try_again; |
8b110e39 MX |
7958 | } |
7959 | ||
2dabb324 CR |
7960 | start += sectorsize; |
7961 | ||
7962 | if (nr_sectors--) { | |
7963 | pgoff += sectorsize; | |
7964 | goto next_block_or_try_again; | |
7965 | } | |
8b110e39 MX |
7966 | } |
7967 | ||
7968 | return 0; | |
7969 | } | |
7970 | ||
4246a0b6 | 7971 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7972 | { |
7973 | struct btrfs_retry_complete *done = bio->bi_private; | |
7974 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2dabb324 | 7975 | struct inode *inode; |
8b110e39 | 7976 | struct bio_vec *bvec; |
2dabb324 | 7977 | u64 start; |
8b110e39 MX |
7978 | int uptodate; |
7979 | int ret; | |
7980 | int i; | |
7981 | ||
4246a0b6 | 7982 | if (bio->bi_error) |
8b110e39 MX |
7983 | goto end; |
7984 | ||
7985 | uptodate = 1; | |
2dabb324 CR |
7986 | |
7987 | start = done->start; | |
7988 | ||
7989 | ASSERT(bio->bi_vcnt == 1); | |
7990 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
7991 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
7992 | ||
8b110e39 MX |
7993 | bio_for_each_segment_all(bvec, bio, i) { |
7994 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
2dabb324 CR |
7995 | bvec->bv_page, bvec->bv_offset, |
7996 | done->start, bvec->bv_len); | |
8b110e39 MX |
7997 | if (!ret) |
7998 | clean_io_failure(done->inode, done->start, | |
2dabb324 | 7999 | bvec->bv_page, bvec->bv_offset); |
8b110e39 MX |
8000 | else |
8001 | uptodate = 0; | |
8002 | } | |
8003 | ||
8004 | done->uptodate = uptodate; | |
8005 | end: | |
8006 | complete(&done->done); | |
8007 | bio_put(bio); | |
8008 | } | |
8009 | ||
8010 | static int __btrfs_subio_endio_read(struct inode *inode, | |
8011 | struct btrfs_io_bio *io_bio, int err) | |
8012 | { | |
2dabb324 | 8013 | struct btrfs_fs_info *fs_info; |
8b110e39 MX |
8014 | struct bio_vec *bvec; |
8015 | struct btrfs_retry_complete done; | |
8016 | u64 start; | |
8017 | u64 offset = 0; | |
2dabb324 CR |
8018 | u32 sectorsize; |
8019 | int nr_sectors; | |
8020 | unsigned int pgoff; | |
8021 | int csum_pos; | |
8b110e39 MX |
8022 | int i; |
8023 | int ret; | |
dc380aea | 8024 | |
2dabb324 CR |
8025 | fs_info = BTRFS_I(inode)->root->fs_info; |
8026 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
8027 | ||
8b110e39 | 8028 | err = 0; |
c1dc0896 | 8029 | start = io_bio->logical; |
8b110e39 MX |
8030 | done.inode = inode; |
8031 | ||
c1dc0896 | 8032 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
2dabb324 CR |
8033 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8034 | ||
8035 | pgoff = bvec->bv_offset; | |
8036 | next_block: | |
8037 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8038 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8039 | bvec->bv_page, pgoff, start, | |
8040 | sectorsize); | |
8b110e39 MX |
8041 | if (likely(!ret)) |
8042 | goto next; | |
8043 | try_again: | |
8044 | done.uptodate = 0; | |
8045 | done.start = start; | |
8046 | init_completion(&done.done); | |
8047 | ||
2dabb324 CR |
8048 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8049 | pgoff, start, start + sectorsize - 1, | |
8050 | io_bio->mirror_num, | |
8051 | btrfs_retry_endio, &done); | |
8b110e39 MX |
8052 | if (ret) { |
8053 | err = ret; | |
8054 | goto next; | |
8055 | } | |
8056 | ||
8057 | wait_for_completion(&done.done); | |
8058 | ||
8059 | if (!done.uptodate) { | |
8060 | /* We might have another mirror, so try again */ | |
8061 | goto try_again; | |
8062 | } | |
8063 | next: | |
2dabb324 CR |
8064 | offset += sectorsize; |
8065 | start += sectorsize; | |
8066 | ||
8067 | ASSERT(nr_sectors); | |
8068 | ||
8069 | if (--nr_sectors) { | |
8070 | pgoff += sectorsize; | |
8071 | goto next_block; | |
8072 | } | |
2c30c71b | 8073 | } |
c1dc0896 MX |
8074 | |
8075 | return err; | |
8076 | } | |
8077 | ||
8b110e39 MX |
8078 | static int btrfs_subio_endio_read(struct inode *inode, |
8079 | struct btrfs_io_bio *io_bio, int err) | |
8080 | { | |
8081 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8082 | ||
8083 | if (skip_csum) { | |
8084 | if (unlikely(err)) | |
8085 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8086 | else | |
8087 | return 0; | |
8088 | } else { | |
8089 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8090 | } | |
8091 | } | |
8092 | ||
4246a0b6 | 8093 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8094 | { |
8095 | struct btrfs_dio_private *dip = bio->bi_private; | |
8096 | struct inode *inode = dip->inode; | |
8097 | struct bio *dio_bio; | |
8098 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 8099 | int err = bio->bi_error; |
c1dc0896 | 8100 | |
8b110e39 MX |
8101 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8102 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 8103 | |
4b46fce2 | 8104 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8105 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8106 | dio_bio = dip->dio_bio; |
4b46fce2 | 8107 | |
4b46fce2 | 8108 | kfree(dip); |
c0da7aa1 | 8109 | |
1636d1d7 | 8110 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8111 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
8112 | |
8113 | if (io_bio->end_io) | |
8114 | io_bio->end_io(io_bio, err); | |
9be3395b | 8115 | bio_put(bio); |
4b46fce2 JB |
8116 | } |
8117 | ||
14543774 FM |
8118 | static void btrfs_endio_direct_write_update_ordered(struct inode *inode, |
8119 | const u64 offset, | |
8120 | const u64 bytes, | |
8121 | const int uptodate) | |
4b46fce2 | 8122 | { |
4b46fce2 | 8123 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4b46fce2 | 8124 | struct btrfs_ordered_extent *ordered = NULL; |
14543774 FM |
8125 | u64 ordered_offset = offset; |
8126 | u64 ordered_bytes = bytes; | |
4b46fce2 JB |
8127 | int ret; |
8128 | ||
163cf09c CM |
8129 | again: |
8130 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8131 | &ordered_offset, | |
4246a0b6 | 8132 | ordered_bytes, |
14543774 | 8133 | uptodate); |
4b46fce2 | 8134 | if (!ret) |
163cf09c | 8135 | goto out_test; |
4b46fce2 | 8136 | |
9e0af237 LB |
8137 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
8138 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
8139 | btrfs_queue_work(root->fs_info->endio_write_workers, |
8140 | &ordered->work); | |
163cf09c CM |
8141 | out_test: |
8142 | /* | |
8143 | * our bio might span multiple ordered extents. If we haven't | |
8144 | * completed the accounting for the whole dio, go back and try again | |
8145 | */ | |
14543774 FM |
8146 | if (ordered_offset < offset + bytes) { |
8147 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8148 | ordered = NULL; |
163cf09c CM |
8149 | goto again; |
8150 | } | |
14543774 FM |
8151 | } |
8152 | ||
8153 | static void btrfs_endio_direct_write(struct bio *bio) | |
8154 | { | |
8155 | struct btrfs_dio_private *dip = bio->bi_private; | |
8156 | struct bio *dio_bio = dip->dio_bio; | |
8157 | ||
8158 | btrfs_endio_direct_write_update_ordered(dip->inode, | |
8159 | dip->logical_offset, | |
8160 | dip->bytes, | |
8161 | !bio->bi_error); | |
4b46fce2 | 8162 | |
4b46fce2 | 8163 | kfree(dip); |
c0da7aa1 | 8164 | |
1636d1d7 | 8165 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8166 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 8167 | bio_put(bio); |
4b46fce2 JB |
8168 | } |
8169 | ||
eaf25d93 CM |
8170 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
8171 | struct bio *bio, int mirror_num, | |
8172 | unsigned long bio_flags, u64 offset) | |
8173 | { | |
8174 | int ret; | |
8175 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8176 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 8177 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8178 | return 0; |
8179 | } | |
8180 | ||
4246a0b6 | 8181 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8182 | { |
8183 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8184 | int err = bio->bi_error; |
e65e1535 | 8185 | |
8b110e39 MX |
8186 | if (err) |
8187 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
8188 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
8189 | btrfs_ino(dip->inode), bio->bi_rw, | |
8190 | (unsigned long long)bio->bi_iter.bi_sector, | |
8191 | bio->bi_iter.bi_size, err); | |
8192 | ||
8193 | if (dip->subio_endio) | |
8194 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8195 | |
8196 | if (err) { | |
e65e1535 MX |
8197 | dip->errors = 1; |
8198 | ||
8199 | /* | |
8200 | * before atomic variable goto zero, we must make sure | |
8201 | * dip->errors is perceived to be set. | |
8202 | */ | |
4e857c58 | 8203 | smp_mb__before_atomic(); |
e65e1535 MX |
8204 | } |
8205 | ||
8206 | /* if there are more bios still pending for this dio, just exit */ | |
8207 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8208 | goto out; | |
8209 | ||
9be3395b | 8210 | if (dip->errors) { |
e65e1535 | 8211 | bio_io_error(dip->orig_bio); |
9be3395b | 8212 | } else { |
4246a0b6 CH |
8213 | dip->dio_bio->bi_error = 0; |
8214 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8215 | } |
8216 | out: | |
8217 | bio_put(bio); | |
8218 | } | |
8219 | ||
8220 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8221 | u64 first_sector, gfp_t gfp_flags) | |
8222 | { | |
da2f0f74 | 8223 | struct bio *bio; |
22365979 | 8224 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8225 | if (bio) |
8226 | bio_associate_current(bio); | |
8227 | return bio; | |
e65e1535 MX |
8228 | } |
8229 | ||
c1dc0896 MX |
8230 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
8231 | struct inode *inode, | |
8232 | struct btrfs_dio_private *dip, | |
8233 | struct bio *bio, | |
8234 | u64 file_offset) | |
8235 | { | |
8236 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8237 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8238 | int ret; | |
8239 | ||
8240 | /* | |
8241 | * We load all the csum data we need when we submit | |
8242 | * the first bio to reduce the csum tree search and | |
8243 | * contention. | |
8244 | */ | |
8245 | if (dip->logical_offset == file_offset) { | |
8246 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
8247 | file_offset); | |
8248 | if (ret) | |
8249 | return ret; | |
8250 | } | |
8251 | ||
8252 | if (bio == dip->orig_bio) | |
8253 | return 0; | |
8254 | ||
8255 | file_offset -= dip->logical_offset; | |
8256 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8257 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8258 | ||
8259 | return 0; | |
8260 | } | |
8261 | ||
e65e1535 MX |
8262 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
8263 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 8264 | int async_submit) |
e65e1535 | 8265 | { |
facc8a22 | 8266 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
8267 | int write = rw & REQ_WRITE; |
8268 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8269 | int ret; | |
8270 | ||
b812ce28 JB |
8271 | if (async_submit) |
8272 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8273 | ||
e65e1535 | 8274 | bio_get(bio); |
5fd02043 JB |
8275 | |
8276 | if (!write) { | |
bfebd8b5 DS |
8277 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
8278 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
8279 | if (ret) |
8280 | goto err; | |
8281 | } | |
e65e1535 | 8282 | |
1ae39938 JB |
8283 | if (skip_sum) |
8284 | goto map; | |
8285 | ||
8286 | if (write && async_submit) { | |
e65e1535 MX |
8287 | ret = btrfs_wq_submit_bio(root->fs_info, |
8288 | inode, rw, bio, 0, 0, | |
8289 | file_offset, | |
8290 | __btrfs_submit_bio_start_direct_io, | |
8291 | __btrfs_submit_bio_done); | |
8292 | goto err; | |
1ae39938 JB |
8293 | } else if (write) { |
8294 | /* | |
8295 | * If we aren't doing async submit, calculate the csum of the | |
8296 | * bio now. | |
8297 | */ | |
8298 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
8299 | if (ret) | |
8300 | goto err; | |
23ea8e5a | 8301 | } else { |
c1dc0896 MX |
8302 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
8303 | file_offset); | |
c2db1073 TI |
8304 | if (ret) |
8305 | goto err; | |
8306 | } | |
1ae39938 JB |
8307 | map: |
8308 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
8309 | err: |
8310 | bio_put(bio); | |
8311 | return ret; | |
8312 | } | |
8313 | ||
8314 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
8315 | int skip_sum) | |
8316 | { | |
8317 | struct inode *inode = dip->inode; | |
8318 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
8319 | struct bio *bio; |
8320 | struct bio *orig_bio = dip->orig_bio; | |
8321 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 8322 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8323 | u64 file_offset = dip->logical_offset; |
8324 | u64 submit_len = 0; | |
8325 | u64 map_length; | |
5f4dc8fc | 8326 | u32 blocksize = root->sectorsize; |
1ae39938 | 8327 | int async_submit = 0; |
5f4dc8fc CR |
8328 | int nr_sectors; |
8329 | int ret; | |
8330 | int i; | |
e65e1535 | 8331 | |
4f024f37 | 8332 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8333 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 8334 | &map_length, NULL, 0); |
7a5c3c9b | 8335 | if (ret) |
e65e1535 | 8336 | return -EIO; |
facc8a22 | 8337 | |
4f024f37 | 8338 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8339 | bio = orig_bio; |
c1dc0896 | 8340 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8341 | goto submit; |
8342 | } | |
8343 | ||
53b381b3 | 8344 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8345 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8346 | async_submit = 0; |
8347 | else | |
8348 | async_submit = 1; | |
8349 | ||
02f57c7a JB |
8350 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8351 | if (!bio) | |
8352 | return -ENOMEM; | |
7a5c3c9b | 8353 | |
02f57c7a JB |
8354 | bio->bi_private = dip; |
8355 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8356 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8357 | atomic_inc(&dip->pending_bios); |
8358 | ||
e65e1535 | 8359 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5f4dc8fc CR |
8360 | nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len); |
8361 | i = 0; | |
8362 | next_block: | |
8363 | if (unlikely(map_length < submit_len + blocksize || | |
8364 | bio_add_page(bio, bvec->bv_page, blocksize, | |
8365 | bvec->bv_offset + (i * blocksize)) < blocksize)) { | |
e65e1535 MX |
8366 | /* |
8367 | * inc the count before we submit the bio so | |
8368 | * we know the end IO handler won't happen before | |
8369 | * we inc the count. Otherwise, the dip might get freed | |
8370 | * before we're done setting it up | |
8371 | */ | |
8372 | atomic_inc(&dip->pending_bios); | |
8373 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
8374 | file_offset, skip_sum, | |
c329861d | 8375 | async_submit); |
e65e1535 MX |
8376 | if (ret) { |
8377 | bio_put(bio); | |
8378 | atomic_dec(&dip->pending_bios); | |
8379 | goto out_err; | |
8380 | } | |
8381 | ||
e65e1535 MX |
8382 | start_sector += submit_len >> 9; |
8383 | file_offset += submit_len; | |
8384 | ||
8385 | submit_len = 0; | |
e65e1535 MX |
8386 | |
8387 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8388 | start_sector, GFP_NOFS); | |
8389 | if (!bio) | |
8390 | goto out_err; | |
8391 | bio->bi_private = dip; | |
8392 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8393 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8394 | |
4f024f37 | 8395 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8396 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 8397 | start_sector << 9, |
e65e1535 MX |
8398 | &map_length, NULL, 0); |
8399 | if (ret) { | |
8400 | bio_put(bio); | |
8401 | goto out_err; | |
8402 | } | |
5f4dc8fc CR |
8403 | |
8404 | goto next_block; | |
e65e1535 | 8405 | } else { |
5f4dc8fc CR |
8406 | submit_len += blocksize; |
8407 | if (--nr_sectors) { | |
8408 | i++; | |
8409 | goto next_block; | |
8410 | } | |
e65e1535 MX |
8411 | bvec++; |
8412 | } | |
8413 | } | |
8414 | ||
02f57c7a | 8415 | submit: |
e65e1535 | 8416 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 8417 | async_submit); |
e65e1535 MX |
8418 | if (!ret) |
8419 | return 0; | |
8420 | ||
8421 | bio_put(bio); | |
8422 | out_err: | |
8423 | dip->errors = 1; | |
8424 | /* | |
8425 | * before atomic variable goto zero, we must | |
8426 | * make sure dip->errors is perceived to be set. | |
8427 | */ | |
4e857c58 | 8428 | smp_mb__before_atomic(); |
e65e1535 MX |
8429 | if (atomic_dec_and_test(&dip->pending_bios)) |
8430 | bio_io_error(dip->orig_bio); | |
8431 | ||
8432 | /* bio_end_io() will handle error, so we needn't return it */ | |
8433 | return 0; | |
8434 | } | |
8435 | ||
9be3395b CM |
8436 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
8437 | struct inode *inode, loff_t file_offset) | |
4b46fce2 | 8438 | { |
61de718f FM |
8439 | struct btrfs_dio_private *dip = NULL; |
8440 | struct bio *io_bio = NULL; | |
23ea8e5a | 8441 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8442 | int skip_sum; |
7b6d91da | 8443 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
8444 | int ret = 0; |
8445 | ||
8446 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8447 | ||
9be3395b | 8448 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8449 | if (!io_bio) { |
8450 | ret = -ENOMEM; | |
8451 | goto free_ordered; | |
8452 | } | |
8453 | ||
c1dc0896 | 8454 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8455 | if (!dip) { |
8456 | ret = -ENOMEM; | |
61de718f | 8457 | goto free_ordered; |
4b46fce2 | 8458 | } |
4b46fce2 | 8459 | |
9be3395b | 8460 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8461 | dip->inode = inode; |
8462 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8463 | dip->bytes = dio_bio->bi_iter.bi_size; |
8464 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8465 | io_bio->bi_private = dip; |
9be3395b CM |
8466 | dip->orig_bio = io_bio; |
8467 | dip->dio_bio = dio_bio; | |
e65e1535 | 8468 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8469 | btrfs_bio = btrfs_io_bio(io_bio); |
8470 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8471 | |
c1dc0896 | 8472 | if (write) { |
9be3395b | 8473 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8474 | } else { |
9be3395b | 8475 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8476 | dip->subio_endio = btrfs_subio_endio_read; |
8477 | } | |
4b46fce2 | 8478 | |
f28a4928 FM |
8479 | /* |
8480 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8481 | * even if we fail to submit a bio, because in such case we do the | |
8482 | * corresponding error handling below and it must not be done a second | |
8483 | * time by btrfs_direct_IO(). | |
8484 | */ | |
8485 | if (write) { | |
8486 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8487 | ||
8488 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8489 | dip->bytes; | |
8490 | dio_data->unsubmitted_oe_range_start = | |
8491 | dio_data->unsubmitted_oe_range_end; | |
8492 | } | |
8493 | ||
e65e1535 MX |
8494 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
8495 | if (!ret) | |
eaf25d93 | 8496 | return; |
9be3395b | 8497 | |
23ea8e5a MX |
8498 | if (btrfs_bio->end_io) |
8499 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8500 | |
4b46fce2 JB |
8501 | free_ordered: |
8502 | /* | |
61de718f FM |
8503 | * If we arrived here it means either we failed to submit the dip |
8504 | * or we either failed to clone the dio_bio or failed to allocate the | |
8505 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8506 | * call bio_endio against our io_bio so that we get proper resource | |
8507 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8508 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8509 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8510 | */ |
61de718f | 8511 | if (io_bio && dip) { |
4246a0b6 CH |
8512 | io_bio->bi_error = -EIO; |
8513 | bio_endio(io_bio); | |
61de718f FM |
8514 | /* |
8515 | * The end io callbacks free our dip, do the final put on io_bio | |
8516 | * and all the cleanup and final put for dio_bio (through | |
8517 | * dio_end_io()). | |
8518 | */ | |
8519 | dip = NULL; | |
8520 | io_bio = NULL; | |
8521 | } else { | |
14543774 FM |
8522 | if (write) |
8523 | btrfs_endio_direct_write_update_ordered(inode, | |
8524 | file_offset, | |
8525 | dio_bio->bi_iter.bi_size, | |
8526 | 0); | |
8527 | else | |
61de718f FM |
8528 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8529 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8530 | |
4246a0b6 | 8531 | dio_bio->bi_error = -EIO; |
61de718f FM |
8532 | /* |
8533 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8534 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8535 | */ | |
8536 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8537 | } |
61de718f FM |
8538 | if (io_bio) |
8539 | bio_put(io_bio); | |
8540 | kfree(dip); | |
4b46fce2 JB |
8541 | } |
8542 | ||
6f673763 | 8543 | static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb, |
28060d5d | 8544 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8545 | { |
8546 | int seg; | |
a1b75f7d | 8547 | int i; |
5a5f79b5 CM |
8548 | unsigned blocksize_mask = root->sectorsize - 1; |
8549 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
8550 | |
8551 | if (offset & blocksize_mask) | |
8552 | goto out; | |
8553 | ||
28060d5d AV |
8554 | if (iov_iter_alignment(iter) & blocksize_mask) |
8555 | goto out; | |
a1b75f7d | 8556 | |
28060d5d | 8557 | /* If this is a write we don't need to check anymore */ |
6f673763 | 8558 | if (iov_iter_rw(iter) == WRITE) |
28060d5d AV |
8559 | return 0; |
8560 | /* | |
8561 | * Check to make sure we don't have duplicate iov_base's in this | |
8562 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8563 | * when reading back. | |
8564 | */ | |
8565 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8566 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8567 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8568 | goto out; |
8569 | } | |
5a5f79b5 CM |
8570 | } |
8571 | retval = 0; | |
8572 | out: | |
8573 | return retval; | |
8574 | } | |
eb838e73 | 8575 | |
c8b8e32d | 8576 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8577 | { |
4b46fce2 JB |
8578 | struct file *file = iocb->ki_filp; |
8579 | struct inode *inode = file->f_mapping->host; | |
50745b0a | 8580 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8581 | struct btrfs_dio_data dio_data = { 0 }; | |
c8b8e32d | 8582 | loff_t offset = iocb->ki_pos; |
0934856d | 8583 | size_t count = 0; |
2e60a51e | 8584 | int flags = 0; |
38851cc1 MX |
8585 | bool wakeup = true; |
8586 | bool relock = false; | |
0934856d | 8587 | ssize_t ret; |
4b46fce2 | 8588 | |
6f673763 | 8589 | if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset)) |
5a5f79b5 | 8590 | return 0; |
3f7c579c | 8591 | |
fe0f07d0 | 8592 | inode_dio_begin(inode); |
4e857c58 | 8593 | smp_mb__after_atomic(); |
38851cc1 | 8594 | |
0e267c44 | 8595 | /* |
41bd9ca4 MX |
8596 | * The generic stuff only does filemap_write_and_wait_range, which |
8597 | * isn't enough if we've written compressed pages to this area, so | |
8598 | * we need to flush the dirty pages again to make absolutely sure | |
8599 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8600 | */ |
a6cbcd4a | 8601 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8602 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8603 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8604 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8605 | offset + count - 1); | |
0e267c44 | 8606 | |
6f673763 | 8607 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8608 | /* |
8609 | * If the write DIO is beyond the EOF, we need update | |
8610 | * the isize, but it is protected by i_mutex. So we can | |
8611 | * not unlock the i_mutex at this case. | |
8612 | */ | |
8613 | if (offset + count <= inode->i_size) { | |
5955102c | 8614 | inode_unlock(inode); |
38851cc1 MX |
8615 | relock = true; |
8616 | } | |
7cf5b976 | 8617 | ret = btrfs_delalloc_reserve_space(inode, offset, count); |
0934856d | 8618 | if (ret) |
38851cc1 | 8619 | goto out; |
50745b0a | 8620 | dio_data.outstanding_extents = div64_u64(count + |
e1cbbfa5 JB |
8621 | BTRFS_MAX_EXTENT_SIZE - 1, |
8622 | BTRFS_MAX_EXTENT_SIZE); | |
8623 | ||
8624 | /* | |
8625 | * We need to know how many extents we reserved so that we can | |
8626 | * do the accounting properly if we go over the number we | |
8627 | * originally calculated. Abuse current->journal_info for this. | |
8628 | */ | |
50745b0a | 8629 | dio_data.reserve = round_up(count, root->sectorsize); |
f28a4928 FM |
8630 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8631 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8632 | current->journal_info = &dio_data; |
ee39b432 DS |
8633 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8634 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8635 | inode_dio_end(inode); |
38851cc1 MX |
8636 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8637 | wakeup = false; | |
0934856d MX |
8638 | } |
8639 | ||
17f8c842 OS |
8640 | ret = __blockdev_direct_IO(iocb, inode, |
8641 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
c8b8e32d | 8642 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8643 | btrfs_submit_direct, flags); |
6f673763 | 8644 | if (iov_iter_rw(iter) == WRITE) { |
e1cbbfa5 | 8645 | current->journal_info = NULL; |
ddba1bfc | 8646 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8647 | if (dio_data.reserve) |
7cf5b976 QW |
8648 | btrfs_delalloc_release_space(inode, offset, |
8649 | dio_data.reserve); | |
f28a4928 FM |
8650 | /* |
8651 | * On error we might have left some ordered extents | |
8652 | * without submitting corresponding bios for them, so | |
8653 | * cleanup them up to avoid other tasks getting them | |
8654 | * and waiting for them to complete forever. | |
8655 | */ | |
8656 | if (dio_data.unsubmitted_oe_range_start < | |
8657 | dio_data.unsubmitted_oe_range_end) | |
8658 | btrfs_endio_direct_write_update_ordered(inode, | |
8659 | dio_data.unsubmitted_oe_range_start, | |
8660 | dio_data.unsubmitted_oe_range_end - | |
8661 | dio_data.unsubmitted_oe_range_start, | |
8662 | 0); | |
ddba1bfc | 8663 | } else if (ret >= 0 && (size_t)ret < count) |
7cf5b976 QW |
8664 | btrfs_delalloc_release_space(inode, offset, |
8665 | count - (size_t)ret); | |
0934856d | 8666 | } |
38851cc1 | 8667 | out: |
2e60a51e | 8668 | if (wakeup) |
fe0f07d0 | 8669 | inode_dio_end(inode); |
38851cc1 | 8670 | if (relock) |
5955102c | 8671 | inode_lock(inode); |
0934856d MX |
8672 | |
8673 | return ret; | |
16432985 CM |
8674 | } |
8675 | ||
05dadc09 TI |
8676 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8677 | ||
1506fcc8 YS |
8678 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8679 | __u64 start, __u64 len) | |
8680 | { | |
05dadc09 TI |
8681 | int ret; |
8682 | ||
8683 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8684 | if (ret) | |
8685 | return ret; | |
8686 | ||
ec29ed5b | 8687 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8688 | } |
8689 | ||
a52d9a80 | 8690 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8691 | { |
d1310b2e CM |
8692 | struct extent_io_tree *tree; |
8693 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8694 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8695 | } |
1832a6d5 | 8696 | |
a52d9a80 | 8697 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8698 | { |
d1310b2e | 8699 | struct extent_io_tree *tree; |
be7bd730 JB |
8700 | struct inode *inode = page->mapping->host; |
8701 | int ret; | |
b888db2b CM |
8702 | |
8703 | if (current->flags & PF_MEMALLOC) { | |
8704 | redirty_page_for_writepage(wbc, page); | |
8705 | unlock_page(page); | |
8706 | return 0; | |
8707 | } | |
be7bd730 JB |
8708 | |
8709 | /* | |
8710 | * If we are under memory pressure we will call this directly from the | |
8711 | * VM, we need to make sure we have the inode referenced for the ordered | |
8712 | * extent. If not just return like we didn't do anything. | |
8713 | */ | |
8714 | if (!igrab(inode)) { | |
8715 | redirty_page_for_writepage(wbc, page); | |
8716 | return AOP_WRITEPAGE_ACTIVATE; | |
8717 | } | |
d1310b2e | 8718 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8719 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8720 | btrfs_add_delayed_iput(inode); | |
8721 | return ret; | |
9ebefb18 CM |
8722 | } |
8723 | ||
48a3b636 ES |
8724 | static int btrfs_writepages(struct address_space *mapping, |
8725 | struct writeback_control *wbc) | |
b293f02e | 8726 | { |
d1310b2e | 8727 | struct extent_io_tree *tree; |
771ed689 | 8728 | |
d1310b2e | 8729 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8730 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8731 | } | |
8732 | ||
3ab2fb5a CM |
8733 | static int |
8734 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8735 | struct list_head *pages, unsigned nr_pages) | |
8736 | { | |
d1310b2e CM |
8737 | struct extent_io_tree *tree; |
8738 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8739 | return extent_readpages(tree, mapping, pages, nr_pages, |
8740 | btrfs_get_extent); | |
8741 | } | |
e6dcd2dc | 8742 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8743 | { |
d1310b2e CM |
8744 | struct extent_io_tree *tree; |
8745 | struct extent_map_tree *map; | |
a52d9a80 | 8746 | int ret; |
8c2383c3 | 8747 | |
d1310b2e CM |
8748 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8749 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8750 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8751 | if (ret == 1) { |
8752 | ClearPagePrivate(page); | |
8753 | set_page_private(page, 0); | |
09cbfeaf | 8754 | put_page(page); |
39279cc3 | 8755 | } |
a52d9a80 | 8756 | return ret; |
39279cc3 CM |
8757 | } |
8758 | ||
e6dcd2dc CM |
8759 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8760 | { | |
98509cfc CM |
8761 | if (PageWriteback(page) || PageDirty(page)) |
8762 | return 0; | |
b335b003 | 8763 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8764 | } |
8765 | ||
d47992f8 LC |
8766 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8767 | unsigned int length) | |
39279cc3 | 8768 | { |
5fd02043 | 8769 | struct inode *inode = page->mapping->host; |
d1310b2e | 8770 | struct extent_io_tree *tree; |
e6dcd2dc | 8771 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8772 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8773 | u64 page_start = page_offset(page); |
09cbfeaf | 8774 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8775 | u64 start; |
8776 | u64 end; | |
131e404a | 8777 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8778 | |
8b62b72b CM |
8779 | /* |
8780 | * we have the page locked, so new writeback can't start, | |
8781 | * and the dirty bit won't be cleared while we are here. | |
8782 | * | |
8783 | * Wait for IO on this page so that we can safely clear | |
8784 | * the PagePrivate2 bit and do ordered accounting | |
8785 | */ | |
e6dcd2dc | 8786 | wait_on_page_writeback(page); |
8b62b72b | 8787 | |
5fd02043 | 8788 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8789 | if (offset) { |
8790 | btrfs_releasepage(page, GFP_NOFS); | |
8791 | return; | |
8792 | } | |
131e404a FDBM |
8793 | |
8794 | if (!inode_evicting) | |
ff13db41 | 8795 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8796 | again: |
8797 | start = page_start; | |
8798 | ordered = btrfs_lookup_ordered_range(inode, start, | |
8799 | page_end - start + 1); | |
e6dcd2dc | 8800 | if (ordered) { |
dbfdb6d1 | 8801 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8802 | /* |
8803 | * IO on this page will never be started, so we need | |
8804 | * to account for any ordered extents now | |
8805 | */ | |
131e404a | 8806 | if (!inode_evicting) |
dbfdb6d1 | 8807 | clear_extent_bit(tree, start, end, |
131e404a FDBM |
8808 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8809 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8810 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8811 | GFP_NOFS); | |
8b62b72b CM |
8812 | /* |
8813 | * whoever cleared the private bit is responsible | |
8814 | * for the finish_ordered_io | |
8815 | */ | |
77cef2ec JB |
8816 | if (TestClearPagePrivate2(page)) { |
8817 | struct btrfs_ordered_inode_tree *tree; | |
8818 | u64 new_len; | |
8819 | ||
8820 | tree = &BTRFS_I(inode)->ordered_tree; | |
8821 | ||
8822 | spin_lock_irq(&tree->lock); | |
8823 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8824 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8825 | if (new_len < ordered->truncated_len) |
8826 | ordered->truncated_len = new_len; | |
8827 | spin_unlock_irq(&tree->lock); | |
8828 | ||
8829 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8830 | start, |
8831 | end - start + 1, 1)) | |
77cef2ec | 8832 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8833 | } |
e6dcd2dc | 8834 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8835 | if (!inode_evicting) { |
8836 | cached_state = NULL; | |
dbfdb6d1 | 8837 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8838 | &cached_state); |
8839 | } | |
dbfdb6d1 CR |
8840 | |
8841 | start = end + 1; | |
8842 | if (start < page_end) | |
8843 | goto again; | |
131e404a FDBM |
8844 | } |
8845 | ||
b9d0b389 QW |
8846 | /* |
8847 | * Qgroup reserved space handler | |
8848 | * Page here will be either | |
8849 | * 1) Already written to disk | |
8850 | * In this case, its reserved space is released from data rsv map | |
8851 | * and will be freed by delayed_ref handler finally. | |
8852 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8853 | * space. | |
8854 | * 2) Not written to disk | |
8855 | * This means the reserved space should be freed here. | |
8856 | */ | |
09cbfeaf | 8857 | btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE); |
131e404a FDBM |
8858 | if (!inode_evicting) { |
8859 | clear_extent_bit(tree, page_start, page_end, | |
8860 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8861 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8862 | EXTENT_DEFRAG, 1, 1, | |
8863 | &cached_state, GFP_NOFS); | |
8864 | ||
8865 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8866 | } |
e6dcd2dc | 8867 | |
4a096752 | 8868 | ClearPageChecked(page); |
9ad6b7bc | 8869 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8870 | ClearPagePrivate(page); |
8871 | set_page_private(page, 0); | |
09cbfeaf | 8872 | put_page(page); |
9ad6b7bc | 8873 | } |
39279cc3 CM |
8874 | } |
8875 | ||
9ebefb18 CM |
8876 | /* |
8877 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8878 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8879 | * be careful to check for EOF conditions here. We set the page up correctly | |
8880 | * for a written page which means we get ENOSPC checking when writing into | |
8881 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8882 | * support these features. | |
8883 | * | |
8884 | * We are not allowed to take the i_mutex here so we have to play games to | |
8885 | * protect against truncate races as the page could now be beyond EOF. Because | |
8886 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8887 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8888 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8889 | * unlock the page. | |
8890 | */ | |
c2ec175c | 8891 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8892 | { |
c2ec175c | 8893 | struct page *page = vmf->page; |
496ad9aa | 8894 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8895 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8896 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8897 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8898 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8899 | char *kaddr; |
8900 | unsigned long zero_start; | |
9ebefb18 | 8901 | loff_t size; |
1832a6d5 | 8902 | int ret; |
9998eb70 | 8903 | int reserved = 0; |
d0b7da88 | 8904 | u64 reserved_space; |
a52d9a80 | 8905 | u64 page_start; |
e6dcd2dc | 8906 | u64 page_end; |
d0b7da88 CR |
8907 | u64 end; |
8908 | ||
09cbfeaf | 8909 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8910 | |
b2b5ef5c | 8911 | sb_start_pagefault(inode->i_sb); |
df480633 | 8912 | page_start = page_offset(page); |
09cbfeaf | 8913 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8914 | end = page_end; |
df480633 | 8915 | |
d0b7da88 CR |
8916 | /* |
8917 | * Reserving delalloc space after obtaining the page lock can lead to | |
8918 | * deadlock. For example, if a dirty page is locked by this function | |
8919 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8920 | * dirty page write out, then the btrfs_writepage() function could | |
8921 | * end up waiting indefinitely to get a lock on the page currently | |
8922 | * being processed by btrfs_page_mkwrite() function. | |
8923 | */ | |
7cf5b976 | 8924 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
d0b7da88 | 8925 | reserved_space); |
9998eb70 | 8926 | if (!ret) { |
e41f941a | 8927 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8928 | reserved = 1; |
8929 | } | |
56a76f82 NP |
8930 | if (ret) { |
8931 | if (ret == -ENOMEM) | |
8932 | ret = VM_FAULT_OOM; | |
8933 | else /* -ENOSPC, -EIO, etc */ | |
8934 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8935 | if (reserved) |
8936 | goto out; | |
8937 | goto out_noreserve; | |
56a76f82 | 8938 | } |
1832a6d5 | 8939 | |
56a76f82 | 8940 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8941 | again: |
9ebefb18 | 8942 | lock_page(page); |
9ebefb18 | 8943 | size = i_size_read(inode); |
a52d9a80 | 8944 | |
9ebefb18 | 8945 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8946 | (page_start >= size)) { |
9ebefb18 CM |
8947 | /* page got truncated out from underneath us */ |
8948 | goto out_unlock; | |
8949 | } | |
e6dcd2dc CM |
8950 | wait_on_page_writeback(page); |
8951 | ||
ff13db41 | 8952 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8953 | set_page_extent_mapped(page); |
8954 | ||
eb84ae03 CM |
8955 | /* |
8956 | * we can't set the delalloc bits if there are pending ordered | |
8957 | * extents. Drop our locks and wait for them to finish | |
8958 | */ | |
d0b7da88 | 8959 | ordered = btrfs_lookup_ordered_range(inode, page_start, page_end); |
e6dcd2dc | 8960 | if (ordered) { |
2ac55d41 JB |
8961 | unlock_extent_cached(io_tree, page_start, page_end, |
8962 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8963 | unlock_page(page); |
eb84ae03 | 8964 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8965 | btrfs_put_ordered_extent(ordered); |
8966 | goto again; | |
8967 | } | |
8968 | ||
09cbfeaf | 8969 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
d0b7da88 | 8970 | reserved_space = round_up(size - page_start, root->sectorsize); |
09cbfeaf | 8971 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
8972 | end = page_start + reserved_space - 1; |
8973 | spin_lock(&BTRFS_I(inode)->lock); | |
8974 | BTRFS_I(inode)->outstanding_extents++; | |
8975 | spin_unlock(&BTRFS_I(inode)->lock); | |
8976 | btrfs_delalloc_release_space(inode, page_start, | |
09cbfeaf | 8977 | PAGE_SIZE - reserved_space); |
d0b7da88 CR |
8978 | } |
8979 | } | |
8980 | ||
fbf19087 JB |
8981 | /* |
8982 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8983 | * if it was already dirty, so for space accounting reasons we need to | |
8984 | * clear any delalloc bits for the range we are fixing to save. There | |
8985 | * is probably a better way to do this, but for now keep consistent with | |
8986 | * prepare_pages in the normal write path. | |
8987 | */ | |
d0b7da88 | 8988 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
8989 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8990 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 8991 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 8992 | |
d0b7da88 | 8993 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
2ac55d41 | 8994 | &cached_state); |
9ed74f2d | 8995 | if (ret) { |
2ac55d41 JB |
8996 | unlock_extent_cached(io_tree, page_start, page_end, |
8997 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
8998 | ret = VM_FAULT_SIGBUS; |
8999 | goto out_unlock; | |
9000 | } | |
e6dcd2dc | 9001 | ret = 0; |
9ebefb18 CM |
9002 | |
9003 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9004 | if (page_start + PAGE_SIZE > size) |
9005 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9006 | else |
09cbfeaf | 9007 | zero_start = PAGE_SIZE; |
9ebefb18 | 9008 | |
09cbfeaf | 9009 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9010 | kaddr = kmap(page); |
09cbfeaf | 9011 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9012 | flush_dcache_page(page); |
9013 | kunmap(page); | |
9014 | } | |
247e743c | 9015 | ClearPageChecked(page); |
e6dcd2dc | 9016 | set_page_dirty(page); |
50a9b214 | 9017 | SetPageUptodate(page); |
5a3f23d5 | 9018 | |
257c62e1 CM |
9019 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
9020 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 9021 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9022 | |
2ac55d41 | 9023 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9024 | |
9025 | out_unlock: | |
b2b5ef5c JK |
9026 | if (!ret) { |
9027 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 9028 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9029 | } |
9ebefb18 | 9030 | unlock_page(page); |
1832a6d5 | 9031 | out: |
d0b7da88 | 9032 | btrfs_delalloc_release_space(inode, page_start, reserved_space); |
9998eb70 | 9033 | out_noreserve: |
b2b5ef5c | 9034 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
9035 | return ret; |
9036 | } | |
9037 | ||
a41ad394 | 9038 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
9039 | { |
9040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 9041 | struct btrfs_block_rsv *rsv; |
a71754fc | 9042 | int ret = 0; |
3893e33b | 9043 | int err = 0; |
39279cc3 | 9044 | struct btrfs_trans_handle *trans; |
dbe674a9 | 9045 | u64 mask = root->sectorsize - 1; |
07127184 | 9046 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 9047 | |
0ef8b726 JB |
9048 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9049 | (u64)-1); | |
9050 | if (ret) | |
9051 | return ret; | |
39279cc3 | 9052 | |
fcb80c2a | 9053 | /* |
01327610 | 9054 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9055 | * 3 things going on here |
9056 | * | |
9057 | * 1) We need to reserve space for our orphan item and the space to | |
9058 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9059 | * orphan item because we didn't reserve space to remove it. | |
9060 | * | |
9061 | * 2) We need to reserve space to update our inode. | |
9062 | * | |
9063 | * 3) We need to have something to cache all the space that is going to | |
9064 | * be free'd up by the truncate operation, but also have some slack | |
9065 | * space reserved in case it uses space during the truncate (thank you | |
9066 | * very much snapshotting). | |
9067 | * | |
01327610 | 9068 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9069 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9070 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9071 | * doesn't end up using space reserved for updating the inode or |
9072 | * removing the orphan item. We also need to be able to stop the | |
9073 | * transaction and start a new one, which means we need to be able to | |
9074 | * update the inode several times, and we have no idea of knowing how | |
9075 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9076 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9077 | * Then there is the orphan item, which does indeed need to be held on |
9078 | * to for the whole operation, and we need nobody to touch this reserved | |
9079 | * space except the orphan code. | |
9080 | * | |
9081 | * So that leaves us with | |
9082 | * | |
9083 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9084 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9085 | * transaction reservation. | |
9086 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9087 | * updating the inode. | |
9088 | */ | |
66d8f3dd | 9089 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9090 | if (!rsv) |
9091 | return -ENOMEM; | |
4a338542 | 9092 | rsv->size = min_size; |
ca7e70f5 | 9093 | rsv->failfast = 1; |
f0cd846e | 9094 | |
907cbceb | 9095 | /* |
07127184 | 9096 | * 1 for the truncate slack space |
907cbceb JB |
9097 | * 1 for updating the inode. |
9098 | */ | |
f3fe820c | 9099 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9100 | if (IS_ERR(trans)) { |
9101 | err = PTR_ERR(trans); | |
9102 | goto out; | |
9103 | } | |
f0cd846e | 9104 | |
907cbceb JB |
9105 | /* Migrate the slack space for the truncate to our reserve */ |
9106 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
9107 | min_size); | |
fcb80c2a | 9108 | BUG_ON(ret); |
f0cd846e | 9109 | |
5dc562c5 JB |
9110 | /* |
9111 | * So if we truncate and then write and fsync we normally would just | |
9112 | * write the extents that changed, which is a problem if we need to | |
9113 | * first truncate that entire inode. So set this flag so we write out | |
9114 | * all of the extents in the inode to the sync log so we're completely | |
9115 | * safe. | |
9116 | */ | |
9117 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9118 | trans->block_rsv = rsv; |
907cbceb | 9119 | |
8082510e YZ |
9120 | while (1) { |
9121 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9122 | inode->i_size, | |
9123 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9124 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9125 | err = ret; |
8082510e | 9126 | break; |
3893e33b | 9127 | } |
39279cc3 | 9128 | |
fcb80c2a | 9129 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 9130 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9131 | if (ret) { |
9132 | err = ret; | |
9133 | break; | |
9134 | } | |
ca7e70f5 | 9135 | |
8082510e | 9136 | btrfs_end_transaction(trans, root); |
b53d3f5d | 9137 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
9138 | |
9139 | trans = btrfs_start_transaction(root, 2); | |
9140 | if (IS_ERR(trans)) { | |
9141 | ret = err = PTR_ERR(trans); | |
9142 | trans = NULL; | |
9143 | break; | |
9144 | } | |
9145 | ||
9146 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
9147 | rsv, min_size); | |
9148 | BUG_ON(ret); /* shouldn't happen */ | |
9149 | trans->block_rsv = rsv; | |
8082510e YZ |
9150 | } |
9151 | ||
9152 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9153 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 9154 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
9155 | if (ret) |
9156 | err = ret; | |
8082510e YZ |
9157 | } |
9158 | ||
917c16b2 CM |
9159 | if (trans) { |
9160 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
9161 | ret = btrfs_update_inode(trans, root, inode); | |
9162 | if (ret && !err) | |
9163 | err = ret; | |
7b128766 | 9164 | |
7ad85bb7 | 9165 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 9166 | btrfs_btree_balance_dirty(root); |
917c16b2 | 9167 | } |
fcb80c2a JB |
9168 | |
9169 | out: | |
9170 | btrfs_free_block_rsv(root, rsv); | |
9171 | ||
3893e33b JB |
9172 | if (ret && !err) |
9173 | err = ret; | |
a41ad394 | 9174 | |
3893e33b | 9175 | return err; |
39279cc3 CM |
9176 | } |
9177 | ||
d352ac68 CM |
9178 | /* |
9179 | * create a new subvolume directory/inode (helper for the ioctl). | |
9180 | */ | |
d2fb3437 | 9181 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9182 | struct btrfs_root *new_root, |
9183 | struct btrfs_root *parent_root, | |
9184 | u64 new_dirid) | |
39279cc3 | 9185 | { |
39279cc3 | 9186 | struct inode *inode; |
76dda93c | 9187 | int err; |
00e4e6b3 | 9188 | u64 index = 0; |
39279cc3 | 9189 | |
12fc9d09 FA |
9190 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9191 | new_dirid, new_dirid, | |
9192 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9193 | &index); | |
54aa1f4d | 9194 | if (IS_ERR(inode)) |
f46b5a66 | 9195 | return PTR_ERR(inode); |
39279cc3 CM |
9196 | inode->i_op = &btrfs_dir_inode_operations; |
9197 | inode->i_fop = &btrfs_dir_file_operations; | |
9198 | ||
bfe86848 | 9199 | set_nlink(inode, 1); |
dbe674a9 | 9200 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 9201 | unlock_new_inode(inode); |
3b96362c | 9202 | |
63541927 FDBM |
9203 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9204 | if (err) | |
9205 | btrfs_err(new_root->fs_info, | |
351fd353 | 9206 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9207 | new_root->root_key.objectid, err); |
9208 | ||
76dda93c | 9209 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9210 | |
76dda93c | 9211 | iput(inode); |
ce598979 | 9212 | return err; |
39279cc3 CM |
9213 | } |
9214 | ||
39279cc3 CM |
9215 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9216 | { | |
9217 | struct btrfs_inode *ei; | |
2ead6ae7 | 9218 | struct inode *inode; |
39279cc3 CM |
9219 | |
9220 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9221 | if (!ei) | |
9222 | return NULL; | |
2ead6ae7 YZ |
9223 | |
9224 | ei->root = NULL; | |
2ead6ae7 | 9225 | ei->generation = 0; |
15ee9bc7 | 9226 | ei->last_trans = 0; |
257c62e1 | 9227 | ei->last_sub_trans = 0; |
e02119d5 | 9228 | ei->logged_trans = 0; |
2ead6ae7 | 9229 | ei->delalloc_bytes = 0; |
47059d93 | 9230 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9231 | ei->disk_i_size = 0; |
9232 | ei->flags = 0; | |
7709cde3 | 9233 | ei->csum_bytes = 0; |
2ead6ae7 | 9234 | ei->index_cnt = (u64)-1; |
67de1176 | 9235 | ei->dir_index = 0; |
2ead6ae7 | 9236 | ei->last_unlink_trans = 0; |
46d8bc34 | 9237 | ei->last_log_commit = 0; |
8089fe62 | 9238 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9239 | |
9e0baf60 JB |
9240 | spin_lock_init(&ei->lock); |
9241 | ei->outstanding_extents = 0; | |
9242 | ei->reserved_extents = 0; | |
2ead6ae7 | 9243 | |
72ac3c0d | 9244 | ei->runtime_flags = 0; |
261507a0 | 9245 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9246 | |
16cdcec7 MX |
9247 | ei->delayed_node = NULL; |
9248 | ||
9cc97d64 | 9249 | ei->i_otime.tv_sec = 0; |
9250 | ei->i_otime.tv_nsec = 0; | |
9251 | ||
2ead6ae7 | 9252 | inode = &ei->vfs_inode; |
a8067e02 | 9253 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
9254 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
9255 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
9256 | ei->io_tree.track_uptodate = 1; |
9257 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9258 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9259 | mutex_init(&ei->log_mutex); |
f248679e | 9260 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9261 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9262 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9263 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9264 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9265 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9266 | |
9267 | return inode; | |
39279cc3 CM |
9268 | } |
9269 | ||
aaedb55b JB |
9270 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9271 | void btrfs_test_destroy_inode(struct inode *inode) | |
9272 | { | |
9273 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9274 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
9275 | } | |
9276 | #endif | |
9277 | ||
fa0d7e3d NP |
9278 | static void btrfs_i_callback(struct rcu_head *head) |
9279 | { | |
9280 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9281 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9282 | } | |
9283 | ||
39279cc3 CM |
9284 | void btrfs_destroy_inode(struct inode *inode) |
9285 | { | |
e6dcd2dc | 9286 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9287 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9288 | ||
b3d9b7a3 | 9289 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9290 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9291 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9292 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9293 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9294 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9295 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9296 | |
a6dbd429 JB |
9297 | /* |
9298 | * This can happen where we create an inode, but somebody else also | |
9299 | * created the same inode and we need to destroy the one we already | |
9300 | * created. | |
9301 | */ | |
9302 | if (!root) | |
9303 | goto free; | |
9304 | ||
8a35d95f JB |
9305 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9306 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 9307 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 9308 | btrfs_ino(inode)); |
8a35d95f | 9309 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9310 | } |
7b128766 | 9311 | |
d397712b | 9312 | while (1) { |
e6dcd2dc CM |
9313 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9314 | if (!ordered) | |
9315 | break; | |
9316 | else { | |
c2cf52eb | 9317 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 9318 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
9319 | btrfs_remove_ordered_extent(inode, ordered); |
9320 | btrfs_put_ordered_extent(ordered); | |
9321 | btrfs_put_ordered_extent(ordered); | |
9322 | } | |
9323 | } | |
56fa9d07 | 9324 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9325 | inode_tree_del(inode); |
5b21f2ed | 9326 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 9327 | free: |
fa0d7e3d | 9328 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9329 | } |
9330 | ||
45321ac5 | 9331 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9332 | { |
9333 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9334 | |
6379ef9f NA |
9335 | if (root == NULL) |
9336 | return 1; | |
9337 | ||
fa6ac876 | 9338 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9339 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9340 | return 1; |
76dda93c | 9341 | else |
45321ac5 | 9342 | return generic_drop_inode(inode); |
76dda93c YZ |
9343 | } |
9344 | ||
0ee0fda0 | 9345 | static void init_once(void *foo) |
39279cc3 CM |
9346 | { |
9347 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9348 | ||
9349 | inode_init_once(&ei->vfs_inode); | |
9350 | } | |
9351 | ||
9352 | void btrfs_destroy_cachep(void) | |
9353 | { | |
8c0a8537 KS |
9354 | /* |
9355 | * Make sure all delayed rcu free inodes are flushed before we | |
9356 | * destroy cache. | |
9357 | */ | |
9358 | rcu_barrier(); | |
5598e900 KM |
9359 | kmem_cache_destroy(btrfs_inode_cachep); |
9360 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9361 | kmem_cache_destroy(btrfs_transaction_cachep); | |
9362 | kmem_cache_destroy(btrfs_path_cachep); | |
9363 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9364 | } |
9365 | ||
9366 | int btrfs_init_cachep(void) | |
9367 | { | |
837e1972 | 9368 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9369 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9370 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9371 | init_once); | |
39279cc3 CM |
9372 | if (!btrfs_inode_cachep) |
9373 | goto fail; | |
9601e3f6 | 9374 | |
837e1972 | 9375 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
9376 | sizeof(struct btrfs_trans_handle), 0, |
9377 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9378 | if (!btrfs_trans_handle_cachep) |
9379 | goto fail; | |
9601e3f6 | 9380 | |
837e1972 | 9381 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
9382 | sizeof(struct btrfs_transaction), 0, |
9383 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9384 | if (!btrfs_transaction_cachep) |
9385 | goto fail; | |
9601e3f6 | 9386 | |
837e1972 | 9387 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
9388 | sizeof(struct btrfs_path), 0, |
9389 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9390 | if (!btrfs_path_cachep) |
9391 | goto fail; | |
9601e3f6 | 9392 | |
837e1972 | 9393 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
9394 | sizeof(struct btrfs_free_space), 0, |
9395 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
9396 | if (!btrfs_free_space_cachep) | |
9397 | goto fail; | |
9398 | ||
39279cc3 CM |
9399 | return 0; |
9400 | fail: | |
9401 | btrfs_destroy_cachep(); | |
9402 | return -ENOMEM; | |
9403 | } | |
9404 | ||
9405 | static int btrfs_getattr(struct vfsmount *mnt, | |
9406 | struct dentry *dentry, struct kstat *stat) | |
9407 | { | |
df0af1a5 | 9408 | u64 delalloc_bytes; |
2b0143b5 | 9409 | struct inode *inode = d_inode(dentry); |
fadc0d8b DS |
9410 | u32 blocksize = inode->i_sb->s_blocksize; |
9411 | ||
39279cc3 | 9412 | generic_fillattr(inode, stat); |
0ee5dc67 | 9413 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9414 | |
9415 | spin_lock(&BTRFS_I(inode)->lock); | |
9416 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9417 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9418 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9419 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9420 | return 0; |
9421 | } | |
9422 | ||
cdd1fedf DF |
9423 | static int btrfs_rename_exchange(struct inode *old_dir, |
9424 | struct dentry *old_dentry, | |
9425 | struct inode *new_dir, | |
9426 | struct dentry *new_dentry) | |
9427 | { | |
9428 | struct btrfs_trans_handle *trans; | |
9429 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9430 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9431 | struct inode *new_inode = new_dentry->d_inode; | |
9432 | struct inode *old_inode = old_dentry->d_inode; | |
9433 | struct timespec ctime = CURRENT_TIME; | |
9434 | struct dentry *parent; | |
9435 | u64 old_ino = btrfs_ino(old_inode); | |
9436 | u64 new_ino = btrfs_ino(new_inode); | |
9437 | u64 old_idx = 0; | |
9438 | u64 new_idx = 0; | |
9439 | u64 root_objectid; | |
9440 | int ret; | |
86e8aa0e FM |
9441 | bool root_log_pinned = false; |
9442 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9443 | |
9444 | /* we only allow rename subvolume link between subvolumes */ | |
9445 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9446 | return -EXDEV; | |
9447 | ||
9448 | /* close the race window with snapshot create/destroy ioctl */ | |
9449 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9450 | down_read(&root->fs_info->subvol_sem); | |
9451 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9452 | down_read(&dest->fs_info->subvol_sem); | |
9453 | ||
9454 | /* | |
9455 | * We want to reserve the absolute worst case amount of items. So if | |
9456 | * both inodes are subvols and we need to unlink them then that would | |
9457 | * require 4 item modifications, but if they are both normal inodes it | |
9458 | * would require 5 item modifications, so we'll assume their normal | |
9459 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9460 | * should cover the worst case number of items we'll modify. | |
9461 | */ | |
9462 | trans = btrfs_start_transaction(root, 12); | |
9463 | if (IS_ERR(trans)) { | |
9464 | ret = PTR_ERR(trans); | |
9465 | goto out_notrans; | |
9466 | } | |
9467 | ||
9468 | /* | |
9469 | * We need to find a free sequence number both in the source and | |
9470 | * in the destination directory for the exchange. | |
9471 | */ | |
9472 | ret = btrfs_set_inode_index(new_dir, &old_idx); | |
9473 | if (ret) | |
9474 | goto out_fail; | |
9475 | ret = btrfs_set_inode_index(old_dir, &new_idx); | |
9476 | if (ret) | |
9477 | goto out_fail; | |
9478 | ||
9479 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9480 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9481 | ||
9482 | /* Reference for the source. */ | |
9483 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9484 | /* force full log commit if subvolume involved. */ | |
9485 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9486 | } else { | |
376e5a57 FM |
9487 | btrfs_pin_log_trans(root); |
9488 | root_log_pinned = true; | |
cdd1fedf DF |
9489 | ret = btrfs_insert_inode_ref(trans, dest, |
9490 | new_dentry->d_name.name, | |
9491 | new_dentry->d_name.len, | |
9492 | old_ino, | |
9493 | btrfs_ino(new_dir), old_idx); | |
9494 | if (ret) | |
9495 | goto out_fail; | |
cdd1fedf DF |
9496 | } |
9497 | ||
9498 | /* And now for the dest. */ | |
9499 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9500 | /* force full log commit if subvolume involved. */ | |
9501 | btrfs_set_log_full_commit(dest->fs_info, trans); | |
9502 | } else { | |
376e5a57 FM |
9503 | btrfs_pin_log_trans(dest); |
9504 | dest_log_pinned = true; | |
cdd1fedf DF |
9505 | ret = btrfs_insert_inode_ref(trans, root, |
9506 | old_dentry->d_name.name, | |
9507 | old_dentry->d_name.len, | |
9508 | new_ino, | |
9509 | btrfs_ino(old_dir), new_idx); | |
9510 | if (ret) | |
9511 | goto out_fail; | |
cdd1fedf DF |
9512 | } |
9513 | ||
9514 | /* Update inode version and ctime/mtime. */ | |
9515 | inode_inc_iversion(old_dir); | |
9516 | inode_inc_iversion(new_dir); | |
9517 | inode_inc_iversion(old_inode); | |
9518 | inode_inc_iversion(new_inode); | |
9519 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9520 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9521 | old_inode->i_ctime = ctime; | |
9522 | new_inode->i_ctime = ctime; | |
9523 | ||
9524 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
9525 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9526 | btrfs_record_unlink_dir(trans, new_dir, new_inode, 1); | |
9527 | } | |
9528 | ||
9529 | /* src is a subvolume */ | |
9530 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9531 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9532 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9533 | root_objectid, | |
9534 | old_dentry->d_name.name, | |
9535 | old_dentry->d_name.len); | |
9536 | } else { /* src is an inode */ | |
9537 | ret = __btrfs_unlink_inode(trans, root, old_dir, | |
9538 | old_dentry->d_inode, | |
9539 | old_dentry->d_name.name, | |
9540 | old_dentry->d_name.len); | |
9541 | if (!ret) | |
9542 | ret = btrfs_update_inode(trans, root, old_inode); | |
9543 | } | |
9544 | if (ret) { | |
9545 | btrfs_abort_transaction(trans, root, ret); | |
9546 | goto out_fail; | |
9547 | } | |
9548 | ||
9549 | /* dest is a subvolume */ | |
9550 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9551 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9552 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9553 | root_objectid, | |
9554 | new_dentry->d_name.name, | |
9555 | new_dentry->d_name.len); | |
9556 | } else { /* dest is an inode */ | |
9557 | ret = __btrfs_unlink_inode(trans, dest, new_dir, | |
9558 | new_dentry->d_inode, | |
9559 | new_dentry->d_name.name, | |
9560 | new_dentry->d_name.len); | |
9561 | if (!ret) | |
9562 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9563 | } | |
9564 | if (ret) { | |
9565 | btrfs_abort_transaction(trans, root, ret); | |
9566 | goto out_fail; | |
9567 | } | |
9568 | ||
9569 | ret = btrfs_add_link(trans, new_dir, old_inode, | |
9570 | new_dentry->d_name.name, | |
9571 | new_dentry->d_name.len, 0, old_idx); | |
9572 | if (ret) { | |
9573 | btrfs_abort_transaction(trans, root, ret); | |
9574 | goto out_fail; | |
9575 | } | |
9576 | ||
9577 | ret = btrfs_add_link(trans, old_dir, new_inode, | |
9578 | old_dentry->d_name.name, | |
9579 | old_dentry->d_name.len, 0, new_idx); | |
9580 | if (ret) { | |
9581 | btrfs_abort_transaction(trans, root, ret); | |
9582 | goto out_fail; | |
9583 | } | |
9584 | ||
9585 | if (old_inode->i_nlink == 1) | |
9586 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9587 | if (new_inode->i_nlink == 1) | |
9588 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9589 | ||
86e8aa0e | 9590 | if (root_log_pinned) { |
cdd1fedf DF |
9591 | parent = new_dentry->d_parent; |
9592 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
9593 | btrfs_end_log_trans(root); | |
86e8aa0e | 9594 | root_log_pinned = false; |
cdd1fedf | 9595 | } |
86e8aa0e | 9596 | if (dest_log_pinned) { |
cdd1fedf DF |
9597 | parent = old_dentry->d_parent; |
9598 | btrfs_log_new_name(trans, new_inode, new_dir, parent); | |
9599 | btrfs_end_log_trans(dest); | |
86e8aa0e | 9600 | dest_log_pinned = false; |
cdd1fedf DF |
9601 | } |
9602 | out_fail: | |
86e8aa0e FM |
9603 | /* |
9604 | * If we have pinned a log and an error happened, we unpin tasks | |
9605 | * trying to sync the log and force them to fallback to a transaction | |
9606 | * commit if the log currently contains any of the inodes involved in | |
9607 | * this rename operation (to ensure we do not persist a log with an | |
9608 | * inconsistent state for any of these inodes or leading to any | |
9609 | * inconsistencies when replayed). If the transaction was aborted, the | |
9610 | * abortion reason is propagated to userspace when attempting to commit | |
9611 | * the transaction. If the log does not contain any of these inodes, we | |
9612 | * allow the tasks to sync it. | |
9613 | */ | |
9614 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
9615 | if (btrfs_inode_in_log(old_dir, root->fs_info->generation) || | |
9616 | btrfs_inode_in_log(new_dir, root->fs_info->generation) || | |
9617 | btrfs_inode_in_log(old_inode, root->fs_info->generation) || | |
9618 | (new_inode && | |
9619 | btrfs_inode_in_log(new_inode, root->fs_info->generation))) | |
9620 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9621 | ||
9622 | if (root_log_pinned) { | |
9623 | btrfs_end_log_trans(root); | |
9624 | root_log_pinned = false; | |
9625 | } | |
9626 | if (dest_log_pinned) { | |
9627 | btrfs_end_log_trans(dest); | |
9628 | dest_log_pinned = false; | |
9629 | } | |
9630 | } | |
cdd1fedf DF |
9631 | ret = btrfs_end_transaction(trans, root); |
9632 | out_notrans: | |
9633 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9634 | up_read(&dest->fs_info->subvol_sem); | |
9635 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9636 | up_read(&root->fs_info->subvol_sem); | |
9637 | ||
9638 | return ret; | |
9639 | } | |
9640 | ||
9641 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9642 | struct btrfs_root *root, | |
9643 | struct inode *dir, | |
9644 | struct dentry *dentry) | |
9645 | { | |
9646 | int ret; | |
9647 | struct inode *inode; | |
9648 | u64 objectid; | |
9649 | u64 index; | |
9650 | ||
9651 | ret = btrfs_find_free_ino(root, &objectid); | |
9652 | if (ret) | |
9653 | return ret; | |
9654 | ||
9655 | inode = btrfs_new_inode(trans, root, dir, | |
9656 | dentry->d_name.name, | |
9657 | dentry->d_name.len, | |
9658 | btrfs_ino(dir), | |
9659 | objectid, | |
9660 | S_IFCHR | WHITEOUT_MODE, | |
9661 | &index); | |
9662 | ||
9663 | if (IS_ERR(inode)) { | |
9664 | ret = PTR_ERR(inode); | |
9665 | return ret; | |
9666 | } | |
9667 | ||
9668 | inode->i_op = &btrfs_special_inode_operations; | |
9669 | init_special_inode(inode, inode->i_mode, | |
9670 | WHITEOUT_DEV); | |
9671 | ||
9672 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9673 | &dentry->d_name); | |
9674 | if (ret) | |
c9901618 | 9675 | goto out; |
cdd1fedf DF |
9676 | |
9677 | ret = btrfs_add_nondir(trans, dir, dentry, | |
9678 | inode, 0, index); | |
9679 | if (ret) | |
c9901618 | 9680 | goto out; |
cdd1fedf DF |
9681 | |
9682 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9683 | out: |
cdd1fedf | 9684 | unlock_new_inode(inode); |
c9901618 FM |
9685 | if (ret) |
9686 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9687 | iput(inode); |
9688 | ||
c9901618 | 9689 | return ret; |
cdd1fedf DF |
9690 | } |
9691 | ||
d397712b | 9692 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9693 | struct inode *new_dir, struct dentry *new_dentry, |
9694 | unsigned int flags) | |
39279cc3 CM |
9695 | { |
9696 | struct btrfs_trans_handle *trans; | |
5062af35 | 9697 | unsigned int trans_num_items; |
39279cc3 | 9698 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9699 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9700 | struct inode *new_inode = d_inode(new_dentry); |
9701 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9702 | u64 index = 0; |
4df27c4d | 9703 | u64 root_objectid; |
39279cc3 | 9704 | int ret; |
33345d01 | 9705 | u64 old_ino = btrfs_ino(old_inode); |
3dc9e8f7 | 9706 | bool log_pinned = false; |
39279cc3 | 9707 | |
33345d01 | 9708 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9709 | return -EPERM; |
9710 | ||
4df27c4d | 9711 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9712 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9713 | return -EXDEV; |
9714 | ||
33345d01 LZ |
9715 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
9716 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 9717 | return -ENOTEMPTY; |
5f39d397 | 9718 | |
4df27c4d YZ |
9719 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9720 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9721 | return -ENOTEMPTY; | |
9c52057c CM |
9722 | |
9723 | ||
9724 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9725 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9726 | new_dentry->d_name.name, |
9727 | new_dentry->d_name.len); | |
9728 | ||
9729 | if (ret) { | |
9730 | if (ret == -EEXIST) { | |
9731 | /* we shouldn't get | |
9732 | * eexist without a new_inode */ | |
fae7f21c | 9733 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9734 | return ret; |
9735 | } | |
9736 | } else { | |
9737 | /* maybe -EOVERFLOW */ | |
9738 | return ret; | |
9739 | } | |
9740 | } | |
9741 | ret = 0; | |
9742 | ||
5a3f23d5 | 9743 | /* |
8d875f95 CM |
9744 | * we're using rename to replace one file with another. Start IO on it |
9745 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9746 | */ |
8d875f95 | 9747 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9748 | filemap_flush(old_inode->i_mapping); |
9749 | ||
76dda93c | 9750 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9751 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9752 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
9753 | /* |
9754 | * We want to reserve the absolute worst case amount of items. So if | |
9755 | * both inodes are subvols and we need to unlink them then that would | |
9756 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9757 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9758 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9759 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9760 | * If our rename has the whiteout flag, we need more 5 units for the |
9761 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9762 | * when selinux is enabled). | |
a22285a6 | 9763 | */ |
5062af35 FM |
9764 | trans_num_items = 11; |
9765 | if (flags & RENAME_WHITEOUT) | |
9766 | trans_num_items += 5; | |
9767 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9768 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9769 | ret = PTR_ERR(trans); |
9770 | goto out_notrans; | |
9771 | } | |
76dda93c | 9772 | |
4df27c4d YZ |
9773 | if (dest != root) |
9774 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9775 | |
a5719521 YZ |
9776 | ret = btrfs_set_inode_index(new_dir, &index); |
9777 | if (ret) | |
9778 | goto out_fail; | |
5a3f23d5 | 9779 | |
67de1176 | 9780 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9781 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9782 | /* force full log commit if subvolume involved. */ |
995946dd | 9783 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 9784 | } else { |
c4aba954 FM |
9785 | btrfs_pin_log_trans(root); |
9786 | log_pinned = true; | |
a5719521 YZ |
9787 | ret = btrfs_insert_inode_ref(trans, dest, |
9788 | new_dentry->d_name.name, | |
9789 | new_dentry->d_name.len, | |
33345d01 LZ |
9790 | old_ino, |
9791 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
9792 | if (ret) |
9793 | goto out_fail; | |
4df27c4d | 9794 | } |
5a3f23d5 | 9795 | |
0c4d2d95 JB |
9796 | inode_inc_iversion(old_dir); |
9797 | inode_inc_iversion(new_dir); | |
9798 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9799 | old_dir->i_ctime = old_dir->i_mtime = |
9800 | new_dir->i_ctime = new_dir->i_mtime = | |
9801 | old_inode->i_ctime = current_fs_time(old_dir->i_sb); | |
5f39d397 | 9802 | |
12fcfd22 CM |
9803 | if (old_dentry->d_parent != new_dentry->d_parent) |
9804 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9805 | ||
33345d01 | 9806 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9807 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9808 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9809 | old_dentry->d_name.name, | |
9810 | old_dentry->d_name.len); | |
9811 | } else { | |
92986796 | 9812 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
2b0143b5 | 9813 | d_inode(old_dentry), |
92986796 AV |
9814 | old_dentry->d_name.name, |
9815 | old_dentry->d_name.len); | |
9816 | if (!ret) | |
9817 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9818 | } |
79787eaa JM |
9819 | if (ret) { |
9820 | btrfs_abort_transaction(trans, root, ret); | |
9821 | goto out_fail; | |
9822 | } | |
39279cc3 CM |
9823 | |
9824 | if (new_inode) { | |
0c4d2d95 | 9825 | inode_inc_iversion(new_inode); |
04b285f3 | 9826 | new_inode->i_ctime = current_fs_time(new_inode->i_sb); |
33345d01 | 9827 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
9828 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9829 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9830 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9831 | root_objectid, | |
9832 | new_dentry->d_name.name, | |
9833 | new_dentry->d_name.len); | |
9834 | BUG_ON(new_inode->i_nlink == 0); | |
9835 | } else { | |
9836 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
2b0143b5 | 9837 | d_inode(new_dentry), |
4df27c4d YZ |
9838 | new_dentry->d_name.name, |
9839 | new_dentry->d_name.len); | |
9840 | } | |
4ef31a45 | 9841 | if (!ret && new_inode->i_nlink == 0) |
2b0143b5 | 9842 | ret = btrfs_orphan_add(trans, d_inode(new_dentry)); |
79787eaa JM |
9843 | if (ret) { |
9844 | btrfs_abort_transaction(trans, root, ret); | |
9845 | goto out_fail; | |
9846 | } | |
39279cc3 | 9847 | } |
aec7477b | 9848 | |
4df27c4d YZ |
9849 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9850 | new_dentry->d_name.name, | |
a5719521 | 9851 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
9852 | if (ret) { |
9853 | btrfs_abort_transaction(trans, root, ret); | |
9854 | goto out_fail; | |
9855 | } | |
39279cc3 | 9856 | |
67de1176 MX |
9857 | if (old_inode->i_nlink == 1) |
9858 | BTRFS_I(old_inode)->dir_index = index; | |
9859 | ||
3dc9e8f7 | 9860 | if (log_pinned) { |
10d9f309 | 9861 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9862 | |
6a912213 | 9863 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d | 9864 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9865 | log_pinned = false; |
4df27c4d | 9866 | } |
cdd1fedf DF |
9867 | |
9868 | if (flags & RENAME_WHITEOUT) { | |
9869 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9870 | old_dentry); | |
9871 | ||
9872 | if (ret) { | |
9873 | btrfs_abort_transaction(trans, root, ret); | |
9874 | goto out_fail; | |
9875 | } | |
4df27c4d | 9876 | } |
39279cc3 | 9877 | out_fail: |
3dc9e8f7 FM |
9878 | /* |
9879 | * If we have pinned the log and an error happened, we unpin tasks | |
9880 | * trying to sync the log and force them to fallback to a transaction | |
9881 | * commit if the log currently contains any of the inodes involved in | |
9882 | * this rename operation (to ensure we do not persist a log with an | |
9883 | * inconsistent state for any of these inodes or leading to any | |
9884 | * inconsistencies when replayed). If the transaction was aborted, the | |
9885 | * abortion reason is propagated to userspace when attempting to commit | |
9886 | * the transaction. If the log does not contain any of these inodes, we | |
9887 | * allow the tasks to sync it. | |
9888 | */ | |
9889 | if (ret && log_pinned) { | |
9890 | if (btrfs_inode_in_log(old_dir, root->fs_info->generation) || | |
9891 | btrfs_inode_in_log(new_dir, root->fs_info->generation) || | |
9892 | btrfs_inode_in_log(old_inode, root->fs_info->generation) || | |
9893 | (new_inode && | |
9894 | btrfs_inode_in_log(new_inode, root->fs_info->generation))) | |
9895 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9896 | ||
9897 | btrfs_end_log_trans(root); | |
9898 | log_pinned = false; | |
9899 | } | |
7ad85bb7 | 9900 | btrfs_end_transaction(trans, root); |
b44c59a8 | 9901 | out_notrans: |
33345d01 | 9902 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9903 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 9904 | |
39279cc3 CM |
9905 | return ret; |
9906 | } | |
9907 | ||
80ace85c MS |
9908 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9909 | struct inode *new_dir, struct dentry *new_dentry, | |
9910 | unsigned int flags) | |
9911 | { | |
cdd1fedf | 9912 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9913 | return -EINVAL; |
9914 | ||
cdd1fedf DF |
9915 | if (flags & RENAME_EXCHANGE) |
9916 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9917 | new_dentry); | |
9918 | ||
9919 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9920 | } |
9921 | ||
8ccf6f19 MX |
9922 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9923 | { | |
9924 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9925 | struct inode *inode; |
8ccf6f19 MX |
9926 | |
9927 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9928 | work); | |
9f23e289 | 9929 | inode = delalloc_work->inode; |
30424601 DS |
9930 | filemap_flush(inode->i_mapping); |
9931 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9932 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9933 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
9934 | |
9935 | if (delalloc_work->delay_iput) | |
9f23e289 | 9936 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9937 | else |
9f23e289 | 9938 | iput(inode); |
8ccf6f19 MX |
9939 | complete(&delalloc_work->completion); |
9940 | } | |
9941 | ||
9942 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 9943 | int delay_iput) |
8ccf6f19 MX |
9944 | { |
9945 | struct btrfs_delalloc_work *work; | |
9946 | ||
100d5702 | 9947 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9948 | if (!work) |
9949 | return NULL; | |
9950 | ||
9951 | init_completion(&work->completion); | |
9952 | INIT_LIST_HEAD(&work->list); | |
9953 | work->inode = inode; | |
8ccf6f19 | 9954 | work->delay_iput = delay_iput; |
9e0af237 LB |
9955 | WARN_ON_ONCE(!inode); |
9956 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9957 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9958 | |
9959 | return work; | |
9960 | } | |
9961 | ||
9962 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9963 | { | |
9964 | wait_for_completion(&work->completion); | |
100d5702 | 9965 | kfree(work); |
8ccf6f19 MX |
9966 | } |
9967 | ||
d352ac68 CM |
9968 | /* |
9969 | * some fairly slow code that needs optimization. This walks the list | |
9970 | * of all the inodes with pending delalloc and forces them to disk. | |
9971 | */ | |
6c255e67 MX |
9972 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
9973 | int nr) | |
ea8c2819 | 9974 | { |
ea8c2819 | 9975 | struct btrfs_inode *binode; |
5b21f2ed | 9976 | struct inode *inode; |
8ccf6f19 MX |
9977 | struct btrfs_delalloc_work *work, *next; |
9978 | struct list_head works; | |
1eafa6c7 | 9979 | struct list_head splice; |
8ccf6f19 | 9980 | int ret = 0; |
ea8c2819 | 9981 | |
8ccf6f19 | 9982 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9983 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9984 | |
573bfb72 | 9985 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9986 | spin_lock(&root->delalloc_lock); |
9987 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9988 | while (!list_empty(&splice)) { |
9989 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9990 | delalloc_inodes); |
1eafa6c7 | 9991 | |
eb73c1b7 MX |
9992 | list_move_tail(&binode->delalloc_inodes, |
9993 | &root->delalloc_inodes); | |
5b21f2ed | 9994 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9995 | if (!inode) { |
eb73c1b7 | 9996 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9997 | continue; |
df0af1a5 | 9998 | } |
eb73c1b7 | 9999 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10000 | |
651d494a | 10001 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10002 | if (!work) { |
f4ab9ea7 JB |
10003 | if (delay_iput) |
10004 | btrfs_add_delayed_iput(inode); | |
10005 | else | |
10006 | iput(inode); | |
1eafa6c7 | 10007 | ret = -ENOMEM; |
a1ecaabb | 10008 | goto out; |
5b21f2ed | 10009 | } |
1eafa6c7 | 10010 | list_add_tail(&work->list, &works); |
a44903ab QW |
10011 | btrfs_queue_work(root->fs_info->flush_workers, |
10012 | &work->work); | |
6c255e67 MX |
10013 | ret++; |
10014 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10015 | goto out; |
5b21f2ed | 10016 | cond_resched(); |
eb73c1b7 | 10017 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10018 | } |
eb73c1b7 | 10019 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10020 | |
a1ecaabb | 10021 | out: |
eb73c1b7 MX |
10022 | list_for_each_entry_safe(work, next, &works, list) { |
10023 | list_del_init(&work->list); | |
10024 | btrfs_wait_and_free_delalloc_work(work); | |
10025 | } | |
10026 | ||
10027 | if (!list_empty_careful(&splice)) { | |
10028 | spin_lock(&root->delalloc_lock); | |
10029 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10030 | spin_unlock(&root->delalloc_lock); | |
10031 | } | |
573bfb72 | 10032 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10033 | return ret; |
10034 | } | |
1eafa6c7 | 10035 | |
eb73c1b7 MX |
10036 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10037 | { | |
10038 | int ret; | |
1eafa6c7 | 10039 | |
2c21b4d7 | 10040 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
10041 | return -EROFS; |
10042 | ||
6c255e67 MX |
10043 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10044 | if (ret > 0) | |
10045 | ret = 0; | |
eb73c1b7 MX |
10046 | /* |
10047 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
10048 | * we have to make sure the IO is actually started and that |
10049 | * ordered extents get created before we return | |
10050 | */ | |
10051 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 10052 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 10053 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 10054 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
10055 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
10056 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
10057 | } |
10058 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
10059 | return ret; |
10060 | } | |
10061 | ||
6c255e67 MX |
10062 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10063 | int nr) | |
eb73c1b7 MX |
10064 | { |
10065 | struct btrfs_root *root; | |
10066 | struct list_head splice; | |
10067 | int ret; | |
10068 | ||
2c21b4d7 | 10069 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10070 | return -EROFS; |
10071 | ||
10072 | INIT_LIST_HEAD(&splice); | |
10073 | ||
573bfb72 | 10074 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10075 | spin_lock(&fs_info->delalloc_root_lock); |
10076 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10077 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10078 | root = list_first_entry(&splice, struct btrfs_root, |
10079 | delalloc_root); | |
10080 | root = btrfs_grab_fs_root(root); | |
10081 | BUG_ON(!root); | |
10082 | list_move_tail(&root->delalloc_root, | |
10083 | &fs_info->delalloc_roots); | |
10084 | spin_unlock(&fs_info->delalloc_root_lock); | |
10085 | ||
6c255e67 | 10086 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10087 | btrfs_put_fs_root(root); |
6c255e67 | 10088 | if (ret < 0) |
eb73c1b7 MX |
10089 | goto out; |
10090 | ||
6c255e67 MX |
10091 | if (nr != -1) { |
10092 | nr -= ret; | |
10093 | WARN_ON(nr < 0); | |
10094 | } | |
eb73c1b7 | 10095 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10096 | } |
eb73c1b7 | 10097 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10098 | |
6c255e67 | 10099 | ret = 0; |
eb73c1b7 MX |
10100 | atomic_inc(&fs_info->async_submit_draining); |
10101 | while (atomic_read(&fs_info->nr_async_submits) || | |
10102 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10103 | wait_event(fs_info->async_submit_wait, | |
10104 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10105 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10106 | } | |
10107 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 10108 | out: |
1eafa6c7 | 10109 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10110 | spin_lock(&fs_info->delalloc_root_lock); |
10111 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10112 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10113 | } |
573bfb72 | 10114 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10115 | return ret; |
ea8c2819 CM |
10116 | } |
10117 | ||
39279cc3 CM |
10118 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10119 | const char *symname) | |
10120 | { | |
10121 | struct btrfs_trans_handle *trans; | |
10122 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10123 | struct btrfs_path *path; | |
10124 | struct btrfs_key key; | |
1832a6d5 | 10125 | struct inode *inode = NULL; |
39279cc3 CM |
10126 | int err; |
10127 | int drop_inode = 0; | |
10128 | u64 objectid; | |
67871254 | 10129 | u64 index = 0; |
39279cc3 CM |
10130 | int name_len; |
10131 | int datasize; | |
5f39d397 | 10132 | unsigned long ptr; |
39279cc3 | 10133 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10134 | struct extent_buffer *leaf; |
39279cc3 | 10135 | |
f06becc4 | 10136 | name_len = strlen(symname); |
39279cc3 CM |
10137 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
10138 | return -ENAMETOOLONG; | |
1832a6d5 | 10139 | |
9ed74f2d JB |
10140 | /* |
10141 | * 2 items for inode item and ref | |
10142 | * 2 items for dir items | |
9269d12b FM |
10143 | * 1 item for updating parent inode item |
10144 | * 1 item for the inline extent item | |
9ed74f2d JB |
10145 | * 1 item for xattr if selinux is on |
10146 | */ | |
9269d12b | 10147 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10148 | if (IS_ERR(trans)) |
10149 | return PTR_ERR(trans); | |
1832a6d5 | 10150 | |
581bb050 LZ |
10151 | err = btrfs_find_free_ino(root, &objectid); |
10152 | if (err) | |
10153 | goto out_unlock; | |
10154 | ||
aec7477b | 10155 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 10156 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 10157 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
10158 | if (IS_ERR(inode)) { |
10159 | err = PTR_ERR(inode); | |
39279cc3 | 10160 | goto out_unlock; |
7cf96da3 | 10161 | } |
39279cc3 | 10162 | |
ad19db71 CS |
10163 | /* |
10164 | * If the active LSM wants to access the inode during | |
10165 | * d_instantiate it needs these. Smack checks to see | |
10166 | * if the filesystem supports xattrs by looking at the | |
10167 | * ops vector. | |
10168 | */ | |
10169 | inode->i_fop = &btrfs_file_operations; | |
10170 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10171 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10172 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10173 | ||
10174 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10175 | if (err) | |
10176 | goto out_unlock_inode; | |
ad19db71 | 10177 | |
39279cc3 | 10178 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10179 | if (!path) { |
10180 | err = -ENOMEM; | |
b0d5d10f | 10181 | goto out_unlock_inode; |
d8926bb3 | 10182 | } |
33345d01 | 10183 | key.objectid = btrfs_ino(inode); |
39279cc3 | 10184 | key.offset = 0; |
962a298f | 10185 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10186 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10187 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10188 | datasize); | |
54aa1f4d | 10189 | if (err) { |
b0839166 | 10190 | btrfs_free_path(path); |
b0d5d10f | 10191 | goto out_unlock_inode; |
54aa1f4d | 10192 | } |
5f39d397 CM |
10193 | leaf = path->nodes[0]; |
10194 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10195 | struct btrfs_file_extent_item); | |
10196 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10197 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10198 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10199 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10200 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10201 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10202 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10203 | ||
39279cc3 | 10204 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10205 | write_extent_buffer(leaf, symname, ptr, name_len); |
10206 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10207 | btrfs_free_path(path); |
5f39d397 | 10208 | |
39279cc3 | 10209 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10210 | inode_nohighmem(inode); |
39279cc3 | 10211 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10212 | inode_set_bytes(inode, name_len); |
f06becc4 | 10213 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 10214 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10215 | /* |
10216 | * Last step, add directory indexes for our symlink inode. This is the | |
10217 | * last step to avoid extra cleanup of these indexes if an error happens | |
10218 | * elsewhere above. | |
10219 | */ | |
10220 | if (!err) | |
10221 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
b0d5d10f | 10222 | if (err) { |
54aa1f4d | 10223 | drop_inode = 1; |
b0d5d10f CM |
10224 | goto out_unlock_inode; |
10225 | } | |
10226 | ||
10227 | unlock_new_inode(inode); | |
10228 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10229 | |
10230 | out_unlock: | |
7ad85bb7 | 10231 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
10232 | if (drop_inode) { |
10233 | inode_dec_link_count(inode); | |
10234 | iput(inode); | |
10235 | } | |
b53d3f5d | 10236 | btrfs_btree_balance_dirty(root); |
39279cc3 | 10237 | return err; |
b0d5d10f CM |
10238 | |
10239 | out_unlock_inode: | |
10240 | drop_inode = 1; | |
10241 | unlock_new_inode(inode); | |
10242 | goto out_unlock; | |
39279cc3 | 10243 | } |
16432985 | 10244 | |
0af3d00b JB |
10245 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10246 | u64 start, u64 num_bytes, u64 min_size, | |
10247 | loff_t actual_len, u64 *alloc_hint, | |
10248 | struct btrfs_trans_handle *trans) | |
d899e052 | 10249 | { |
5dc562c5 JB |
10250 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10251 | struct extent_map *em; | |
d899e052 YZ |
10252 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10253 | struct btrfs_key ins; | |
d899e052 | 10254 | u64 cur_offset = start; |
55a61d1d | 10255 | u64 i_size; |
154ea289 | 10256 | u64 cur_bytes; |
0b670dc4 | 10257 | u64 last_alloc = (u64)-1; |
d899e052 | 10258 | int ret = 0; |
0af3d00b | 10259 | bool own_trans = true; |
d899e052 | 10260 | |
0af3d00b JB |
10261 | if (trans) |
10262 | own_trans = false; | |
d899e052 | 10263 | while (num_bytes > 0) { |
0af3d00b JB |
10264 | if (own_trans) { |
10265 | trans = btrfs_start_transaction(root, 3); | |
10266 | if (IS_ERR(trans)) { | |
10267 | ret = PTR_ERR(trans); | |
10268 | break; | |
10269 | } | |
5a303d5d YZ |
10270 | } |
10271 | ||
ee22184b | 10272 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10273 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10274 | /* |
10275 | * If we are severely fragmented we could end up with really | |
10276 | * small allocations, so if the allocator is returning small | |
10277 | * chunks lets make its job easier by only searching for those | |
10278 | * sized chunks. | |
10279 | */ | |
10280 | cur_bytes = min(cur_bytes, last_alloc); | |
00361589 | 10281 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 10282 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 10283 | if (ret) { |
0af3d00b JB |
10284 | if (own_trans) |
10285 | btrfs_end_transaction(trans, root); | |
a22285a6 | 10286 | break; |
d899e052 | 10287 | } |
9cfa3e34 | 10288 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
5a303d5d | 10289 | |
0b670dc4 | 10290 | last_alloc = ins.offset; |
d899e052 YZ |
10291 | ret = insert_reserved_file_extent(trans, inode, |
10292 | cur_offset, ins.objectid, | |
10293 | ins.offset, ins.offset, | |
920bbbfb | 10294 | ins.offset, 0, 0, 0, |
d899e052 | 10295 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10296 | if (ret) { |
857cc2fc | 10297 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 10298 | ins.offset, 0); |
79787eaa JM |
10299 | btrfs_abort_transaction(trans, root, ret); |
10300 | if (own_trans) | |
10301 | btrfs_end_transaction(trans, root); | |
10302 | break; | |
10303 | } | |
31193213 | 10304 | |
a1ed835e CM |
10305 | btrfs_drop_extent_cache(inode, cur_offset, |
10306 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 10307 | |
5dc562c5 JB |
10308 | em = alloc_extent_map(); |
10309 | if (!em) { | |
10310 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10311 | &BTRFS_I(inode)->runtime_flags); | |
10312 | goto next; | |
10313 | } | |
10314 | ||
10315 | em->start = cur_offset; | |
10316 | em->orig_start = cur_offset; | |
10317 | em->len = ins.offset; | |
10318 | em->block_start = ins.objectid; | |
10319 | em->block_len = ins.offset; | |
b4939680 | 10320 | em->orig_block_len = ins.offset; |
cc95bef6 | 10321 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
10322 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
10323 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
10324 | em->generation = trans->transid; | |
10325 | ||
10326 | while (1) { | |
10327 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10328 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10329 | write_unlock(&em_tree->lock); |
10330 | if (ret != -EEXIST) | |
10331 | break; | |
10332 | btrfs_drop_extent_cache(inode, cur_offset, | |
10333 | cur_offset + ins.offset - 1, | |
10334 | 0); | |
10335 | } | |
10336 | free_extent_map(em); | |
10337 | next: | |
d899e052 YZ |
10338 | num_bytes -= ins.offset; |
10339 | cur_offset += ins.offset; | |
efa56464 | 10340 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10341 | |
0c4d2d95 | 10342 | inode_inc_iversion(inode); |
04b285f3 | 10343 | inode->i_ctime = current_fs_time(inode->i_sb); |
6cbff00f | 10344 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10345 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10346 | (actual_len > inode->i_size) && |
10347 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10348 | if (cur_offset > actual_len) |
55a61d1d | 10349 | i_size = actual_len; |
d1ea6a61 | 10350 | else |
55a61d1d JB |
10351 | i_size = cur_offset; |
10352 | i_size_write(inode, i_size); | |
10353 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10354 | } |
10355 | ||
d899e052 | 10356 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10357 | |
10358 | if (ret) { | |
10359 | btrfs_abort_transaction(trans, root, ret); | |
10360 | if (own_trans) | |
10361 | btrfs_end_transaction(trans, root); | |
10362 | break; | |
10363 | } | |
d899e052 | 10364 | |
0af3d00b JB |
10365 | if (own_trans) |
10366 | btrfs_end_transaction(trans, root); | |
5a303d5d | 10367 | } |
d899e052 YZ |
10368 | return ret; |
10369 | } | |
10370 | ||
0af3d00b JB |
10371 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10372 | u64 start, u64 num_bytes, u64 min_size, | |
10373 | loff_t actual_len, u64 *alloc_hint) | |
10374 | { | |
10375 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10376 | min_size, actual_len, alloc_hint, | |
10377 | NULL); | |
10378 | } | |
10379 | ||
10380 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10381 | struct btrfs_trans_handle *trans, int mode, | |
10382 | u64 start, u64 num_bytes, u64 min_size, | |
10383 | loff_t actual_len, u64 *alloc_hint) | |
10384 | { | |
10385 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10386 | min_size, actual_len, alloc_hint, trans); | |
10387 | } | |
10388 | ||
e6dcd2dc CM |
10389 | static int btrfs_set_page_dirty(struct page *page) |
10390 | { | |
e6dcd2dc CM |
10391 | return __set_page_dirty_nobuffers(page); |
10392 | } | |
10393 | ||
10556cb2 | 10394 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10395 | { |
b83cc969 | 10396 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10397 | umode_t mode = inode->i_mode; |
b83cc969 | 10398 | |
cb6db4e5 JM |
10399 | if (mask & MAY_WRITE && |
10400 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10401 | if (btrfs_root_readonly(root)) | |
10402 | return -EROFS; | |
10403 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10404 | return -EACCES; | |
10405 | } | |
2830ba7f | 10406 | return generic_permission(inode, mask); |
fdebe2bd | 10407 | } |
39279cc3 | 10408 | |
ef3b9af5 FM |
10409 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10410 | { | |
10411 | struct btrfs_trans_handle *trans; | |
10412 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10413 | struct inode *inode = NULL; | |
10414 | u64 objectid; | |
10415 | u64 index; | |
10416 | int ret = 0; | |
10417 | ||
10418 | /* | |
10419 | * 5 units required for adding orphan entry | |
10420 | */ | |
10421 | trans = btrfs_start_transaction(root, 5); | |
10422 | if (IS_ERR(trans)) | |
10423 | return PTR_ERR(trans); | |
10424 | ||
10425 | ret = btrfs_find_free_ino(root, &objectid); | |
10426 | if (ret) | |
10427 | goto out; | |
10428 | ||
10429 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
10430 | btrfs_ino(dir), objectid, mode, &index); | |
10431 | if (IS_ERR(inode)) { | |
10432 | ret = PTR_ERR(inode); | |
10433 | inode = NULL; | |
10434 | goto out; | |
10435 | } | |
10436 | ||
ef3b9af5 FM |
10437 | inode->i_fop = &btrfs_file_operations; |
10438 | inode->i_op = &btrfs_file_inode_operations; | |
10439 | ||
10440 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10441 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10442 | ||
b0d5d10f CM |
10443 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10444 | if (ret) | |
10445 | goto out_inode; | |
10446 | ||
10447 | ret = btrfs_update_inode(trans, root, inode); | |
10448 | if (ret) | |
10449 | goto out_inode; | |
ef3b9af5 FM |
10450 | ret = btrfs_orphan_add(trans, inode); |
10451 | if (ret) | |
b0d5d10f | 10452 | goto out_inode; |
ef3b9af5 | 10453 | |
5762b5c9 FM |
10454 | /* |
10455 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10456 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10457 | * through: | |
10458 | * | |
10459 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10460 | */ | |
10461 | set_nlink(inode, 1); | |
b0d5d10f | 10462 | unlock_new_inode(inode); |
ef3b9af5 FM |
10463 | d_tmpfile(dentry, inode); |
10464 | mark_inode_dirty(inode); | |
10465 | ||
10466 | out: | |
10467 | btrfs_end_transaction(trans, root); | |
10468 | if (ret) | |
10469 | iput(inode); | |
10470 | btrfs_balance_delayed_items(root); | |
10471 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 10472 | return ret; |
b0d5d10f CM |
10473 | |
10474 | out_inode: | |
10475 | unlock_new_inode(inode); | |
10476 | goto out; | |
10477 | ||
ef3b9af5 FM |
10478 | } |
10479 | ||
b38ef71c FM |
10480 | /* Inspired by filemap_check_errors() */ |
10481 | int btrfs_inode_check_errors(struct inode *inode) | |
10482 | { | |
10483 | int ret = 0; | |
10484 | ||
10485 | if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) && | |
10486 | test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags)) | |
10487 | ret = -ENOSPC; | |
10488 | if (test_bit(AS_EIO, &inode->i_mapping->flags) && | |
10489 | test_and_clear_bit(AS_EIO, &inode->i_mapping->flags)) | |
10490 | ret = -EIO; | |
10491 | ||
10492 | return ret; | |
10493 | } | |
10494 | ||
6e1d5dcc | 10495 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10496 | .getattr = btrfs_getattr, |
39279cc3 CM |
10497 | .lookup = btrfs_lookup, |
10498 | .create = btrfs_create, | |
10499 | .unlink = btrfs_unlink, | |
10500 | .link = btrfs_link, | |
10501 | .mkdir = btrfs_mkdir, | |
10502 | .rmdir = btrfs_rmdir, | |
80ace85c | 10503 | .rename2 = btrfs_rename2, |
39279cc3 CM |
10504 | .symlink = btrfs_symlink, |
10505 | .setattr = btrfs_setattr, | |
618e21d5 | 10506 | .mknod = btrfs_mknod, |
e0d46f5c | 10507 | .setxattr = generic_setxattr, |
9172abbc | 10508 | .getxattr = generic_getxattr, |
5103e947 | 10509 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10510 | .removexattr = generic_removexattr, |
fdebe2bd | 10511 | .permission = btrfs_permission, |
4e34e719 | 10512 | .get_acl = btrfs_get_acl, |
996a710d | 10513 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10514 | .update_time = btrfs_update_time, |
ef3b9af5 | 10515 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10516 | }; |
6e1d5dcc | 10517 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10518 | .lookup = btrfs_lookup, |
fdebe2bd | 10519 | .permission = btrfs_permission, |
4e34e719 | 10520 | .get_acl = btrfs_get_acl, |
996a710d | 10521 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10522 | .update_time = btrfs_update_time, |
39279cc3 | 10523 | }; |
76dda93c | 10524 | |
828c0950 | 10525 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10526 | .llseek = generic_file_llseek, |
10527 | .read = generic_read_dir, | |
fe742fd4 | 10528 | .iterate = btrfs_real_readdir, |
34287aa3 | 10529 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10530 | #ifdef CONFIG_COMPAT |
4c63c245 | 10531 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10532 | #endif |
6bf13c0c | 10533 | .release = btrfs_release_file, |
e02119d5 | 10534 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10535 | }; |
10536 | ||
20e5506b | 10537 | static const struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 10538 | .fill_delalloc = run_delalloc_range, |
065631f6 | 10539 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 10540 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 10541 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 10542 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10543 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10544 | .set_bit_hook = btrfs_set_bit_hook, |
10545 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10546 | .merge_extent_hook = btrfs_merge_extent_hook, |
10547 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
10548 | }; |
10549 | ||
35054394 CM |
10550 | /* |
10551 | * btrfs doesn't support the bmap operation because swapfiles | |
10552 | * use bmap to make a mapping of extents in the file. They assume | |
10553 | * these extents won't change over the life of the file and they | |
10554 | * use the bmap result to do IO directly to the drive. | |
10555 | * | |
10556 | * the btrfs bmap call would return logical addresses that aren't | |
10557 | * suitable for IO and they also will change frequently as COW | |
10558 | * operations happen. So, swapfile + btrfs == corruption. | |
10559 | * | |
10560 | * For now we're avoiding this by dropping bmap. | |
10561 | */ | |
7f09410b | 10562 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10563 | .readpage = btrfs_readpage, |
10564 | .writepage = btrfs_writepage, | |
b293f02e | 10565 | .writepages = btrfs_writepages, |
3ab2fb5a | 10566 | .readpages = btrfs_readpages, |
16432985 | 10567 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10568 | .invalidatepage = btrfs_invalidatepage, |
10569 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10570 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10571 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10572 | }; |
10573 | ||
7f09410b | 10574 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10575 | .readpage = btrfs_readpage, |
10576 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10577 | .invalidatepage = btrfs_invalidatepage, |
10578 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10579 | }; |
10580 | ||
6e1d5dcc | 10581 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10582 | .getattr = btrfs_getattr, |
10583 | .setattr = btrfs_setattr, | |
e0d46f5c | 10584 | .setxattr = generic_setxattr, |
9172abbc | 10585 | .getxattr = generic_getxattr, |
5103e947 | 10586 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10587 | .removexattr = generic_removexattr, |
fdebe2bd | 10588 | .permission = btrfs_permission, |
1506fcc8 | 10589 | .fiemap = btrfs_fiemap, |
4e34e719 | 10590 | .get_acl = btrfs_get_acl, |
996a710d | 10591 | .set_acl = btrfs_set_acl, |
e41f941a | 10592 | .update_time = btrfs_update_time, |
39279cc3 | 10593 | }; |
6e1d5dcc | 10594 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10595 | .getattr = btrfs_getattr, |
10596 | .setattr = btrfs_setattr, | |
fdebe2bd | 10597 | .permission = btrfs_permission, |
e0d46f5c | 10598 | .setxattr = generic_setxattr, |
9172abbc | 10599 | .getxattr = generic_getxattr, |
33268eaf | 10600 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10601 | .removexattr = generic_removexattr, |
4e34e719 | 10602 | .get_acl = btrfs_get_acl, |
996a710d | 10603 | .set_acl = btrfs_set_acl, |
e41f941a | 10604 | .update_time = btrfs_update_time, |
618e21d5 | 10605 | }; |
6e1d5dcc | 10606 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 | 10607 | .readlink = generic_readlink, |
6b255391 | 10608 | .get_link = page_get_link, |
f209561a | 10609 | .getattr = btrfs_getattr, |
22c44fe6 | 10610 | .setattr = btrfs_setattr, |
fdebe2bd | 10611 | .permission = btrfs_permission, |
e0d46f5c | 10612 | .setxattr = generic_setxattr, |
9172abbc | 10613 | .getxattr = generic_getxattr, |
0279b4cd | 10614 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10615 | .removexattr = generic_removexattr, |
e41f941a | 10616 | .update_time = btrfs_update_time, |
39279cc3 | 10617 | }; |
76dda93c | 10618 | |
82d339d9 | 10619 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10620 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10621 | .d_release = btrfs_dentry_release, |
76dda93c | 10622 | }; |