]>
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> | |
a27bb332 | 35 | #include <linux/aio.h> |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
5103e947 | 37 | #include <linux/xattr.h> |
33268eaf | 38 | #include <linux/posix_acl.h> |
d899e052 | 39 | #include <linux/falloc.h> |
5a0e3ad6 | 40 | #include <linux/slab.h> |
7a36ddec | 41 | #include <linux/ratelimit.h> |
22c44fe6 | 42 | #include <linux/mount.h> |
55e301fd | 43 | #include <linux/btrfs.h> |
53b381b3 | 44 | #include <linux/blkdev.h> |
f23b5a59 | 45 | #include <linux/posix_acl_xattr.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" |
39279cc3 CM |
62 | |
63 | struct btrfs_iget_args { | |
90d3e592 | 64 | struct btrfs_key *location; |
39279cc3 CM |
65 | struct btrfs_root *root; |
66 | }; | |
67 | ||
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_dir_inode_operations; |
69 | static const struct inode_operations btrfs_symlink_inode_operations; | |
70 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
71 | static const struct inode_operations btrfs_special_inode_operations; | |
72 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
73 | static const struct address_space_operations btrfs_aops; |
74 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 75 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 76 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
77 | |
78 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 79 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
80 | struct kmem_cache *btrfs_trans_handle_cachep; |
81 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 82 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 83 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
84 | |
85 | #define S_SHIFT 12 | |
86 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
87 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
88 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
89 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
90 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
91 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
92 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
93 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
94 | }; | |
95 | ||
3972f260 | 96 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 97 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 98 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
99 | static noinline int cow_file_range(struct inode *inode, |
100 | struct page *locked_page, | |
101 | u64 start, u64 end, int *page_started, | |
102 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
103 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
104 | u64 len, u64 orig_start, | |
105 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
106 | u64 orig_block_len, u64 ram_bytes, |
107 | int type); | |
7b128766 | 108 | |
48a3b636 | 109 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 110 | |
f34f57a3 | 111 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
112 | struct inode *inode, struct inode *dir, |
113 | const struct qstr *qstr) | |
0279b4cd JO |
114 | { |
115 | int err; | |
116 | ||
f34f57a3 | 117 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 118 | if (!err) |
2a7dba39 | 119 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
120 | return err; |
121 | } | |
122 | ||
c8b97818 CM |
123 | /* |
124 | * this does all the hard work for inserting an inline extent into | |
125 | * the btree. The caller should have done a btrfs_drop_extents so that | |
126 | * no overlapping inline items exist in the btree | |
127 | */ | |
40f76580 | 128 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 129 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
130 | struct btrfs_root *root, struct inode *inode, |
131 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 132 | int compress_type, |
c8b97818 CM |
133 | struct page **compressed_pages) |
134 | { | |
c8b97818 CM |
135 | struct extent_buffer *leaf; |
136 | struct page *page = NULL; | |
137 | char *kaddr; | |
138 | unsigned long ptr; | |
139 | struct btrfs_file_extent_item *ei; | |
140 | int err = 0; | |
141 | int ret; | |
142 | size_t cur_size = size; | |
c8b97818 | 143 | unsigned long offset; |
c8b97818 | 144 | |
fe3f566c | 145 | if (compressed_size && compressed_pages) |
c8b97818 | 146 | cur_size = compressed_size; |
c8b97818 | 147 | |
1acae57b | 148 | inode_add_bytes(inode, size); |
c8b97818 | 149 | |
1acae57b FDBM |
150 | if (!extent_inserted) { |
151 | struct btrfs_key key; | |
152 | size_t datasize; | |
c8b97818 | 153 | |
1acae57b FDBM |
154 | key.objectid = btrfs_ino(inode); |
155 | key.offset = start; | |
962a298f | 156 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 157 | |
1acae57b FDBM |
158 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
159 | path->leave_spinning = 1; | |
160 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
161 | datasize); | |
162 | if (ret) { | |
163 | err = ret; | |
164 | goto fail; | |
165 | } | |
c8b97818 CM |
166 | } |
167 | leaf = path->nodes[0]; | |
168 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
169 | struct btrfs_file_extent_item); | |
170 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
171 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
172 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
173 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
174 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
175 | ptr = btrfs_file_extent_inline_start(ei); | |
176 | ||
261507a0 | 177 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
178 | struct page *cpage; |
179 | int i = 0; | |
d397712b | 180 | while (compressed_size > 0) { |
c8b97818 | 181 | cpage = compressed_pages[i]; |
5b050f04 | 182 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
183 | PAGE_CACHE_SIZE); |
184 | ||
7ac687d9 | 185 | kaddr = kmap_atomic(cpage); |
c8b97818 | 186 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 187 | kunmap_atomic(kaddr); |
c8b97818 CM |
188 | |
189 | i++; | |
190 | ptr += cur_size; | |
191 | compressed_size -= cur_size; | |
192 | } | |
193 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 194 | compress_type); |
c8b97818 CM |
195 | } else { |
196 | page = find_get_page(inode->i_mapping, | |
197 | start >> PAGE_CACHE_SHIFT); | |
198 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 199 | kaddr = kmap_atomic(page); |
c8b97818 CM |
200 | offset = start & (PAGE_CACHE_SIZE - 1); |
201 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 202 | kunmap_atomic(kaddr); |
c8b97818 CM |
203 | page_cache_release(page); |
204 | } | |
205 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 206 | btrfs_release_path(path); |
c8b97818 | 207 | |
c2167754 YZ |
208 | /* |
209 | * we're an inline extent, so nobody can | |
210 | * extend the file past i_size without locking | |
211 | * a page we already have locked. | |
212 | * | |
213 | * We must do any isize and inode updates | |
214 | * before we unlock the pages. Otherwise we | |
215 | * could end up racing with unlink. | |
216 | */ | |
c8b97818 | 217 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 218 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 219 | |
79787eaa | 220 | return ret; |
c8b97818 | 221 | fail: |
c8b97818 CM |
222 | return err; |
223 | } | |
224 | ||
225 | ||
226 | /* | |
227 | * conditionally insert an inline extent into the file. This | |
228 | * does the checks required to make sure the data is small enough | |
229 | * to fit as an inline extent. | |
230 | */ | |
00361589 JB |
231 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
232 | struct inode *inode, u64 start, | |
233 | u64 end, size_t compressed_size, | |
234 | int compress_type, | |
235 | struct page **compressed_pages) | |
c8b97818 | 236 | { |
00361589 | 237 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
238 | u64 isize = i_size_read(inode); |
239 | u64 actual_end = min(end + 1, isize); | |
240 | u64 inline_len = actual_end - start; | |
fda2832f | 241 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
242 | u64 data_len = inline_len; |
243 | int ret; | |
1acae57b FDBM |
244 | struct btrfs_path *path; |
245 | int extent_inserted = 0; | |
246 | u32 extent_item_size; | |
c8b97818 CM |
247 | |
248 | if (compressed_size) | |
249 | data_len = compressed_size; | |
250 | ||
251 | if (start > 0 || | |
70b99e69 | 252 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
253 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
254 | (!compressed_size && | |
255 | (actual_end & (root->sectorsize - 1)) == 0) || | |
256 | end + 1 < isize || | |
257 | data_len > root->fs_info->max_inline) { | |
258 | return 1; | |
259 | } | |
260 | ||
1acae57b FDBM |
261 | path = btrfs_alloc_path(); |
262 | if (!path) | |
263 | return -ENOMEM; | |
264 | ||
00361589 | 265 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
266 | if (IS_ERR(trans)) { |
267 | btrfs_free_path(path); | |
00361589 | 268 | return PTR_ERR(trans); |
1acae57b | 269 | } |
00361589 JB |
270 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
271 | ||
1acae57b FDBM |
272 | if (compressed_size && compressed_pages) |
273 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
274 | compressed_size); | |
275 | else | |
276 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
277 | inline_len); | |
278 | ||
279 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
280 | start, aligned_end, NULL, | |
281 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
282 | if (ret) { |
283 | btrfs_abort_transaction(trans, root, ret); | |
284 | goto out; | |
285 | } | |
c8b97818 CM |
286 | |
287 | if (isize > actual_end) | |
288 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
289 | ret = insert_inline_extent(trans, path, extent_inserted, |
290 | root, inode, start, | |
c8b97818 | 291 | inline_len, compressed_size, |
fe3f566c | 292 | compress_type, compressed_pages); |
2adcac1a | 293 | if (ret && ret != -ENOSPC) { |
79787eaa | 294 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 295 | goto out; |
2adcac1a | 296 | } else if (ret == -ENOSPC) { |
00361589 JB |
297 | ret = 1; |
298 | goto out; | |
79787eaa | 299 | } |
2adcac1a | 300 | |
bdc20e67 | 301 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 302 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 303 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 304 | out: |
1acae57b | 305 | btrfs_free_path(path); |
00361589 JB |
306 | btrfs_end_transaction(trans, root); |
307 | return ret; | |
c8b97818 CM |
308 | } |
309 | ||
771ed689 CM |
310 | struct async_extent { |
311 | u64 start; | |
312 | u64 ram_size; | |
313 | u64 compressed_size; | |
314 | struct page **pages; | |
315 | unsigned long nr_pages; | |
261507a0 | 316 | int compress_type; |
771ed689 CM |
317 | struct list_head list; |
318 | }; | |
319 | ||
320 | struct async_cow { | |
321 | struct inode *inode; | |
322 | struct btrfs_root *root; | |
323 | struct page *locked_page; | |
324 | u64 start; | |
325 | u64 end; | |
326 | struct list_head extents; | |
327 | struct btrfs_work work; | |
328 | }; | |
329 | ||
330 | static noinline int add_async_extent(struct async_cow *cow, | |
331 | u64 start, u64 ram_size, | |
332 | u64 compressed_size, | |
333 | struct page **pages, | |
261507a0 LZ |
334 | unsigned long nr_pages, |
335 | int compress_type) | |
771ed689 CM |
336 | { |
337 | struct async_extent *async_extent; | |
338 | ||
339 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 340 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
341 | async_extent->start = start; |
342 | async_extent->ram_size = ram_size; | |
343 | async_extent->compressed_size = compressed_size; | |
344 | async_extent->pages = pages; | |
345 | async_extent->nr_pages = nr_pages; | |
261507a0 | 346 | async_extent->compress_type = compress_type; |
771ed689 CM |
347 | list_add_tail(&async_extent->list, &cow->extents); |
348 | return 0; | |
349 | } | |
350 | ||
f79707b0 WS |
351 | static inline int inode_need_compress(struct inode *inode) |
352 | { | |
353 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
354 | ||
355 | /* force compress */ | |
356 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
357 | return 1; | |
358 | /* bad compression ratios */ | |
359 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
360 | return 0; | |
361 | if (btrfs_test_opt(root, COMPRESS) || | |
362 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
363 | BTRFS_I(inode)->force_compress) | |
364 | return 1; | |
365 | return 0; | |
366 | } | |
367 | ||
d352ac68 | 368 | /* |
771ed689 CM |
369 | * we create compressed extents in two phases. The first |
370 | * phase compresses a range of pages that have already been | |
371 | * locked (both pages and state bits are locked). | |
c8b97818 | 372 | * |
771ed689 CM |
373 | * This is done inside an ordered work queue, and the compression |
374 | * is spread across many cpus. The actual IO submission is step | |
375 | * two, and the ordered work queue takes care of making sure that | |
376 | * happens in the same order things were put onto the queue by | |
377 | * writepages and friends. | |
c8b97818 | 378 | * |
771ed689 CM |
379 | * If this code finds it can't get good compression, it puts an |
380 | * entry onto the work queue to write the uncompressed bytes. This | |
381 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
382 | * are written in the same order that the flusher thread sent them |
383 | * down. | |
d352ac68 | 384 | */ |
771ed689 CM |
385 | static noinline int compress_file_range(struct inode *inode, |
386 | struct page *locked_page, | |
387 | u64 start, u64 end, | |
388 | struct async_cow *async_cow, | |
389 | int *num_added) | |
b888db2b CM |
390 | { |
391 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 392 | u64 num_bytes; |
db94535d | 393 | u64 blocksize = root->sectorsize; |
c8b97818 | 394 | u64 actual_end; |
42dc7bab | 395 | u64 isize = i_size_read(inode); |
e6dcd2dc | 396 | int ret = 0; |
c8b97818 CM |
397 | struct page **pages = NULL; |
398 | unsigned long nr_pages; | |
399 | unsigned long nr_pages_ret = 0; | |
400 | unsigned long total_compressed = 0; | |
401 | unsigned long total_in = 0; | |
402 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 403 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
404 | int i; |
405 | int will_compress; | |
261507a0 | 406 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 407 | int redirty = 0; |
b888db2b | 408 | |
4cb13e5d LB |
409 | /* if this is a small write inside eof, kick off a defrag */ |
410 | if ((end - start + 1) < 16 * 1024 && | |
411 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
412 | btrfs_add_inode_defrag(NULL, inode); |
413 | ||
68bb462d WS |
414 | /* |
415 | * skip compression for a small file range(<=blocksize) that | |
416 | * isn't an inline extent, since it dosen't save disk space at all. | |
417 | */ | |
418 | if ((end - start + 1) <= blocksize && | |
419 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
420 | goto cleanup_and_bail_uncompressed; | |
421 | ||
42dc7bab | 422 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
423 | again: |
424 | will_compress = 0; | |
425 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
426 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 427 | |
f03d9301 CM |
428 | /* |
429 | * we don't want to send crud past the end of i_size through | |
430 | * compression, that's just a waste of CPU time. So, if the | |
431 | * end of the file is before the start of our current | |
432 | * requested range of bytes, we bail out to the uncompressed | |
433 | * cleanup code that can deal with all of this. | |
434 | * | |
435 | * It isn't really the fastest way to fix things, but this is a | |
436 | * very uncommon corner. | |
437 | */ | |
438 | if (actual_end <= start) | |
439 | goto cleanup_and_bail_uncompressed; | |
440 | ||
c8b97818 CM |
441 | total_compressed = actual_end - start; |
442 | ||
443 | /* we want to make sure that amount of ram required to uncompress | |
444 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
445 | * of a compressed extent to 128k. This is a crucial number |
446 | * because it also controls how easily we can spread reads across | |
447 | * cpus for decompression. | |
448 | * | |
449 | * We also want to make sure the amount of IO required to do | |
450 | * a random read is reasonably small, so we limit the size of | |
451 | * a compressed extent to 128k. | |
c8b97818 CM |
452 | */ |
453 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 454 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 455 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
456 | total_in = 0; |
457 | ret = 0; | |
db94535d | 458 | |
771ed689 CM |
459 | /* |
460 | * we do compression for mount -o compress and when the | |
461 | * inode has not been flagged as nocompress. This flag can | |
462 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 463 | */ |
f79707b0 | 464 | if (inode_need_compress(inode)) { |
c8b97818 | 465 | WARN_ON(pages); |
cfbc246e | 466 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
467 | if (!pages) { |
468 | /* just bail out to the uncompressed code */ | |
469 | goto cont; | |
470 | } | |
c8b97818 | 471 | |
261507a0 LZ |
472 | if (BTRFS_I(inode)->force_compress) |
473 | compress_type = BTRFS_I(inode)->force_compress; | |
474 | ||
4adaa611 CM |
475 | /* |
476 | * we need to call clear_page_dirty_for_io on each | |
477 | * page in the range. Otherwise applications with the file | |
478 | * mmap'd can wander in and change the page contents while | |
479 | * we are compressing them. | |
480 | * | |
481 | * If the compression fails for any reason, we set the pages | |
482 | * dirty again later on. | |
483 | */ | |
484 | extent_range_clear_dirty_for_io(inode, start, end); | |
485 | redirty = 1; | |
261507a0 LZ |
486 | ret = btrfs_compress_pages(compress_type, |
487 | inode->i_mapping, start, | |
488 | total_compressed, pages, | |
489 | nr_pages, &nr_pages_ret, | |
490 | &total_in, | |
491 | &total_compressed, | |
492 | max_compressed); | |
c8b97818 CM |
493 | |
494 | if (!ret) { | |
495 | unsigned long offset = total_compressed & | |
496 | (PAGE_CACHE_SIZE - 1); | |
497 | struct page *page = pages[nr_pages_ret - 1]; | |
498 | char *kaddr; | |
499 | ||
500 | /* zero the tail end of the last page, we might be | |
501 | * sending it down to disk | |
502 | */ | |
503 | if (offset) { | |
7ac687d9 | 504 | kaddr = kmap_atomic(page); |
c8b97818 CM |
505 | memset(kaddr + offset, 0, |
506 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 507 | kunmap_atomic(kaddr); |
c8b97818 CM |
508 | } |
509 | will_compress = 1; | |
510 | } | |
511 | } | |
560f7d75 | 512 | cont: |
c8b97818 CM |
513 | if (start == 0) { |
514 | /* lets try to make an inline extent */ | |
771ed689 | 515 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 516 | /* we didn't compress the entire range, try |
771ed689 | 517 | * to make an uncompressed inline extent. |
c8b97818 | 518 | */ |
00361589 JB |
519 | ret = cow_file_range_inline(root, inode, start, end, |
520 | 0, 0, NULL); | |
c8b97818 | 521 | } else { |
771ed689 | 522 | /* try making a compressed inline extent */ |
00361589 | 523 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
524 | total_compressed, |
525 | compress_type, pages); | |
c8b97818 | 526 | } |
79787eaa | 527 | if (ret <= 0) { |
151a41bc JB |
528 | unsigned long clear_flags = EXTENT_DELALLOC | |
529 | EXTENT_DEFRAG; | |
530 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; | |
531 | ||
771ed689 | 532 | /* |
79787eaa JM |
533 | * inline extent creation worked or returned error, |
534 | * we don't need to create any more async work items. | |
535 | * Unlock and free up our temp pages. | |
771ed689 | 536 | */ |
c2790a2e | 537 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 538 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
539 | PAGE_CLEAR_DIRTY | |
540 | PAGE_SET_WRITEBACK | | |
541 | PAGE_END_WRITEBACK); | |
c8b97818 CM |
542 | goto free_pages_out; |
543 | } | |
544 | } | |
545 | ||
546 | if (will_compress) { | |
547 | /* | |
548 | * we aren't doing an inline extent round the compressed size | |
549 | * up to a block size boundary so the allocator does sane | |
550 | * things | |
551 | */ | |
fda2832f | 552 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
553 | |
554 | /* | |
555 | * one last check to make sure the compression is really a | |
556 | * win, compare the page count read with the blocks on disk | |
557 | */ | |
fda2832f | 558 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
559 | if (total_compressed >= total_in) { |
560 | will_compress = 0; | |
561 | } else { | |
c8b97818 CM |
562 | num_bytes = total_in; |
563 | } | |
564 | } | |
565 | if (!will_compress && pages) { | |
566 | /* | |
567 | * the compression code ran but failed to make things smaller, | |
568 | * free any pages it allocated and our page pointer array | |
569 | */ | |
570 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 571 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
572 | page_cache_release(pages[i]); |
573 | } | |
574 | kfree(pages); | |
575 | pages = NULL; | |
576 | total_compressed = 0; | |
577 | nr_pages_ret = 0; | |
578 | ||
579 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
580 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
581 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 582 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 583 | } |
c8b97818 | 584 | } |
771ed689 CM |
585 | if (will_compress) { |
586 | *num_added += 1; | |
c8b97818 | 587 | |
771ed689 CM |
588 | /* the async work queues will take care of doing actual |
589 | * allocation on disk for these compressed pages, | |
590 | * and will submit them to the elevator. | |
591 | */ | |
592 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
593 | total_compressed, pages, nr_pages_ret, |
594 | compress_type); | |
179e29e4 | 595 | |
24ae6365 | 596 | if (start + num_bytes < end) { |
771ed689 CM |
597 | start += num_bytes; |
598 | pages = NULL; | |
599 | cond_resched(); | |
600 | goto again; | |
601 | } | |
602 | } else { | |
f03d9301 | 603 | cleanup_and_bail_uncompressed: |
771ed689 CM |
604 | /* |
605 | * No compression, but we still need to write the pages in | |
606 | * the file we've been given so far. redirty the locked | |
607 | * page if it corresponds to our extent and set things up | |
608 | * for the async work queue to run cow_file_range to do | |
609 | * the normal delalloc dance | |
610 | */ | |
611 | if (page_offset(locked_page) >= start && | |
612 | page_offset(locked_page) <= end) { | |
613 | __set_page_dirty_nobuffers(locked_page); | |
614 | /* unlocked later on in the async handlers */ | |
615 | } | |
4adaa611 CM |
616 | if (redirty) |
617 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
618 | add_async_extent(async_cow, start, end - start + 1, |
619 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
620 | *num_added += 1; |
621 | } | |
3b951516 | 622 | |
771ed689 | 623 | out: |
79787eaa | 624 | return ret; |
771ed689 CM |
625 | |
626 | free_pages_out: | |
627 | for (i = 0; i < nr_pages_ret; i++) { | |
628 | WARN_ON(pages[i]->mapping); | |
629 | page_cache_release(pages[i]); | |
630 | } | |
d397712b | 631 | kfree(pages); |
771ed689 CM |
632 | |
633 | goto out; | |
634 | } | |
635 | ||
636 | /* | |
637 | * phase two of compressed writeback. This is the ordered portion | |
638 | * of the code, which only gets called in the order the work was | |
639 | * queued. We walk all the async extents created by compress_file_range | |
640 | * and send them down to the disk. | |
641 | */ | |
642 | static noinline int submit_compressed_extents(struct inode *inode, | |
643 | struct async_cow *async_cow) | |
644 | { | |
645 | struct async_extent *async_extent; | |
646 | u64 alloc_hint = 0; | |
771ed689 CM |
647 | struct btrfs_key ins; |
648 | struct extent_map *em; | |
649 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
650 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
651 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 652 | int ret = 0; |
771ed689 CM |
653 | |
654 | if (list_empty(&async_cow->extents)) | |
655 | return 0; | |
656 | ||
3e04e7f1 | 657 | again: |
d397712b | 658 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
659 | async_extent = list_entry(async_cow->extents.next, |
660 | struct async_extent, list); | |
661 | list_del(&async_extent->list); | |
c8b97818 | 662 | |
771ed689 CM |
663 | io_tree = &BTRFS_I(inode)->io_tree; |
664 | ||
f5a84ee3 | 665 | retry: |
771ed689 CM |
666 | /* did the compression code fall back to uncompressed IO? */ |
667 | if (!async_extent->pages) { | |
668 | int page_started = 0; | |
669 | unsigned long nr_written = 0; | |
670 | ||
671 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 672 | async_extent->start + |
d0082371 | 673 | async_extent->ram_size - 1); |
771ed689 CM |
674 | |
675 | /* allocate blocks */ | |
f5a84ee3 JB |
676 | ret = cow_file_range(inode, async_cow->locked_page, |
677 | async_extent->start, | |
678 | async_extent->start + | |
679 | async_extent->ram_size - 1, | |
680 | &page_started, &nr_written, 0); | |
771ed689 | 681 | |
79787eaa JM |
682 | /* JDM XXX */ |
683 | ||
771ed689 CM |
684 | /* |
685 | * if page_started, cow_file_range inserted an | |
686 | * inline extent and took care of all the unlocking | |
687 | * and IO for us. Otherwise, we need to submit | |
688 | * all those pages down to the drive. | |
689 | */ | |
f5a84ee3 | 690 | if (!page_started && !ret) |
771ed689 CM |
691 | extent_write_locked_range(io_tree, |
692 | inode, async_extent->start, | |
d397712b | 693 | async_extent->start + |
771ed689 CM |
694 | async_extent->ram_size - 1, |
695 | btrfs_get_extent, | |
696 | WB_SYNC_ALL); | |
3e04e7f1 JB |
697 | else if (ret) |
698 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
699 | kfree(async_extent); |
700 | cond_resched(); | |
701 | continue; | |
702 | } | |
703 | ||
704 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 705 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 706 | |
00361589 | 707 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
708 | async_extent->compressed_size, |
709 | async_extent->compressed_size, | |
e570fd27 | 710 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 JB |
711 | if (ret) { |
712 | int i; | |
3e04e7f1 | 713 | |
f5a84ee3 JB |
714 | for (i = 0; i < async_extent->nr_pages; i++) { |
715 | WARN_ON(async_extent->pages[i]->mapping); | |
716 | page_cache_release(async_extent->pages[i]); | |
717 | } | |
718 | kfree(async_extent->pages); | |
719 | async_extent->nr_pages = 0; | |
720 | async_extent->pages = NULL; | |
3e04e7f1 | 721 | |
fdf8e2ea JB |
722 | if (ret == -ENOSPC) { |
723 | unlock_extent(io_tree, async_extent->start, | |
724 | async_extent->start + | |
725 | async_extent->ram_size - 1); | |
ce62003f LB |
726 | |
727 | /* | |
728 | * we need to redirty the pages if we decide to | |
729 | * fallback to uncompressed IO, otherwise we | |
730 | * will not submit these pages down to lower | |
731 | * layers. | |
732 | */ | |
733 | extent_range_redirty_for_io(inode, | |
734 | async_extent->start, | |
735 | async_extent->start + | |
736 | async_extent->ram_size - 1); | |
737 | ||
79787eaa | 738 | goto retry; |
fdf8e2ea | 739 | } |
3e04e7f1 | 740 | goto out_free; |
f5a84ee3 JB |
741 | } |
742 | ||
c2167754 YZ |
743 | /* |
744 | * here we're doing allocation and writeback of the | |
745 | * compressed pages | |
746 | */ | |
747 | btrfs_drop_extent_cache(inode, async_extent->start, | |
748 | async_extent->start + | |
749 | async_extent->ram_size - 1, 0); | |
750 | ||
172ddd60 | 751 | em = alloc_extent_map(); |
b9aa55be LB |
752 | if (!em) { |
753 | ret = -ENOMEM; | |
3e04e7f1 | 754 | goto out_free_reserve; |
b9aa55be | 755 | } |
771ed689 CM |
756 | em->start = async_extent->start; |
757 | em->len = async_extent->ram_size; | |
445a6944 | 758 | em->orig_start = em->start; |
2ab28f32 JB |
759 | em->mod_start = em->start; |
760 | em->mod_len = em->len; | |
c8b97818 | 761 | |
771ed689 CM |
762 | em->block_start = ins.objectid; |
763 | em->block_len = ins.offset; | |
b4939680 | 764 | em->orig_block_len = ins.offset; |
cc95bef6 | 765 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 766 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 767 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
768 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
769 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 770 | em->generation = -1; |
771ed689 | 771 | |
d397712b | 772 | while (1) { |
890871be | 773 | write_lock(&em_tree->lock); |
09a2a8f9 | 774 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 775 | write_unlock(&em_tree->lock); |
771ed689 CM |
776 | if (ret != -EEXIST) { |
777 | free_extent_map(em); | |
778 | break; | |
779 | } | |
780 | btrfs_drop_extent_cache(inode, async_extent->start, | |
781 | async_extent->start + | |
782 | async_extent->ram_size - 1, 0); | |
783 | } | |
784 | ||
3e04e7f1 JB |
785 | if (ret) |
786 | goto out_free_reserve; | |
787 | ||
261507a0 LZ |
788 | ret = btrfs_add_ordered_extent_compress(inode, |
789 | async_extent->start, | |
790 | ins.objectid, | |
791 | async_extent->ram_size, | |
792 | ins.offset, | |
793 | BTRFS_ORDERED_COMPRESSED, | |
794 | async_extent->compress_type); | |
d9f85963 FM |
795 | if (ret) { |
796 | btrfs_drop_extent_cache(inode, async_extent->start, | |
797 | async_extent->start + | |
798 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 799 | goto out_free_reserve; |
d9f85963 | 800 | } |
771ed689 | 801 | |
771ed689 CM |
802 | /* |
803 | * clear dirty, set writeback and unlock the pages. | |
804 | */ | |
c2790a2e | 805 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
806 | async_extent->start + |
807 | async_extent->ram_size - 1, | |
151a41bc JB |
808 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
809 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 810 | PAGE_SET_WRITEBACK); |
771ed689 | 811 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
812 | async_extent->start, |
813 | async_extent->ram_size, | |
814 | ins.objectid, | |
815 | ins.offset, async_extent->pages, | |
816 | async_extent->nr_pages); | |
771ed689 CM |
817 | alloc_hint = ins.objectid + ins.offset; |
818 | kfree(async_extent); | |
3e04e7f1 JB |
819 | if (ret) |
820 | goto out; | |
771ed689 CM |
821 | cond_resched(); |
822 | } | |
79787eaa JM |
823 | ret = 0; |
824 | out: | |
825 | return ret; | |
3e04e7f1 | 826 | out_free_reserve: |
e570fd27 | 827 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 828 | out_free: |
c2790a2e | 829 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
830 | async_extent->start + |
831 | async_extent->ram_size - 1, | |
c2790a2e | 832 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
833 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
834 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
835 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 836 | kfree(async_extent); |
3e04e7f1 | 837 | goto again; |
771ed689 CM |
838 | } |
839 | ||
4b46fce2 JB |
840 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
841 | u64 num_bytes) | |
842 | { | |
843 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
844 | struct extent_map *em; | |
845 | u64 alloc_hint = 0; | |
846 | ||
847 | read_lock(&em_tree->lock); | |
848 | em = search_extent_mapping(em_tree, start, num_bytes); | |
849 | if (em) { | |
850 | /* | |
851 | * if block start isn't an actual block number then find the | |
852 | * first block in this inode and use that as a hint. If that | |
853 | * block is also bogus then just don't worry about it. | |
854 | */ | |
855 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
856 | free_extent_map(em); | |
857 | em = search_extent_mapping(em_tree, 0, 0); | |
858 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
859 | alloc_hint = em->block_start; | |
860 | if (em) | |
861 | free_extent_map(em); | |
862 | } else { | |
863 | alloc_hint = em->block_start; | |
864 | free_extent_map(em); | |
865 | } | |
866 | } | |
867 | read_unlock(&em_tree->lock); | |
868 | ||
869 | return alloc_hint; | |
870 | } | |
871 | ||
771ed689 CM |
872 | /* |
873 | * when extent_io.c finds a delayed allocation range in the file, | |
874 | * the call backs end up in this code. The basic idea is to | |
875 | * allocate extents on disk for the range, and create ordered data structs | |
876 | * in ram to track those extents. | |
877 | * | |
878 | * locked_page is the page that writepage had locked already. We use | |
879 | * it to make sure we don't do extra locks or unlocks. | |
880 | * | |
881 | * *page_started is set to one if we unlock locked_page and do everything | |
882 | * required to start IO on it. It may be clean and already done with | |
883 | * IO when we return. | |
884 | */ | |
00361589 JB |
885 | static noinline int cow_file_range(struct inode *inode, |
886 | struct page *locked_page, | |
887 | u64 start, u64 end, int *page_started, | |
888 | unsigned long *nr_written, | |
889 | int unlock) | |
771ed689 | 890 | { |
00361589 | 891 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
892 | u64 alloc_hint = 0; |
893 | u64 num_bytes; | |
894 | unsigned long ram_size; | |
895 | u64 disk_num_bytes; | |
896 | u64 cur_alloc_size; | |
897 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
898 | struct btrfs_key ins; |
899 | struct extent_map *em; | |
900 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
901 | int ret = 0; | |
902 | ||
02ecd2c2 JB |
903 | if (btrfs_is_free_space_inode(inode)) { |
904 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
905 | ret = -EINVAL; |
906 | goto out_unlock; | |
02ecd2c2 | 907 | } |
771ed689 | 908 | |
fda2832f | 909 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
910 | num_bytes = max(blocksize, num_bytes); |
911 | disk_num_bytes = num_bytes; | |
771ed689 | 912 | |
4cb5300b | 913 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
914 | if (num_bytes < 64 * 1024 && |
915 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 916 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 917 | |
771ed689 CM |
918 | if (start == 0) { |
919 | /* lets try to make an inline extent */ | |
00361589 JB |
920 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
921 | NULL); | |
771ed689 | 922 | if (ret == 0) { |
c2790a2e JB |
923 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
924 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 925 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
926 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
927 | PAGE_END_WRITEBACK); | |
c2167754 | 928 | |
771ed689 CM |
929 | *nr_written = *nr_written + |
930 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
931 | *page_started = 1; | |
771ed689 | 932 | goto out; |
79787eaa | 933 | } else if (ret < 0) { |
79787eaa | 934 | goto out_unlock; |
771ed689 CM |
935 | } |
936 | } | |
937 | ||
938 | BUG_ON(disk_num_bytes > | |
6c41761f | 939 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 940 | |
4b46fce2 | 941 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
942 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
943 | ||
d397712b | 944 | while (disk_num_bytes > 0) { |
a791e35e CM |
945 | unsigned long op; |
946 | ||
287a0ab9 | 947 | cur_alloc_size = disk_num_bytes; |
00361589 | 948 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 949 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 950 | &ins, 1, 1); |
00361589 | 951 | if (ret < 0) |
79787eaa | 952 | goto out_unlock; |
d397712b | 953 | |
172ddd60 | 954 | em = alloc_extent_map(); |
b9aa55be LB |
955 | if (!em) { |
956 | ret = -ENOMEM; | |
ace68bac | 957 | goto out_reserve; |
b9aa55be | 958 | } |
e6dcd2dc | 959 | em->start = start; |
445a6944 | 960 | em->orig_start = em->start; |
771ed689 CM |
961 | ram_size = ins.offset; |
962 | em->len = ins.offset; | |
2ab28f32 JB |
963 | em->mod_start = em->start; |
964 | em->mod_len = em->len; | |
c8b97818 | 965 | |
e6dcd2dc | 966 | em->block_start = ins.objectid; |
c8b97818 | 967 | em->block_len = ins.offset; |
b4939680 | 968 | em->orig_block_len = ins.offset; |
cc95bef6 | 969 | em->ram_bytes = ram_size; |
e6dcd2dc | 970 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 971 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 972 | em->generation = -1; |
c8b97818 | 973 | |
d397712b | 974 | while (1) { |
890871be | 975 | write_lock(&em_tree->lock); |
09a2a8f9 | 976 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 977 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
978 | if (ret != -EEXIST) { |
979 | free_extent_map(em); | |
980 | break; | |
981 | } | |
982 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 983 | start + ram_size - 1, 0); |
e6dcd2dc | 984 | } |
ace68bac LB |
985 | if (ret) |
986 | goto out_reserve; | |
e6dcd2dc | 987 | |
98d20f67 | 988 | cur_alloc_size = ins.offset; |
e6dcd2dc | 989 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 990 | ram_size, cur_alloc_size, 0); |
ace68bac | 991 | if (ret) |
d9f85963 | 992 | goto out_drop_extent_cache; |
c8b97818 | 993 | |
17d217fe YZ |
994 | if (root->root_key.objectid == |
995 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
996 | ret = btrfs_reloc_clone_csums(inode, start, | |
997 | cur_alloc_size); | |
00361589 | 998 | if (ret) |
d9f85963 | 999 | goto out_drop_extent_cache; |
17d217fe YZ |
1000 | } |
1001 | ||
d397712b | 1002 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1003 | break; |
d397712b | 1004 | |
c8b97818 CM |
1005 | /* we're not doing compressed IO, don't unlock the first |
1006 | * page (which the caller expects to stay locked), don't | |
1007 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1008 | * |
1009 | * Do set the Private2 bit so we know this page was properly | |
1010 | * setup for writepage | |
c8b97818 | 1011 | */ |
c2790a2e JB |
1012 | op = unlock ? PAGE_UNLOCK : 0; |
1013 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1014 | |
c2790a2e JB |
1015 | extent_clear_unlock_delalloc(inode, start, |
1016 | start + ram_size - 1, locked_page, | |
1017 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1018 | op); | |
c8b97818 | 1019 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1020 | num_bytes -= cur_alloc_size; |
1021 | alloc_hint = ins.objectid + ins.offset; | |
1022 | start += cur_alloc_size; | |
b888db2b | 1023 | } |
79787eaa | 1024 | out: |
be20aa9d | 1025 | return ret; |
b7d5b0a8 | 1026 | |
d9f85963 FM |
1027 | out_drop_extent_cache: |
1028 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1029 | out_reserve: |
e570fd27 | 1030 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1031 | out_unlock: |
c2790a2e | 1032 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1033 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1034 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1035 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1036 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1037 | goto out; |
771ed689 | 1038 | } |
c8b97818 | 1039 | |
771ed689 CM |
1040 | /* |
1041 | * work queue call back to started compression on a file and pages | |
1042 | */ | |
1043 | static noinline void async_cow_start(struct btrfs_work *work) | |
1044 | { | |
1045 | struct async_cow *async_cow; | |
1046 | int num_added = 0; | |
1047 | async_cow = container_of(work, struct async_cow, work); | |
1048 | ||
1049 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1050 | async_cow->start, async_cow->end, async_cow, | |
1051 | &num_added); | |
8180ef88 | 1052 | if (num_added == 0) { |
cb77fcd8 | 1053 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1054 | async_cow->inode = NULL; |
8180ef88 | 1055 | } |
771ed689 CM |
1056 | } |
1057 | ||
1058 | /* | |
1059 | * work queue call back to submit previously compressed pages | |
1060 | */ | |
1061 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1062 | { | |
1063 | struct async_cow *async_cow; | |
1064 | struct btrfs_root *root; | |
1065 | unsigned long nr_pages; | |
1066 | ||
1067 | async_cow = container_of(work, struct async_cow, work); | |
1068 | ||
1069 | root = async_cow->root; | |
1070 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1071 | PAGE_CACHE_SHIFT; | |
1072 | ||
66657b31 | 1073 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1074 | 5 * 1024 * 1024 && |
771ed689 CM |
1075 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1076 | wake_up(&root->fs_info->async_submit_wait); | |
1077 | ||
d397712b | 1078 | if (async_cow->inode) |
771ed689 | 1079 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1080 | } |
c8b97818 | 1081 | |
771ed689 CM |
1082 | static noinline void async_cow_free(struct btrfs_work *work) |
1083 | { | |
1084 | struct async_cow *async_cow; | |
1085 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1086 | if (async_cow->inode) |
cb77fcd8 | 1087 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1088 | kfree(async_cow); |
1089 | } | |
1090 | ||
1091 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1092 | u64 start, u64 end, int *page_started, | |
1093 | unsigned long *nr_written) | |
1094 | { | |
1095 | struct async_cow *async_cow; | |
1096 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1097 | unsigned long nr_pages; | |
1098 | u64 cur_end; | |
287082b0 | 1099 | int limit = 10 * 1024 * 1024; |
771ed689 | 1100 | |
a3429ab7 CM |
1101 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1102 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1103 | while (start < end) { |
771ed689 | 1104 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1105 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1106 | async_cow->inode = igrab(inode); |
771ed689 CM |
1107 | async_cow->root = root; |
1108 | async_cow->locked_page = locked_page; | |
1109 | async_cow->start = start; | |
1110 | ||
f79707b0 WS |
1111 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1112 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1113 | cur_end = end; |
1114 | else | |
1115 | cur_end = min(end, start + 512 * 1024 - 1); | |
1116 | ||
1117 | async_cow->end = cur_end; | |
1118 | INIT_LIST_HEAD(&async_cow->extents); | |
1119 | ||
9e0af237 LB |
1120 | btrfs_init_work(&async_cow->work, |
1121 | btrfs_delalloc_helper, | |
1122 | async_cow_start, async_cow_submit, | |
1123 | async_cow_free); | |
771ed689 | 1124 | |
771ed689 CM |
1125 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1126 | PAGE_CACHE_SHIFT; | |
1127 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1128 | ||
afe3d242 QW |
1129 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1130 | &async_cow->work); | |
771ed689 CM |
1131 | |
1132 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1133 | wait_event(root->fs_info->async_submit_wait, | |
1134 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1135 | limit)); | |
1136 | } | |
1137 | ||
d397712b | 1138 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1139 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1140 | wait_event(root->fs_info->async_submit_wait, | |
1141 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1142 | 0)); | |
1143 | } | |
1144 | ||
1145 | *nr_written += nr_pages; | |
1146 | start = cur_end + 1; | |
1147 | } | |
1148 | *page_started = 1; | |
1149 | return 0; | |
be20aa9d CM |
1150 | } |
1151 | ||
d397712b | 1152 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1153 | u64 bytenr, u64 num_bytes) |
1154 | { | |
1155 | int ret; | |
1156 | struct btrfs_ordered_sum *sums; | |
1157 | LIST_HEAD(list); | |
1158 | ||
07d400a6 | 1159 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1160 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1161 | if (ret == 0 && list_empty(&list)) |
1162 | return 0; | |
1163 | ||
1164 | while (!list_empty(&list)) { | |
1165 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1166 | list_del(&sums->list); | |
1167 | kfree(sums); | |
1168 | } | |
1169 | return 1; | |
1170 | } | |
1171 | ||
d352ac68 CM |
1172 | /* |
1173 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1174 | * of the extents that exist in the file, and COWs the file as required. | |
1175 | * | |
1176 | * If no cow copies or snapshots exist, we write directly to the existing | |
1177 | * blocks on disk | |
1178 | */ | |
7f366cfe CM |
1179 | static noinline int run_delalloc_nocow(struct inode *inode, |
1180 | struct page *locked_page, | |
771ed689 CM |
1181 | u64 start, u64 end, int *page_started, int force, |
1182 | unsigned long *nr_written) | |
be20aa9d | 1183 | { |
be20aa9d | 1184 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1185 | struct btrfs_trans_handle *trans; |
be20aa9d | 1186 | struct extent_buffer *leaf; |
be20aa9d | 1187 | struct btrfs_path *path; |
80ff3856 | 1188 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1189 | struct btrfs_key found_key; |
80ff3856 YZ |
1190 | u64 cow_start; |
1191 | u64 cur_offset; | |
1192 | u64 extent_end; | |
5d4f98a2 | 1193 | u64 extent_offset; |
80ff3856 YZ |
1194 | u64 disk_bytenr; |
1195 | u64 num_bytes; | |
b4939680 | 1196 | u64 disk_num_bytes; |
cc95bef6 | 1197 | u64 ram_bytes; |
80ff3856 | 1198 | int extent_type; |
79787eaa | 1199 | int ret, err; |
d899e052 | 1200 | int type; |
80ff3856 YZ |
1201 | int nocow; |
1202 | int check_prev = 1; | |
82d5902d | 1203 | bool nolock; |
33345d01 | 1204 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1205 | |
1206 | path = btrfs_alloc_path(); | |
17ca04af | 1207 | if (!path) { |
c2790a2e JB |
1208 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1209 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1210 | EXTENT_DO_ACCOUNTING | |
1211 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1212 | PAGE_CLEAR_DIRTY | |
1213 | PAGE_SET_WRITEBACK | | |
1214 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1215 | return -ENOMEM; |
17ca04af | 1216 | } |
82d5902d | 1217 | |
83eea1f1 | 1218 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1219 | |
1220 | if (nolock) | |
7a7eaa40 | 1221 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1222 | else |
7a7eaa40 | 1223 | trans = btrfs_join_transaction(root); |
ff5714cc | 1224 | |
79787eaa | 1225 | if (IS_ERR(trans)) { |
c2790a2e JB |
1226 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1227 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1228 | EXTENT_DO_ACCOUNTING | |
1229 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1230 | PAGE_CLEAR_DIRTY | |
1231 | PAGE_SET_WRITEBACK | | |
1232 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1233 | btrfs_free_path(path); |
1234 | return PTR_ERR(trans); | |
1235 | } | |
1236 | ||
74b21075 | 1237 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1238 | |
80ff3856 YZ |
1239 | cow_start = (u64)-1; |
1240 | cur_offset = start; | |
1241 | while (1) { | |
33345d01 | 1242 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1243 | cur_offset, 0); |
d788a349 | 1244 | if (ret < 0) |
79787eaa | 1245 | goto error; |
80ff3856 YZ |
1246 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1247 | leaf = path->nodes[0]; | |
1248 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1249 | path->slots[0] - 1); | |
33345d01 | 1250 | if (found_key.objectid == ino && |
80ff3856 YZ |
1251 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1252 | path->slots[0]--; | |
1253 | } | |
1254 | check_prev = 0; | |
1255 | next_slot: | |
1256 | leaf = path->nodes[0]; | |
1257 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1258 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1259 | if (ret < 0) |
79787eaa | 1260 | goto error; |
80ff3856 YZ |
1261 | if (ret > 0) |
1262 | break; | |
1263 | leaf = path->nodes[0]; | |
1264 | } | |
be20aa9d | 1265 | |
80ff3856 YZ |
1266 | nocow = 0; |
1267 | disk_bytenr = 0; | |
17d217fe | 1268 | num_bytes = 0; |
80ff3856 YZ |
1269 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1270 | ||
33345d01 | 1271 | if (found_key.objectid > ino || |
80ff3856 YZ |
1272 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1273 | found_key.offset > end) | |
1274 | break; | |
1275 | ||
1276 | if (found_key.offset > cur_offset) { | |
1277 | extent_end = found_key.offset; | |
e9061e21 | 1278 | extent_type = 0; |
80ff3856 YZ |
1279 | goto out_check; |
1280 | } | |
1281 | ||
1282 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1283 | struct btrfs_file_extent_item); | |
1284 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1285 | ||
cc95bef6 | 1286 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1287 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1288 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1289 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1290 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1291 | extent_end = found_key.offset + |
1292 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1293 | disk_num_bytes = |
1294 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1295 | if (extent_end <= start) { |
1296 | path->slots[0]++; | |
1297 | goto next_slot; | |
1298 | } | |
17d217fe YZ |
1299 | if (disk_bytenr == 0) |
1300 | goto out_check; | |
80ff3856 YZ |
1301 | if (btrfs_file_extent_compression(leaf, fi) || |
1302 | btrfs_file_extent_encryption(leaf, fi) || | |
1303 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1304 | goto out_check; | |
d899e052 YZ |
1305 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1306 | goto out_check; | |
d2fb3437 | 1307 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1308 | goto out_check; |
33345d01 | 1309 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1310 | found_key.offset - |
1311 | extent_offset, disk_bytenr)) | |
17d217fe | 1312 | goto out_check; |
5d4f98a2 | 1313 | disk_bytenr += extent_offset; |
17d217fe YZ |
1314 | disk_bytenr += cur_offset - found_key.offset; |
1315 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1316 | /* |
1317 | * if there are pending snapshots for this root, | |
1318 | * we fall into common COW way. | |
1319 | */ | |
1320 | if (!nolock) { | |
1321 | err = btrfs_start_nocow_write(root); | |
1322 | if (!err) | |
1323 | goto out_check; | |
1324 | } | |
17d217fe YZ |
1325 | /* |
1326 | * force cow if csum exists in the range. | |
1327 | * this ensure that csum for a given extent are | |
1328 | * either valid or do not exist. | |
1329 | */ | |
1330 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1331 | goto out_check; | |
80ff3856 YZ |
1332 | nocow = 1; |
1333 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1334 | extent_end = found_key.offset + | |
514ac8ad CM |
1335 | btrfs_file_extent_inline_len(leaf, |
1336 | path->slots[0], fi); | |
80ff3856 YZ |
1337 | extent_end = ALIGN(extent_end, root->sectorsize); |
1338 | } else { | |
1339 | BUG_ON(1); | |
1340 | } | |
1341 | out_check: | |
1342 | if (extent_end <= start) { | |
1343 | path->slots[0]++; | |
e9894fd3 WS |
1344 | if (!nolock && nocow) |
1345 | btrfs_end_nocow_write(root); | |
80ff3856 YZ |
1346 | goto next_slot; |
1347 | } | |
1348 | if (!nocow) { | |
1349 | if (cow_start == (u64)-1) | |
1350 | cow_start = cur_offset; | |
1351 | cur_offset = extent_end; | |
1352 | if (cur_offset > end) | |
1353 | break; | |
1354 | path->slots[0]++; | |
1355 | goto next_slot; | |
7ea394f1 YZ |
1356 | } |
1357 | ||
b3b4aa74 | 1358 | btrfs_release_path(path); |
80ff3856 | 1359 | if (cow_start != (u64)-1) { |
00361589 JB |
1360 | ret = cow_file_range(inode, locked_page, |
1361 | cow_start, found_key.offset - 1, | |
1362 | page_started, nr_written, 1); | |
e9894fd3 WS |
1363 | if (ret) { |
1364 | if (!nolock && nocow) | |
1365 | btrfs_end_nocow_write(root); | |
79787eaa | 1366 | goto error; |
e9894fd3 | 1367 | } |
80ff3856 | 1368 | cow_start = (u64)-1; |
7ea394f1 | 1369 | } |
80ff3856 | 1370 | |
d899e052 YZ |
1371 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1372 | struct extent_map *em; | |
1373 | struct extent_map_tree *em_tree; | |
1374 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1375 | em = alloc_extent_map(); |
79787eaa | 1376 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1377 | em->start = cur_offset; |
70c8a91c | 1378 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1379 | em->len = num_bytes; |
1380 | em->block_len = num_bytes; | |
1381 | em->block_start = disk_bytenr; | |
b4939680 | 1382 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1383 | em->ram_bytes = ram_bytes; |
d899e052 | 1384 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1385 | em->mod_start = em->start; |
1386 | em->mod_len = em->len; | |
d899e052 | 1387 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1388 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1389 | em->generation = -1; |
d899e052 | 1390 | while (1) { |
890871be | 1391 | write_lock(&em_tree->lock); |
09a2a8f9 | 1392 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1393 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1394 | if (ret != -EEXIST) { |
1395 | free_extent_map(em); | |
1396 | break; | |
1397 | } | |
1398 | btrfs_drop_extent_cache(inode, em->start, | |
1399 | em->start + em->len - 1, 0); | |
1400 | } | |
1401 | type = BTRFS_ORDERED_PREALLOC; | |
1402 | } else { | |
1403 | type = BTRFS_ORDERED_NOCOW; | |
1404 | } | |
80ff3856 YZ |
1405 | |
1406 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1407 | num_bytes, num_bytes, type); |
79787eaa | 1408 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1409 | |
efa56464 YZ |
1410 | if (root->root_key.objectid == |
1411 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1412 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1413 | num_bytes); | |
e9894fd3 WS |
1414 | if (ret) { |
1415 | if (!nolock && nocow) | |
1416 | btrfs_end_nocow_write(root); | |
79787eaa | 1417 | goto error; |
e9894fd3 | 1418 | } |
efa56464 YZ |
1419 | } |
1420 | ||
c2790a2e JB |
1421 | extent_clear_unlock_delalloc(inode, cur_offset, |
1422 | cur_offset + num_bytes - 1, | |
1423 | locked_page, EXTENT_LOCKED | | |
1424 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1425 | PAGE_SET_PRIVATE2); | |
e9894fd3 WS |
1426 | if (!nolock && nocow) |
1427 | btrfs_end_nocow_write(root); | |
80ff3856 YZ |
1428 | cur_offset = extent_end; |
1429 | if (cur_offset > end) | |
1430 | break; | |
be20aa9d | 1431 | } |
b3b4aa74 | 1432 | btrfs_release_path(path); |
80ff3856 | 1433 | |
17ca04af | 1434 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1435 | cow_start = cur_offset; |
17ca04af JB |
1436 | cur_offset = end; |
1437 | } | |
1438 | ||
80ff3856 | 1439 | if (cow_start != (u64)-1) { |
00361589 JB |
1440 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1441 | page_started, nr_written, 1); | |
d788a349 | 1442 | if (ret) |
79787eaa | 1443 | goto error; |
80ff3856 YZ |
1444 | } |
1445 | ||
79787eaa | 1446 | error: |
a698d075 | 1447 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1448 | if (!ret) |
1449 | ret = err; | |
1450 | ||
17ca04af | 1451 | if (ret && cur_offset < end) |
c2790a2e JB |
1452 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1453 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1454 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1455 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1456 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1457 | PAGE_SET_WRITEBACK | |
1458 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1459 | btrfs_free_path(path); |
79787eaa | 1460 | return ret; |
be20aa9d CM |
1461 | } |
1462 | ||
47059d93 WS |
1463 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1464 | { | |
1465 | ||
1466 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1467 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1468 | return 0; | |
1469 | ||
1470 | /* | |
1471 | * @defrag_bytes is a hint value, no spinlock held here, | |
1472 | * if is not zero, it means the file is defragging. | |
1473 | * Force cow if given extent needs to be defragged. | |
1474 | */ | |
1475 | if (BTRFS_I(inode)->defrag_bytes && | |
1476 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1477 | EXTENT_DEFRAG, 0, NULL)) | |
1478 | return 1; | |
1479 | ||
1480 | return 0; | |
1481 | } | |
1482 | ||
d352ac68 CM |
1483 | /* |
1484 | * extent_io.c call back to do delayed allocation processing | |
1485 | */ | |
c8b97818 | 1486 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1487 | u64 start, u64 end, int *page_started, |
1488 | unsigned long *nr_written) | |
be20aa9d | 1489 | { |
be20aa9d | 1490 | int ret; |
7f366cfe | 1491 | struct btrfs_root *root = BTRFS_I(inode)->root; |
47059d93 | 1492 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1493 | |
47059d93 | 1494 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1495 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1496 | page_started, 1, nr_written); |
47059d93 | 1497 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1498 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1499 | page_started, 0, nr_written); |
7ddf5a42 JB |
1500 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1501 | !(BTRFS_I(inode)->force_compress) && | |
1502 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1503 | ret = cow_file_range(inode, locked_page, start, end, |
1504 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1505 | } else { |
1506 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1507 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1508 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1509 | page_started, nr_written); |
7ddf5a42 | 1510 | } |
b888db2b CM |
1511 | return ret; |
1512 | } | |
1513 | ||
1bf85046 JM |
1514 | static void btrfs_split_extent_hook(struct inode *inode, |
1515 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1516 | { |
0ca1f7ce | 1517 | /* not delalloc, ignore it */ |
9ed74f2d | 1518 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1519 | return; |
9ed74f2d | 1520 | |
9e0baf60 JB |
1521 | spin_lock(&BTRFS_I(inode)->lock); |
1522 | BTRFS_I(inode)->outstanding_extents++; | |
1523 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1524 | } |
1525 | ||
1526 | /* | |
1527 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1528 | * extents so we can keep track of new extents that are just merged onto old | |
1529 | * extents, such as when we are doing sequential writes, so we can properly | |
1530 | * account for the metadata space we'll need. | |
1531 | */ | |
1bf85046 JM |
1532 | static void btrfs_merge_extent_hook(struct inode *inode, |
1533 | struct extent_state *new, | |
1534 | struct extent_state *other) | |
9ed74f2d | 1535 | { |
9ed74f2d JB |
1536 | /* not delalloc, ignore it */ |
1537 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1538 | return; |
9ed74f2d | 1539 | |
9e0baf60 JB |
1540 | spin_lock(&BTRFS_I(inode)->lock); |
1541 | BTRFS_I(inode)->outstanding_extents--; | |
1542 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1543 | } |
1544 | ||
eb73c1b7 MX |
1545 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1546 | struct inode *inode) | |
1547 | { | |
1548 | spin_lock(&root->delalloc_lock); | |
1549 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1550 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1551 | &root->delalloc_inodes); | |
1552 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1553 | &BTRFS_I(inode)->runtime_flags); | |
1554 | root->nr_delalloc_inodes++; | |
1555 | if (root->nr_delalloc_inodes == 1) { | |
1556 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1557 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1558 | list_add_tail(&root->delalloc_root, | |
1559 | &root->fs_info->delalloc_roots); | |
1560 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1561 | } | |
1562 | } | |
1563 | spin_unlock(&root->delalloc_lock); | |
1564 | } | |
1565 | ||
1566 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1567 | struct inode *inode) | |
1568 | { | |
1569 | spin_lock(&root->delalloc_lock); | |
1570 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1571 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1572 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1573 | &BTRFS_I(inode)->runtime_flags); | |
1574 | root->nr_delalloc_inodes--; | |
1575 | if (!root->nr_delalloc_inodes) { | |
1576 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1577 | BUG_ON(list_empty(&root->delalloc_root)); | |
1578 | list_del_init(&root->delalloc_root); | |
1579 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1580 | } | |
1581 | } | |
1582 | spin_unlock(&root->delalloc_lock); | |
1583 | } | |
1584 | ||
d352ac68 CM |
1585 | /* |
1586 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1587 | * bytes in this file, and to maintain the list of inodes that | |
1588 | * have pending delalloc work to be done. | |
1589 | */ | |
1bf85046 | 1590 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1591 | struct extent_state *state, unsigned long *bits) |
291d673e | 1592 | { |
9ed74f2d | 1593 | |
47059d93 WS |
1594 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1595 | WARN_ON(1); | |
75eff68e CM |
1596 | /* |
1597 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1598 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1599 | * bit, which is only set or cleared with irqs on |
1600 | */ | |
0ca1f7ce | 1601 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1602 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1603 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1604 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1605 | |
9e0baf60 | 1606 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1607 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1608 | } else { |
1609 | spin_lock(&BTRFS_I(inode)->lock); | |
1610 | BTRFS_I(inode)->outstanding_extents++; | |
1611 | spin_unlock(&BTRFS_I(inode)->lock); | |
1612 | } | |
287a0ab9 | 1613 | |
963d678b MX |
1614 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1615 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1616 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1617 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1618 | if (*bits & EXTENT_DEFRAG) |
1619 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1620 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1621 | &BTRFS_I(inode)->runtime_flags)) |
1622 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1623 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1624 | } |
291d673e CM |
1625 | } |
1626 | ||
d352ac68 CM |
1627 | /* |
1628 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1629 | */ | |
1bf85046 | 1630 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1631 | struct extent_state *state, |
1632 | unsigned long *bits) | |
291d673e | 1633 | { |
47059d93 WS |
1634 | u64 len = state->end + 1 - state->start; |
1635 | ||
1636 | spin_lock(&BTRFS_I(inode)->lock); | |
1637 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1638 | BTRFS_I(inode)->defrag_bytes -= len; | |
1639 | spin_unlock(&BTRFS_I(inode)->lock); | |
1640 | ||
75eff68e CM |
1641 | /* |
1642 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1643 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1644 | * bit, which is only set or cleared with irqs on |
1645 | */ | |
0ca1f7ce | 1646 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1647 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1648 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1649 | |
9e0baf60 | 1650 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1651 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1652 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1653 | spin_lock(&BTRFS_I(inode)->lock); | |
1654 | BTRFS_I(inode)->outstanding_extents--; | |
1655 | spin_unlock(&BTRFS_I(inode)->lock); | |
1656 | } | |
0ca1f7ce | 1657 | |
b6d08f06 JB |
1658 | /* |
1659 | * We don't reserve metadata space for space cache inodes so we | |
1660 | * don't need to call dellalloc_release_metadata if there is an | |
1661 | * error. | |
1662 | */ | |
1663 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1664 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1665 | btrfs_delalloc_release_metadata(inode, len); |
1666 | ||
0cb59c99 | 1667 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1668 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1669 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1670 | |
963d678b MX |
1671 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1672 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1673 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1674 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1675 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1676 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1677 | &BTRFS_I(inode)->runtime_flags)) |
1678 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1679 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1680 | } |
291d673e CM |
1681 | } |
1682 | ||
d352ac68 CM |
1683 | /* |
1684 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1685 | * we don't create bios that span stripes or chunks | |
1686 | */ | |
64a16701 | 1687 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1688 | size_t size, struct bio *bio, |
1689 | unsigned long bio_flags) | |
239b14b3 CM |
1690 | { |
1691 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1692 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1693 | u64 length = 0; |
1694 | u64 map_length; | |
239b14b3 CM |
1695 | int ret; |
1696 | ||
771ed689 CM |
1697 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1698 | return 0; | |
1699 | ||
4f024f37 | 1700 | length = bio->bi_iter.bi_size; |
239b14b3 | 1701 | map_length = length; |
64a16701 | 1702 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1703 | &map_length, NULL, 0); |
3ec706c8 | 1704 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1705 | BUG_ON(ret < 0); |
d397712b | 1706 | if (map_length < length + size) |
239b14b3 | 1707 | return 1; |
3444a972 | 1708 | return 0; |
239b14b3 CM |
1709 | } |
1710 | ||
d352ac68 CM |
1711 | /* |
1712 | * in order to insert checksums into the metadata in large chunks, | |
1713 | * we wait until bio submission time. All the pages in the bio are | |
1714 | * checksummed and sums are attached onto the ordered extent record. | |
1715 | * | |
1716 | * At IO completion time the cums attached on the ordered extent record | |
1717 | * are inserted into the btree | |
1718 | */ | |
d397712b CM |
1719 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1720 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1721 | unsigned long bio_flags, |
1722 | u64 bio_offset) | |
065631f6 | 1723 | { |
065631f6 | 1724 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1725 | int ret = 0; |
e015640f | 1726 | |
d20f7043 | 1727 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1728 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1729 | return 0; |
1730 | } | |
e015640f | 1731 | |
4a69a410 CM |
1732 | /* |
1733 | * in order to insert checksums into the metadata in large chunks, | |
1734 | * we wait until bio submission time. All the pages in the bio are | |
1735 | * checksummed and sums are attached onto the ordered extent record. | |
1736 | * | |
1737 | * At IO completion time the cums attached on the ordered extent record | |
1738 | * are inserted into the btree | |
1739 | */ | |
b2950863 | 1740 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1741 | int mirror_num, unsigned long bio_flags, |
1742 | u64 bio_offset) | |
4a69a410 CM |
1743 | { |
1744 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1745 | int ret; |
1746 | ||
1747 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1748 | if (ret) | |
1749 | bio_endio(bio, ret); | |
1750 | return ret; | |
44b8bd7e CM |
1751 | } |
1752 | ||
d352ac68 | 1753 | /* |
cad321ad CM |
1754 | * extent_io.c submission hook. This does the right thing for csum calculation |
1755 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1756 | */ |
b2950863 | 1757 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1758 | int mirror_num, unsigned long bio_flags, |
1759 | u64 bio_offset) | |
44b8bd7e CM |
1760 | { |
1761 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1762 | int ret = 0; | |
19b9bdb0 | 1763 | int skip_sum; |
0417341e | 1764 | int metadata = 0; |
b812ce28 | 1765 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1766 | |
6cbff00f | 1767 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1768 | |
83eea1f1 | 1769 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1770 | metadata = 2; |
1771 | ||
7b6d91da | 1772 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1773 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1774 | if (ret) | |
61891923 | 1775 | goto out; |
5fd02043 | 1776 | |
d20f7043 | 1777 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1778 | ret = btrfs_submit_compressed_read(inode, bio, |
1779 | mirror_num, | |
1780 | bio_flags); | |
1781 | goto out; | |
c2db1073 TI |
1782 | } else if (!skip_sum) { |
1783 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1784 | if (ret) | |
61891923 | 1785 | goto out; |
c2db1073 | 1786 | } |
4d1b5fb4 | 1787 | goto mapit; |
b812ce28 | 1788 | } else if (async && !skip_sum) { |
17d217fe YZ |
1789 | /* csum items have already been cloned */ |
1790 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1791 | goto mapit; | |
19b9bdb0 | 1792 | /* we're doing a write, do the async checksumming */ |
61891923 | 1793 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1794 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1795 | bio_flags, bio_offset, |
1796 | __btrfs_submit_bio_start, | |
4a69a410 | 1797 | __btrfs_submit_bio_done); |
61891923 | 1798 | goto out; |
b812ce28 JB |
1799 | } else if (!skip_sum) { |
1800 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1801 | if (ret) | |
1802 | goto out; | |
19b9bdb0 CM |
1803 | } |
1804 | ||
0b86a832 | 1805 | mapit: |
61891923 SB |
1806 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1807 | ||
1808 | out: | |
1809 | if (ret < 0) | |
1810 | bio_endio(bio, ret); | |
1811 | return ret; | |
065631f6 | 1812 | } |
6885f308 | 1813 | |
d352ac68 CM |
1814 | /* |
1815 | * given a list of ordered sums record them in the inode. This happens | |
1816 | * at IO completion time based on sums calculated at bio submission time. | |
1817 | */ | |
ba1da2f4 | 1818 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1819 | struct inode *inode, u64 file_offset, |
1820 | struct list_head *list) | |
1821 | { | |
e6dcd2dc CM |
1822 | struct btrfs_ordered_sum *sum; |
1823 | ||
c6e30871 | 1824 | list_for_each_entry(sum, list, list) { |
39847c4d | 1825 | trans->adding_csums = 1; |
d20f7043 CM |
1826 | btrfs_csum_file_blocks(trans, |
1827 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1828 | trans->adding_csums = 0; |
e6dcd2dc CM |
1829 | } |
1830 | return 0; | |
1831 | } | |
1832 | ||
2ac55d41 JB |
1833 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1834 | struct extent_state **cached_state) | |
ea8c2819 | 1835 | { |
6c1500f2 | 1836 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1837 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1838 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1839 | } |
1840 | ||
d352ac68 | 1841 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1842 | struct btrfs_writepage_fixup { |
1843 | struct page *page; | |
1844 | struct btrfs_work work; | |
1845 | }; | |
1846 | ||
b2950863 | 1847 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1848 | { |
1849 | struct btrfs_writepage_fixup *fixup; | |
1850 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1851 | struct extent_state *cached_state = NULL; |
247e743c CM |
1852 | struct page *page; |
1853 | struct inode *inode; | |
1854 | u64 page_start; | |
1855 | u64 page_end; | |
87826df0 | 1856 | int ret; |
247e743c CM |
1857 | |
1858 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1859 | page = fixup->page; | |
4a096752 | 1860 | again: |
247e743c CM |
1861 | lock_page(page); |
1862 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1863 | ClearPageChecked(page); | |
1864 | goto out_page; | |
1865 | } | |
1866 | ||
1867 | inode = page->mapping->host; | |
1868 | page_start = page_offset(page); | |
1869 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1870 | ||
2ac55d41 | 1871 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1872 | &cached_state); |
4a096752 CM |
1873 | |
1874 | /* already ordered? We're done */ | |
8b62b72b | 1875 | if (PagePrivate2(page)) |
247e743c | 1876 | goto out; |
4a096752 CM |
1877 | |
1878 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1879 | if (ordered) { | |
2ac55d41 JB |
1880 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1881 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1882 | unlock_page(page); |
1883 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1884 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1885 | goto again; |
1886 | } | |
247e743c | 1887 | |
87826df0 JM |
1888 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1889 | if (ret) { | |
1890 | mapping_set_error(page->mapping, ret); | |
1891 | end_extent_writepage(page, ret, page_start, page_end); | |
1892 | ClearPageChecked(page); | |
1893 | goto out; | |
1894 | } | |
1895 | ||
2ac55d41 | 1896 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1897 | ClearPageChecked(page); |
87826df0 | 1898 | set_page_dirty(page); |
247e743c | 1899 | out: |
2ac55d41 JB |
1900 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1901 | &cached_state, GFP_NOFS); | |
247e743c CM |
1902 | out_page: |
1903 | unlock_page(page); | |
1904 | page_cache_release(page); | |
b897abec | 1905 | kfree(fixup); |
247e743c CM |
1906 | } |
1907 | ||
1908 | /* | |
1909 | * There are a few paths in the higher layers of the kernel that directly | |
1910 | * set the page dirty bit without asking the filesystem if it is a | |
1911 | * good idea. This causes problems because we want to make sure COW | |
1912 | * properly happens and the data=ordered rules are followed. | |
1913 | * | |
c8b97818 | 1914 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1915 | * hasn't been properly setup for IO. We kick off an async process |
1916 | * to fix it up. The async helper will wait for ordered extents, set | |
1917 | * the delalloc bit and make it safe to write the page. | |
1918 | */ | |
b2950863 | 1919 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1920 | { |
1921 | struct inode *inode = page->mapping->host; | |
1922 | struct btrfs_writepage_fixup *fixup; | |
1923 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1924 | |
8b62b72b CM |
1925 | /* this page is properly in the ordered list */ |
1926 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1927 | return 0; |
1928 | ||
1929 | if (PageChecked(page)) | |
1930 | return -EAGAIN; | |
1931 | ||
1932 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1933 | if (!fixup) | |
1934 | return -EAGAIN; | |
f421950f | 1935 | |
247e743c CM |
1936 | SetPageChecked(page); |
1937 | page_cache_get(page); | |
9e0af237 LB |
1938 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
1939 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 1940 | fixup->page = page; |
dc6e3209 | 1941 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 1942 | return -EBUSY; |
247e743c CM |
1943 | } |
1944 | ||
d899e052 YZ |
1945 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1946 | struct inode *inode, u64 file_pos, | |
1947 | u64 disk_bytenr, u64 disk_num_bytes, | |
1948 | u64 num_bytes, u64 ram_bytes, | |
1949 | u8 compression, u8 encryption, | |
1950 | u16 other_encoding, int extent_type) | |
1951 | { | |
1952 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1953 | struct btrfs_file_extent_item *fi; | |
1954 | struct btrfs_path *path; | |
1955 | struct extent_buffer *leaf; | |
1956 | struct btrfs_key ins; | |
1acae57b | 1957 | int extent_inserted = 0; |
d899e052 YZ |
1958 | int ret; |
1959 | ||
1960 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1961 | if (!path) |
1962 | return -ENOMEM; | |
d899e052 | 1963 | |
a1ed835e CM |
1964 | /* |
1965 | * we may be replacing one extent in the tree with another. | |
1966 | * The new extent is pinned in the extent map, and we don't want | |
1967 | * to drop it from the cache until it is completely in the btree. | |
1968 | * | |
1969 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1970 | * the caller is expected to unpin it and allow it to be merged | |
1971 | * with the others. | |
1972 | */ | |
1acae57b FDBM |
1973 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
1974 | file_pos + num_bytes, NULL, 0, | |
1975 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
1976 | if (ret) |
1977 | goto out; | |
d899e052 | 1978 | |
1acae57b FDBM |
1979 | if (!extent_inserted) { |
1980 | ins.objectid = btrfs_ino(inode); | |
1981 | ins.offset = file_pos; | |
1982 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1983 | ||
1984 | path->leave_spinning = 1; | |
1985 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
1986 | sizeof(*fi)); | |
1987 | if (ret) | |
1988 | goto out; | |
1989 | } | |
d899e052 YZ |
1990 | leaf = path->nodes[0]; |
1991 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1992 | struct btrfs_file_extent_item); | |
1993 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1994 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1995 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1996 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1997 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1998 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1999 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2000 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2001 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2002 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2003 | |
d899e052 | 2004 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2005 | btrfs_release_path(path); |
d899e052 YZ |
2006 | |
2007 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2008 | |
2009 | ins.objectid = disk_bytenr; | |
2010 | ins.offset = disk_num_bytes; | |
2011 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2012 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2013 | root->root_key.objectid, | |
33345d01 | 2014 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 2015 | out: |
d899e052 | 2016 | btrfs_free_path(path); |
b9473439 | 2017 | |
79787eaa | 2018 | return ret; |
d899e052 YZ |
2019 | } |
2020 | ||
38c227d8 LB |
2021 | /* snapshot-aware defrag */ |
2022 | struct sa_defrag_extent_backref { | |
2023 | struct rb_node node; | |
2024 | struct old_sa_defrag_extent *old; | |
2025 | u64 root_id; | |
2026 | u64 inum; | |
2027 | u64 file_pos; | |
2028 | u64 extent_offset; | |
2029 | u64 num_bytes; | |
2030 | u64 generation; | |
2031 | }; | |
2032 | ||
2033 | struct old_sa_defrag_extent { | |
2034 | struct list_head list; | |
2035 | struct new_sa_defrag_extent *new; | |
2036 | ||
2037 | u64 extent_offset; | |
2038 | u64 bytenr; | |
2039 | u64 offset; | |
2040 | u64 len; | |
2041 | int count; | |
2042 | }; | |
2043 | ||
2044 | struct new_sa_defrag_extent { | |
2045 | struct rb_root root; | |
2046 | struct list_head head; | |
2047 | struct btrfs_path *path; | |
2048 | struct inode *inode; | |
2049 | u64 file_pos; | |
2050 | u64 len; | |
2051 | u64 bytenr; | |
2052 | u64 disk_len; | |
2053 | u8 compress_type; | |
2054 | }; | |
2055 | ||
2056 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2057 | struct sa_defrag_extent_backref *b2) | |
2058 | { | |
2059 | if (b1->root_id < b2->root_id) | |
2060 | return -1; | |
2061 | else if (b1->root_id > b2->root_id) | |
2062 | return 1; | |
2063 | ||
2064 | if (b1->inum < b2->inum) | |
2065 | return -1; | |
2066 | else if (b1->inum > b2->inum) | |
2067 | return 1; | |
2068 | ||
2069 | if (b1->file_pos < b2->file_pos) | |
2070 | return -1; | |
2071 | else if (b1->file_pos > b2->file_pos) | |
2072 | return 1; | |
2073 | ||
2074 | /* | |
2075 | * [------------------------------] ===> (a range of space) | |
2076 | * |<--->| |<---->| =============> (fs/file tree A) | |
2077 | * |<---------------------------->| ===> (fs/file tree B) | |
2078 | * | |
2079 | * A range of space can refer to two file extents in one tree while | |
2080 | * refer to only one file extent in another tree. | |
2081 | * | |
2082 | * So we may process a disk offset more than one time(two extents in A) | |
2083 | * and locate at the same extent(one extent in B), then insert two same | |
2084 | * backrefs(both refer to the extent in B). | |
2085 | */ | |
2086 | return 0; | |
2087 | } | |
2088 | ||
2089 | static void backref_insert(struct rb_root *root, | |
2090 | struct sa_defrag_extent_backref *backref) | |
2091 | { | |
2092 | struct rb_node **p = &root->rb_node; | |
2093 | struct rb_node *parent = NULL; | |
2094 | struct sa_defrag_extent_backref *entry; | |
2095 | int ret; | |
2096 | ||
2097 | while (*p) { | |
2098 | parent = *p; | |
2099 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2100 | ||
2101 | ret = backref_comp(backref, entry); | |
2102 | if (ret < 0) | |
2103 | p = &(*p)->rb_left; | |
2104 | else | |
2105 | p = &(*p)->rb_right; | |
2106 | } | |
2107 | ||
2108 | rb_link_node(&backref->node, parent, p); | |
2109 | rb_insert_color(&backref->node, root); | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Note the backref might has changed, and in this case we just return 0. | |
2114 | */ | |
2115 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2116 | void *ctx) | |
2117 | { | |
2118 | struct btrfs_file_extent_item *extent; | |
2119 | struct btrfs_fs_info *fs_info; | |
2120 | struct old_sa_defrag_extent *old = ctx; | |
2121 | struct new_sa_defrag_extent *new = old->new; | |
2122 | struct btrfs_path *path = new->path; | |
2123 | struct btrfs_key key; | |
2124 | struct btrfs_root *root; | |
2125 | struct sa_defrag_extent_backref *backref; | |
2126 | struct extent_buffer *leaf; | |
2127 | struct inode *inode = new->inode; | |
2128 | int slot; | |
2129 | int ret; | |
2130 | u64 extent_offset; | |
2131 | u64 num_bytes; | |
2132 | ||
2133 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2134 | inum == btrfs_ino(inode)) | |
2135 | return 0; | |
2136 | ||
2137 | key.objectid = root_id; | |
2138 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2139 | key.offset = (u64)-1; | |
2140 | ||
2141 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2142 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2143 | if (IS_ERR(root)) { | |
2144 | if (PTR_ERR(root) == -ENOENT) | |
2145 | return 0; | |
2146 | WARN_ON(1); | |
2147 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2148 | inum, offset, root_id); | |
2149 | return PTR_ERR(root); | |
2150 | } | |
2151 | ||
2152 | key.objectid = inum; | |
2153 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2154 | if (offset > (u64)-1 << 32) | |
2155 | key.offset = 0; | |
2156 | else | |
2157 | key.offset = offset; | |
2158 | ||
2159 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2160 | if (WARN_ON(ret < 0)) |
38c227d8 | 2161 | return ret; |
50f1319c | 2162 | ret = 0; |
38c227d8 LB |
2163 | |
2164 | while (1) { | |
2165 | cond_resched(); | |
2166 | ||
2167 | leaf = path->nodes[0]; | |
2168 | slot = path->slots[0]; | |
2169 | ||
2170 | if (slot >= btrfs_header_nritems(leaf)) { | |
2171 | ret = btrfs_next_leaf(root, path); | |
2172 | if (ret < 0) { | |
2173 | goto out; | |
2174 | } else if (ret > 0) { | |
2175 | ret = 0; | |
2176 | goto out; | |
2177 | } | |
2178 | continue; | |
2179 | } | |
2180 | ||
2181 | path->slots[0]++; | |
2182 | ||
2183 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2184 | ||
2185 | if (key.objectid > inum) | |
2186 | goto out; | |
2187 | ||
2188 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2189 | continue; | |
2190 | ||
2191 | extent = btrfs_item_ptr(leaf, slot, | |
2192 | struct btrfs_file_extent_item); | |
2193 | ||
2194 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2195 | continue; | |
2196 | ||
e68afa49 LB |
2197 | /* |
2198 | * 'offset' refers to the exact key.offset, | |
2199 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2200 | * (key.offset - extent_offset). | |
2201 | */ | |
2202 | if (key.offset != offset) | |
38c227d8 LB |
2203 | continue; |
2204 | ||
e68afa49 | 2205 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2206 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2207 | |
38c227d8 LB |
2208 | if (extent_offset >= old->extent_offset + old->offset + |
2209 | old->len || extent_offset + num_bytes <= | |
2210 | old->extent_offset + old->offset) | |
2211 | continue; | |
38c227d8 LB |
2212 | break; |
2213 | } | |
2214 | ||
2215 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2216 | if (!backref) { | |
2217 | ret = -ENOENT; | |
2218 | goto out; | |
2219 | } | |
2220 | ||
2221 | backref->root_id = root_id; | |
2222 | backref->inum = inum; | |
e68afa49 | 2223 | backref->file_pos = offset; |
38c227d8 LB |
2224 | backref->num_bytes = num_bytes; |
2225 | backref->extent_offset = extent_offset; | |
2226 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2227 | backref->old = old; | |
2228 | backref_insert(&new->root, backref); | |
2229 | old->count++; | |
2230 | out: | |
2231 | btrfs_release_path(path); | |
2232 | WARN_ON(ret); | |
2233 | return ret; | |
2234 | } | |
2235 | ||
2236 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2237 | struct new_sa_defrag_extent *new) | |
2238 | { | |
2239 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2240 | struct old_sa_defrag_extent *old, *tmp; | |
2241 | int ret; | |
2242 | ||
2243 | new->path = path; | |
2244 | ||
2245 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2246 | ret = iterate_inodes_from_logical(old->bytenr + |
2247 | old->extent_offset, fs_info, | |
38c227d8 LB |
2248 | path, record_one_backref, |
2249 | old); | |
4724b106 JB |
2250 | if (ret < 0 && ret != -ENOENT) |
2251 | return false; | |
38c227d8 LB |
2252 | |
2253 | /* no backref to be processed for this extent */ | |
2254 | if (!old->count) { | |
2255 | list_del(&old->list); | |
2256 | kfree(old); | |
2257 | } | |
2258 | } | |
2259 | ||
2260 | if (list_empty(&new->head)) | |
2261 | return false; | |
2262 | ||
2263 | return true; | |
2264 | } | |
2265 | ||
2266 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2267 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2268 | struct new_sa_defrag_extent *new) |
38c227d8 | 2269 | { |
116e0024 | 2270 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2271 | return 0; |
2272 | ||
2273 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2274 | return 0; | |
2275 | ||
116e0024 LB |
2276 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2277 | return 0; | |
2278 | ||
2279 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2280 | btrfs_file_extent_other_encoding(leaf, fi)) |
2281 | return 0; | |
2282 | ||
2283 | return 1; | |
2284 | } | |
2285 | ||
2286 | /* | |
2287 | * Note the backref might has changed, and in this case we just return 0. | |
2288 | */ | |
2289 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2290 | struct sa_defrag_extent_backref *prev, | |
2291 | struct sa_defrag_extent_backref *backref) | |
2292 | { | |
2293 | struct btrfs_file_extent_item *extent; | |
2294 | struct btrfs_file_extent_item *item; | |
2295 | struct btrfs_ordered_extent *ordered; | |
2296 | struct btrfs_trans_handle *trans; | |
2297 | struct btrfs_fs_info *fs_info; | |
2298 | struct btrfs_root *root; | |
2299 | struct btrfs_key key; | |
2300 | struct extent_buffer *leaf; | |
2301 | struct old_sa_defrag_extent *old = backref->old; | |
2302 | struct new_sa_defrag_extent *new = old->new; | |
2303 | struct inode *src_inode = new->inode; | |
2304 | struct inode *inode; | |
2305 | struct extent_state *cached = NULL; | |
2306 | int ret = 0; | |
2307 | u64 start; | |
2308 | u64 len; | |
2309 | u64 lock_start; | |
2310 | u64 lock_end; | |
2311 | bool merge = false; | |
2312 | int index; | |
2313 | ||
2314 | if (prev && prev->root_id == backref->root_id && | |
2315 | prev->inum == backref->inum && | |
2316 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2317 | merge = true; | |
2318 | ||
2319 | /* step 1: get root */ | |
2320 | key.objectid = backref->root_id; | |
2321 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2322 | key.offset = (u64)-1; | |
2323 | ||
2324 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2325 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2326 | ||
2327 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2328 | if (IS_ERR(root)) { | |
2329 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2330 | if (PTR_ERR(root) == -ENOENT) | |
2331 | return 0; | |
2332 | return PTR_ERR(root); | |
2333 | } | |
38c227d8 | 2334 | |
bcbba5e6 WS |
2335 | if (btrfs_root_readonly(root)) { |
2336 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2337 | return 0; | |
2338 | } | |
2339 | ||
38c227d8 LB |
2340 | /* step 2: get inode */ |
2341 | key.objectid = backref->inum; | |
2342 | key.type = BTRFS_INODE_ITEM_KEY; | |
2343 | key.offset = 0; | |
2344 | ||
2345 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2346 | if (IS_ERR(inode)) { | |
2347 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2348 | return 0; | |
2349 | } | |
2350 | ||
2351 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2352 | ||
2353 | /* step 3: relink backref */ | |
2354 | lock_start = backref->file_pos; | |
2355 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2356 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2357 | 0, &cached); | |
2358 | ||
2359 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2360 | if (ordered) { | |
2361 | btrfs_put_ordered_extent(ordered); | |
2362 | goto out_unlock; | |
2363 | } | |
2364 | ||
2365 | trans = btrfs_join_transaction(root); | |
2366 | if (IS_ERR(trans)) { | |
2367 | ret = PTR_ERR(trans); | |
2368 | goto out_unlock; | |
2369 | } | |
2370 | ||
2371 | key.objectid = backref->inum; | |
2372 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2373 | key.offset = backref->file_pos; | |
2374 | ||
2375 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2376 | if (ret < 0) { | |
2377 | goto out_free_path; | |
2378 | } else if (ret > 0) { | |
2379 | ret = 0; | |
2380 | goto out_free_path; | |
2381 | } | |
2382 | ||
2383 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2384 | struct btrfs_file_extent_item); | |
2385 | ||
2386 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2387 | backref->generation) | |
2388 | goto out_free_path; | |
2389 | ||
2390 | btrfs_release_path(path); | |
2391 | ||
2392 | start = backref->file_pos; | |
2393 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2394 | start += old->extent_offset + old->offset - | |
2395 | backref->extent_offset; | |
2396 | ||
2397 | len = min(backref->extent_offset + backref->num_bytes, | |
2398 | old->extent_offset + old->offset + old->len); | |
2399 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2400 | ||
2401 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2402 | start + len, 1); | |
2403 | if (ret) | |
2404 | goto out_free_path; | |
2405 | again: | |
2406 | key.objectid = btrfs_ino(inode); | |
2407 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2408 | key.offset = start; | |
2409 | ||
a09a0a70 | 2410 | path->leave_spinning = 1; |
38c227d8 LB |
2411 | if (merge) { |
2412 | struct btrfs_file_extent_item *fi; | |
2413 | u64 extent_len; | |
2414 | struct btrfs_key found_key; | |
2415 | ||
3c9665df | 2416 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2417 | if (ret < 0) |
2418 | goto out_free_path; | |
2419 | ||
2420 | path->slots[0]--; | |
2421 | leaf = path->nodes[0]; | |
2422 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2423 | ||
2424 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2425 | struct btrfs_file_extent_item); | |
2426 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2427 | ||
116e0024 LB |
2428 | if (extent_len + found_key.offset == start && |
2429 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2430 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2431 | extent_len + len); | |
2432 | btrfs_mark_buffer_dirty(leaf); | |
2433 | inode_add_bytes(inode, len); | |
2434 | ||
2435 | ret = 1; | |
2436 | goto out_free_path; | |
2437 | } else { | |
2438 | merge = false; | |
2439 | btrfs_release_path(path); | |
2440 | goto again; | |
2441 | } | |
2442 | } | |
2443 | ||
2444 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2445 | sizeof(*extent)); | |
2446 | if (ret) { | |
2447 | btrfs_abort_transaction(trans, root, ret); | |
2448 | goto out_free_path; | |
2449 | } | |
2450 | ||
2451 | leaf = path->nodes[0]; | |
2452 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2453 | struct btrfs_file_extent_item); | |
2454 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2455 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2456 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2457 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2458 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2459 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2460 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2461 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2462 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2463 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2464 | ||
2465 | btrfs_mark_buffer_dirty(leaf); | |
2466 | inode_add_bytes(inode, len); | |
a09a0a70 | 2467 | btrfs_release_path(path); |
38c227d8 LB |
2468 | |
2469 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2470 | new->disk_len, 0, | |
2471 | backref->root_id, backref->inum, | |
2472 | new->file_pos, 0); /* start - extent_offset */ | |
2473 | if (ret) { | |
2474 | btrfs_abort_transaction(trans, root, ret); | |
2475 | goto out_free_path; | |
2476 | } | |
2477 | ||
2478 | ret = 1; | |
2479 | out_free_path: | |
2480 | btrfs_release_path(path); | |
a09a0a70 | 2481 | path->leave_spinning = 0; |
38c227d8 LB |
2482 | btrfs_end_transaction(trans, root); |
2483 | out_unlock: | |
2484 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2485 | &cached, GFP_NOFS); | |
2486 | iput(inode); | |
2487 | return ret; | |
2488 | } | |
2489 | ||
6f519564 LB |
2490 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2491 | { | |
2492 | struct old_sa_defrag_extent *old, *tmp; | |
2493 | ||
2494 | if (!new) | |
2495 | return; | |
2496 | ||
2497 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2498 | list_del(&old->list); | |
2499 | kfree(old); | |
2500 | } | |
2501 | kfree(new); | |
2502 | } | |
2503 | ||
38c227d8 LB |
2504 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2505 | { | |
2506 | struct btrfs_path *path; | |
38c227d8 LB |
2507 | struct sa_defrag_extent_backref *backref; |
2508 | struct sa_defrag_extent_backref *prev = NULL; | |
2509 | struct inode *inode; | |
2510 | struct btrfs_root *root; | |
2511 | struct rb_node *node; | |
2512 | int ret; | |
2513 | ||
2514 | inode = new->inode; | |
2515 | root = BTRFS_I(inode)->root; | |
2516 | ||
2517 | path = btrfs_alloc_path(); | |
2518 | if (!path) | |
2519 | return; | |
2520 | ||
2521 | if (!record_extent_backrefs(path, new)) { | |
2522 | btrfs_free_path(path); | |
2523 | goto out; | |
2524 | } | |
2525 | btrfs_release_path(path); | |
2526 | ||
2527 | while (1) { | |
2528 | node = rb_first(&new->root); | |
2529 | if (!node) | |
2530 | break; | |
2531 | rb_erase(node, &new->root); | |
2532 | ||
2533 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2534 | ||
2535 | ret = relink_extent_backref(path, prev, backref); | |
2536 | WARN_ON(ret < 0); | |
2537 | ||
2538 | kfree(prev); | |
2539 | ||
2540 | if (ret == 1) | |
2541 | prev = backref; | |
2542 | else | |
2543 | prev = NULL; | |
2544 | cond_resched(); | |
2545 | } | |
2546 | kfree(prev); | |
2547 | ||
2548 | btrfs_free_path(path); | |
38c227d8 | 2549 | out: |
6f519564 LB |
2550 | free_sa_defrag_extent(new); |
2551 | ||
38c227d8 LB |
2552 | atomic_dec(&root->fs_info->defrag_running); |
2553 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2554 | } |
2555 | ||
2556 | static struct new_sa_defrag_extent * | |
2557 | record_old_file_extents(struct inode *inode, | |
2558 | struct btrfs_ordered_extent *ordered) | |
2559 | { | |
2560 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2561 | struct btrfs_path *path; | |
2562 | struct btrfs_key key; | |
6f519564 | 2563 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2564 | struct new_sa_defrag_extent *new; |
2565 | int ret; | |
2566 | ||
2567 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2568 | if (!new) | |
2569 | return NULL; | |
2570 | ||
2571 | new->inode = inode; | |
2572 | new->file_pos = ordered->file_offset; | |
2573 | new->len = ordered->len; | |
2574 | new->bytenr = ordered->start; | |
2575 | new->disk_len = ordered->disk_len; | |
2576 | new->compress_type = ordered->compress_type; | |
2577 | new->root = RB_ROOT; | |
2578 | INIT_LIST_HEAD(&new->head); | |
2579 | ||
2580 | path = btrfs_alloc_path(); | |
2581 | if (!path) | |
2582 | goto out_kfree; | |
2583 | ||
2584 | key.objectid = btrfs_ino(inode); | |
2585 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2586 | key.offset = new->file_pos; | |
2587 | ||
2588 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2589 | if (ret < 0) | |
2590 | goto out_free_path; | |
2591 | if (ret > 0 && path->slots[0] > 0) | |
2592 | path->slots[0]--; | |
2593 | ||
2594 | /* find out all the old extents for the file range */ | |
2595 | while (1) { | |
2596 | struct btrfs_file_extent_item *extent; | |
2597 | struct extent_buffer *l; | |
2598 | int slot; | |
2599 | u64 num_bytes; | |
2600 | u64 offset; | |
2601 | u64 end; | |
2602 | u64 disk_bytenr; | |
2603 | u64 extent_offset; | |
2604 | ||
2605 | l = path->nodes[0]; | |
2606 | slot = path->slots[0]; | |
2607 | ||
2608 | if (slot >= btrfs_header_nritems(l)) { | |
2609 | ret = btrfs_next_leaf(root, path); | |
2610 | if (ret < 0) | |
6f519564 | 2611 | goto out_free_path; |
38c227d8 LB |
2612 | else if (ret > 0) |
2613 | break; | |
2614 | continue; | |
2615 | } | |
2616 | ||
2617 | btrfs_item_key_to_cpu(l, &key, slot); | |
2618 | ||
2619 | if (key.objectid != btrfs_ino(inode)) | |
2620 | break; | |
2621 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2622 | break; | |
2623 | if (key.offset >= new->file_pos + new->len) | |
2624 | break; | |
2625 | ||
2626 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2627 | ||
2628 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2629 | if (key.offset + num_bytes < new->file_pos) | |
2630 | goto next; | |
2631 | ||
2632 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2633 | if (!disk_bytenr) | |
2634 | goto next; | |
2635 | ||
2636 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2637 | ||
2638 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2639 | if (!old) | |
6f519564 | 2640 | goto out_free_path; |
38c227d8 LB |
2641 | |
2642 | offset = max(new->file_pos, key.offset); | |
2643 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2644 | ||
2645 | old->bytenr = disk_bytenr; | |
2646 | old->extent_offset = extent_offset; | |
2647 | old->offset = offset - key.offset; | |
2648 | old->len = end - offset; | |
2649 | old->new = new; | |
2650 | old->count = 0; | |
2651 | list_add_tail(&old->list, &new->head); | |
2652 | next: | |
2653 | path->slots[0]++; | |
2654 | cond_resched(); | |
2655 | } | |
2656 | ||
2657 | btrfs_free_path(path); | |
2658 | atomic_inc(&root->fs_info->defrag_running); | |
2659 | ||
2660 | return new; | |
2661 | ||
38c227d8 LB |
2662 | out_free_path: |
2663 | btrfs_free_path(path); | |
2664 | out_kfree: | |
6f519564 | 2665 | free_sa_defrag_extent(new); |
38c227d8 LB |
2666 | return NULL; |
2667 | } | |
2668 | ||
e570fd27 MX |
2669 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2670 | u64 start, u64 len) | |
2671 | { | |
2672 | struct btrfs_block_group_cache *cache; | |
2673 | ||
2674 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2675 | ASSERT(cache); | |
2676 | ||
2677 | spin_lock(&cache->lock); | |
2678 | cache->delalloc_bytes -= len; | |
2679 | spin_unlock(&cache->lock); | |
2680 | ||
2681 | btrfs_put_block_group(cache); | |
2682 | } | |
2683 | ||
d352ac68 CM |
2684 | /* as ordered data IO finishes, this gets called so we can finish |
2685 | * an ordered extent if the range of bytes in the file it covers are | |
2686 | * fully written. | |
2687 | */ | |
5fd02043 | 2688 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2689 | { |
5fd02043 | 2690 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2691 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2692 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2693 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2694 | struct extent_state *cached_state = NULL; |
38c227d8 | 2695 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2696 | int compress_type = 0; |
77cef2ec JB |
2697 | int ret = 0; |
2698 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2699 | bool nolock; |
77cef2ec | 2700 | bool truncated = false; |
e6dcd2dc | 2701 | |
83eea1f1 | 2702 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2703 | |
5fd02043 JB |
2704 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2705 | ret = -EIO; | |
2706 | goto out; | |
2707 | } | |
2708 | ||
77cef2ec JB |
2709 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2710 | truncated = true; | |
2711 | logical_len = ordered_extent->truncated_len; | |
2712 | /* Truncated the entire extent, don't bother adding */ | |
2713 | if (!logical_len) | |
2714 | goto out; | |
2715 | } | |
2716 | ||
c2167754 | 2717 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2718 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2719 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2720 | if (nolock) | |
2721 | trans = btrfs_join_transaction_nolock(root); | |
2722 | else | |
2723 | trans = btrfs_join_transaction(root); | |
2724 | if (IS_ERR(trans)) { | |
2725 | ret = PTR_ERR(trans); | |
2726 | trans = NULL; | |
2727 | goto out; | |
c2167754 | 2728 | } |
6c760c07 JB |
2729 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2730 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2731 | if (ret) /* -ENOMEM or corruption */ | |
2732 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2733 | goto out; |
2734 | } | |
e6dcd2dc | 2735 | |
2ac55d41 JB |
2736 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2737 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2738 | 0, &cached_state); |
e6dcd2dc | 2739 | |
38c227d8 LB |
2740 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2741 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2742 | EXTENT_DEFRAG, 1, cached_state); | |
2743 | if (ret) { | |
2744 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2745 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2746 | /* the inode is shared */ |
2747 | new = record_old_file_extents(inode, ordered_extent); | |
2748 | ||
2749 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2750 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2751 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2752 | } | |
2753 | ||
0cb59c99 | 2754 | if (nolock) |
7a7eaa40 | 2755 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2756 | else |
7a7eaa40 | 2757 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2758 | if (IS_ERR(trans)) { |
2759 | ret = PTR_ERR(trans); | |
2760 | trans = NULL; | |
2761 | goto out_unlock; | |
2762 | } | |
a79b7d4b | 2763 | |
0ca1f7ce | 2764 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2765 | |
c8b97818 | 2766 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2767 | compress_type = ordered_extent->compress_type; |
d899e052 | 2768 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2769 | BUG_ON(compress_type); |
920bbbfb | 2770 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2771 | ordered_extent->file_offset, |
2772 | ordered_extent->file_offset + | |
77cef2ec | 2773 | logical_len); |
d899e052 | 2774 | } else { |
0af3d00b | 2775 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2776 | ret = insert_reserved_file_extent(trans, inode, |
2777 | ordered_extent->file_offset, | |
2778 | ordered_extent->start, | |
2779 | ordered_extent->disk_len, | |
77cef2ec | 2780 | logical_len, logical_len, |
261507a0 | 2781 | compress_type, 0, 0, |
d899e052 | 2782 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2783 | if (!ret) |
2784 | btrfs_release_delalloc_bytes(root, | |
2785 | ordered_extent->start, | |
2786 | ordered_extent->disk_len); | |
d899e052 | 2787 | } |
5dc562c5 JB |
2788 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2789 | ordered_extent->file_offset, ordered_extent->len, | |
2790 | trans->transid); | |
79787eaa JM |
2791 | if (ret < 0) { |
2792 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2793 | goto out_unlock; |
79787eaa | 2794 | } |
2ac55d41 | 2795 | |
e6dcd2dc CM |
2796 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2797 | &ordered_extent->list); | |
2798 | ||
6c760c07 JB |
2799 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2800 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2801 | if (ret) { /* -ENOMEM or corruption */ | |
2802 | btrfs_abort_transaction(trans, root, ret); | |
2803 | goto out_unlock; | |
1ef30be1 JB |
2804 | } |
2805 | ret = 0; | |
5fd02043 JB |
2806 | out_unlock: |
2807 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2808 | ordered_extent->file_offset + | |
2809 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2810 | out: |
5b0e95bf | 2811 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2812 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2813 | if (trans) |
2814 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2815 | |
77cef2ec JB |
2816 | if (ret || truncated) { |
2817 | u64 start, end; | |
2818 | ||
2819 | if (truncated) | |
2820 | start = ordered_extent->file_offset + logical_len; | |
2821 | else | |
2822 | start = ordered_extent->file_offset; | |
2823 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2824 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2825 | ||
2826 | /* Drop the cache for the part of the extent we didn't write. */ | |
2827 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2828 | |
0bec9ef5 JB |
2829 | /* |
2830 | * If the ordered extent had an IOERR or something else went | |
2831 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2832 | * back to the allocator. We only free the extent in the |
2833 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2834 | */ |
77cef2ec JB |
2835 | if ((ret || !logical_len) && |
2836 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2837 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2838 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 2839 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2840 | } |
2841 | ||
2842 | ||
5fd02043 | 2843 | /* |
8bad3c02 LB |
2844 | * This needs to be done to make sure anybody waiting knows we are done |
2845 | * updating everything for this ordered extent. | |
5fd02043 JB |
2846 | */ |
2847 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2848 | ||
38c227d8 | 2849 | /* for snapshot-aware defrag */ |
6f519564 LB |
2850 | if (new) { |
2851 | if (ret) { | |
2852 | free_sa_defrag_extent(new); | |
2853 | atomic_dec(&root->fs_info->defrag_running); | |
2854 | } else { | |
2855 | relink_file_extents(new); | |
2856 | } | |
2857 | } | |
38c227d8 | 2858 | |
e6dcd2dc CM |
2859 | /* once for us */ |
2860 | btrfs_put_ordered_extent(ordered_extent); | |
2861 | /* once for the tree */ | |
2862 | btrfs_put_ordered_extent(ordered_extent); | |
2863 | ||
5fd02043 JB |
2864 | return ret; |
2865 | } | |
2866 | ||
2867 | static void finish_ordered_fn(struct btrfs_work *work) | |
2868 | { | |
2869 | struct btrfs_ordered_extent *ordered_extent; | |
2870 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2871 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2872 | } |
2873 | ||
b2950863 | 2874 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2875 | struct extent_state *state, int uptodate) |
2876 | { | |
5fd02043 JB |
2877 | struct inode *inode = page->mapping->host; |
2878 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2879 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
2880 | struct btrfs_workqueue *wq; |
2881 | btrfs_work_func_t func; | |
5fd02043 | 2882 | |
1abe9b8a | 2883 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2884 | ||
8b62b72b | 2885 | ClearPagePrivate2(page); |
5fd02043 JB |
2886 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2887 | end - start + 1, uptodate)) | |
2888 | return 0; | |
2889 | ||
9e0af237 LB |
2890 | if (btrfs_is_free_space_inode(inode)) { |
2891 | wq = root->fs_info->endio_freespace_worker; | |
2892 | func = btrfs_freespace_write_helper; | |
2893 | } else { | |
2894 | wq = root->fs_info->endio_write_workers; | |
2895 | func = btrfs_endio_write_helper; | |
2896 | } | |
5fd02043 | 2897 | |
9e0af237 LB |
2898 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
2899 | NULL); | |
2900 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
2901 | |
2902 | return 0; | |
211f90e6 CM |
2903 | } |
2904 | ||
d352ac68 CM |
2905 | /* |
2906 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2907 | * if there's a match, we allow the bio to finish. If not, the code in |
2908 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2909 | */ |
facc8a22 MX |
2910 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2911 | u64 phy_offset, struct page *page, | |
2912 | u64 start, u64 end, int mirror) | |
07157aac | 2913 | { |
4eee4fa4 | 2914 | size_t offset = start - page_offset(page); |
07157aac | 2915 | struct inode *inode = page->mapping->host; |
d1310b2e | 2916 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2917 | char *kaddr; |
ff79f819 | 2918 | struct btrfs_root *root = BTRFS_I(inode)->root; |
facc8a22 | 2919 | u32 csum_expected; |
ff79f819 | 2920 | u32 csum = ~(u32)0; |
c2cf52eb SK |
2921 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
2922 | DEFAULT_RATELIMIT_BURST); | |
d1310b2e | 2923 | |
d20f7043 CM |
2924 | if (PageChecked(page)) { |
2925 | ClearPageChecked(page); | |
2926 | goto good; | |
2927 | } | |
6cbff00f CH |
2928 | |
2929 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2930 | goto good; |
17d217fe YZ |
2931 | |
2932 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2933 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2934 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2935 | GFP_NOFS); | |
b6cda9bc | 2936 | return 0; |
17d217fe | 2937 | } |
d20f7043 | 2938 | |
facc8a22 MX |
2939 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
2940 | csum_expected = *(((u32 *)io_bio->csum) + phy_offset); | |
d397712b | 2941 | |
facc8a22 | 2942 | kaddr = kmap_atomic(page); |
b0496686 | 2943 | csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1); |
ff79f819 | 2944 | btrfs_csum_final(csum, (char *)&csum); |
facc8a22 | 2945 | if (csum != csum_expected) |
07157aac | 2946 | goto zeroit; |
d397712b | 2947 | |
7ac687d9 | 2948 | kunmap_atomic(kaddr); |
d20f7043 | 2949 | good: |
07157aac CM |
2950 | return 0; |
2951 | ||
2952 | zeroit: | |
c2cf52eb | 2953 | if (__ratelimit(&_rs)) |
facc8a22 | 2954 | btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", |
c1c9ff7c | 2955 | btrfs_ino(page->mapping->host), start, csum, csum_expected); |
db94535d CM |
2956 | memset(kaddr + offset, 1, end - start + 1); |
2957 | flush_dcache_page(page); | |
7ac687d9 | 2958 | kunmap_atomic(kaddr); |
facc8a22 | 2959 | if (csum_expected == 0) |
3b951516 | 2960 | return 0; |
7e38326f | 2961 | return -EIO; |
07157aac | 2962 | } |
b888db2b | 2963 | |
24bbcf04 YZ |
2964 | struct delayed_iput { |
2965 | struct list_head list; | |
2966 | struct inode *inode; | |
2967 | }; | |
2968 | ||
79787eaa JM |
2969 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2970 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2971 | void btrfs_add_delayed_iput(struct inode *inode) |
2972 | { | |
2973 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2974 | struct delayed_iput *delayed; | |
2975 | ||
2976 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2977 | return; | |
2978 | ||
2979 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2980 | delayed->inode = inode; | |
2981 | ||
2982 | spin_lock(&fs_info->delayed_iput_lock); | |
2983 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2984 | spin_unlock(&fs_info->delayed_iput_lock); | |
2985 | } | |
2986 | ||
2987 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2988 | { | |
2989 | LIST_HEAD(list); | |
2990 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2991 | struct delayed_iput *delayed; | |
2992 | int empty; | |
2993 | ||
2994 | spin_lock(&fs_info->delayed_iput_lock); | |
2995 | empty = list_empty(&fs_info->delayed_iputs); | |
2996 | spin_unlock(&fs_info->delayed_iput_lock); | |
2997 | if (empty) | |
2998 | return; | |
2999 | ||
24bbcf04 YZ |
3000 | spin_lock(&fs_info->delayed_iput_lock); |
3001 | list_splice_init(&fs_info->delayed_iputs, &list); | |
3002 | spin_unlock(&fs_info->delayed_iput_lock); | |
3003 | ||
3004 | while (!list_empty(&list)) { | |
3005 | delayed = list_entry(list.next, struct delayed_iput, list); | |
3006 | list_del(&delayed->list); | |
3007 | iput(delayed->inode); | |
3008 | kfree(delayed); | |
3009 | } | |
24bbcf04 YZ |
3010 | } |
3011 | ||
d68fc57b | 3012 | /* |
42b2aa86 | 3013 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3014 | * files in the subvolume, it removes orphan item and frees block_rsv |
3015 | * structure. | |
3016 | */ | |
3017 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3018 | struct btrfs_root *root) | |
3019 | { | |
90290e19 | 3020 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3021 | int ret; |
3022 | ||
8a35d95f | 3023 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3024 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3025 | return; | |
3026 | ||
90290e19 | 3027 | spin_lock(&root->orphan_lock); |
8a35d95f | 3028 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3029 | spin_unlock(&root->orphan_lock); |
3030 | return; | |
3031 | } | |
3032 | ||
3033 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3034 | spin_unlock(&root->orphan_lock); | |
3035 | return; | |
3036 | } | |
3037 | ||
3038 | block_rsv = root->orphan_block_rsv; | |
3039 | root->orphan_block_rsv = NULL; | |
3040 | spin_unlock(&root->orphan_lock); | |
3041 | ||
27cdeb70 | 3042 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3043 | btrfs_root_refs(&root->root_item) > 0) { |
3044 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3045 | root->root_key.objectid); | |
4ef31a45 JB |
3046 | if (ret) |
3047 | btrfs_abort_transaction(trans, root, ret); | |
3048 | else | |
27cdeb70 MX |
3049 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3050 | &root->state); | |
d68fc57b YZ |
3051 | } |
3052 | ||
90290e19 JB |
3053 | if (block_rsv) { |
3054 | WARN_ON(block_rsv->size > 0); | |
3055 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3056 | } |
3057 | } | |
3058 | ||
7b128766 JB |
3059 | /* |
3060 | * This creates an orphan entry for the given inode in case something goes | |
3061 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3062 | * |
3063 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3064 | * this function. | |
7b128766 JB |
3065 | */ |
3066 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3067 | { | |
3068 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3069 | struct btrfs_block_rsv *block_rsv = NULL; |
3070 | int reserve = 0; | |
3071 | int insert = 0; | |
3072 | int ret; | |
7b128766 | 3073 | |
d68fc57b | 3074 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3075 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3076 | if (!block_rsv) |
3077 | return -ENOMEM; | |
d68fc57b | 3078 | } |
7b128766 | 3079 | |
d68fc57b YZ |
3080 | spin_lock(&root->orphan_lock); |
3081 | if (!root->orphan_block_rsv) { | |
3082 | root->orphan_block_rsv = block_rsv; | |
3083 | } else if (block_rsv) { | |
3084 | btrfs_free_block_rsv(root, block_rsv); | |
3085 | block_rsv = NULL; | |
7b128766 | 3086 | } |
7b128766 | 3087 | |
8a35d95f JB |
3088 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3089 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3090 | #if 0 |
3091 | /* | |
3092 | * For proper ENOSPC handling, we should do orphan | |
3093 | * cleanup when mounting. But this introduces backward | |
3094 | * compatibility issue. | |
3095 | */ | |
3096 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3097 | insert = 2; | |
3098 | else | |
3099 | insert = 1; | |
3100 | #endif | |
3101 | insert = 1; | |
321f0e70 | 3102 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3103 | } |
3104 | ||
72ac3c0d JB |
3105 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3106 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3107 | reserve = 1; |
d68fc57b | 3108 | spin_unlock(&root->orphan_lock); |
7b128766 | 3109 | |
d68fc57b YZ |
3110 | /* grab metadata reservation from transaction handle */ |
3111 | if (reserve) { | |
3112 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3113 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3114 | } |
7b128766 | 3115 | |
d68fc57b YZ |
3116 | /* insert an orphan item to track this unlinked/truncated file */ |
3117 | if (insert >= 1) { | |
33345d01 | 3118 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3119 | if (ret) { |
703c88e0 | 3120 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3121 | if (reserve) { |
3122 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3123 | &BTRFS_I(inode)->runtime_flags); | |
3124 | btrfs_orphan_release_metadata(inode); | |
3125 | } | |
3126 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3127 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3128 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3129 | btrfs_abort_transaction(trans, root, ret); |
3130 | return ret; | |
3131 | } | |
79787eaa JM |
3132 | } |
3133 | ret = 0; | |
d68fc57b YZ |
3134 | } |
3135 | ||
3136 | /* insert an orphan item to track subvolume contains orphan files */ | |
3137 | if (insert >= 2) { | |
3138 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3139 | root->root_key.objectid); | |
79787eaa JM |
3140 | if (ret && ret != -EEXIST) { |
3141 | btrfs_abort_transaction(trans, root, ret); | |
3142 | return ret; | |
3143 | } | |
d68fc57b YZ |
3144 | } |
3145 | return 0; | |
7b128766 JB |
3146 | } |
3147 | ||
3148 | /* | |
3149 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3150 | * item for this particular inode. | |
3151 | */ | |
48a3b636 ES |
3152 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3153 | struct inode *inode) | |
7b128766 JB |
3154 | { |
3155 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3156 | int delete_item = 0; |
3157 | int release_rsv = 0; | |
7b128766 JB |
3158 | int ret = 0; |
3159 | ||
d68fc57b | 3160 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3161 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3162 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3163 | delete_item = 1; |
7b128766 | 3164 | |
72ac3c0d JB |
3165 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3166 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3167 | release_rsv = 1; |
d68fc57b | 3168 | spin_unlock(&root->orphan_lock); |
7b128766 | 3169 | |
703c88e0 | 3170 | if (delete_item) { |
8a35d95f | 3171 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3172 | if (trans) |
3173 | ret = btrfs_del_orphan_item(trans, root, | |
3174 | btrfs_ino(inode)); | |
8a35d95f | 3175 | } |
7b128766 | 3176 | |
703c88e0 FDBM |
3177 | if (release_rsv) |
3178 | btrfs_orphan_release_metadata(inode); | |
3179 | ||
4ef31a45 | 3180 | return ret; |
7b128766 JB |
3181 | } |
3182 | ||
3183 | /* | |
3184 | * this cleans up any orphans that may be left on the list from the last use | |
3185 | * of this root. | |
3186 | */ | |
66b4ffd1 | 3187 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3188 | { |
3189 | struct btrfs_path *path; | |
3190 | struct extent_buffer *leaf; | |
7b128766 JB |
3191 | struct btrfs_key key, found_key; |
3192 | struct btrfs_trans_handle *trans; | |
3193 | struct inode *inode; | |
8f6d7f4f | 3194 | u64 last_objectid = 0; |
7b128766 JB |
3195 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3196 | ||
d68fc57b | 3197 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3198 | return 0; |
c71bf099 YZ |
3199 | |
3200 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3201 | if (!path) { |
3202 | ret = -ENOMEM; | |
3203 | goto out; | |
3204 | } | |
7b128766 JB |
3205 | path->reada = -1; |
3206 | ||
3207 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3208 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3209 | key.offset = (u64)-1; |
3210 | ||
7b128766 JB |
3211 | while (1) { |
3212 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3213 | if (ret < 0) |
3214 | goto out; | |
7b128766 JB |
3215 | |
3216 | /* | |
3217 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3218 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3219 | * find the key and see if we have stuff that matches |
3220 | */ | |
3221 | if (ret > 0) { | |
66b4ffd1 | 3222 | ret = 0; |
7b128766 JB |
3223 | if (path->slots[0] == 0) |
3224 | break; | |
3225 | path->slots[0]--; | |
3226 | } | |
3227 | ||
3228 | /* pull out the item */ | |
3229 | leaf = path->nodes[0]; | |
7b128766 JB |
3230 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3231 | ||
3232 | /* make sure the item matches what we want */ | |
3233 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3234 | break; | |
962a298f | 3235 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3236 | break; |
3237 | ||
3238 | /* release the path since we're done with it */ | |
b3b4aa74 | 3239 | btrfs_release_path(path); |
7b128766 JB |
3240 | |
3241 | /* | |
3242 | * this is where we are basically btrfs_lookup, without the | |
3243 | * crossing root thing. we store the inode number in the | |
3244 | * offset of the orphan item. | |
3245 | */ | |
8f6d7f4f JB |
3246 | |
3247 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3248 | btrfs_err(root->fs_info, |
3249 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3250 | ret = -EINVAL; |
3251 | goto out; | |
3252 | } | |
3253 | ||
3254 | last_objectid = found_key.offset; | |
3255 | ||
5d4f98a2 YZ |
3256 | found_key.objectid = found_key.offset; |
3257 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3258 | found_key.offset = 0; | |
73f73415 | 3259 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3260 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3261 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3262 | goto out; |
7b128766 | 3263 | |
f8e9e0b0 AJ |
3264 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3265 | struct btrfs_root *dead_root; | |
3266 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3267 | int is_dead_root = 0; | |
3268 | ||
3269 | /* | |
3270 | * this is an orphan in the tree root. Currently these | |
3271 | * could come from 2 sources: | |
3272 | * a) a snapshot deletion in progress | |
3273 | * b) a free space cache inode | |
3274 | * We need to distinguish those two, as the snapshot | |
3275 | * orphan must not get deleted. | |
3276 | * find_dead_roots already ran before us, so if this | |
3277 | * is a snapshot deletion, we should find the root | |
3278 | * in the dead_roots list | |
3279 | */ | |
3280 | spin_lock(&fs_info->trans_lock); | |
3281 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3282 | root_list) { | |
3283 | if (dead_root->root_key.objectid == | |
3284 | found_key.objectid) { | |
3285 | is_dead_root = 1; | |
3286 | break; | |
3287 | } | |
3288 | } | |
3289 | spin_unlock(&fs_info->trans_lock); | |
3290 | if (is_dead_root) { | |
3291 | /* prevent this orphan from being found again */ | |
3292 | key.offset = found_key.objectid - 1; | |
3293 | continue; | |
3294 | } | |
3295 | } | |
7b128766 | 3296 | /* |
a8c9e576 JB |
3297 | * Inode is already gone but the orphan item is still there, |
3298 | * kill the orphan item. | |
7b128766 | 3299 | */ |
a8c9e576 JB |
3300 | if (ret == -ESTALE) { |
3301 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3302 | if (IS_ERR(trans)) { |
3303 | ret = PTR_ERR(trans); | |
3304 | goto out; | |
3305 | } | |
c2cf52eb SK |
3306 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3307 | found_key.objectid); | |
a8c9e576 JB |
3308 | ret = btrfs_del_orphan_item(trans, root, |
3309 | found_key.objectid); | |
5b21f2ed | 3310 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3311 | if (ret) |
3312 | goto out; | |
7b128766 JB |
3313 | continue; |
3314 | } | |
3315 | ||
a8c9e576 JB |
3316 | /* |
3317 | * add this inode to the orphan list so btrfs_orphan_del does | |
3318 | * the proper thing when we hit it | |
3319 | */ | |
8a35d95f JB |
3320 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3321 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3322 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3323 | |
7b128766 JB |
3324 | /* if we have links, this was a truncate, lets do that */ |
3325 | if (inode->i_nlink) { | |
fae7f21c | 3326 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3327 | iput(inode); |
3328 | continue; | |
3329 | } | |
7b128766 | 3330 | nr_truncate++; |
f3fe820c JB |
3331 | |
3332 | /* 1 for the orphan item deletion. */ | |
3333 | trans = btrfs_start_transaction(root, 1); | |
3334 | if (IS_ERR(trans)) { | |
c69b26b0 | 3335 | iput(inode); |
f3fe820c JB |
3336 | ret = PTR_ERR(trans); |
3337 | goto out; | |
3338 | } | |
3339 | ret = btrfs_orphan_add(trans, inode); | |
3340 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3341 | if (ret) { |
3342 | iput(inode); | |
f3fe820c | 3343 | goto out; |
c69b26b0 | 3344 | } |
f3fe820c | 3345 | |
66b4ffd1 | 3346 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3347 | if (ret) |
3348 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3349 | } else { |
3350 | nr_unlink++; | |
3351 | } | |
3352 | ||
3353 | /* this will do delete_inode and everything for us */ | |
3354 | iput(inode); | |
66b4ffd1 JB |
3355 | if (ret) |
3356 | goto out; | |
7b128766 | 3357 | } |
3254c876 MX |
3358 | /* release the path since we're done with it */ |
3359 | btrfs_release_path(path); | |
3360 | ||
d68fc57b YZ |
3361 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3362 | ||
3363 | if (root->orphan_block_rsv) | |
3364 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3365 | (u64)-1); | |
3366 | ||
27cdeb70 MX |
3367 | if (root->orphan_block_rsv || |
3368 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3369 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3370 | if (!IS_ERR(trans)) |
3371 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3372 | } |
7b128766 JB |
3373 | |
3374 | if (nr_unlink) | |
4884b476 | 3375 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3376 | if (nr_truncate) |
4884b476 | 3377 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3378 | |
3379 | out: | |
3380 | if (ret) | |
c2cf52eb SK |
3381 | btrfs_crit(root->fs_info, |
3382 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3383 | btrfs_free_path(path); |
3384 | return ret; | |
7b128766 JB |
3385 | } |
3386 | ||
46a53cca CM |
3387 | /* |
3388 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3389 | * don't find any xattrs, we know there can't be any acls. | |
3390 | * | |
3391 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3392 | */ | |
3393 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3394 | int slot, u64 objectid, |
3395 | int *first_xattr_slot) | |
46a53cca CM |
3396 | { |
3397 | u32 nritems = btrfs_header_nritems(leaf); | |
3398 | struct btrfs_key found_key; | |
f23b5a59 JB |
3399 | static u64 xattr_access = 0; |
3400 | static u64 xattr_default = 0; | |
46a53cca CM |
3401 | int scanned = 0; |
3402 | ||
f23b5a59 JB |
3403 | if (!xattr_access) { |
3404 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3405 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3406 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3407 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3408 | } | |
3409 | ||
46a53cca | 3410 | slot++; |
63541927 | 3411 | *first_xattr_slot = -1; |
46a53cca CM |
3412 | while (slot < nritems) { |
3413 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3414 | ||
3415 | /* we found a different objectid, there must not be acls */ | |
3416 | if (found_key.objectid != objectid) | |
3417 | return 0; | |
3418 | ||
3419 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3420 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3421 | if (*first_xattr_slot == -1) |
3422 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3423 | if (found_key.offset == xattr_access || |
3424 | found_key.offset == xattr_default) | |
3425 | return 1; | |
3426 | } | |
46a53cca CM |
3427 | |
3428 | /* | |
3429 | * we found a key greater than an xattr key, there can't | |
3430 | * be any acls later on | |
3431 | */ | |
3432 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3433 | return 0; | |
3434 | ||
3435 | slot++; | |
3436 | scanned++; | |
3437 | ||
3438 | /* | |
3439 | * it goes inode, inode backrefs, xattrs, extents, | |
3440 | * so if there are a ton of hard links to an inode there can | |
3441 | * be a lot of backrefs. Don't waste time searching too hard, | |
3442 | * this is just an optimization | |
3443 | */ | |
3444 | if (scanned >= 8) | |
3445 | break; | |
3446 | } | |
3447 | /* we hit the end of the leaf before we found an xattr or | |
3448 | * something larger than an xattr. We have to assume the inode | |
3449 | * has acls | |
3450 | */ | |
63541927 FDBM |
3451 | if (*first_xattr_slot == -1) |
3452 | *first_xattr_slot = slot; | |
46a53cca CM |
3453 | return 1; |
3454 | } | |
3455 | ||
d352ac68 CM |
3456 | /* |
3457 | * read an inode from the btree into the in-memory inode | |
3458 | */ | |
5d4f98a2 | 3459 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3460 | { |
3461 | struct btrfs_path *path; | |
5f39d397 | 3462 | struct extent_buffer *leaf; |
39279cc3 | 3463 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3464 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3465 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3466 | struct btrfs_key location; | |
67de1176 | 3467 | unsigned long ptr; |
46a53cca | 3468 | int maybe_acls; |
618e21d5 | 3469 | u32 rdev; |
39279cc3 | 3470 | int ret; |
2f7e33d4 | 3471 | bool filled = false; |
63541927 | 3472 | int first_xattr_slot; |
2f7e33d4 MX |
3473 | |
3474 | ret = btrfs_fill_inode(inode, &rdev); | |
3475 | if (!ret) | |
3476 | filled = true; | |
39279cc3 CM |
3477 | |
3478 | path = btrfs_alloc_path(); | |
1748f843 MF |
3479 | if (!path) |
3480 | goto make_bad; | |
3481 | ||
39279cc3 | 3482 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3483 | |
39279cc3 | 3484 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3485 | if (ret) |
39279cc3 | 3486 | goto make_bad; |
39279cc3 | 3487 | |
5f39d397 | 3488 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3489 | |
3490 | if (filled) | |
67de1176 | 3491 | goto cache_index; |
2f7e33d4 | 3492 | |
5f39d397 CM |
3493 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3494 | struct btrfs_inode_item); | |
5f39d397 | 3495 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3496 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3497 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3498 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3499 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3500 | |
3501 | tspec = btrfs_inode_atime(inode_item); | |
3502 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3503 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3504 | ||
3505 | tspec = btrfs_inode_mtime(inode_item); | |
3506 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3507 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3508 | ||
3509 | tspec = btrfs_inode_ctime(inode_item); | |
3510 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3511 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3512 | ||
a76a3cd4 | 3513 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3514 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3515 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3516 | ||
3517 | /* | |
3518 | * If we were modified in the current generation and evicted from memory | |
3519 | * and then re-read we need to do a full sync since we don't have any | |
3520 | * idea about which extents were modified before we were evicted from | |
3521 | * cache. | |
3522 | */ | |
3523 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3524 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3525 | &BTRFS_I(inode)->runtime_flags); | |
3526 | ||
0c4d2d95 | 3527 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3528 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3529 | inode->i_rdev = 0; |
5f39d397 CM |
3530 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3531 | ||
aec7477b | 3532 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3533 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
67de1176 MX |
3534 | |
3535 | cache_index: | |
3536 | path->slots[0]++; | |
3537 | if (inode->i_nlink != 1 || | |
3538 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3539 | goto cache_acl; | |
3540 | ||
3541 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3542 | if (location.objectid != btrfs_ino(inode)) | |
3543 | goto cache_acl; | |
3544 | ||
3545 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3546 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3547 | struct btrfs_inode_ref *ref; | |
3548 | ||
3549 | ref = (struct btrfs_inode_ref *)ptr; | |
3550 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3551 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3552 | struct btrfs_inode_extref *extref; | |
3553 | ||
3554 | extref = (struct btrfs_inode_extref *)ptr; | |
3555 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3556 | extref); | |
3557 | } | |
2f7e33d4 | 3558 | cache_acl: |
46a53cca CM |
3559 | /* |
3560 | * try to precache a NULL acl entry for files that don't have | |
3561 | * any xattrs or acls | |
3562 | */ | |
33345d01 | 3563 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3564 | btrfs_ino(inode), &first_xattr_slot); |
3565 | if (first_xattr_slot != -1) { | |
3566 | path->slots[0] = first_xattr_slot; | |
3567 | ret = btrfs_load_inode_props(inode, path); | |
3568 | if (ret) | |
3569 | btrfs_err(root->fs_info, | |
351fd353 | 3570 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3571 | btrfs_ino(inode), |
3572 | root->root_key.objectid, ret); | |
3573 | } | |
3574 | btrfs_free_path(path); | |
3575 | ||
72c04902 AV |
3576 | if (!maybe_acls) |
3577 | cache_no_acl(inode); | |
46a53cca | 3578 | |
39279cc3 | 3579 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3580 | case S_IFREG: |
3581 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3582 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3583 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3584 | inode->i_fop = &btrfs_file_operations; |
3585 | inode->i_op = &btrfs_file_inode_operations; | |
3586 | break; | |
3587 | case S_IFDIR: | |
3588 | inode->i_fop = &btrfs_dir_file_operations; | |
3589 | if (root == root->fs_info->tree_root) | |
3590 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3591 | else | |
3592 | inode->i_op = &btrfs_dir_inode_operations; | |
3593 | break; | |
3594 | case S_IFLNK: | |
3595 | inode->i_op = &btrfs_symlink_inode_operations; | |
3596 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3597 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3598 | break; |
618e21d5 | 3599 | default: |
0279b4cd | 3600 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3601 | init_special_inode(inode, inode->i_mode, rdev); |
3602 | break; | |
39279cc3 | 3603 | } |
6cbff00f CH |
3604 | |
3605 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3606 | return; |
3607 | ||
3608 | make_bad: | |
39279cc3 | 3609 | btrfs_free_path(path); |
39279cc3 CM |
3610 | make_bad_inode(inode); |
3611 | } | |
3612 | ||
d352ac68 CM |
3613 | /* |
3614 | * given a leaf and an inode, copy the inode fields into the leaf | |
3615 | */ | |
e02119d5 CM |
3616 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3617 | struct extent_buffer *leaf, | |
5f39d397 | 3618 | struct btrfs_inode_item *item, |
39279cc3 CM |
3619 | struct inode *inode) |
3620 | { | |
51fab693 LB |
3621 | struct btrfs_map_token token; |
3622 | ||
3623 | btrfs_init_map_token(&token); | |
5f39d397 | 3624 | |
51fab693 LB |
3625 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3626 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3627 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3628 | &token); | |
3629 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3630 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3631 | |
51fab693 LB |
3632 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3633 | inode->i_atime.tv_sec, &token); | |
3634 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3635 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3636 | |
51fab693 LB |
3637 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3638 | inode->i_mtime.tv_sec, &token); | |
3639 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3640 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3641 | |
51fab693 LB |
3642 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3643 | inode->i_ctime.tv_sec, &token); | |
3644 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3645 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3646 | |
51fab693 LB |
3647 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3648 | &token); | |
3649 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3650 | &token); | |
3651 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3652 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3653 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3654 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3655 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3656 | } |
3657 | ||
d352ac68 CM |
3658 | /* |
3659 | * copy everything in the in-memory inode into the btree. | |
3660 | */ | |
2115133f | 3661 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3662 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3663 | { |
3664 | struct btrfs_inode_item *inode_item; | |
3665 | struct btrfs_path *path; | |
5f39d397 | 3666 | struct extent_buffer *leaf; |
39279cc3 CM |
3667 | int ret; |
3668 | ||
3669 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3670 | if (!path) |
3671 | return -ENOMEM; | |
3672 | ||
b9473439 | 3673 | path->leave_spinning = 1; |
16cdcec7 MX |
3674 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3675 | 1); | |
39279cc3 CM |
3676 | if (ret) { |
3677 | if (ret > 0) | |
3678 | ret = -ENOENT; | |
3679 | goto failed; | |
3680 | } | |
3681 | ||
5f39d397 CM |
3682 | leaf = path->nodes[0]; |
3683 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3684 | struct btrfs_inode_item); |
39279cc3 | 3685 | |
e02119d5 | 3686 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3687 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3688 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3689 | ret = 0; |
3690 | failed: | |
39279cc3 CM |
3691 | btrfs_free_path(path); |
3692 | return ret; | |
3693 | } | |
3694 | ||
2115133f CM |
3695 | /* |
3696 | * copy everything in the in-memory inode into the btree. | |
3697 | */ | |
3698 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3699 | struct btrfs_root *root, struct inode *inode) | |
3700 | { | |
3701 | int ret; | |
3702 | ||
3703 | /* | |
3704 | * If the inode is a free space inode, we can deadlock during commit | |
3705 | * if we put it into the delayed code. | |
3706 | * | |
3707 | * The data relocation inode should also be directly updated | |
3708 | * without delay | |
3709 | */ | |
83eea1f1 | 3710 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 3711 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
3712 | btrfs_update_root_times(trans, root); |
3713 | ||
2115133f CM |
3714 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3715 | if (!ret) | |
3716 | btrfs_set_inode_last_trans(trans, inode); | |
3717 | return ret; | |
3718 | } | |
3719 | ||
3720 | return btrfs_update_inode_item(trans, root, inode); | |
3721 | } | |
3722 | ||
be6aef60 JB |
3723 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3724 | struct btrfs_root *root, | |
3725 | struct inode *inode) | |
2115133f CM |
3726 | { |
3727 | int ret; | |
3728 | ||
3729 | ret = btrfs_update_inode(trans, root, inode); | |
3730 | if (ret == -ENOSPC) | |
3731 | return btrfs_update_inode_item(trans, root, inode); | |
3732 | return ret; | |
3733 | } | |
3734 | ||
d352ac68 CM |
3735 | /* |
3736 | * unlink helper that gets used here in inode.c and in the tree logging | |
3737 | * recovery code. It remove a link in a directory with a given name, and | |
3738 | * also drops the back refs in the inode to the directory | |
3739 | */ | |
92986796 AV |
3740 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3741 | struct btrfs_root *root, | |
3742 | struct inode *dir, struct inode *inode, | |
3743 | const char *name, int name_len) | |
39279cc3 CM |
3744 | { |
3745 | struct btrfs_path *path; | |
39279cc3 | 3746 | int ret = 0; |
5f39d397 | 3747 | struct extent_buffer *leaf; |
39279cc3 | 3748 | struct btrfs_dir_item *di; |
5f39d397 | 3749 | struct btrfs_key key; |
aec7477b | 3750 | u64 index; |
33345d01 LZ |
3751 | u64 ino = btrfs_ino(inode); |
3752 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3753 | |
3754 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3755 | if (!path) { |
3756 | ret = -ENOMEM; | |
554233a6 | 3757 | goto out; |
54aa1f4d CM |
3758 | } |
3759 | ||
b9473439 | 3760 | path->leave_spinning = 1; |
33345d01 | 3761 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3762 | name, name_len, -1); |
3763 | if (IS_ERR(di)) { | |
3764 | ret = PTR_ERR(di); | |
3765 | goto err; | |
3766 | } | |
3767 | if (!di) { | |
3768 | ret = -ENOENT; | |
3769 | goto err; | |
3770 | } | |
5f39d397 CM |
3771 | leaf = path->nodes[0]; |
3772 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3773 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3774 | if (ret) |
3775 | goto err; | |
b3b4aa74 | 3776 | btrfs_release_path(path); |
39279cc3 | 3777 | |
67de1176 MX |
3778 | /* |
3779 | * If we don't have dir index, we have to get it by looking up | |
3780 | * the inode ref, since we get the inode ref, remove it directly, | |
3781 | * it is unnecessary to do delayed deletion. | |
3782 | * | |
3783 | * But if we have dir index, needn't search inode ref to get it. | |
3784 | * Since the inode ref is close to the inode item, it is better | |
3785 | * that we delay to delete it, and just do this deletion when | |
3786 | * we update the inode item. | |
3787 | */ | |
3788 | if (BTRFS_I(inode)->dir_index) { | |
3789 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3790 | if (!ret) { | |
3791 | index = BTRFS_I(inode)->dir_index; | |
3792 | goto skip_backref; | |
3793 | } | |
3794 | } | |
3795 | ||
33345d01 LZ |
3796 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3797 | dir_ino, &index); | |
aec7477b | 3798 | if (ret) { |
c2cf52eb SK |
3799 | btrfs_info(root->fs_info, |
3800 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3801 | name_len, name, ino, dir_ino); |
79787eaa | 3802 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3803 | goto err; |
3804 | } | |
67de1176 | 3805 | skip_backref: |
16cdcec7 | 3806 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3807 | if (ret) { |
3808 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3809 | goto err; |
79787eaa | 3810 | } |
39279cc3 | 3811 | |
e02119d5 | 3812 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3813 | inode, dir_ino); |
79787eaa JM |
3814 | if (ret != 0 && ret != -ENOENT) { |
3815 | btrfs_abort_transaction(trans, root, ret); | |
3816 | goto err; | |
3817 | } | |
e02119d5 CM |
3818 | |
3819 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3820 | dir, index); | |
6418c961 CM |
3821 | if (ret == -ENOENT) |
3822 | ret = 0; | |
d4e3991b ZB |
3823 | else if (ret) |
3824 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3825 | err: |
3826 | btrfs_free_path(path); | |
e02119d5 CM |
3827 | if (ret) |
3828 | goto out; | |
3829 | ||
3830 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3831 | inode_inc_iversion(inode); |
3832 | inode_inc_iversion(dir); | |
e02119d5 | 3833 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3834 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3835 | out: |
39279cc3 CM |
3836 | return ret; |
3837 | } | |
3838 | ||
92986796 AV |
3839 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3840 | struct btrfs_root *root, | |
3841 | struct inode *dir, struct inode *inode, | |
3842 | const char *name, int name_len) | |
3843 | { | |
3844 | int ret; | |
3845 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3846 | if (!ret) { | |
8b558c5f | 3847 | drop_nlink(inode); |
92986796 AV |
3848 | ret = btrfs_update_inode(trans, root, inode); |
3849 | } | |
3850 | return ret; | |
3851 | } | |
39279cc3 | 3852 | |
a22285a6 YZ |
3853 | /* |
3854 | * helper to start transaction for unlink and rmdir. | |
3855 | * | |
d52be818 JB |
3856 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3857 | * if we cannot make our reservations the normal way try and see if there is | |
3858 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3859 | * allow the unlink to occur. | |
a22285a6 | 3860 | */ |
d52be818 | 3861 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3862 | { |
39279cc3 | 3863 | struct btrfs_trans_handle *trans; |
a22285a6 | 3864 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3865 | int ret; |
3866 | ||
e70bea5f JB |
3867 | /* |
3868 | * 1 for the possible orphan item | |
3869 | * 1 for the dir item | |
3870 | * 1 for the dir index | |
3871 | * 1 for the inode ref | |
e70bea5f JB |
3872 | * 1 for the inode |
3873 | */ | |
6e137ed3 | 3874 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3875 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3876 | return trans; | |
4df27c4d | 3877 | |
d52be818 JB |
3878 | if (PTR_ERR(trans) == -ENOSPC) { |
3879 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3880 | |
d52be818 JB |
3881 | trans = btrfs_start_transaction(root, 0); |
3882 | if (IS_ERR(trans)) | |
3883 | return trans; | |
3884 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3885 | &root->fs_info->trans_block_rsv, | |
3886 | num_bytes, 5); | |
3887 | if (ret) { | |
3888 | btrfs_end_transaction(trans, root); | |
3889 | return ERR_PTR(ret); | |
a22285a6 | 3890 | } |
5a77d76c | 3891 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3892 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3893 | } |
d52be818 | 3894 | return trans; |
a22285a6 YZ |
3895 | } |
3896 | ||
3897 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3898 | { | |
3899 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3900 | struct btrfs_trans_handle *trans; | |
3901 | struct inode *inode = dentry->d_inode; | |
3902 | int ret; | |
a22285a6 | 3903 | |
d52be818 | 3904 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3905 | if (IS_ERR(trans)) |
3906 | return PTR_ERR(trans); | |
5f39d397 | 3907 | |
12fcfd22 CM |
3908 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3909 | ||
e02119d5 CM |
3910 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3911 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3912 | if (ret) |
3913 | goto out; | |
7b128766 | 3914 | |
a22285a6 | 3915 | if (inode->i_nlink == 0) { |
7b128766 | 3916 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3917 | if (ret) |
3918 | goto out; | |
a22285a6 | 3919 | } |
7b128766 | 3920 | |
b532402e | 3921 | out: |
d52be818 | 3922 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3923 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3924 | return ret; |
3925 | } | |
3926 | ||
4df27c4d YZ |
3927 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3928 | struct btrfs_root *root, | |
3929 | struct inode *dir, u64 objectid, | |
3930 | const char *name, int name_len) | |
3931 | { | |
3932 | struct btrfs_path *path; | |
3933 | struct extent_buffer *leaf; | |
3934 | struct btrfs_dir_item *di; | |
3935 | struct btrfs_key key; | |
3936 | u64 index; | |
3937 | int ret; | |
33345d01 | 3938 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3939 | |
3940 | path = btrfs_alloc_path(); | |
3941 | if (!path) | |
3942 | return -ENOMEM; | |
3943 | ||
33345d01 | 3944 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3945 | name, name_len, -1); |
79787eaa JM |
3946 | if (IS_ERR_OR_NULL(di)) { |
3947 | if (!di) | |
3948 | ret = -ENOENT; | |
3949 | else | |
3950 | ret = PTR_ERR(di); | |
3951 | goto out; | |
3952 | } | |
4df27c4d YZ |
3953 | |
3954 | leaf = path->nodes[0]; | |
3955 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3956 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3957 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3958 | if (ret) { |
3959 | btrfs_abort_transaction(trans, root, ret); | |
3960 | goto out; | |
3961 | } | |
b3b4aa74 | 3962 | btrfs_release_path(path); |
4df27c4d YZ |
3963 | |
3964 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3965 | objectid, root->root_key.objectid, | |
33345d01 | 3966 | dir_ino, &index, name, name_len); |
4df27c4d | 3967 | if (ret < 0) { |
79787eaa JM |
3968 | if (ret != -ENOENT) { |
3969 | btrfs_abort_transaction(trans, root, ret); | |
3970 | goto out; | |
3971 | } | |
33345d01 | 3972 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3973 | name, name_len); |
79787eaa JM |
3974 | if (IS_ERR_OR_NULL(di)) { |
3975 | if (!di) | |
3976 | ret = -ENOENT; | |
3977 | else | |
3978 | ret = PTR_ERR(di); | |
3979 | btrfs_abort_transaction(trans, root, ret); | |
3980 | goto out; | |
3981 | } | |
4df27c4d YZ |
3982 | |
3983 | leaf = path->nodes[0]; | |
3984 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3985 | btrfs_release_path(path); |
4df27c4d YZ |
3986 | index = key.offset; |
3987 | } | |
945d8962 | 3988 | btrfs_release_path(path); |
4df27c4d | 3989 | |
16cdcec7 | 3990 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3991 | if (ret) { |
3992 | btrfs_abort_transaction(trans, root, ret); | |
3993 | goto out; | |
3994 | } | |
4df27c4d YZ |
3995 | |
3996 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3997 | inode_inc_iversion(dir); |
4df27c4d | 3998 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3999 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4000 | if (ret) |
4001 | btrfs_abort_transaction(trans, root, ret); | |
4002 | out: | |
71d7aed0 | 4003 | btrfs_free_path(path); |
79787eaa | 4004 | return ret; |
4df27c4d YZ |
4005 | } |
4006 | ||
39279cc3 CM |
4007 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4008 | { | |
4009 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 4010 | int err = 0; |
39279cc3 | 4011 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4012 | struct btrfs_trans_handle *trans; |
39279cc3 | 4013 | |
b3ae244e | 4014 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4015 | return -ENOTEMPTY; |
b3ae244e DS |
4016 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4017 | return -EPERM; | |
134d4512 | 4018 | |
d52be818 | 4019 | trans = __unlink_start_trans(dir); |
a22285a6 | 4020 | if (IS_ERR(trans)) |
5df6a9f6 | 4021 | return PTR_ERR(trans); |
5df6a9f6 | 4022 | |
33345d01 | 4023 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4024 | err = btrfs_unlink_subvol(trans, root, dir, |
4025 | BTRFS_I(inode)->location.objectid, | |
4026 | dentry->d_name.name, | |
4027 | dentry->d_name.len); | |
4028 | goto out; | |
4029 | } | |
4030 | ||
7b128766 JB |
4031 | err = btrfs_orphan_add(trans, inode); |
4032 | if (err) | |
4df27c4d | 4033 | goto out; |
7b128766 | 4034 | |
39279cc3 | 4035 | /* now the directory is empty */ |
e02119d5 CM |
4036 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
4037 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 4038 | if (!err) |
dbe674a9 | 4039 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4040 | out: |
d52be818 | 4041 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4042 | btrfs_btree_balance_dirty(root); |
3954401f | 4043 | |
39279cc3 CM |
4044 | return err; |
4045 | } | |
4046 | ||
39279cc3 CM |
4047 | /* |
4048 | * this can truncate away extent items, csum items and directory items. | |
4049 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4050 | * any higher than new_size |
39279cc3 CM |
4051 | * |
4052 | * csum items that cross the new i_size are truncated to the new size | |
4053 | * as well. | |
7b128766 JB |
4054 | * |
4055 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4056 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4057 | */ |
8082510e YZ |
4058 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4059 | struct btrfs_root *root, | |
4060 | struct inode *inode, | |
4061 | u64 new_size, u32 min_type) | |
39279cc3 | 4062 | { |
39279cc3 | 4063 | struct btrfs_path *path; |
5f39d397 | 4064 | struct extent_buffer *leaf; |
39279cc3 | 4065 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4066 | struct btrfs_key key; |
4067 | struct btrfs_key found_key; | |
39279cc3 | 4068 | u64 extent_start = 0; |
db94535d | 4069 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4070 | u64 extent_offset = 0; |
39279cc3 | 4071 | u64 item_end = 0; |
7f4f6e0a | 4072 | u64 last_size = (u64)-1; |
8082510e | 4073 | u32 found_type = (u8)-1; |
39279cc3 CM |
4074 | int found_extent; |
4075 | int del_item; | |
85e21bac CM |
4076 | int pending_del_nr = 0; |
4077 | int pending_del_slot = 0; | |
179e29e4 | 4078 | int extent_type = -1; |
8082510e YZ |
4079 | int ret; |
4080 | int err = 0; | |
33345d01 | 4081 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
4082 | |
4083 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4084 | |
0eb0e19c MF |
4085 | path = btrfs_alloc_path(); |
4086 | if (!path) | |
4087 | return -ENOMEM; | |
4088 | path->reada = -1; | |
4089 | ||
5dc562c5 JB |
4090 | /* |
4091 | * We want to drop from the next block forward in case this new size is | |
4092 | * not block aligned since we will be keeping the last block of the | |
4093 | * extent just the way it is. | |
4094 | */ | |
27cdeb70 MX |
4095 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4096 | root == root->fs_info->tree_root) | |
fda2832f QW |
4097 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4098 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4099 | |
16cdcec7 MX |
4100 | /* |
4101 | * This function is also used to drop the items in the log tree before | |
4102 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4103 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4104 | * items. | |
4105 | */ | |
4106 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4107 | btrfs_kill_delayed_inode_items(inode); | |
4108 | ||
33345d01 | 4109 | key.objectid = ino; |
39279cc3 | 4110 | key.offset = (u64)-1; |
5f39d397 CM |
4111 | key.type = (u8)-1; |
4112 | ||
85e21bac | 4113 | search_again: |
b9473439 | 4114 | path->leave_spinning = 1; |
85e21bac | 4115 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4116 | if (ret < 0) { |
4117 | err = ret; | |
4118 | goto out; | |
4119 | } | |
d397712b | 4120 | |
85e21bac | 4121 | if (ret > 0) { |
e02119d5 CM |
4122 | /* there are no items in the tree for us to truncate, we're |
4123 | * done | |
4124 | */ | |
8082510e YZ |
4125 | if (path->slots[0] == 0) |
4126 | goto out; | |
85e21bac CM |
4127 | path->slots[0]--; |
4128 | } | |
4129 | ||
d397712b | 4130 | while (1) { |
39279cc3 | 4131 | fi = NULL; |
5f39d397 CM |
4132 | leaf = path->nodes[0]; |
4133 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4134 | found_type = found_key.type; |
39279cc3 | 4135 | |
33345d01 | 4136 | if (found_key.objectid != ino) |
39279cc3 | 4137 | break; |
5f39d397 | 4138 | |
85e21bac | 4139 | if (found_type < min_type) |
39279cc3 CM |
4140 | break; |
4141 | ||
5f39d397 | 4142 | item_end = found_key.offset; |
39279cc3 | 4143 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4144 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4145 | struct btrfs_file_extent_item); |
179e29e4 CM |
4146 | extent_type = btrfs_file_extent_type(leaf, fi); |
4147 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4148 | item_end += |
db94535d | 4149 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4150 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4151 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4152 | path->slots[0], fi); |
39279cc3 | 4153 | } |
008630c1 | 4154 | item_end--; |
39279cc3 | 4155 | } |
8082510e YZ |
4156 | if (found_type > min_type) { |
4157 | del_item = 1; | |
4158 | } else { | |
4159 | if (item_end < new_size) | |
b888db2b | 4160 | break; |
8082510e YZ |
4161 | if (found_key.offset >= new_size) |
4162 | del_item = 1; | |
4163 | else | |
4164 | del_item = 0; | |
39279cc3 | 4165 | } |
39279cc3 | 4166 | found_extent = 0; |
39279cc3 | 4167 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4168 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4169 | goto delete; | |
4170 | ||
7f4f6e0a JB |
4171 | if (del_item) |
4172 | last_size = found_key.offset; | |
4173 | else | |
4174 | last_size = new_size; | |
4175 | ||
179e29e4 | 4176 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4177 | u64 num_dec; |
db94535d | 4178 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4179 | if (!del_item) { |
db94535d CM |
4180 | u64 orig_num_bytes = |
4181 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4182 | extent_num_bytes = ALIGN(new_size - |
4183 | found_key.offset, | |
4184 | root->sectorsize); | |
db94535d CM |
4185 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4186 | extent_num_bytes); | |
4187 | num_dec = (orig_num_bytes - | |
9069218d | 4188 | extent_num_bytes); |
27cdeb70 MX |
4189 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4190 | &root->state) && | |
4191 | extent_start != 0) | |
a76a3cd4 | 4192 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4193 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4194 | } else { |
db94535d CM |
4195 | extent_num_bytes = |
4196 | btrfs_file_extent_disk_num_bytes(leaf, | |
4197 | fi); | |
5d4f98a2 YZ |
4198 | extent_offset = found_key.offset - |
4199 | btrfs_file_extent_offset(leaf, fi); | |
4200 | ||
39279cc3 | 4201 | /* FIXME blocksize != 4096 */ |
9069218d | 4202 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4203 | if (extent_start != 0) { |
4204 | found_extent = 1; | |
27cdeb70 MX |
4205 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4206 | &root->state)) | |
a76a3cd4 | 4207 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4208 | } |
39279cc3 | 4209 | } |
9069218d | 4210 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4211 | /* |
4212 | * we can't truncate inline items that have had | |
4213 | * special encodings | |
4214 | */ | |
4215 | if (!del_item && | |
4216 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4217 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4218 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4219 | u32 size = new_size - found_key.offset; |
4220 | ||
27cdeb70 | 4221 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
a76a3cd4 YZ |
4222 | inode_sub_bytes(inode, item_end + 1 - |
4223 | new_size); | |
514ac8ad CM |
4224 | |
4225 | /* | |
4226 | * update the ram bytes to properly reflect | |
4227 | * the new size of our item | |
4228 | */ | |
4229 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
e02119d5 CM |
4230 | size = |
4231 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4232 | btrfs_truncate_item(root, path, size, 1); |
27cdeb70 MX |
4233 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4234 | &root->state)) { | |
a76a3cd4 YZ |
4235 | inode_sub_bytes(inode, item_end + 1 - |
4236 | found_key.offset); | |
9069218d | 4237 | } |
39279cc3 | 4238 | } |
179e29e4 | 4239 | delete: |
39279cc3 | 4240 | if (del_item) { |
85e21bac CM |
4241 | if (!pending_del_nr) { |
4242 | /* no pending yet, add ourselves */ | |
4243 | pending_del_slot = path->slots[0]; | |
4244 | pending_del_nr = 1; | |
4245 | } else if (pending_del_nr && | |
4246 | path->slots[0] + 1 == pending_del_slot) { | |
4247 | /* hop on the pending chunk */ | |
4248 | pending_del_nr++; | |
4249 | pending_del_slot = path->slots[0]; | |
4250 | } else { | |
d397712b | 4251 | BUG(); |
85e21bac | 4252 | } |
39279cc3 CM |
4253 | } else { |
4254 | break; | |
4255 | } | |
27cdeb70 MX |
4256 | if (found_extent && |
4257 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4258 | root == root->fs_info->tree_root)) { | |
b9473439 | 4259 | btrfs_set_path_blocking(path); |
39279cc3 | 4260 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4261 | extent_num_bytes, 0, |
4262 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4263 | ino, extent_offset, 0); |
39279cc3 CM |
4264 | BUG_ON(ret); |
4265 | } | |
85e21bac | 4266 | |
8082510e YZ |
4267 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4268 | break; | |
4269 | ||
4270 | if (path->slots[0] == 0 || | |
4271 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4272 | if (pending_del_nr) { |
4273 | ret = btrfs_del_items(trans, root, path, | |
4274 | pending_del_slot, | |
4275 | pending_del_nr); | |
79787eaa JM |
4276 | if (ret) { |
4277 | btrfs_abort_transaction(trans, | |
4278 | root, ret); | |
4279 | goto error; | |
4280 | } | |
8082510e YZ |
4281 | pending_del_nr = 0; |
4282 | } | |
b3b4aa74 | 4283 | btrfs_release_path(path); |
85e21bac | 4284 | goto search_again; |
8082510e YZ |
4285 | } else { |
4286 | path->slots[0]--; | |
85e21bac | 4287 | } |
39279cc3 | 4288 | } |
8082510e | 4289 | out: |
85e21bac CM |
4290 | if (pending_del_nr) { |
4291 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4292 | pending_del_nr); | |
79787eaa JM |
4293 | if (ret) |
4294 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4295 | } |
79787eaa | 4296 | error: |
dac5705c FM |
4297 | if (last_size != (u64)-1 && |
4298 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
7f4f6e0a | 4299 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
39279cc3 | 4300 | btrfs_free_path(path); |
8082510e | 4301 | return err; |
39279cc3 CM |
4302 | } |
4303 | ||
4304 | /* | |
2aaa6655 JB |
4305 | * btrfs_truncate_page - read, zero a chunk and write a page |
4306 | * @inode - inode that we're zeroing | |
4307 | * @from - the offset to start zeroing | |
4308 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4309 | * offset | |
4310 | * @front - zero up to the offset instead of from the offset on | |
4311 | * | |
4312 | * This will find the page for the "from" offset and cow the page and zero the | |
4313 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4314 | */ |
2aaa6655 JB |
4315 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4316 | int front) | |
39279cc3 | 4317 | { |
2aaa6655 | 4318 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4319 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4320 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4321 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4322 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4323 | char *kaddr; |
db94535d | 4324 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4325 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4326 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4327 | struct page *page; | |
3b16a4e3 | 4328 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4329 | int ret = 0; |
a52d9a80 | 4330 | u64 page_start; |
e6dcd2dc | 4331 | u64 page_end; |
39279cc3 | 4332 | |
2aaa6655 JB |
4333 | if ((offset & (blocksize - 1)) == 0 && |
4334 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4335 | goto out; |
0ca1f7ce | 4336 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4337 | if (ret) |
4338 | goto out; | |
39279cc3 | 4339 | |
211c17f5 | 4340 | again: |
3b16a4e3 | 4341 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4342 | if (!page) { |
0ca1f7ce | 4343 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4344 | ret = -ENOMEM; |
39279cc3 | 4345 | goto out; |
5d5e103a | 4346 | } |
e6dcd2dc CM |
4347 | |
4348 | page_start = page_offset(page); | |
4349 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4350 | ||
39279cc3 | 4351 | if (!PageUptodate(page)) { |
9ebefb18 | 4352 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4353 | lock_page(page); |
211c17f5 CM |
4354 | if (page->mapping != mapping) { |
4355 | unlock_page(page); | |
4356 | page_cache_release(page); | |
4357 | goto again; | |
4358 | } | |
39279cc3 CM |
4359 | if (!PageUptodate(page)) { |
4360 | ret = -EIO; | |
89642229 | 4361 | goto out_unlock; |
39279cc3 CM |
4362 | } |
4363 | } | |
211c17f5 | 4364 | wait_on_page_writeback(page); |
e6dcd2dc | 4365 | |
d0082371 | 4366 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4367 | set_page_extent_mapped(page); |
4368 | ||
4369 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4370 | if (ordered) { | |
2ac55d41 JB |
4371 | unlock_extent_cached(io_tree, page_start, page_end, |
4372 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4373 | unlock_page(page); |
4374 | page_cache_release(page); | |
eb84ae03 | 4375 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4376 | btrfs_put_ordered_extent(ordered); |
4377 | goto again; | |
4378 | } | |
4379 | ||
2ac55d41 | 4380 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4381 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4382 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4383 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4384 | |
2ac55d41 JB |
4385 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4386 | &cached_state); | |
9ed74f2d | 4387 | if (ret) { |
2ac55d41 JB |
4388 | unlock_extent_cached(io_tree, page_start, page_end, |
4389 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4390 | goto out_unlock; |
4391 | } | |
4392 | ||
e6dcd2dc | 4393 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4394 | if (!len) |
4395 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4396 | kaddr = kmap(page); |
2aaa6655 JB |
4397 | if (front) |
4398 | memset(kaddr, 0, offset); | |
4399 | else | |
4400 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4401 | flush_dcache_page(page); |
4402 | kunmap(page); | |
4403 | } | |
247e743c | 4404 | ClearPageChecked(page); |
e6dcd2dc | 4405 | set_page_dirty(page); |
2ac55d41 JB |
4406 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4407 | GFP_NOFS); | |
39279cc3 | 4408 | |
89642229 | 4409 | out_unlock: |
5d5e103a | 4410 | if (ret) |
0ca1f7ce | 4411 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4412 | unlock_page(page); |
4413 | page_cache_release(page); | |
4414 | out: | |
4415 | return ret; | |
4416 | } | |
4417 | ||
16e7549f JB |
4418 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4419 | u64 offset, u64 len) | |
4420 | { | |
4421 | struct btrfs_trans_handle *trans; | |
4422 | int ret; | |
4423 | ||
4424 | /* | |
4425 | * Still need to make sure the inode looks like it's been updated so | |
4426 | * that any holes get logged if we fsync. | |
4427 | */ | |
4428 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4429 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4430 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4431 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4432 | return 0; | |
4433 | } | |
4434 | ||
4435 | /* | |
4436 | * 1 - for the one we're dropping | |
4437 | * 1 - for the one we're adding | |
4438 | * 1 - for updating the inode. | |
4439 | */ | |
4440 | trans = btrfs_start_transaction(root, 3); | |
4441 | if (IS_ERR(trans)) | |
4442 | return PTR_ERR(trans); | |
4443 | ||
4444 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4445 | if (ret) { | |
4446 | btrfs_abort_transaction(trans, root, ret); | |
4447 | btrfs_end_transaction(trans, root); | |
4448 | return ret; | |
4449 | } | |
4450 | ||
4451 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4452 | 0, 0, len, 0, len, 0, 0, 0); | |
4453 | if (ret) | |
4454 | btrfs_abort_transaction(trans, root, ret); | |
4455 | else | |
4456 | btrfs_update_inode(trans, root, inode); | |
4457 | btrfs_end_transaction(trans, root); | |
4458 | return ret; | |
4459 | } | |
4460 | ||
695a0d0d JB |
4461 | /* |
4462 | * This function puts in dummy file extents for the area we're creating a hole | |
4463 | * for. So if we are truncating this file to a larger size we need to insert | |
4464 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4465 | * the range between oldsize and size | |
4466 | */ | |
a41ad394 | 4467 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4468 | { |
9036c102 YZ |
4469 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4470 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4471 | struct extent_map *em = NULL; |
2ac55d41 | 4472 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4473 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4474 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4475 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4476 | u64 last_byte; |
4477 | u64 cur_offset; | |
4478 | u64 hole_size; | |
9ed74f2d | 4479 | int err = 0; |
39279cc3 | 4480 | |
a71754fc JB |
4481 | /* |
4482 | * If our size started in the middle of a page we need to zero out the | |
4483 | * rest of the page before we expand the i_size, otherwise we could | |
4484 | * expose stale data. | |
4485 | */ | |
4486 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4487 | if (err) | |
4488 | return err; | |
4489 | ||
9036c102 YZ |
4490 | if (size <= hole_start) |
4491 | return 0; | |
4492 | ||
9036c102 YZ |
4493 | while (1) { |
4494 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4495 | |
2ac55d41 | 4496 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4497 | &cached_state); |
fa7c1494 MX |
4498 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4499 | block_end - hole_start); | |
9036c102 YZ |
4500 | if (!ordered) |
4501 | break; | |
2ac55d41 JB |
4502 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4503 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4504 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4505 | btrfs_put_ordered_extent(ordered); |
4506 | } | |
39279cc3 | 4507 | |
9036c102 YZ |
4508 | cur_offset = hole_start; |
4509 | while (1) { | |
4510 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4511 | block_end - cur_offset, 0); | |
79787eaa JM |
4512 | if (IS_ERR(em)) { |
4513 | err = PTR_ERR(em); | |
f2767956 | 4514 | em = NULL; |
79787eaa JM |
4515 | break; |
4516 | } | |
9036c102 | 4517 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4518 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4519 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4520 | struct extent_map *hole_em; |
9036c102 | 4521 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4522 | |
16e7549f JB |
4523 | err = maybe_insert_hole(root, inode, cur_offset, |
4524 | hole_size); | |
4525 | if (err) | |
3893e33b | 4526 | break; |
5dc562c5 JB |
4527 | btrfs_drop_extent_cache(inode, cur_offset, |
4528 | cur_offset + hole_size - 1, 0); | |
4529 | hole_em = alloc_extent_map(); | |
4530 | if (!hole_em) { | |
4531 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4532 | &BTRFS_I(inode)->runtime_flags); | |
4533 | goto next; | |
4534 | } | |
4535 | hole_em->start = cur_offset; | |
4536 | hole_em->len = hole_size; | |
4537 | hole_em->orig_start = cur_offset; | |
8082510e | 4538 | |
5dc562c5 JB |
4539 | hole_em->block_start = EXTENT_MAP_HOLE; |
4540 | hole_em->block_len = 0; | |
b4939680 | 4541 | hole_em->orig_block_len = 0; |
cc95bef6 | 4542 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4543 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4544 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4545 | hole_em->generation = root->fs_info->generation; |
8082510e | 4546 | |
5dc562c5 JB |
4547 | while (1) { |
4548 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4549 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4550 | write_unlock(&em_tree->lock); |
4551 | if (err != -EEXIST) | |
4552 | break; | |
4553 | btrfs_drop_extent_cache(inode, cur_offset, | |
4554 | cur_offset + | |
4555 | hole_size - 1, 0); | |
4556 | } | |
4557 | free_extent_map(hole_em); | |
9036c102 | 4558 | } |
16e7549f | 4559 | next: |
9036c102 | 4560 | free_extent_map(em); |
a22285a6 | 4561 | em = NULL; |
9036c102 | 4562 | cur_offset = last_byte; |
8082510e | 4563 | if (cur_offset >= block_end) |
9036c102 YZ |
4564 | break; |
4565 | } | |
a22285a6 | 4566 | free_extent_map(em); |
2ac55d41 JB |
4567 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4568 | GFP_NOFS); | |
9036c102 YZ |
4569 | return err; |
4570 | } | |
39279cc3 | 4571 | |
3972f260 | 4572 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4573 | { |
f4a2f4c5 MX |
4574 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4575 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4576 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4577 | loff_t newsize = attr->ia_size; |
4578 | int mask = attr->ia_valid; | |
8082510e YZ |
4579 | int ret; |
4580 | ||
3972f260 ES |
4581 | /* |
4582 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4583 | * special case where we need to update the times despite not having | |
4584 | * these flags set. For all other operations the VFS set these flags | |
4585 | * explicitly if it wants a timestamp update. | |
4586 | */ | |
dff6efc3 CH |
4587 | if (newsize != oldsize) { |
4588 | inode_inc_iversion(inode); | |
4589 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4590 | inode->i_ctime = inode->i_mtime = | |
4591 | current_fs_time(inode->i_sb); | |
4592 | } | |
3972f260 | 4593 | |
a41ad394 | 4594 | if (newsize > oldsize) { |
7caef267 | 4595 | truncate_pagecache(inode, newsize); |
a41ad394 | 4596 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
f4a2f4c5 | 4597 | if (ret) |
8082510e | 4598 | return ret; |
8082510e | 4599 | |
f4a2f4c5 MX |
4600 | trans = btrfs_start_transaction(root, 1); |
4601 | if (IS_ERR(trans)) | |
4602 | return PTR_ERR(trans); | |
4603 | ||
4604 | i_size_write(inode, newsize); | |
4605 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4606 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4607 | btrfs_end_transaction(trans, root); |
a41ad394 | 4608 | } else { |
8082510e | 4609 | |
a41ad394 JB |
4610 | /* |
4611 | * We're truncating a file that used to have good data down to | |
4612 | * zero. Make sure it gets into the ordered flush list so that | |
4613 | * any new writes get down to disk quickly. | |
4614 | */ | |
4615 | if (newsize == 0) | |
72ac3c0d JB |
4616 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4617 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4618 | |
f3fe820c JB |
4619 | /* |
4620 | * 1 for the orphan item we're going to add | |
4621 | * 1 for the orphan item deletion. | |
4622 | */ | |
4623 | trans = btrfs_start_transaction(root, 2); | |
4624 | if (IS_ERR(trans)) | |
4625 | return PTR_ERR(trans); | |
4626 | ||
4627 | /* | |
4628 | * We need to do this in case we fail at _any_ point during the | |
4629 | * actual truncate. Once we do the truncate_setsize we could | |
4630 | * invalidate pages which forces any outstanding ordered io to | |
4631 | * be instantly completed which will give us extents that need | |
4632 | * to be truncated. If we fail to get an orphan inode down we | |
4633 | * could have left over extents that were never meant to live, | |
4634 | * so we need to garuntee from this point on that everything | |
4635 | * will be consistent. | |
4636 | */ | |
4637 | ret = btrfs_orphan_add(trans, inode); | |
4638 | btrfs_end_transaction(trans, root); | |
4639 | if (ret) | |
4640 | return ret; | |
4641 | ||
a41ad394 JB |
4642 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4643 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4644 | |
4645 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4646 | btrfs_inode_block_unlocked_dio(inode); | |
4647 | inode_dio_wait(inode); | |
4648 | btrfs_inode_resume_unlocked_dio(inode); | |
4649 | ||
a41ad394 | 4650 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4651 | if (ret && inode->i_nlink) { |
4652 | int err; | |
4653 | ||
4654 | /* | |
4655 | * failed to truncate, disk_i_size is only adjusted down | |
4656 | * as we remove extents, so it should represent the true | |
4657 | * size of the inode, so reset the in memory size and | |
4658 | * delete our orphan entry. | |
4659 | */ | |
4660 | trans = btrfs_join_transaction(root); | |
4661 | if (IS_ERR(trans)) { | |
4662 | btrfs_orphan_del(NULL, inode); | |
4663 | return ret; | |
4664 | } | |
4665 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4666 | err = btrfs_orphan_del(trans, inode); | |
4667 | if (err) | |
4668 | btrfs_abort_transaction(trans, root, err); | |
4669 | btrfs_end_transaction(trans, root); | |
4670 | } | |
8082510e YZ |
4671 | } |
4672 | ||
a41ad394 | 4673 | return ret; |
8082510e YZ |
4674 | } |
4675 | ||
9036c102 YZ |
4676 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4677 | { | |
4678 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4679 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4680 | int err; |
39279cc3 | 4681 | |
b83cc969 LZ |
4682 | if (btrfs_root_readonly(root)) |
4683 | return -EROFS; | |
4684 | ||
9036c102 YZ |
4685 | err = inode_change_ok(inode, attr); |
4686 | if (err) | |
4687 | return err; | |
2bf5a725 | 4688 | |
5a3f23d5 | 4689 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4690 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4691 | if (err) |
4692 | return err; | |
39279cc3 | 4693 | } |
9036c102 | 4694 | |
1025774c CH |
4695 | if (attr->ia_valid) { |
4696 | setattr_copy(inode, attr); | |
0c4d2d95 | 4697 | inode_inc_iversion(inode); |
22c44fe6 | 4698 | err = btrfs_dirty_inode(inode); |
1025774c | 4699 | |
22c44fe6 | 4700 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 4701 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 4702 | } |
33268eaf | 4703 | |
39279cc3 CM |
4704 | return err; |
4705 | } | |
61295eb8 | 4706 | |
131e404a FDBM |
4707 | /* |
4708 | * While truncating the inode pages during eviction, we get the VFS calling | |
4709 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
4710 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
4711 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
4712 | * extent_state structures over and over, wasting lots of time. | |
4713 | * | |
4714 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
4715 | * those expensive operations on a per page basis and do only the ordered io | |
4716 | * finishing, while we release here the extent_map and extent_state structures, | |
4717 | * without the excessive merging and splitting. | |
4718 | */ | |
4719 | static void evict_inode_truncate_pages(struct inode *inode) | |
4720 | { | |
4721 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
4722 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
4723 | struct rb_node *node; | |
4724 | ||
4725 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 4726 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
4727 | |
4728 | write_lock(&map_tree->lock); | |
4729 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
4730 | struct extent_map *em; | |
4731 | ||
4732 | node = rb_first(&map_tree->map); | |
4733 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
4734 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
4735 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
4736 | remove_extent_mapping(map_tree, em); |
4737 | free_extent_map(em); | |
7064dd5c FM |
4738 | if (need_resched()) { |
4739 | write_unlock(&map_tree->lock); | |
4740 | cond_resched(); | |
4741 | write_lock(&map_tree->lock); | |
4742 | } | |
131e404a FDBM |
4743 | } |
4744 | write_unlock(&map_tree->lock); | |
4745 | ||
4746 | spin_lock(&io_tree->lock); | |
4747 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
4748 | struct extent_state *state; | |
4749 | struct extent_state *cached_state = NULL; | |
4750 | ||
4751 | node = rb_first(&io_tree->state); | |
4752 | state = rb_entry(node, struct extent_state, rb_node); | |
4753 | atomic_inc(&state->refs); | |
4754 | spin_unlock(&io_tree->lock); | |
4755 | ||
4756 | lock_extent_bits(io_tree, state->start, state->end, | |
4757 | 0, &cached_state); | |
4758 | clear_extent_bit(io_tree, state->start, state->end, | |
4759 | EXTENT_LOCKED | EXTENT_DIRTY | | |
4760 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
4761 | EXTENT_DEFRAG, 1, 1, | |
4762 | &cached_state, GFP_NOFS); | |
4763 | free_extent_state(state); | |
4764 | ||
7064dd5c | 4765 | cond_resched(); |
131e404a FDBM |
4766 | spin_lock(&io_tree->lock); |
4767 | } | |
4768 | spin_unlock(&io_tree->lock); | |
4769 | } | |
4770 | ||
bd555975 | 4771 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4772 | { |
4773 | struct btrfs_trans_handle *trans; | |
4774 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4775 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4776 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4777 | int ret; |
4778 | ||
1abe9b8a | 4779 | trace_btrfs_inode_evict(inode); |
4780 | ||
131e404a FDBM |
4781 | evict_inode_truncate_pages(inode); |
4782 | ||
69e9c6c6 SB |
4783 | if (inode->i_nlink && |
4784 | ((btrfs_root_refs(&root->root_item) != 0 && | |
4785 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
4786 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
4787 | goto no_delete; |
4788 | ||
39279cc3 | 4789 | if (is_bad_inode(inode)) { |
7b128766 | 4790 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4791 | goto no_delete; |
4792 | } | |
bd555975 | 4793 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4794 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4795 | |
c71bf099 | 4796 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4797 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4798 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4799 | goto no_delete; |
4800 | } | |
4801 | ||
76dda93c | 4802 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
4803 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
4804 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
4805 | goto no_delete; |
4806 | } | |
4807 | ||
0e8c36a9 MX |
4808 | ret = btrfs_commit_inode_delayed_inode(inode); |
4809 | if (ret) { | |
4810 | btrfs_orphan_del(NULL, inode); | |
4811 | goto no_delete; | |
4812 | } | |
4813 | ||
66d8f3dd | 4814 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4815 | if (!rsv) { |
4816 | btrfs_orphan_del(NULL, inode); | |
4817 | goto no_delete; | |
4818 | } | |
4a338542 | 4819 | rsv->size = min_size; |
ca7e70f5 | 4820 | rsv->failfast = 1; |
726c35fa | 4821 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4822 | |
dbe674a9 | 4823 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4824 | |
4289a667 | 4825 | /* |
8407aa46 MX |
4826 | * This is a bit simpler than btrfs_truncate since we've already |
4827 | * reserved our space for our orphan item in the unlink, so we just | |
4828 | * need to reserve some slack space in case we add bytes and update | |
4829 | * inode item when doing the truncate. | |
4289a667 | 4830 | */ |
8082510e | 4831 | while (1) { |
08e007d2 MX |
4832 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4833 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4834 | |
4835 | /* | |
4836 | * Try and steal from the global reserve since we will | |
4837 | * likely not use this space anyway, we want to try as | |
4838 | * hard as possible to get this to work. | |
4839 | */ | |
4840 | if (ret) | |
4841 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4842 | |
d68fc57b | 4843 | if (ret) { |
c2cf52eb SK |
4844 | btrfs_warn(root->fs_info, |
4845 | "Could not get space for a delete, will truncate on mount %d", | |
4846 | ret); | |
4289a667 JB |
4847 | btrfs_orphan_del(NULL, inode); |
4848 | btrfs_free_block_rsv(root, rsv); | |
4849 | goto no_delete; | |
d68fc57b | 4850 | } |
7b128766 | 4851 | |
0e8c36a9 | 4852 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4853 | if (IS_ERR(trans)) { |
4854 | btrfs_orphan_del(NULL, inode); | |
4855 | btrfs_free_block_rsv(root, rsv); | |
4856 | goto no_delete; | |
d68fc57b | 4857 | } |
7b128766 | 4858 | |
4289a667 JB |
4859 | trans->block_rsv = rsv; |
4860 | ||
d68fc57b | 4861 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4862 | if (ret != -ENOSPC) |
8082510e | 4863 | break; |
85e21bac | 4864 | |
8407aa46 | 4865 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4866 | btrfs_end_transaction(trans, root); |
4867 | trans = NULL; | |
b53d3f5d | 4868 | btrfs_btree_balance_dirty(root); |
8082510e | 4869 | } |
5f39d397 | 4870 | |
4289a667 JB |
4871 | btrfs_free_block_rsv(root, rsv); |
4872 | ||
4ef31a45 JB |
4873 | /* |
4874 | * Errors here aren't a big deal, it just means we leave orphan items | |
4875 | * in the tree. They will be cleaned up on the next mount. | |
4876 | */ | |
8082510e | 4877 | if (ret == 0) { |
4289a667 | 4878 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
4879 | btrfs_orphan_del(trans, inode); |
4880 | } else { | |
4881 | btrfs_orphan_del(NULL, inode); | |
8082510e | 4882 | } |
54aa1f4d | 4883 | |
4289a667 | 4884 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4885 | if (!(root == root->fs_info->tree_root || |
4886 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4887 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4888 | |
54aa1f4d | 4889 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4890 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4891 | no_delete: |
89042e5a | 4892 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4893 | clear_inode(inode); |
8082510e | 4894 | return; |
39279cc3 CM |
4895 | } |
4896 | ||
4897 | /* | |
4898 | * this returns the key found in the dir entry in the location pointer. | |
4899 | * If no dir entries were found, location->objectid is 0. | |
4900 | */ | |
4901 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4902 | struct btrfs_key *location) | |
4903 | { | |
4904 | const char *name = dentry->d_name.name; | |
4905 | int namelen = dentry->d_name.len; | |
4906 | struct btrfs_dir_item *di; | |
4907 | struct btrfs_path *path; | |
4908 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4909 | int ret = 0; |
39279cc3 CM |
4910 | |
4911 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4912 | if (!path) |
4913 | return -ENOMEM; | |
3954401f | 4914 | |
33345d01 | 4915 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4916 | namelen, 0); |
0d9f7f3e Y |
4917 | if (IS_ERR(di)) |
4918 | ret = PTR_ERR(di); | |
d397712b | 4919 | |
c704005d | 4920 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4921 | goto out_err; |
d397712b | 4922 | |
5f39d397 | 4923 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4924 | out: |
39279cc3 CM |
4925 | btrfs_free_path(path); |
4926 | return ret; | |
3954401f CM |
4927 | out_err: |
4928 | location->objectid = 0; | |
4929 | goto out; | |
39279cc3 CM |
4930 | } |
4931 | ||
4932 | /* | |
4933 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4934 | * needs to be changed to reflect the root directory of the tree root. This | |
4935 | * is kind of like crossing a mount point. | |
4936 | */ | |
4937 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4938 | struct inode *dir, |
4939 | struct dentry *dentry, | |
4940 | struct btrfs_key *location, | |
4941 | struct btrfs_root **sub_root) | |
39279cc3 | 4942 | { |
4df27c4d YZ |
4943 | struct btrfs_path *path; |
4944 | struct btrfs_root *new_root; | |
4945 | struct btrfs_root_ref *ref; | |
4946 | struct extent_buffer *leaf; | |
4947 | int ret; | |
4948 | int err = 0; | |
39279cc3 | 4949 | |
4df27c4d YZ |
4950 | path = btrfs_alloc_path(); |
4951 | if (!path) { | |
4952 | err = -ENOMEM; | |
4953 | goto out; | |
4954 | } | |
39279cc3 | 4955 | |
4df27c4d | 4956 | err = -ENOENT; |
75ac2dd9 KN |
4957 | ret = btrfs_find_item(root->fs_info->tree_root, path, |
4958 | BTRFS_I(dir)->root->root_key.objectid, | |
4959 | location->objectid, BTRFS_ROOT_REF_KEY, NULL); | |
4df27c4d YZ |
4960 | if (ret) { |
4961 | if (ret < 0) | |
4962 | err = ret; | |
4963 | goto out; | |
4964 | } | |
39279cc3 | 4965 | |
4df27c4d YZ |
4966 | leaf = path->nodes[0]; |
4967 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4968 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4969 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4970 | goto out; | |
39279cc3 | 4971 | |
4df27c4d YZ |
4972 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4973 | (unsigned long)(ref + 1), | |
4974 | dentry->d_name.len); | |
4975 | if (ret) | |
4976 | goto out; | |
4977 | ||
b3b4aa74 | 4978 | btrfs_release_path(path); |
4df27c4d YZ |
4979 | |
4980 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4981 | if (IS_ERR(new_root)) { | |
4982 | err = PTR_ERR(new_root); | |
4983 | goto out; | |
4984 | } | |
4985 | ||
4df27c4d YZ |
4986 | *sub_root = new_root; |
4987 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4988 | location->type = BTRFS_INODE_ITEM_KEY; | |
4989 | location->offset = 0; | |
4990 | err = 0; | |
4991 | out: | |
4992 | btrfs_free_path(path); | |
4993 | return err; | |
39279cc3 CM |
4994 | } |
4995 | ||
5d4f98a2 YZ |
4996 | static void inode_tree_add(struct inode *inode) |
4997 | { | |
4998 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4999 | struct btrfs_inode *entry; | |
03e860bd FNP |
5000 | struct rb_node **p; |
5001 | struct rb_node *parent; | |
cef21937 | 5002 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5003 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5004 | |
1d3382cb | 5005 | if (inode_unhashed(inode)) |
76dda93c | 5006 | return; |
e1409cef | 5007 | parent = NULL; |
5d4f98a2 | 5008 | spin_lock(&root->inode_lock); |
e1409cef | 5009 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5010 | while (*p) { |
5011 | parent = *p; | |
5012 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5013 | ||
33345d01 | 5014 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5015 | p = &parent->rb_left; |
33345d01 | 5016 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5017 | p = &parent->rb_right; |
5d4f98a2 YZ |
5018 | else { |
5019 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5020 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5021 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5022 | RB_CLEAR_NODE(parent); |
5023 | spin_unlock(&root->inode_lock); | |
cef21937 | 5024 | return; |
5d4f98a2 YZ |
5025 | } |
5026 | } | |
cef21937 FDBM |
5027 | rb_link_node(new, parent, p); |
5028 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5029 | spin_unlock(&root->inode_lock); |
5030 | } | |
5031 | ||
5032 | static void inode_tree_del(struct inode *inode) | |
5033 | { | |
5034 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5035 | int empty = 0; |
5d4f98a2 | 5036 | |
03e860bd | 5037 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5038 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5039 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5040 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5041 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5042 | } |
03e860bd | 5043 | spin_unlock(&root->inode_lock); |
76dda93c | 5044 | |
69e9c6c6 | 5045 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5046 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5047 | spin_lock(&root->inode_lock); | |
5048 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5049 | spin_unlock(&root->inode_lock); | |
5050 | if (empty) | |
5051 | btrfs_add_dead_root(root); | |
5052 | } | |
5053 | } | |
5054 | ||
143bede5 | 5055 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5056 | { |
5057 | struct rb_node *node; | |
5058 | struct rb_node *prev; | |
5059 | struct btrfs_inode *entry; | |
5060 | struct inode *inode; | |
5061 | u64 objectid = 0; | |
5062 | ||
7813b3db LB |
5063 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5064 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5065 | |
5066 | spin_lock(&root->inode_lock); | |
5067 | again: | |
5068 | node = root->inode_tree.rb_node; | |
5069 | prev = NULL; | |
5070 | while (node) { | |
5071 | prev = node; | |
5072 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5073 | ||
33345d01 | 5074 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5075 | node = node->rb_left; |
33345d01 | 5076 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5077 | node = node->rb_right; |
5078 | else | |
5079 | break; | |
5080 | } | |
5081 | if (!node) { | |
5082 | while (prev) { | |
5083 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5084 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5085 | node = prev; |
5086 | break; | |
5087 | } | |
5088 | prev = rb_next(prev); | |
5089 | } | |
5090 | } | |
5091 | while (node) { | |
5092 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5093 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5094 | inode = igrab(&entry->vfs_inode); |
5095 | if (inode) { | |
5096 | spin_unlock(&root->inode_lock); | |
5097 | if (atomic_read(&inode->i_count) > 1) | |
5098 | d_prune_aliases(inode); | |
5099 | /* | |
45321ac5 | 5100 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5101 | * the inode cache when its usage count |
5102 | * hits zero. | |
5103 | */ | |
5104 | iput(inode); | |
5105 | cond_resched(); | |
5106 | spin_lock(&root->inode_lock); | |
5107 | goto again; | |
5108 | } | |
5109 | ||
5110 | if (cond_resched_lock(&root->inode_lock)) | |
5111 | goto again; | |
5112 | ||
5113 | node = rb_next(node); | |
5114 | } | |
5115 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5116 | } |
5117 | ||
e02119d5 CM |
5118 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5119 | { | |
5120 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5121 | inode->i_ino = args->location->objectid; |
5122 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5123 | sizeof(*args->location)); | |
e02119d5 | 5124 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5125 | return 0; |
5126 | } | |
5127 | ||
5128 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5129 | { | |
5130 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5131 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5132 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5133 | } |
5134 | ||
5d4f98a2 | 5135 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5136 | struct btrfs_key *location, |
5d4f98a2 | 5137 | struct btrfs_root *root) |
39279cc3 CM |
5138 | { |
5139 | struct inode *inode; | |
5140 | struct btrfs_iget_args args; | |
90d3e592 | 5141 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5142 | |
90d3e592 | 5143 | args.location = location; |
39279cc3 CM |
5144 | args.root = root; |
5145 | ||
778ba82b | 5146 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5147 | btrfs_init_locked_inode, |
5148 | (void *)&args); | |
5149 | return inode; | |
5150 | } | |
5151 | ||
1a54ef8c BR |
5152 | /* Get an inode object given its location and corresponding root. |
5153 | * Returns in *is_new if the inode was read from disk | |
5154 | */ | |
5155 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5156 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5157 | { |
5158 | struct inode *inode; | |
5159 | ||
90d3e592 | 5160 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5161 | if (!inode) |
5d4f98a2 | 5162 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5163 | |
5164 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5165 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5166 | if (!is_bad_inode(inode)) { |
5167 | inode_tree_add(inode); | |
5168 | unlock_new_inode(inode); | |
5169 | if (new) | |
5170 | *new = 1; | |
5171 | } else { | |
e0b6d65b ST |
5172 | unlock_new_inode(inode); |
5173 | iput(inode); | |
5174 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5175 | } |
5176 | } | |
5177 | ||
1a54ef8c BR |
5178 | return inode; |
5179 | } | |
5180 | ||
4df27c4d YZ |
5181 | static struct inode *new_simple_dir(struct super_block *s, |
5182 | struct btrfs_key *key, | |
5183 | struct btrfs_root *root) | |
5184 | { | |
5185 | struct inode *inode = new_inode(s); | |
5186 | ||
5187 | if (!inode) | |
5188 | return ERR_PTR(-ENOMEM); | |
5189 | ||
4df27c4d YZ |
5190 | BTRFS_I(inode)->root = root; |
5191 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5192 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5193 | |
5194 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5195 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5196 | inode->i_fop = &simple_dir_operations; |
5197 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
5198 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
5199 | ||
5200 | return inode; | |
5201 | } | |
5202 | ||
3de4586c | 5203 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5204 | { |
d397712b | 5205 | struct inode *inode; |
4df27c4d | 5206 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5207 | struct btrfs_root *sub_root = root; |
5208 | struct btrfs_key location; | |
76dda93c | 5209 | int index; |
b4aff1f8 | 5210 | int ret = 0; |
39279cc3 CM |
5211 | |
5212 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5213 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5214 | |
39e3c955 | 5215 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5216 | if (ret < 0) |
5217 | return ERR_PTR(ret); | |
5f39d397 | 5218 | |
4df27c4d | 5219 | if (location.objectid == 0) |
5662344b | 5220 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5221 | |
5222 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5223 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5224 | return inode; |
5225 | } | |
5226 | ||
5227 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5228 | ||
76dda93c | 5229 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5230 | ret = fixup_tree_root_location(root, dir, dentry, |
5231 | &location, &sub_root); | |
5232 | if (ret < 0) { | |
5233 | if (ret != -ENOENT) | |
5234 | inode = ERR_PTR(ret); | |
5235 | else | |
5236 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5237 | } else { | |
73f73415 | 5238 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5239 | } |
76dda93c YZ |
5240 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5241 | ||
34d19bad | 5242 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5243 | down_read(&root->fs_info->cleanup_work_sem); |
5244 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5245 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5246 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5247 | if (ret) { |
5248 | iput(inode); | |
66b4ffd1 | 5249 | inode = ERR_PTR(ret); |
01cd3367 | 5250 | } |
9c3b306e FM |
5251 | /* |
5252 | * If orphan cleanup did remove any orphans, it means the tree | |
5253 | * was modified and therefore the commit root is not the same as | |
5254 | * the current root anymore. This is a problem, because send | |
5255 | * uses the commit root and therefore can see inode items that | |
5256 | * don't exist in the current root anymore, and for example make | |
5257 | * calls to btrfs_iget, which will do tree lookups based on the | |
5258 | * current root and not on the commit root. Those lookups will | |
5259 | * fail, returning a -ESTALE error, and making send fail with | |
5260 | * that error. So make sure a send does not see any orphans we | |
5261 | * have just removed, and that it will see the same inodes | |
5262 | * regardless of whether a transaction commit happened before | |
5263 | * it started (meaning that the commit root will be the same as | |
5264 | * the current root) or not. | |
5265 | */ | |
5266 | if (sub_root->node != sub_root->commit_root) { | |
5267 | u64 sub_flags = btrfs_root_flags(&sub_root->root_item); | |
5268 | ||
5269 | if (sub_flags & BTRFS_ROOT_SUBVOL_RDONLY) { | |
5270 | struct extent_buffer *eb; | |
5271 | ||
5272 | /* | |
5273 | * Assert we can't have races between dentry | |
5274 | * lookup called through the snapshot creation | |
5275 | * ioctl and the VFS. | |
5276 | */ | |
5277 | ASSERT(mutex_is_locked(&dir->i_mutex)); | |
5278 | ||
5279 | down_write(&root->fs_info->commit_root_sem); | |
5280 | eb = sub_root->commit_root; | |
5281 | sub_root->commit_root = | |
5282 | btrfs_root_node(sub_root); | |
5283 | up_write(&root->fs_info->commit_root_sem); | |
5284 | free_extent_buffer(eb); | |
5285 | } | |
5286 | } | |
c71bf099 YZ |
5287 | } |
5288 | ||
3de4586c CM |
5289 | return inode; |
5290 | } | |
5291 | ||
fe15ce44 | 5292 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5293 | { |
5294 | struct btrfs_root *root; | |
848cce0d | 5295 | struct inode *inode = dentry->d_inode; |
76dda93c | 5296 | |
848cce0d LZ |
5297 | if (!inode && !IS_ROOT(dentry)) |
5298 | inode = dentry->d_parent->d_inode; | |
76dda93c | 5299 | |
848cce0d LZ |
5300 | if (inode) { |
5301 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5302 | if (btrfs_root_refs(&root->root_item) == 0) |
5303 | return 1; | |
848cce0d LZ |
5304 | |
5305 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5306 | return 1; | |
efefb143 | 5307 | } |
76dda93c YZ |
5308 | return 0; |
5309 | } | |
5310 | ||
b4aff1f8 JB |
5311 | static void btrfs_dentry_release(struct dentry *dentry) |
5312 | { | |
944a4515 | 5313 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5314 | } |
5315 | ||
3de4586c | 5316 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5317 | unsigned int flags) |
3de4586c | 5318 | { |
5662344b | 5319 | struct inode *inode; |
a66e7cc6 | 5320 | |
5662344b TI |
5321 | inode = btrfs_lookup_dentry(dir, dentry); |
5322 | if (IS_ERR(inode)) { | |
5323 | if (PTR_ERR(inode) == -ENOENT) | |
5324 | inode = NULL; | |
5325 | else | |
5326 | return ERR_CAST(inode); | |
5327 | } | |
5328 | ||
3a0dfa6a | 5329 | return d_materialise_unique(dentry, inode); |
39279cc3 CM |
5330 | } |
5331 | ||
16cdcec7 | 5332 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5333 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5334 | }; | |
5335 | ||
9cdda8d3 | 5336 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5337 | { |
9cdda8d3 | 5338 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5339 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5340 | struct btrfs_item *item; | |
5341 | struct btrfs_dir_item *di; | |
5342 | struct btrfs_key key; | |
5f39d397 | 5343 | struct btrfs_key found_key; |
39279cc3 | 5344 | struct btrfs_path *path; |
16cdcec7 MX |
5345 | struct list_head ins_list; |
5346 | struct list_head del_list; | |
39279cc3 | 5347 | int ret; |
5f39d397 | 5348 | struct extent_buffer *leaf; |
39279cc3 | 5349 | int slot; |
39279cc3 CM |
5350 | unsigned char d_type; |
5351 | int over = 0; | |
5352 | u32 di_cur; | |
5353 | u32 di_total; | |
5354 | u32 di_len; | |
5355 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5356 | char tmp_name[32]; |
5357 | char *name_ptr; | |
5358 | int name_len; | |
9cdda8d3 | 5359 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5360 | |
5361 | /* FIXME, use a real flag for deciding about the key type */ | |
5362 | if (root->fs_info->tree_root == root) | |
5363 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5364 | |
9cdda8d3 AV |
5365 | if (!dir_emit_dots(file, ctx)) |
5366 | return 0; | |
5367 | ||
49593bfa | 5368 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5369 | if (!path) |
5370 | return -ENOMEM; | |
ff5714cc | 5371 | |
026fd317 | 5372 | path->reada = 1; |
49593bfa | 5373 | |
16cdcec7 MX |
5374 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5375 | INIT_LIST_HEAD(&ins_list); | |
5376 | INIT_LIST_HEAD(&del_list); | |
5377 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5378 | } | |
5379 | ||
962a298f | 5380 | key.type = key_type; |
9cdda8d3 | 5381 | key.offset = ctx->pos; |
33345d01 | 5382 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5383 | |
39279cc3 CM |
5384 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5385 | if (ret < 0) | |
5386 | goto err; | |
49593bfa DW |
5387 | |
5388 | while (1) { | |
5f39d397 | 5389 | leaf = path->nodes[0]; |
39279cc3 | 5390 | slot = path->slots[0]; |
b9e03af0 LZ |
5391 | if (slot >= btrfs_header_nritems(leaf)) { |
5392 | ret = btrfs_next_leaf(root, path); | |
5393 | if (ret < 0) | |
5394 | goto err; | |
5395 | else if (ret > 0) | |
5396 | break; | |
5397 | continue; | |
39279cc3 | 5398 | } |
3de4586c | 5399 | |
dd3cc16b | 5400 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5401 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5402 | ||
5403 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5404 | break; |
962a298f | 5405 | if (found_key.type != key_type) |
39279cc3 | 5406 | break; |
9cdda8d3 | 5407 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5408 | goto next; |
16cdcec7 MX |
5409 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5410 | btrfs_should_delete_dir_index(&del_list, | |
5411 | found_key.offset)) | |
5412 | goto next; | |
5f39d397 | 5413 | |
9cdda8d3 | 5414 | ctx->pos = found_key.offset; |
16cdcec7 | 5415 | is_curr = 1; |
49593bfa | 5416 | |
39279cc3 CM |
5417 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5418 | di_cur = 0; | |
5f39d397 | 5419 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5420 | |
5421 | while (di_cur < di_total) { | |
5f39d397 CM |
5422 | struct btrfs_key location; |
5423 | ||
22a94d44 JB |
5424 | if (verify_dir_item(root, leaf, di)) |
5425 | break; | |
5426 | ||
5f39d397 | 5427 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5428 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5429 | name_ptr = tmp_name; |
5430 | } else { | |
5431 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5432 | if (!name_ptr) { |
5433 | ret = -ENOMEM; | |
5434 | goto err; | |
5435 | } | |
5f39d397 CM |
5436 | } |
5437 | read_extent_buffer(leaf, name_ptr, | |
5438 | (unsigned long)(di + 1), name_len); | |
5439 | ||
5440 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5441 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5442 | |
fede766f | 5443 | |
3de4586c | 5444 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5445 | * skip it. |
5446 | * | |
5447 | * In contrast to old kernels, we insert the snapshot's | |
5448 | * dir item and dir index after it has been created, so | |
5449 | * we won't find a reference to our own snapshot. We | |
5450 | * still keep the following code for backward | |
5451 | * compatibility. | |
3de4586c CM |
5452 | */ |
5453 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5454 | location.objectid == root->root_key.objectid) { | |
5455 | over = 0; | |
5456 | goto skip; | |
5457 | } | |
9cdda8d3 AV |
5458 | over = !dir_emit(ctx, name_ptr, name_len, |
5459 | location.objectid, d_type); | |
5f39d397 | 5460 | |
3de4586c | 5461 | skip: |
5f39d397 CM |
5462 | if (name_ptr != tmp_name) |
5463 | kfree(name_ptr); | |
5464 | ||
39279cc3 CM |
5465 | if (over) |
5466 | goto nopos; | |
5103e947 | 5467 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5468 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5469 | di_cur += di_len; |
5470 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5471 | } | |
b9e03af0 LZ |
5472 | next: |
5473 | path->slots[0]++; | |
39279cc3 | 5474 | } |
49593bfa | 5475 | |
16cdcec7 MX |
5476 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5477 | if (is_curr) | |
9cdda8d3 AV |
5478 | ctx->pos++; |
5479 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5480 | if (ret) |
5481 | goto nopos; | |
5482 | } | |
5483 | ||
49593bfa | 5484 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5485 | ctx->pos++; |
5486 | ||
5487 | /* | |
5488 | * Stop new entries from being returned after we return the last | |
5489 | * entry. | |
5490 | * | |
5491 | * New directory entries are assigned a strictly increasing | |
5492 | * offset. This means that new entries created during readdir | |
5493 | * are *guaranteed* to be seen in the future by that readdir. | |
5494 | * This has broken buggy programs which operate on names as | |
5495 | * they're returned by readdir. Until we re-use freed offsets | |
5496 | * we have this hack to stop new entries from being returned | |
5497 | * under the assumption that they'll never reach this huge | |
5498 | * offset. | |
5499 | * | |
5500 | * This is being careful not to overflow 32bit loff_t unless the | |
5501 | * last entry requires it because doing so has broken 32bit apps | |
5502 | * in the past. | |
5503 | */ | |
5504 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5505 | if (ctx->pos >= INT_MAX) | |
5506 | ctx->pos = LLONG_MAX; | |
5507 | else | |
5508 | ctx->pos = INT_MAX; | |
5509 | } | |
39279cc3 CM |
5510 | nopos: |
5511 | ret = 0; | |
5512 | err: | |
16cdcec7 MX |
5513 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5514 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5515 | btrfs_free_path(path); |
39279cc3 CM |
5516 | return ret; |
5517 | } | |
5518 | ||
a9185b41 | 5519 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5520 | { |
5521 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5522 | struct btrfs_trans_handle *trans; | |
5523 | int ret = 0; | |
0af3d00b | 5524 | bool nolock = false; |
39279cc3 | 5525 | |
72ac3c0d | 5526 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5527 | return 0; |
5528 | ||
83eea1f1 | 5529 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5530 | nolock = true; |
0af3d00b | 5531 | |
a9185b41 | 5532 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5533 | if (nolock) |
7a7eaa40 | 5534 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5535 | else |
7a7eaa40 | 5536 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5537 | if (IS_ERR(trans)) |
5538 | return PTR_ERR(trans); | |
a698d075 | 5539 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5540 | } |
5541 | return ret; | |
5542 | } | |
5543 | ||
5544 | /* | |
54aa1f4d | 5545 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5546 | * inode changes. But, it is most likely to find the inode in cache. |
5547 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5548 | * to keep or drop this code. | |
5549 | */ | |
48a3b636 | 5550 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5551 | { |
5552 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5553 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5554 | int ret; |
5555 | ||
72ac3c0d | 5556 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5557 | return 0; |
39279cc3 | 5558 | |
7a7eaa40 | 5559 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5560 | if (IS_ERR(trans)) |
5561 | return PTR_ERR(trans); | |
8929ecfa YZ |
5562 | |
5563 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5564 | if (ret && ret == -ENOSPC) { |
5565 | /* whoops, lets try again with the full transaction */ | |
5566 | btrfs_end_transaction(trans, root); | |
5567 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5568 | if (IS_ERR(trans)) |
5569 | return PTR_ERR(trans); | |
8929ecfa | 5570 | |
94b60442 | 5571 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5572 | } |
39279cc3 | 5573 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5574 | if (BTRFS_I(inode)->delayed_node) |
5575 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5576 | |
5577 | return ret; | |
5578 | } | |
5579 | ||
5580 | /* | |
5581 | * This is a copy of file_update_time. We need this so we can return error on | |
5582 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5583 | */ | |
e41f941a JB |
5584 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5585 | int flags) | |
22c44fe6 | 5586 | { |
2bc55652 AB |
5587 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5588 | ||
5589 | if (btrfs_root_readonly(root)) | |
5590 | return -EROFS; | |
5591 | ||
e41f941a | 5592 | if (flags & S_VERSION) |
22c44fe6 | 5593 | inode_inc_iversion(inode); |
e41f941a JB |
5594 | if (flags & S_CTIME) |
5595 | inode->i_ctime = *now; | |
5596 | if (flags & S_MTIME) | |
5597 | inode->i_mtime = *now; | |
5598 | if (flags & S_ATIME) | |
5599 | inode->i_atime = *now; | |
5600 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5601 | } |
5602 | ||
d352ac68 CM |
5603 | /* |
5604 | * find the highest existing sequence number in a directory | |
5605 | * and then set the in-memory index_cnt variable to reflect | |
5606 | * free sequence numbers | |
5607 | */ | |
aec7477b JB |
5608 | static int btrfs_set_inode_index_count(struct inode *inode) |
5609 | { | |
5610 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5611 | struct btrfs_key key, found_key; | |
5612 | struct btrfs_path *path; | |
5613 | struct extent_buffer *leaf; | |
5614 | int ret; | |
5615 | ||
33345d01 | 5616 | key.objectid = btrfs_ino(inode); |
962a298f | 5617 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5618 | key.offset = (u64)-1; |
5619 | ||
5620 | path = btrfs_alloc_path(); | |
5621 | if (!path) | |
5622 | return -ENOMEM; | |
5623 | ||
5624 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5625 | if (ret < 0) | |
5626 | goto out; | |
5627 | /* FIXME: we should be able to handle this */ | |
5628 | if (ret == 0) | |
5629 | goto out; | |
5630 | ret = 0; | |
5631 | ||
5632 | /* | |
5633 | * MAGIC NUMBER EXPLANATION: | |
5634 | * since we search a directory based on f_pos we have to start at 2 | |
5635 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5636 | * else has to start at 2 | |
5637 | */ | |
5638 | if (path->slots[0] == 0) { | |
5639 | BTRFS_I(inode)->index_cnt = 2; | |
5640 | goto out; | |
5641 | } | |
5642 | ||
5643 | path->slots[0]--; | |
5644 | ||
5645 | leaf = path->nodes[0]; | |
5646 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5647 | ||
33345d01 | 5648 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 5649 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
5650 | BTRFS_I(inode)->index_cnt = 2; |
5651 | goto out; | |
5652 | } | |
5653 | ||
5654 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5655 | out: | |
5656 | btrfs_free_path(path); | |
5657 | return ret; | |
5658 | } | |
5659 | ||
d352ac68 CM |
5660 | /* |
5661 | * helper to find a free sequence number in a given directory. This current | |
5662 | * code is very simple, later versions will do smarter things in the btree | |
5663 | */ | |
3de4586c | 5664 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5665 | { |
5666 | int ret = 0; | |
5667 | ||
5668 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5669 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5670 | if (ret) { | |
5671 | ret = btrfs_set_inode_index_count(dir); | |
5672 | if (ret) | |
5673 | return ret; | |
5674 | } | |
aec7477b JB |
5675 | } |
5676 | ||
00e4e6b3 | 5677 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5678 | BTRFS_I(dir)->index_cnt++; |
5679 | ||
5680 | return ret; | |
5681 | } | |
5682 | ||
b0d5d10f CM |
5683 | static int btrfs_insert_inode_locked(struct inode *inode) |
5684 | { | |
5685 | struct btrfs_iget_args args; | |
5686 | args.location = &BTRFS_I(inode)->location; | |
5687 | args.root = BTRFS_I(inode)->root; | |
5688 | ||
5689 | return insert_inode_locked4(inode, | |
5690 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
5691 | btrfs_find_actor, &args); | |
5692 | } | |
5693 | ||
39279cc3 CM |
5694 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5695 | struct btrfs_root *root, | |
aec7477b | 5696 | struct inode *dir, |
9c58309d | 5697 | const char *name, int name_len, |
175a4eb7 AV |
5698 | u64 ref_objectid, u64 objectid, |
5699 | umode_t mode, u64 *index) | |
39279cc3 CM |
5700 | { |
5701 | struct inode *inode; | |
5f39d397 | 5702 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5703 | struct btrfs_key *location; |
5f39d397 | 5704 | struct btrfs_path *path; |
9c58309d CM |
5705 | struct btrfs_inode_ref *ref; |
5706 | struct btrfs_key key[2]; | |
5707 | u32 sizes[2]; | |
ef3b9af5 | 5708 | int nitems = name ? 2 : 1; |
9c58309d | 5709 | unsigned long ptr; |
39279cc3 | 5710 | int ret; |
39279cc3 | 5711 | |
5f39d397 | 5712 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5713 | if (!path) |
5714 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5715 | |
39279cc3 | 5716 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5717 | if (!inode) { |
5718 | btrfs_free_path(path); | |
39279cc3 | 5719 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5720 | } |
39279cc3 | 5721 | |
5762b5c9 FM |
5722 | /* |
5723 | * O_TMPFILE, set link count to 0, so that after this point, | |
5724 | * we fill in an inode item with the correct link count. | |
5725 | */ | |
5726 | if (!name) | |
5727 | set_nlink(inode, 0); | |
5728 | ||
581bb050 LZ |
5729 | /* |
5730 | * we have to initialize this early, so we can reclaim the inode | |
5731 | * number if we fail afterwards in this function. | |
5732 | */ | |
5733 | inode->i_ino = objectid; | |
5734 | ||
ef3b9af5 | 5735 | if (dir && name) { |
1abe9b8a | 5736 | trace_btrfs_inode_request(dir); |
5737 | ||
3de4586c | 5738 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5739 | if (ret) { |
8fb27640 | 5740 | btrfs_free_path(path); |
09771430 | 5741 | iput(inode); |
aec7477b | 5742 | return ERR_PTR(ret); |
09771430 | 5743 | } |
ef3b9af5 FM |
5744 | } else if (dir) { |
5745 | *index = 0; | |
aec7477b JB |
5746 | } |
5747 | /* | |
5748 | * index_cnt is ignored for everything but a dir, | |
5749 | * btrfs_get_inode_index_count has an explanation for the magic | |
5750 | * number | |
5751 | */ | |
5752 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 5753 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 5754 | BTRFS_I(inode)->root = root; |
e02119d5 | 5755 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5756 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5757 | |
5dc562c5 JB |
5758 | /* |
5759 | * We could have gotten an inode number from somebody who was fsynced | |
5760 | * and then removed in this same transaction, so let's just set full | |
5761 | * sync since it will be a full sync anyway and this will blow away the | |
5762 | * old info in the log. | |
5763 | */ | |
5764 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5765 | ||
9c58309d | 5766 | key[0].objectid = objectid; |
962a298f | 5767 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
5768 | key[0].offset = 0; |
5769 | ||
9c58309d | 5770 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
5771 | |
5772 | if (name) { | |
5773 | /* | |
5774 | * Start new inodes with an inode_ref. This is slightly more | |
5775 | * efficient for small numbers of hard links since they will | |
5776 | * be packed into one item. Extended refs will kick in if we | |
5777 | * add more hard links than can fit in the ref item. | |
5778 | */ | |
5779 | key[1].objectid = objectid; | |
962a298f | 5780 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
5781 | key[1].offset = ref_objectid; |
5782 | ||
5783 | sizes[1] = name_len + sizeof(*ref); | |
5784 | } | |
9c58309d | 5785 | |
b0d5d10f CM |
5786 | location = &BTRFS_I(inode)->location; |
5787 | location->objectid = objectid; | |
5788 | location->offset = 0; | |
962a298f | 5789 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
5790 | |
5791 | ret = btrfs_insert_inode_locked(inode); | |
5792 | if (ret < 0) | |
5793 | goto fail; | |
5794 | ||
b9473439 | 5795 | path->leave_spinning = 1; |
ef3b9af5 | 5796 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 5797 | if (ret != 0) |
b0d5d10f | 5798 | goto fail_unlock; |
5f39d397 | 5799 | |
ecc11fab | 5800 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5801 | inode_set_bytes(inode, 0); |
39279cc3 | 5802 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5803 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5804 | struct btrfs_inode_item); | |
293f7e07 LZ |
5805 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5806 | sizeof(*inode_item)); | |
e02119d5 | 5807 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 5808 | |
ef3b9af5 FM |
5809 | if (name) { |
5810 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5811 | struct btrfs_inode_ref); | |
5812 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
5813 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
5814 | ptr = (unsigned long)(ref + 1); | |
5815 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5816 | } | |
9c58309d | 5817 | |
5f39d397 CM |
5818 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5819 | btrfs_free_path(path); | |
5820 | ||
6cbff00f CH |
5821 | btrfs_inherit_iflags(inode, dir); |
5822 | ||
569254b0 | 5823 | if (S_ISREG(mode)) { |
94272164 CM |
5824 | if (btrfs_test_opt(root, NODATASUM)) |
5825 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5826 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5827 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5828 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5829 | } |
5830 | ||
5d4f98a2 | 5831 | inode_tree_add(inode); |
1abe9b8a | 5832 | |
5833 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5834 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5835 | |
8ea05e3a AB |
5836 | btrfs_update_root_times(trans, root); |
5837 | ||
63541927 FDBM |
5838 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
5839 | if (ret) | |
5840 | btrfs_err(root->fs_info, | |
5841 | "error inheriting props for ino %llu (root %llu): %d", | |
5842 | btrfs_ino(inode), root->root_key.objectid, ret); | |
5843 | ||
39279cc3 | 5844 | return inode; |
b0d5d10f CM |
5845 | |
5846 | fail_unlock: | |
5847 | unlock_new_inode(inode); | |
5f39d397 | 5848 | fail: |
ef3b9af5 | 5849 | if (dir && name) |
aec7477b | 5850 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 5851 | btrfs_free_path(path); |
09771430 | 5852 | iput(inode); |
5f39d397 | 5853 | return ERR_PTR(ret); |
39279cc3 CM |
5854 | } |
5855 | ||
5856 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5857 | { | |
5858 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5859 | } | |
5860 | ||
d352ac68 CM |
5861 | /* |
5862 | * utility function to add 'inode' into 'parent_inode' with | |
5863 | * a give name and a given sequence number. | |
5864 | * if 'add_backref' is true, also insert a backref from the | |
5865 | * inode to the parent directory. | |
5866 | */ | |
e02119d5 CM |
5867 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5868 | struct inode *parent_inode, struct inode *inode, | |
5869 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5870 | { |
4df27c4d | 5871 | int ret = 0; |
39279cc3 | 5872 | struct btrfs_key key; |
e02119d5 | 5873 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5874 | u64 ino = btrfs_ino(inode); |
5875 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5876 | |
33345d01 | 5877 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5878 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5879 | } else { | |
33345d01 | 5880 | key.objectid = ino; |
962a298f | 5881 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
5882 | key.offset = 0; |
5883 | } | |
5884 | ||
33345d01 | 5885 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5886 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5887 | key.objectid, root->root_key.objectid, | |
33345d01 | 5888 | parent_ino, index, name, name_len); |
4df27c4d | 5889 | } else if (add_backref) { |
33345d01 LZ |
5890 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5891 | parent_ino, index); | |
4df27c4d | 5892 | } |
39279cc3 | 5893 | |
79787eaa JM |
5894 | /* Nothing to clean up yet */ |
5895 | if (ret) | |
5896 | return ret; | |
4df27c4d | 5897 | |
79787eaa JM |
5898 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5899 | parent_inode, &key, | |
5900 | btrfs_inode_type(inode), index); | |
9c52057c | 5901 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5902 | goto fail_dir_item; |
5903 | else if (ret) { | |
5904 | btrfs_abort_transaction(trans, root, ret); | |
5905 | return ret; | |
39279cc3 | 5906 | } |
79787eaa JM |
5907 | |
5908 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5909 | name_len * 2); | |
0c4d2d95 | 5910 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5911 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5912 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5913 | if (ret) | |
5914 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5915 | return ret; |
fe66a05a CM |
5916 | |
5917 | fail_dir_item: | |
5918 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5919 | u64 local_index; | |
5920 | int err; | |
5921 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5922 | key.objectid, root->root_key.objectid, | |
5923 | parent_ino, &local_index, name, name_len); | |
5924 | ||
5925 | } else if (add_backref) { | |
5926 | u64 local_index; | |
5927 | int err; | |
5928 | ||
5929 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5930 | ino, parent_ino, &local_index); | |
5931 | } | |
5932 | return ret; | |
39279cc3 CM |
5933 | } |
5934 | ||
5935 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5936 | struct inode *dir, struct dentry *dentry, |
5937 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5938 | { |
a1b075d2 JB |
5939 | int err = btrfs_add_link(trans, dir, inode, |
5940 | dentry->d_name.name, dentry->d_name.len, | |
5941 | backref, index); | |
39279cc3 CM |
5942 | if (err > 0) |
5943 | err = -EEXIST; | |
5944 | return err; | |
5945 | } | |
5946 | ||
618e21d5 | 5947 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5948 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5949 | { |
5950 | struct btrfs_trans_handle *trans; | |
5951 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5952 | struct inode *inode = NULL; |
618e21d5 JB |
5953 | int err; |
5954 | int drop_inode = 0; | |
5955 | u64 objectid; | |
00e4e6b3 | 5956 | u64 index = 0; |
618e21d5 JB |
5957 | |
5958 | if (!new_valid_dev(rdev)) | |
5959 | return -EINVAL; | |
5960 | ||
9ed74f2d JB |
5961 | /* |
5962 | * 2 for inode item and ref | |
5963 | * 2 for dir items | |
5964 | * 1 for xattr if selinux is on | |
5965 | */ | |
a22285a6 YZ |
5966 | trans = btrfs_start_transaction(root, 5); |
5967 | if (IS_ERR(trans)) | |
5968 | return PTR_ERR(trans); | |
1832a6d5 | 5969 | |
581bb050 LZ |
5970 | err = btrfs_find_free_ino(root, &objectid); |
5971 | if (err) | |
5972 | goto out_unlock; | |
5973 | ||
aec7477b | 5974 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5975 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5976 | mode, &index); |
7cf96da3 TI |
5977 | if (IS_ERR(inode)) { |
5978 | err = PTR_ERR(inode); | |
618e21d5 | 5979 | goto out_unlock; |
7cf96da3 | 5980 | } |
618e21d5 | 5981 | |
ad19db71 CS |
5982 | /* |
5983 | * If the active LSM wants to access the inode during | |
5984 | * d_instantiate it needs these. Smack checks to see | |
5985 | * if the filesystem supports xattrs by looking at the | |
5986 | * ops vector. | |
5987 | */ | |
ad19db71 | 5988 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
5989 | init_special_inode(inode, inode->i_mode, rdev); |
5990 | ||
5991 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 5992 | if (err) |
b0d5d10f CM |
5993 | goto out_unlock_inode; |
5994 | ||
5995 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
5996 | if (err) { | |
5997 | goto out_unlock_inode; | |
5998 | } else { | |
1b4ab1bb | 5999 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6000 | unlock_new_inode(inode); |
08c422c2 | 6001 | d_instantiate(dentry, inode); |
618e21d5 | 6002 | } |
b0d5d10f | 6003 | |
618e21d5 | 6004 | out_unlock: |
7ad85bb7 | 6005 | btrfs_end_transaction(trans, root); |
c581afc8 | 6006 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6007 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6008 | if (drop_inode) { |
6009 | inode_dec_link_count(inode); | |
6010 | iput(inode); | |
6011 | } | |
618e21d5 | 6012 | return err; |
b0d5d10f CM |
6013 | |
6014 | out_unlock_inode: | |
6015 | drop_inode = 1; | |
6016 | unlock_new_inode(inode); | |
6017 | goto out_unlock; | |
6018 | ||
618e21d5 JB |
6019 | } |
6020 | ||
39279cc3 | 6021 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6022 | umode_t mode, bool excl) |
39279cc3 CM |
6023 | { |
6024 | struct btrfs_trans_handle *trans; | |
6025 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6026 | struct inode *inode = NULL; |
43baa579 | 6027 | int drop_inode_on_err = 0; |
a22285a6 | 6028 | int err; |
39279cc3 | 6029 | u64 objectid; |
00e4e6b3 | 6030 | u64 index = 0; |
39279cc3 | 6031 | |
9ed74f2d JB |
6032 | /* |
6033 | * 2 for inode item and ref | |
6034 | * 2 for dir items | |
6035 | * 1 for xattr if selinux is on | |
6036 | */ | |
a22285a6 YZ |
6037 | trans = btrfs_start_transaction(root, 5); |
6038 | if (IS_ERR(trans)) | |
6039 | return PTR_ERR(trans); | |
9ed74f2d | 6040 | |
581bb050 LZ |
6041 | err = btrfs_find_free_ino(root, &objectid); |
6042 | if (err) | |
6043 | goto out_unlock; | |
6044 | ||
aec7477b | 6045 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6046 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6047 | mode, &index); |
7cf96da3 TI |
6048 | if (IS_ERR(inode)) { |
6049 | err = PTR_ERR(inode); | |
39279cc3 | 6050 | goto out_unlock; |
7cf96da3 | 6051 | } |
43baa579 | 6052 | drop_inode_on_err = 1; |
ad19db71 CS |
6053 | /* |
6054 | * If the active LSM wants to access the inode during | |
6055 | * d_instantiate it needs these. Smack checks to see | |
6056 | * if the filesystem supports xattrs by looking at the | |
6057 | * ops vector. | |
6058 | */ | |
6059 | inode->i_fop = &btrfs_file_operations; | |
6060 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
6061 | inode->i_mapping->a_ops = &btrfs_aops; |
6062 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
6063 | ||
6064 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6065 | if (err) | |
6066 | goto out_unlock_inode; | |
6067 | ||
6068 | err = btrfs_update_inode(trans, root, inode); | |
6069 | if (err) | |
6070 | goto out_unlock_inode; | |
ad19db71 | 6071 | |
a1b075d2 | 6072 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6073 | if (err) |
b0d5d10f | 6074 | goto out_unlock_inode; |
43baa579 | 6075 | |
43baa579 | 6076 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6077 | unlock_new_inode(inode); |
43baa579 FB |
6078 | d_instantiate(dentry, inode); |
6079 | ||
39279cc3 | 6080 | out_unlock: |
7ad85bb7 | 6081 | btrfs_end_transaction(trans, root); |
43baa579 | 6082 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6083 | inode_dec_link_count(inode); |
6084 | iput(inode); | |
6085 | } | |
c581afc8 | 6086 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6087 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6088 | return err; |
b0d5d10f CM |
6089 | |
6090 | out_unlock_inode: | |
6091 | unlock_new_inode(inode); | |
6092 | goto out_unlock; | |
6093 | ||
39279cc3 CM |
6094 | } |
6095 | ||
6096 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6097 | struct dentry *dentry) | |
6098 | { | |
6099 | struct btrfs_trans_handle *trans; | |
6100 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6101 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 6102 | u64 index; |
39279cc3 CM |
6103 | int err; |
6104 | int drop_inode = 0; | |
6105 | ||
4a8be425 TH |
6106 | /* do not allow sys_link's with other subvols of the same device */ |
6107 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6108 | return -EXDEV; |
4a8be425 | 6109 | |
f186373f | 6110 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6111 | return -EMLINK; |
4a8be425 | 6112 | |
3de4586c | 6113 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6114 | if (err) |
6115 | goto fail; | |
6116 | ||
a22285a6 | 6117 | /* |
7e6b6465 | 6118 | * 2 items for inode and inode ref |
a22285a6 | 6119 | * 2 items for dir items |
7e6b6465 | 6120 | * 1 item for parent inode |
a22285a6 | 6121 | */ |
7e6b6465 | 6122 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6123 | if (IS_ERR(trans)) { |
6124 | err = PTR_ERR(trans); | |
6125 | goto fail; | |
6126 | } | |
5f39d397 | 6127 | |
67de1176 MX |
6128 | /* There are several dir indexes for this inode, clear the cache. */ |
6129 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6130 | inc_nlink(inode); |
0c4d2d95 | 6131 | inode_inc_iversion(inode); |
3153495d | 6132 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 6133 | ihold(inode); |
e9976151 | 6134 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6135 | |
a1b075d2 | 6136 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6137 | |
a5719521 | 6138 | if (err) { |
54aa1f4d | 6139 | drop_inode = 1; |
a5719521 | 6140 | } else { |
10d9f309 | 6141 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6142 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6143 | if (err) |
6144 | goto fail; | |
ef3b9af5 FM |
6145 | if (inode->i_nlink == 1) { |
6146 | /* | |
6147 | * If new hard link count is 1, it's a file created | |
6148 | * with open(2) O_TMPFILE flag. | |
6149 | */ | |
6150 | err = btrfs_orphan_del(trans, inode); | |
6151 | if (err) | |
6152 | goto fail; | |
6153 | } | |
08c422c2 | 6154 | d_instantiate(dentry, inode); |
6a912213 | 6155 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6156 | } |
39279cc3 | 6157 | |
7ad85bb7 | 6158 | btrfs_end_transaction(trans, root); |
c581afc8 | 6159 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6160 | fail: |
39279cc3 CM |
6161 | if (drop_inode) { |
6162 | inode_dec_link_count(inode); | |
6163 | iput(inode); | |
6164 | } | |
b53d3f5d | 6165 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6166 | return err; |
6167 | } | |
6168 | ||
18bb1db3 | 6169 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6170 | { |
b9d86667 | 6171 | struct inode *inode = NULL; |
39279cc3 CM |
6172 | struct btrfs_trans_handle *trans; |
6173 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6174 | int err = 0; | |
6175 | int drop_on_err = 0; | |
b9d86667 | 6176 | u64 objectid = 0; |
00e4e6b3 | 6177 | u64 index = 0; |
39279cc3 | 6178 | |
9ed74f2d JB |
6179 | /* |
6180 | * 2 items for inode and ref | |
6181 | * 2 items for dir items | |
6182 | * 1 for xattr if selinux is on | |
6183 | */ | |
a22285a6 YZ |
6184 | trans = btrfs_start_transaction(root, 5); |
6185 | if (IS_ERR(trans)) | |
6186 | return PTR_ERR(trans); | |
39279cc3 | 6187 | |
581bb050 LZ |
6188 | err = btrfs_find_free_ino(root, &objectid); |
6189 | if (err) | |
6190 | goto out_fail; | |
6191 | ||
aec7477b | 6192 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6193 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6194 | S_IFDIR | mode, &index); |
39279cc3 CM |
6195 | if (IS_ERR(inode)) { |
6196 | err = PTR_ERR(inode); | |
6197 | goto out_fail; | |
6198 | } | |
5f39d397 | 6199 | |
39279cc3 | 6200 | drop_on_err = 1; |
b0d5d10f CM |
6201 | /* these must be set before we unlock the inode */ |
6202 | inode->i_op = &btrfs_dir_inode_operations; | |
6203 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6204 | |
2a7dba39 | 6205 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6206 | if (err) |
b0d5d10f | 6207 | goto out_fail_inode; |
39279cc3 | 6208 | |
dbe674a9 | 6209 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6210 | err = btrfs_update_inode(trans, root, inode); |
6211 | if (err) | |
b0d5d10f | 6212 | goto out_fail_inode; |
5f39d397 | 6213 | |
a1b075d2 JB |
6214 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6215 | dentry->d_name.len, 0, index); | |
39279cc3 | 6216 | if (err) |
b0d5d10f | 6217 | goto out_fail_inode; |
5f39d397 | 6218 | |
39279cc3 | 6219 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6220 | /* |
6221 | * mkdir is special. We're unlocking after we call d_instantiate | |
6222 | * to avoid a race with nfsd calling d_instantiate. | |
6223 | */ | |
6224 | unlock_new_inode(inode); | |
39279cc3 | 6225 | drop_on_err = 0; |
39279cc3 CM |
6226 | |
6227 | out_fail: | |
7ad85bb7 | 6228 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
6229 | if (drop_on_err) |
6230 | iput(inode); | |
c581afc8 | 6231 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6232 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6233 | return err; |
b0d5d10f CM |
6234 | |
6235 | out_fail_inode: | |
6236 | unlock_new_inode(inode); | |
6237 | goto out_fail; | |
39279cc3 CM |
6238 | } |
6239 | ||
d352ac68 CM |
6240 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
6241 | * and an extent that you want to insert, deal with overlap and insert | |
6242 | * the new extent into the tree. | |
6243 | */ | |
3b951516 CM |
6244 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6245 | struct extent_map *existing, | |
e6dcd2dc | 6246 | struct extent_map *em, |
51f395ad | 6247 | u64 map_start) |
3b951516 CM |
6248 | { |
6249 | u64 start_diff; | |
3b951516 | 6250 | |
e6dcd2dc CM |
6251 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
6252 | start_diff = map_start - em->start; | |
6253 | em->start = map_start; | |
51f395ad | 6254 | em->len = existing->start - em->start; |
c8b97818 CM |
6255 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6256 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6257 | em->block_start += start_diff; |
c8b97818 CM |
6258 | em->block_len -= start_diff; |
6259 | } | |
09a2a8f9 | 6260 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6261 | } |
6262 | ||
c8b97818 CM |
6263 | static noinline int uncompress_inline(struct btrfs_path *path, |
6264 | struct inode *inode, struct page *page, | |
6265 | size_t pg_offset, u64 extent_offset, | |
6266 | struct btrfs_file_extent_item *item) | |
6267 | { | |
6268 | int ret; | |
6269 | struct extent_buffer *leaf = path->nodes[0]; | |
6270 | char *tmp; | |
6271 | size_t max_size; | |
6272 | unsigned long inline_size; | |
6273 | unsigned long ptr; | |
261507a0 | 6274 | int compress_type; |
c8b97818 CM |
6275 | |
6276 | WARN_ON(pg_offset != 0); | |
261507a0 | 6277 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6278 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6279 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6280 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6281 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6282 | if (!tmp) |
6283 | return -ENOMEM; | |
c8b97818 CM |
6284 | ptr = btrfs_file_extent_inline_start(item); |
6285 | ||
6286 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6287 | ||
5b050f04 | 6288 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
6289 | ret = btrfs_decompress(compress_type, tmp, page, |
6290 | extent_offset, inline_size, max_size); | |
c8b97818 | 6291 | kfree(tmp); |
166ae5a4 | 6292 | return ret; |
c8b97818 CM |
6293 | } |
6294 | ||
d352ac68 CM |
6295 | /* |
6296 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6297 | * the ugly parts come from merging extents from the disk with the in-ram |
6298 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6299 | * where the in-ram extents might be locked pending data=ordered completion. |
6300 | * | |
6301 | * This also copies inline extents directly into the page. | |
6302 | */ | |
d397712b | 6303 | |
a52d9a80 | 6304 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6305 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6306 | int create) |
6307 | { | |
6308 | int ret; | |
6309 | int err = 0; | |
a52d9a80 CM |
6310 | u64 extent_start = 0; |
6311 | u64 extent_end = 0; | |
33345d01 | 6312 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6313 | u32 found_type; |
f421950f | 6314 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6315 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6316 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6317 | struct extent_buffer *leaf; |
6318 | struct btrfs_key found_key; | |
a52d9a80 CM |
6319 | struct extent_map *em = NULL; |
6320 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6321 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6322 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6323 | const bool new_inline = !page || create; |
a52d9a80 | 6324 | |
a52d9a80 | 6325 | again: |
890871be | 6326 | read_lock(&em_tree->lock); |
d1310b2e | 6327 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6328 | if (em) |
6329 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6330 | read_unlock(&em_tree->lock); |
d1310b2e | 6331 | |
a52d9a80 | 6332 | if (em) { |
e1c4b745 CM |
6333 | if (em->start > start || em->start + em->len <= start) |
6334 | free_extent_map(em); | |
6335 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6336 | free_extent_map(em); |
6337 | else | |
6338 | goto out; | |
a52d9a80 | 6339 | } |
172ddd60 | 6340 | em = alloc_extent_map(); |
a52d9a80 | 6341 | if (!em) { |
d1310b2e CM |
6342 | err = -ENOMEM; |
6343 | goto out; | |
a52d9a80 | 6344 | } |
e6dcd2dc | 6345 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6346 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6347 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6348 | em->len = (u64)-1; |
c8b97818 | 6349 | em->block_len = (u64)-1; |
f421950f CM |
6350 | |
6351 | if (!path) { | |
6352 | path = btrfs_alloc_path(); | |
026fd317 JB |
6353 | if (!path) { |
6354 | err = -ENOMEM; | |
6355 | goto out; | |
6356 | } | |
6357 | /* | |
6358 | * Chances are we'll be called again, so go ahead and do | |
6359 | * readahead | |
6360 | */ | |
6361 | path->reada = 1; | |
f421950f CM |
6362 | } |
6363 | ||
179e29e4 CM |
6364 | ret = btrfs_lookup_file_extent(trans, root, path, |
6365 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6366 | if (ret < 0) { |
6367 | err = ret; | |
6368 | goto out; | |
6369 | } | |
6370 | ||
6371 | if (ret != 0) { | |
6372 | if (path->slots[0] == 0) | |
6373 | goto not_found; | |
6374 | path->slots[0]--; | |
6375 | } | |
6376 | ||
5f39d397 CM |
6377 | leaf = path->nodes[0]; |
6378 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6379 | struct btrfs_file_extent_item); |
a52d9a80 | 6380 | /* are we inside the extent that was found? */ |
5f39d397 | 6381 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6382 | found_type = found_key.type; |
5f39d397 | 6383 | if (found_key.objectid != objectid || |
a52d9a80 | 6384 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6385 | /* |
6386 | * If we backup past the first extent we want to move forward | |
6387 | * and see if there is an extent in front of us, otherwise we'll | |
6388 | * say there is a hole for our whole search range which can | |
6389 | * cause problems. | |
6390 | */ | |
6391 | extent_end = start; | |
6392 | goto next; | |
a52d9a80 CM |
6393 | } |
6394 | ||
5f39d397 CM |
6395 | found_type = btrfs_file_extent_type(leaf, item); |
6396 | extent_start = found_key.offset; | |
d899e052 YZ |
6397 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6398 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6399 | extent_end = extent_start + |
db94535d | 6400 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6401 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6402 | size_t size; | |
514ac8ad | 6403 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6404 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6405 | } |
25a50341 | 6406 | next: |
9036c102 YZ |
6407 | if (start >= extent_end) { |
6408 | path->slots[0]++; | |
6409 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6410 | ret = btrfs_next_leaf(root, path); | |
6411 | if (ret < 0) { | |
6412 | err = ret; | |
6413 | goto out; | |
a52d9a80 | 6414 | } |
9036c102 YZ |
6415 | if (ret > 0) |
6416 | goto not_found; | |
6417 | leaf = path->nodes[0]; | |
a52d9a80 | 6418 | } |
9036c102 YZ |
6419 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6420 | if (found_key.objectid != objectid || | |
6421 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6422 | goto not_found; | |
6423 | if (start + len <= found_key.offset) | |
6424 | goto not_found; | |
e2eca69d WS |
6425 | if (start > found_key.offset) |
6426 | goto next; | |
9036c102 | 6427 | em->start = start; |
70c8a91c | 6428 | em->orig_start = start; |
9036c102 YZ |
6429 | em->len = found_key.offset - start; |
6430 | goto not_found_em; | |
6431 | } | |
6432 | ||
7ffbb598 FM |
6433 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6434 | ||
d899e052 YZ |
6435 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6436 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6437 | goto insert; |
6438 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6439 | unsigned long ptr; |
a52d9a80 | 6440 | char *map; |
3326d1b0 CM |
6441 | size_t size; |
6442 | size_t extent_offset; | |
6443 | size_t copy_size; | |
a52d9a80 | 6444 | |
7ffbb598 | 6445 | if (new_inline) |
689f9346 | 6446 | goto out; |
5f39d397 | 6447 | |
514ac8ad | 6448 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6449 | extent_offset = page_offset(page) + pg_offset - extent_start; |
70dec807 | 6450 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6451 | size - extent_offset); |
3326d1b0 | 6452 | em->start = extent_start + extent_offset; |
fda2832f | 6453 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6454 | em->orig_block_len = em->len; |
70c8a91c | 6455 | em->orig_start = em->start; |
689f9346 | 6456 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6457 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6458 | if (btrfs_file_extent_compression(leaf, item) != |
6459 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6460 | ret = uncompress_inline(path, inode, page, |
6461 | pg_offset, | |
6462 | extent_offset, item); | |
166ae5a4 ZB |
6463 | if (ret) { |
6464 | err = ret; | |
6465 | goto out; | |
6466 | } | |
c8b97818 CM |
6467 | } else { |
6468 | map = kmap(page); | |
6469 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6470 | copy_size); | |
93c82d57 CM |
6471 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6472 | memset(map + pg_offset + copy_size, 0, | |
6473 | PAGE_CACHE_SIZE - pg_offset - | |
6474 | copy_size); | |
6475 | } | |
c8b97818 CM |
6476 | kunmap(page); |
6477 | } | |
179e29e4 CM |
6478 | flush_dcache_page(page); |
6479 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6480 | BUG(); |
179e29e4 CM |
6481 | if (!trans) { |
6482 | kunmap(page); | |
6483 | free_extent_map(em); | |
6484 | em = NULL; | |
ff5714cc | 6485 | |
b3b4aa74 | 6486 | btrfs_release_path(path); |
7a7eaa40 | 6487 | trans = btrfs_join_transaction(root); |
ff5714cc | 6488 | |
3612b495 TI |
6489 | if (IS_ERR(trans)) |
6490 | return ERR_CAST(trans); | |
179e29e4 CM |
6491 | goto again; |
6492 | } | |
c8b97818 | 6493 | map = kmap(page); |
70dec807 | 6494 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6495 | copy_size); |
c8b97818 | 6496 | kunmap(page); |
179e29e4 | 6497 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6498 | } |
d1310b2e | 6499 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6500 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6501 | goto insert; |
a52d9a80 CM |
6502 | } |
6503 | not_found: | |
6504 | em->start = start; | |
70c8a91c | 6505 | em->orig_start = start; |
d1310b2e | 6506 | em->len = len; |
a52d9a80 | 6507 | not_found_em: |
5f39d397 | 6508 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6509 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6510 | insert: |
b3b4aa74 | 6511 | btrfs_release_path(path); |
d1310b2e | 6512 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6513 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6514 | em->start, em->len, start, len); |
a52d9a80 CM |
6515 | err = -EIO; |
6516 | goto out; | |
6517 | } | |
d1310b2e CM |
6518 | |
6519 | err = 0; | |
890871be | 6520 | write_lock(&em_tree->lock); |
09a2a8f9 | 6521 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6522 | /* it is possible that someone inserted the extent into the tree |
6523 | * while we had the lock dropped. It is also possible that | |
6524 | * an overlapping map exists in the tree | |
6525 | */ | |
a52d9a80 | 6526 | if (ret == -EEXIST) { |
3b951516 | 6527 | struct extent_map *existing; |
e6dcd2dc CM |
6528 | |
6529 | ret = 0; | |
6530 | ||
3b951516 | 6531 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
6532 | if (existing && (existing->start > start || |
6533 | existing->start + existing->len <= start)) { | |
6534 | free_extent_map(existing); | |
6535 | existing = NULL; | |
6536 | } | |
3b951516 CM |
6537 | if (!existing) { |
6538 | existing = lookup_extent_mapping(em_tree, em->start, | |
6539 | em->len); | |
6540 | if (existing) { | |
6541 | err = merge_extent_mapping(em_tree, existing, | |
51f395ad | 6542 | em, start); |
3b951516 CM |
6543 | free_extent_map(existing); |
6544 | if (err) { | |
6545 | free_extent_map(em); | |
6546 | em = NULL; | |
6547 | } | |
6548 | } else { | |
6549 | err = -EIO; | |
3b951516 CM |
6550 | free_extent_map(em); |
6551 | em = NULL; | |
6552 | } | |
6553 | } else { | |
6554 | free_extent_map(em); | |
6555 | em = existing; | |
e6dcd2dc | 6556 | err = 0; |
a52d9a80 | 6557 | } |
a52d9a80 | 6558 | } |
890871be | 6559 | write_unlock(&em_tree->lock); |
a52d9a80 | 6560 | out: |
1abe9b8a | 6561 | |
4cd8587c | 6562 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6563 | |
f421950f CM |
6564 | if (path) |
6565 | btrfs_free_path(path); | |
a52d9a80 CM |
6566 | if (trans) { |
6567 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6568 | if (!err) |
a52d9a80 CM |
6569 | err = ret; |
6570 | } | |
a52d9a80 CM |
6571 | if (err) { |
6572 | free_extent_map(em); | |
a52d9a80 CM |
6573 | return ERR_PTR(err); |
6574 | } | |
79787eaa | 6575 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6576 | return em; |
6577 | } | |
6578 | ||
ec29ed5b CM |
6579 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6580 | size_t pg_offset, u64 start, u64 len, | |
6581 | int create) | |
6582 | { | |
6583 | struct extent_map *em; | |
6584 | struct extent_map *hole_em = NULL; | |
6585 | u64 range_start = start; | |
6586 | u64 end; | |
6587 | u64 found; | |
6588 | u64 found_end; | |
6589 | int err = 0; | |
6590 | ||
6591 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6592 | if (IS_ERR(em)) | |
6593 | return em; | |
6594 | if (em) { | |
6595 | /* | |
f9e4fb53 LB |
6596 | * if our em maps to |
6597 | * - a hole or | |
6598 | * - a pre-alloc extent, | |
6599 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6600 | */ |
f9e4fb53 LB |
6601 | if (em->block_start != EXTENT_MAP_HOLE && |
6602 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6603 | return em; |
6604 | else | |
6605 | hole_em = em; | |
6606 | } | |
6607 | ||
6608 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6609 | end = start + len; | |
6610 | if (end < start) | |
6611 | end = (u64)-1; | |
6612 | else | |
6613 | end -= 1; | |
6614 | ||
6615 | em = NULL; | |
6616 | ||
6617 | /* ok, we didn't find anything, lets look for delalloc */ | |
6618 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6619 | end, len, EXTENT_DELALLOC, 1); | |
6620 | found_end = range_start + found; | |
6621 | if (found_end < range_start) | |
6622 | found_end = (u64)-1; | |
6623 | ||
6624 | /* | |
6625 | * we didn't find anything useful, return | |
6626 | * the original results from get_extent() | |
6627 | */ | |
6628 | if (range_start > end || found_end <= start) { | |
6629 | em = hole_em; | |
6630 | hole_em = NULL; | |
6631 | goto out; | |
6632 | } | |
6633 | ||
6634 | /* adjust the range_start to make sure it doesn't | |
6635 | * go backwards from the start they passed in | |
6636 | */ | |
67871254 | 6637 | range_start = max(start, range_start); |
ec29ed5b CM |
6638 | found = found_end - range_start; |
6639 | ||
6640 | if (found > 0) { | |
6641 | u64 hole_start = start; | |
6642 | u64 hole_len = len; | |
6643 | ||
172ddd60 | 6644 | em = alloc_extent_map(); |
ec29ed5b CM |
6645 | if (!em) { |
6646 | err = -ENOMEM; | |
6647 | goto out; | |
6648 | } | |
6649 | /* | |
6650 | * when btrfs_get_extent can't find anything it | |
6651 | * returns one huge hole | |
6652 | * | |
6653 | * make sure what it found really fits our range, and | |
6654 | * adjust to make sure it is based on the start from | |
6655 | * the caller | |
6656 | */ | |
6657 | if (hole_em) { | |
6658 | u64 calc_end = extent_map_end(hole_em); | |
6659 | ||
6660 | if (calc_end <= start || (hole_em->start > end)) { | |
6661 | free_extent_map(hole_em); | |
6662 | hole_em = NULL; | |
6663 | } else { | |
6664 | hole_start = max(hole_em->start, start); | |
6665 | hole_len = calc_end - hole_start; | |
6666 | } | |
6667 | } | |
6668 | em->bdev = NULL; | |
6669 | if (hole_em && range_start > hole_start) { | |
6670 | /* our hole starts before our delalloc, so we | |
6671 | * have to return just the parts of the hole | |
6672 | * that go until the delalloc starts | |
6673 | */ | |
6674 | em->len = min(hole_len, | |
6675 | range_start - hole_start); | |
6676 | em->start = hole_start; | |
6677 | em->orig_start = hole_start; | |
6678 | /* | |
6679 | * don't adjust block start at all, | |
6680 | * it is fixed at EXTENT_MAP_HOLE | |
6681 | */ | |
6682 | em->block_start = hole_em->block_start; | |
6683 | em->block_len = hole_len; | |
f9e4fb53 LB |
6684 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6685 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6686 | } else { |
6687 | em->start = range_start; | |
6688 | em->len = found; | |
6689 | em->orig_start = range_start; | |
6690 | em->block_start = EXTENT_MAP_DELALLOC; | |
6691 | em->block_len = found; | |
6692 | } | |
6693 | } else if (hole_em) { | |
6694 | return hole_em; | |
6695 | } | |
6696 | out: | |
6697 | ||
6698 | free_extent_map(hole_em); | |
6699 | if (err) { | |
6700 | free_extent_map(em); | |
6701 | return ERR_PTR(err); | |
6702 | } | |
6703 | return em; | |
6704 | } | |
6705 | ||
4b46fce2 JB |
6706 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6707 | u64 start, u64 len) | |
6708 | { | |
6709 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 6710 | struct extent_map *em; |
4b46fce2 JB |
6711 | struct btrfs_key ins; |
6712 | u64 alloc_hint; | |
6713 | int ret; | |
4b46fce2 | 6714 | |
4b46fce2 | 6715 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 6716 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 6717 | alloc_hint, &ins, 1, 1); |
00361589 JB |
6718 | if (ret) |
6719 | return ERR_PTR(ret); | |
4b46fce2 | 6720 | |
70c8a91c | 6721 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6722 | ins.offset, ins.offset, ins.offset, 0); |
00361589 | 6723 | if (IS_ERR(em)) { |
e570fd27 | 6724 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6725 | return em; |
6726 | } | |
4b46fce2 JB |
6727 | |
6728 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6729 | ins.offset, ins.offset, 0); | |
6730 | if (ret) { | |
e570fd27 | 6731 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6732 | free_extent_map(em); |
6733 | return ERR_PTR(ret); | |
4b46fce2 | 6734 | } |
00361589 | 6735 | |
4b46fce2 JB |
6736 | return em; |
6737 | } | |
6738 | ||
46bfbb5c CM |
6739 | /* |
6740 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6741 | * block must be cow'd | |
6742 | */ | |
00361589 | 6743 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6744 | u64 *orig_start, u64 *orig_block_len, |
6745 | u64 *ram_bytes) | |
46bfbb5c | 6746 | { |
00361589 | 6747 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
6748 | struct btrfs_path *path; |
6749 | int ret; | |
6750 | struct extent_buffer *leaf; | |
6751 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 6752 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
6753 | struct btrfs_file_extent_item *fi; |
6754 | struct btrfs_key key; | |
6755 | u64 disk_bytenr; | |
6756 | u64 backref_offset; | |
6757 | u64 extent_end; | |
6758 | u64 num_bytes; | |
6759 | int slot; | |
6760 | int found_type; | |
7ee9e440 | 6761 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 6762 | |
46bfbb5c CM |
6763 | path = btrfs_alloc_path(); |
6764 | if (!path) | |
6765 | return -ENOMEM; | |
6766 | ||
00361589 | 6767 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6768 | offset, 0); |
6769 | if (ret < 0) | |
6770 | goto out; | |
6771 | ||
6772 | slot = path->slots[0]; | |
6773 | if (ret == 1) { | |
6774 | if (slot == 0) { | |
6775 | /* can't find the item, must cow */ | |
6776 | ret = 0; | |
6777 | goto out; | |
6778 | } | |
6779 | slot--; | |
6780 | } | |
6781 | ret = 0; | |
6782 | leaf = path->nodes[0]; | |
6783 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6784 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6785 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6786 | /* not our file or wrong item type, must cow */ | |
6787 | goto out; | |
6788 | } | |
6789 | ||
6790 | if (key.offset > offset) { | |
6791 | /* Wrong offset, must cow */ | |
6792 | goto out; | |
6793 | } | |
6794 | ||
6795 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6796 | found_type = btrfs_file_extent_type(leaf, fi); | |
6797 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6798 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6799 | /* not a regular extent, must cow */ | |
6800 | goto out; | |
6801 | } | |
7ee9e440 JB |
6802 | |
6803 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6804 | goto out; | |
6805 | ||
e77751aa MX |
6806 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
6807 | if (extent_end <= offset) | |
6808 | goto out; | |
6809 | ||
46bfbb5c | 6810 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6811 | if (disk_bytenr == 0) |
6812 | goto out; | |
6813 | ||
6814 | if (btrfs_file_extent_compression(leaf, fi) || | |
6815 | btrfs_file_extent_encryption(leaf, fi) || | |
6816 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6817 | goto out; | |
6818 | ||
46bfbb5c CM |
6819 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6820 | ||
7ee9e440 JB |
6821 | if (orig_start) { |
6822 | *orig_start = key.offset - backref_offset; | |
6823 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6824 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6825 | } | |
eb384b55 | 6826 | |
46bfbb5c CM |
6827 | if (btrfs_extent_readonly(root, disk_bytenr)) |
6828 | goto out; | |
7b2b7085 MX |
6829 | |
6830 | num_bytes = min(offset + *len, extent_end) - offset; | |
6831 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6832 | u64 range_end; | |
6833 | ||
6834 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
6835 | ret = test_range_bit(io_tree, offset, range_end, | |
6836 | EXTENT_DELALLOC, 0, NULL); | |
6837 | if (ret) { | |
6838 | ret = -EAGAIN; | |
6839 | goto out; | |
6840 | } | |
6841 | } | |
6842 | ||
1bda19eb | 6843 | btrfs_release_path(path); |
46bfbb5c CM |
6844 | |
6845 | /* | |
6846 | * look for other files referencing this extent, if we | |
6847 | * find any we must cow | |
6848 | */ | |
00361589 JB |
6849 | trans = btrfs_join_transaction(root); |
6850 | if (IS_ERR(trans)) { | |
6851 | ret = 0; | |
46bfbb5c | 6852 | goto out; |
00361589 JB |
6853 | } |
6854 | ||
6855 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
6856 | key.offset - backref_offset, disk_bytenr); | |
6857 | btrfs_end_transaction(trans, root); | |
6858 | if (ret) { | |
6859 | ret = 0; | |
6860 | goto out; | |
6861 | } | |
46bfbb5c CM |
6862 | |
6863 | /* | |
6864 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6865 | * in this extent we are about to write. If there | |
6866 | * are any csums in that range we have to cow in order | |
6867 | * to keep the csums correct | |
6868 | */ | |
6869 | disk_bytenr += backref_offset; | |
6870 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
6871 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6872 | goto out; | |
6873 | /* | |
6874 | * all of the above have passed, it is safe to overwrite this extent | |
6875 | * without cow | |
6876 | */ | |
eb384b55 | 6877 | *len = num_bytes; |
46bfbb5c CM |
6878 | ret = 1; |
6879 | out: | |
6880 | btrfs_free_path(path); | |
6881 | return ret; | |
6882 | } | |
6883 | ||
fc4adbff AG |
6884 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
6885 | { | |
6886 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
6887 | int found = false; | |
6888 | void **pagep = NULL; | |
6889 | struct page *page = NULL; | |
6890 | int start_idx; | |
6891 | int end_idx; | |
6892 | ||
6893 | start_idx = start >> PAGE_CACHE_SHIFT; | |
6894 | ||
6895 | /* | |
6896 | * end is the last byte in the last page. end == start is legal | |
6897 | */ | |
6898 | end_idx = end >> PAGE_CACHE_SHIFT; | |
6899 | ||
6900 | rcu_read_lock(); | |
6901 | ||
6902 | /* Most of the code in this while loop is lifted from | |
6903 | * find_get_page. It's been modified to begin searching from a | |
6904 | * page and return just the first page found in that range. If the | |
6905 | * found idx is less than or equal to the end idx then we know that | |
6906 | * a page exists. If no pages are found or if those pages are | |
6907 | * outside of the range then we're fine (yay!) */ | |
6908 | while (page == NULL && | |
6909 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
6910 | page = radix_tree_deref_slot(pagep); | |
6911 | if (unlikely(!page)) | |
6912 | break; | |
6913 | ||
6914 | if (radix_tree_exception(page)) { | |
809f9016 FM |
6915 | if (radix_tree_deref_retry(page)) { |
6916 | page = NULL; | |
fc4adbff | 6917 | continue; |
809f9016 | 6918 | } |
fc4adbff AG |
6919 | /* |
6920 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
6921 | * here as an exceptional entry: so return it without | |
6922 | * attempting to raise page count. | |
6923 | */ | |
6fdef6d4 | 6924 | page = NULL; |
fc4adbff AG |
6925 | break; /* TODO: Is this relevant for this use case? */ |
6926 | } | |
6927 | ||
91405151 FM |
6928 | if (!page_cache_get_speculative(page)) { |
6929 | page = NULL; | |
fc4adbff | 6930 | continue; |
91405151 | 6931 | } |
fc4adbff AG |
6932 | |
6933 | /* | |
6934 | * Has the page moved? | |
6935 | * This is part of the lockless pagecache protocol. See | |
6936 | * include/linux/pagemap.h for details. | |
6937 | */ | |
6938 | if (unlikely(page != *pagep)) { | |
6939 | page_cache_release(page); | |
6940 | page = NULL; | |
6941 | } | |
6942 | } | |
6943 | ||
6944 | if (page) { | |
6945 | if (page->index <= end_idx) | |
6946 | found = true; | |
6947 | page_cache_release(page); | |
6948 | } | |
6949 | ||
6950 | rcu_read_unlock(); | |
6951 | return found; | |
6952 | } | |
6953 | ||
eb838e73 JB |
6954 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6955 | struct extent_state **cached_state, int writing) | |
6956 | { | |
6957 | struct btrfs_ordered_extent *ordered; | |
6958 | int ret = 0; | |
6959 | ||
6960 | while (1) { | |
6961 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6962 | 0, cached_state); | |
6963 | /* | |
6964 | * We're concerned with the entire range that we're going to be | |
6965 | * doing DIO to, so we need to make sure theres no ordered | |
6966 | * extents in this range. | |
6967 | */ | |
6968 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6969 | lockend - lockstart + 1); | |
6970 | ||
6971 | /* | |
6972 | * We need to make sure there are no buffered pages in this | |
6973 | * range either, we could have raced between the invalidate in | |
6974 | * generic_file_direct_write and locking the extent. The | |
6975 | * invalidate needs to happen so that reads after a write do not | |
6976 | * get stale data. | |
6977 | */ | |
fc4adbff AG |
6978 | if (!ordered && |
6979 | (!writing || | |
6980 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
6981 | break; |
6982 | ||
6983 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6984 | cached_state, GFP_NOFS); | |
6985 | ||
6986 | if (ordered) { | |
6987 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6988 | btrfs_put_ordered_extent(ordered); | |
6989 | } else { | |
6990 | /* Screw you mmap */ | |
6991 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
6992 | lockstart, | |
6993 | lockend); | |
6994 | if (ret) | |
6995 | break; | |
6996 | ||
6997 | /* | |
6998 | * If we found a page that couldn't be invalidated just | |
6999 | * fall back to buffered. | |
7000 | */ | |
7001 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
7002 | lockstart >> PAGE_CACHE_SHIFT, | |
7003 | lockend >> PAGE_CACHE_SHIFT); | |
7004 | if (ret) | |
7005 | break; | |
7006 | } | |
7007 | ||
7008 | cond_resched(); | |
7009 | } | |
7010 | ||
7011 | return ret; | |
7012 | } | |
7013 | ||
69ffb543 JB |
7014 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7015 | u64 len, u64 orig_start, | |
7016 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7017 | u64 orig_block_len, u64 ram_bytes, |
7018 | int type) | |
69ffb543 JB |
7019 | { |
7020 | struct extent_map_tree *em_tree; | |
7021 | struct extent_map *em; | |
7022 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7023 | int ret; | |
7024 | ||
7025 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7026 | em = alloc_extent_map(); | |
7027 | if (!em) | |
7028 | return ERR_PTR(-ENOMEM); | |
7029 | ||
7030 | em->start = start; | |
7031 | em->orig_start = orig_start; | |
2ab28f32 JB |
7032 | em->mod_start = start; |
7033 | em->mod_len = len; | |
69ffb543 JB |
7034 | em->len = len; |
7035 | em->block_len = block_len; | |
7036 | em->block_start = block_start; | |
7037 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7038 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7039 | em->ram_bytes = ram_bytes; |
70c8a91c | 7040 | em->generation = -1; |
69ffb543 JB |
7041 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7042 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7043 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7044 | |
7045 | do { | |
7046 | btrfs_drop_extent_cache(inode, em->start, | |
7047 | em->start + em->len - 1, 0); | |
7048 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7049 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7050 | write_unlock(&em_tree->lock); |
7051 | } while (ret == -EEXIST); | |
7052 | ||
7053 | if (ret) { | |
7054 | free_extent_map(em); | |
7055 | return ERR_PTR(ret); | |
7056 | } | |
7057 | ||
7058 | return em; | |
7059 | } | |
7060 | ||
7061 | ||
4b46fce2 JB |
7062 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7063 | struct buffer_head *bh_result, int create) | |
7064 | { | |
7065 | struct extent_map *em; | |
7066 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7067 | struct extent_state *cached_state = NULL; |
4b46fce2 | 7068 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7069 | u64 lockstart, lockend; |
4b46fce2 | 7070 | u64 len = bh_result->b_size; |
eb838e73 | 7071 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7072 | int ret = 0; |
eb838e73 | 7073 | |
172a5049 | 7074 | if (create) |
eb838e73 | 7075 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 7076 | else |
c329861d | 7077 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7078 | |
c329861d JB |
7079 | lockstart = start; |
7080 | lockend = start + len - 1; | |
7081 | ||
eb838e73 JB |
7082 | /* |
7083 | * If this errors out it's because we couldn't invalidate pagecache for | |
7084 | * this range and we need to fallback to buffered. | |
7085 | */ | |
7086 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
7087 | return -ENOTBLK; | |
7088 | ||
4b46fce2 | 7089 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7090 | if (IS_ERR(em)) { |
7091 | ret = PTR_ERR(em); | |
7092 | goto unlock_err; | |
7093 | } | |
4b46fce2 JB |
7094 | |
7095 | /* | |
7096 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7097 | * io. INLINE is special, and we could probably kludge it in here, but | |
7098 | * it's still buffered so for safety lets just fall back to the generic | |
7099 | * buffered path. | |
7100 | * | |
7101 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7102 | * decompress it, so there will be buffering required no matter what we | |
7103 | * do, so go ahead and fallback to buffered. | |
7104 | * | |
7105 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
7106 | * to buffered IO. Don't blame me, this is the price we pay for using | |
7107 | * the generic code. | |
7108 | */ | |
7109 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7110 | em->block_start == EXTENT_MAP_INLINE) { | |
7111 | free_extent_map(em); | |
eb838e73 JB |
7112 | ret = -ENOTBLK; |
7113 | goto unlock_err; | |
4b46fce2 JB |
7114 | } |
7115 | ||
7116 | /* Just a good old fashioned hole, return */ | |
7117 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7118 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7119 | free_extent_map(em); | |
eb838e73 | 7120 | goto unlock_err; |
4b46fce2 JB |
7121 | } |
7122 | ||
7123 | /* | |
7124 | * We don't allocate a new extent in the following cases | |
7125 | * | |
7126 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7127 | * existing extent. | |
7128 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7129 | * just use the extent. | |
7130 | * | |
7131 | */ | |
46bfbb5c | 7132 | if (!create) { |
eb838e73 JB |
7133 | len = min(len, em->len - (start - em->start)); |
7134 | lockstart = start + len; | |
7135 | goto unlock; | |
46bfbb5c | 7136 | } |
4b46fce2 JB |
7137 | |
7138 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7139 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7140 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
7141 | int type; |
7142 | int ret; | |
eb384b55 | 7143 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7144 | |
7145 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7146 | type = BTRFS_ORDERED_PREALLOC; | |
7147 | else | |
7148 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7149 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7150 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7151 | |
00361589 | 7152 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 7153 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
7154 | if (type == BTRFS_ORDERED_PREALLOC) { |
7155 | free_extent_map(em); | |
7156 | em = create_pinned_em(inode, start, len, | |
7157 | orig_start, | |
b4939680 | 7158 | block_start, len, |
cc95bef6 JB |
7159 | orig_block_len, |
7160 | ram_bytes, type); | |
555e1286 FM |
7161 | if (IS_ERR(em)) { |
7162 | ret = PTR_ERR(em); | |
69ffb543 | 7163 | goto unlock_err; |
555e1286 | 7164 | } |
69ffb543 JB |
7165 | } |
7166 | ||
46bfbb5c CM |
7167 | ret = btrfs_add_ordered_extent_dio(inode, start, |
7168 | block_start, len, len, type); | |
46bfbb5c CM |
7169 | if (ret) { |
7170 | free_extent_map(em); | |
eb838e73 | 7171 | goto unlock_err; |
46bfbb5c CM |
7172 | } |
7173 | goto unlock; | |
4b46fce2 | 7174 | } |
4b46fce2 | 7175 | } |
00361589 | 7176 | |
46bfbb5c CM |
7177 | /* |
7178 | * this will cow the extent, reset the len in case we changed | |
7179 | * it above | |
7180 | */ | |
7181 | len = bh_result->b_size; | |
70c8a91c JB |
7182 | free_extent_map(em); |
7183 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7184 | if (IS_ERR(em)) { |
7185 | ret = PTR_ERR(em); | |
7186 | goto unlock_err; | |
7187 | } | |
46bfbb5c CM |
7188 | len = min(len, em->len - (start - em->start)); |
7189 | unlock: | |
4b46fce2 JB |
7190 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7191 | inode->i_blkbits; | |
46bfbb5c | 7192 | bh_result->b_size = len; |
4b46fce2 JB |
7193 | bh_result->b_bdev = em->bdev; |
7194 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7195 | if (create) { |
7196 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7197 | set_buffer_new(bh_result); | |
7198 | ||
7199 | /* | |
7200 | * Need to update the i_size under the extent lock so buffered | |
7201 | * readers will get the updated i_size when we unlock. | |
7202 | */ | |
7203 | if (start + len > i_size_read(inode)) | |
7204 | i_size_write(inode, start + len); | |
0934856d | 7205 | |
172a5049 MX |
7206 | spin_lock(&BTRFS_I(inode)->lock); |
7207 | BTRFS_I(inode)->outstanding_extents++; | |
7208 | spin_unlock(&BTRFS_I(inode)->lock); | |
7209 | ||
0934856d MX |
7210 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7211 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
7212 | &cached_state, GFP_NOFS); | |
7213 | BUG_ON(ret); | |
c3473e83 | 7214 | } |
4b46fce2 | 7215 | |
eb838e73 JB |
7216 | /* |
7217 | * In the case of write we need to clear and unlock the entire range, | |
7218 | * in the case of read we need to unlock only the end area that we | |
7219 | * aren't using if there is any left over space. | |
7220 | */ | |
24c03fa5 | 7221 | if (lockstart < lockend) { |
0934856d MX |
7222 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7223 | lockend, unlock_bits, 1, 0, | |
7224 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7225 | } else { |
eb838e73 | 7226 | free_extent_state(cached_state); |
24c03fa5 | 7227 | } |
eb838e73 | 7228 | |
4b46fce2 JB |
7229 | free_extent_map(em); |
7230 | ||
7231 | return 0; | |
eb838e73 JB |
7232 | |
7233 | unlock_err: | |
eb838e73 JB |
7234 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7235 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
7236 | return ret; | |
4b46fce2 JB |
7237 | } |
7238 | ||
4b46fce2 JB |
7239 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
7240 | { | |
e65e1535 | 7241 | struct btrfs_dio_private *dip = bio->bi_private; |
2c30c71b | 7242 | struct bio_vec *bvec; |
4b46fce2 JB |
7243 | struct inode *inode = dip->inode; |
7244 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9be3395b | 7245 | struct bio *dio_bio; |
facc8a22 | 7246 | u32 *csums = (u32 *)dip->csum; |
4b46fce2 | 7247 | u64 start; |
2c30c71b | 7248 | int i; |
4b46fce2 JB |
7249 | |
7250 | start = dip->logical_offset; | |
2c30c71b | 7251 | bio_for_each_segment_all(bvec, bio, i) { |
4b46fce2 JB |
7252 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
7253 | struct page *page = bvec->bv_page; | |
7254 | char *kaddr; | |
7255 | u32 csum = ~(u32)0; | |
7256 | unsigned long flags; | |
7257 | ||
7258 | local_irq_save(flags); | |
7ac687d9 | 7259 | kaddr = kmap_atomic(page); |
b0496686 | 7260 | csum = btrfs_csum_data(kaddr + bvec->bv_offset, |
4b46fce2 JB |
7261 | csum, bvec->bv_len); |
7262 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 7263 | kunmap_atomic(kaddr); |
4b46fce2 JB |
7264 | local_irq_restore(flags); |
7265 | ||
7266 | flush_dcache_page(bvec->bv_page); | |
2c30c71b | 7267 | if (csum != csums[i]) { |
facc8a22 | 7268 | btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", |
c1c9ff7c | 7269 | btrfs_ino(inode), start, csum, |
2c30c71b | 7270 | csums[i]); |
4b46fce2 JB |
7271 | err = -EIO; |
7272 | } | |
7273 | } | |
7274 | ||
7275 | start += bvec->bv_len; | |
2c30c71b | 7276 | } |
4b46fce2 JB |
7277 | |
7278 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 7279 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 7280 | dio_bio = dip->dio_bio; |
4b46fce2 | 7281 | |
4b46fce2 | 7282 | kfree(dip); |
c0da7aa1 JB |
7283 | |
7284 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
7285 | if (err) | |
9be3395b CM |
7286 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7287 | dio_end_io(dio_bio, err); | |
7288 | bio_put(bio); | |
4b46fce2 JB |
7289 | } |
7290 | ||
7291 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
7292 | { | |
7293 | struct btrfs_dio_private *dip = bio->bi_private; | |
7294 | struct inode *inode = dip->inode; | |
7295 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 7296 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
7297 | u64 ordered_offset = dip->logical_offset; |
7298 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 7299 | struct bio *dio_bio; |
4b46fce2 JB |
7300 | int ret; |
7301 | ||
7302 | if (err) | |
7303 | goto out_done; | |
163cf09c CM |
7304 | again: |
7305 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
7306 | &ordered_offset, | |
5fd02043 | 7307 | ordered_bytes, !err); |
4b46fce2 | 7308 | if (!ret) |
163cf09c | 7309 | goto out_test; |
4b46fce2 | 7310 | |
9e0af237 LB |
7311 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
7312 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
7313 | btrfs_queue_work(root->fs_info->endio_write_workers, |
7314 | &ordered->work); | |
163cf09c CM |
7315 | out_test: |
7316 | /* | |
7317 | * our bio might span multiple ordered extents. If we haven't | |
7318 | * completed the accounting for the whole dio, go back and try again | |
7319 | */ | |
7320 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
7321 | ordered_bytes = dip->logical_offset + dip->bytes - | |
7322 | ordered_offset; | |
5fd02043 | 7323 | ordered = NULL; |
163cf09c CM |
7324 | goto again; |
7325 | } | |
4b46fce2 | 7326 | out_done: |
9be3395b | 7327 | dio_bio = dip->dio_bio; |
4b46fce2 | 7328 | |
4b46fce2 | 7329 | kfree(dip); |
c0da7aa1 JB |
7330 | |
7331 | /* If we had an error make sure to clear the uptodate flag */ | |
7332 | if (err) | |
9be3395b CM |
7333 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7334 | dio_end_io(dio_bio, err); | |
7335 | bio_put(bio); | |
4b46fce2 JB |
7336 | } |
7337 | ||
eaf25d93 CM |
7338 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
7339 | struct bio *bio, int mirror_num, | |
7340 | unsigned long bio_flags, u64 offset) | |
7341 | { | |
7342 | int ret; | |
7343 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7344 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 7345 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
7346 | return 0; |
7347 | } | |
7348 | ||
e65e1535 MX |
7349 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
7350 | { | |
7351 | struct btrfs_dio_private *dip = bio->bi_private; | |
7352 | ||
7353 | if (err) { | |
efe120a0 FH |
7354 | btrfs_err(BTRFS_I(dip->inode)->root->fs_info, |
7355 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
c1c9ff7c | 7356 | btrfs_ino(dip->inode), bio->bi_rw, |
4f024f37 KO |
7357 | (unsigned long long)bio->bi_iter.bi_sector, |
7358 | bio->bi_iter.bi_size, err); | |
e65e1535 MX |
7359 | dip->errors = 1; |
7360 | ||
7361 | /* | |
7362 | * before atomic variable goto zero, we must make sure | |
7363 | * dip->errors is perceived to be set. | |
7364 | */ | |
4e857c58 | 7365 | smp_mb__before_atomic(); |
e65e1535 MX |
7366 | } |
7367 | ||
7368 | /* if there are more bios still pending for this dio, just exit */ | |
7369 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
7370 | goto out; | |
7371 | ||
9be3395b | 7372 | if (dip->errors) { |
e65e1535 | 7373 | bio_io_error(dip->orig_bio); |
9be3395b CM |
7374 | } else { |
7375 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
7376 | bio_endio(dip->orig_bio, 0); |
7377 | } | |
7378 | out: | |
7379 | bio_put(bio); | |
7380 | } | |
7381 | ||
7382 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
7383 | u64 first_sector, gfp_t gfp_flags) | |
7384 | { | |
7385 | int nr_vecs = bio_get_nr_vecs(bdev); | |
7386 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
7387 | } | |
7388 | ||
7389 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
7390 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 7391 | int async_submit) |
e65e1535 | 7392 | { |
facc8a22 | 7393 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
7394 | int write = rw & REQ_WRITE; |
7395 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7396 | int ret; | |
7397 | ||
b812ce28 JB |
7398 | if (async_submit) |
7399 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7400 | ||
e65e1535 | 7401 | bio_get(bio); |
5fd02043 JB |
7402 | |
7403 | if (!write) { | |
7404 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
7405 | if (ret) | |
7406 | goto err; | |
7407 | } | |
e65e1535 | 7408 | |
1ae39938 JB |
7409 | if (skip_sum) |
7410 | goto map; | |
7411 | ||
7412 | if (write && async_submit) { | |
e65e1535 MX |
7413 | ret = btrfs_wq_submit_bio(root->fs_info, |
7414 | inode, rw, bio, 0, 0, | |
7415 | file_offset, | |
7416 | __btrfs_submit_bio_start_direct_io, | |
7417 | __btrfs_submit_bio_done); | |
7418 | goto err; | |
1ae39938 JB |
7419 | } else if (write) { |
7420 | /* | |
7421 | * If we aren't doing async submit, calculate the csum of the | |
7422 | * bio now. | |
7423 | */ | |
7424 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
7425 | if (ret) | |
7426 | goto err; | |
c2db1073 | 7427 | } else if (!skip_sum) { |
facc8a22 MX |
7428 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio, |
7429 | file_offset); | |
c2db1073 TI |
7430 | if (ret) |
7431 | goto err; | |
7432 | } | |
e65e1535 | 7433 | |
1ae39938 JB |
7434 | map: |
7435 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
7436 | err: |
7437 | bio_put(bio); | |
7438 | return ret; | |
7439 | } | |
7440 | ||
7441 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
7442 | int skip_sum) | |
7443 | { | |
7444 | struct inode *inode = dip->inode; | |
7445 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7446 | struct bio *bio; |
7447 | struct bio *orig_bio = dip->orig_bio; | |
7448 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 7449 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
7450 | u64 file_offset = dip->logical_offset; |
7451 | u64 submit_len = 0; | |
7452 | u64 map_length; | |
7453 | int nr_pages = 0; | |
e65e1535 | 7454 | int ret = 0; |
1ae39938 | 7455 | int async_submit = 0; |
e65e1535 | 7456 | |
4f024f37 | 7457 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7458 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 7459 | &map_length, NULL, 0); |
7a5c3c9b | 7460 | if (ret) |
e65e1535 | 7461 | return -EIO; |
facc8a22 | 7462 | |
4f024f37 | 7463 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a JB |
7464 | bio = orig_bio; |
7465 | goto submit; | |
7466 | } | |
7467 | ||
53b381b3 DW |
7468 | /* async crcs make it difficult to collect full stripe writes. */ |
7469 | if (btrfs_get_alloc_profile(root, 1) & | |
7470 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
7471 | async_submit = 0; | |
7472 | else | |
7473 | async_submit = 1; | |
7474 | ||
02f57c7a JB |
7475 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7476 | if (!bio) | |
7477 | return -ENOMEM; | |
7a5c3c9b | 7478 | |
02f57c7a JB |
7479 | bio->bi_private = dip; |
7480 | bio->bi_end_io = btrfs_end_dio_bio; | |
7481 | atomic_inc(&dip->pending_bios); | |
7482 | ||
e65e1535 MX |
7483 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
7484 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
7485 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
7486 | bvec->bv_offset) < bvec->bv_len)) { | |
7487 | /* | |
7488 | * inc the count before we submit the bio so | |
7489 | * we know the end IO handler won't happen before | |
7490 | * we inc the count. Otherwise, the dip might get freed | |
7491 | * before we're done setting it up | |
7492 | */ | |
7493 | atomic_inc(&dip->pending_bios); | |
7494 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7495 | file_offset, skip_sum, | |
c329861d | 7496 | async_submit); |
e65e1535 MX |
7497 | if (ret) { |
7498 | bio_put(bio); | |
7499 | atomic_dec(&dip->pending_bios); | |
7500 | goto out_err; | |
7501 | } | |
7502 | ||
e65e1535 MX |
7503 | start_sector += submit_len >> 9; |
7504 | file_offset += submit_len; | |
7505 | ||
7506 | submit_len = 0; | |
7507 | nr_pages = 0; | |
7508 | ||
7509 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7510 | start_sector, GFP_NOFS); | |
7511 | if (!bio) | |
7512 | goto out_err; | |
7513 | bio->bi_private = dip; | |
7514 | bio->bi_end_io = btrfs_end_dio_bio; | |
7515 | ||
4f024f37 | 7516 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7517 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7518 | start_sector << 9, |
e65e1535 MX |
7519 | &map_length, NULL, 0); |
7520 | if (ret) { | |
7521 | bio_put(bio); | |
7522 | goto out_err; | |
7523 | } | |
7524 | } else { | |
7525 | submit_len += bvec->bv_len; | |
67871254 | 7526 | nr_pages++; |
e65e1535 MX |
7527 | bvec++; |
7528 | } | |
7529 | } | |
7530 | ||
02f57c7a | 7531 | submit: |
e65e1535 | 7532 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7533 | async_submit); |
e65e1535 MX |
7534 | if (!ret) |
7535 | return 0; | |
7536 | ||
7537 | bio_put(bio); | |
7538 | out_err: | |
7539 | dip->errors = 1; | |
7540 | /* | |
7541 | * before atomic variable goto zero, we must | |
7542 | * make sure dip->errors is perceived to be set. | |
7543 | */ | |
4e857c58 | 7544 | smp_mb__before_atomic(); |
e65e1535 MX |
7545 | if (atomic_dec_and_test(&dip->pending_bios)) |
7546 | bio_io_error(dip->orig_bio); | |
7547 | ||
7548 | /* bio_end_io() will handle error, so we needn't return it */ | |
7549 | return 0; | |
7550 | } | |
7551 | ||
9be3395b CM |
7552 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7553 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7554 | { |
7555 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7556 | struct btrfs_dio_private *dip; | |
9be3395b | 7557 | struct bio *io_bio; |
4b46fce2 | 7558 | int skip_sum; |
facc8a22 | 7559 | int sum_len; |
7b6d91da | 7560 | int write = rw & REQ_WRITE; |
4b46fce2 | 7561 | int ret = 0; |
facc8a22 | 7562 | u16 csum_size; |
4b46fce2 JB |
7563 | |
7564 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7565 | ||
9be3395b | 7566 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7567 | if (!io_bio) { |
7568 | ret = -ENOMEM; | |
7569 | goto free_ordered; | |
7570 | } | |
7571 | ||
facc8a22 MX |
7572 | if (!skip_sum && !write) { |
7573 | csum_size = btrfs_super_csum_size(root->fs_info->super_copy); | |
4f024f37 KO |
7574 | sum_len = dio_bio->bi_iter.bi_size >> |
7575 | inode->i_sb->s_blocksize_bits; | |
facc8a22 MX |
7576 | sum_len *= csum_size; |
7577 | } else { | |
7578 | sum_len = 0; | |
7579 | } | |
7580 | ||
7581 | dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS); | |
4b46fce2 JB |
7582 | if (!dip) { |
7583 | ret = -ENOMEM; | |
9be3395b | 7584 | goto free_io_bio; |
4b46fce2 | 7585 | } |
4b46fce2 | 7586 | |
9be3395b | 7587 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
7588 | dip->inode = inode; |
7589 | dip->logical_offset = file_offset; | |
4f024f37 KO |
7590 | dip->bytes = dio_bio->bi_iter.bi_size; |
7591 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 7592 | io_bio->bi_private = dip; |
e65e1535 | 7593 | dip->errors = 0; |
9be3395b CM |
7594 | dip->orig_bio = io_bio; |
7595 | dip->dio_bio = dio_bio; | |
e65e1535 | 7596 | atomic_set(&dip->pending_bios, 0); |
4b46fce2 JB |
7597 | |
7598 | if (write) | |
9be3395b | 7599 | io_bio->bi_end_io = btrfs_endio_direct_write; |
4b46fce2 | 7600 | else |
9be3395b | 7601 | io_bio->bi_end_io = btrfs_endio_direct_read; |
4b46fce2 | 7602 | |
e65e1535 MX |
7603 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7604 | if (!ret) | |
eaf25d93 | 7605 | return; |
9be3395b CM |
7606 | |
7607 | free_io_bio: | |
7608 | bio_put(io_bio); | |
7609 | ||
4b46fce2 JB |
7610 | free_ordered: |
7611 | /* | |
7612 | * If this is a write, we need to clean up the reserved space and kill | |
7613 | * the ordered extent. | |
7614 | */ | |
7615 | if (write) { | |
7616 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7617 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7618 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7619 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7620 | btrfs_free_reserved_extent(root, ordered->start, | |
e570fd27 | 7621 | ordered->disk_len, 1); |
4b46fce2 JB |
7622 | btrfs_put_ordered_extent(ordered); |
7623 | btrfs_put_ordered_extent(ordered); | |
7624 | } | |
9be3395b | 7625 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7626 | } |
7627 | ||
5a5f79b5 | 7628 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
28060d5d | 7629 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
7630 | { |
7631 | int seg; | |
a1b75f7d | 7632 | int i; |
5a5f79b5 CM |
7633 | unsigned blocksize_mask = root->sectorsize - 1; |
7634 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
7635 | |
7636 | if (offset & blocksize_mask) | |
7637 | goto out; | |
7638 | ||
28060d5d AV |
7639 | if (iov_iter_alignment(iter) & blocksize_mask) |
7640 | goto out; | |
a1b75f7d | 7641 | |
28060d5d AV |
7642 | /* If this is a write we don't need to check anymore */ |
7643 | if (rw & WRITE) | |
7644 | return 0; | |
7645 | /* | |
7646 | * Check to make sure we don't have duplicate iov_base's in this | |
7647 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7648 | * when reading back. | |
7649 | */ | |
7650 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
7651 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
7652 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
7653 | goto out; |
7654 | } | |
5a5f79b5 CM |
7655 | } |
7656 | retval = 0; | |
7657 | out: | |
7658 | return retval; | |
7659 | } | |
eb838e73 | 7660 | |
16432985 | 7661 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 7662 | struct iov_iter *iter, loff_t offset) |
16432985 | 7663 | { |
4b46fce2 JB |
7664 | struct file *file = iocb->ki_filp; |
7665 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7666 | size_t count = 0; |
2e60a51e | 7667 | int flags = 0; |
38851cc1 MX |
7668 | bool wakeup = true; |
7669 | bool relock = false; | |
0934856d | 7670 | ssize_t ret; |
4b46fce2 | 7671 | |
28060d5d | 7672 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset)) |
5a5f79b5 | 7673 | return 0; |
3f7c579c | 7674 | |
38851cc1 | 7675 | atomic_inc(&inode->i_dio_count); |
4e857c58 | 7676 | smp_mb__after_atomic(); |
38851cc1 | 7677 | |
0e267c44 | 7678 | /* |
41bd9ca4 MX |
7679 | * The generic stuff only does filemap_write_and_wait_range, which |
7680 | * isn't enough if we've written compressed pages to this area, so | |
7681 | * we need to flush the dirty pages again to make absolutely sure | |
7682 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 7683 | */ |
a6cbcd4a | 7684 | count = iov_iter_count(iter); |
41bd9ca4 MX |
7685 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
7686 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
7687 | filemap_fdatawrite_range(inode->i_mapping, offset, |
7688 | offset + count - 1); | |
0e267c44 | 7689 | |
0934856d | 7690 | if (rw & WRITE) { |
38851cc1 MX |
7691 | /* |
7692 | * If the write DIO is beyond the EOF, we need update | |
7693 | * the isize, but it is protected by i_mutex. So we can | |
7694 | * not unlock the i_mutex at this case. | |
7695 | */ | |
7696 | if (offset + count <= inode->i_size) { | |
7697 | mutex_unlock(&inode->i_mutex); | |
7698 | relock = true; | |
7699 | } | |
0934856d MX |
7700 | ret = btrfs_delalloc_reserve_space(inode, count); |
7701 | if (ret) | |
38851cc1 MX |
7702 | goto out; |
7703 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
7704 | &BTRFS_I(inode)->runtime_flags))) { | |
7705 | inode_dio_done(inode); | |
7706 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7707 | wakeup = false; | |
0934856d MX |
7708 | } |
7709 | ||
7710 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
7711 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
31b14039 | 7712 | iter, offset, btrfs_get_blocks_direct, NULL, |
2e60a51e | 7713 | btrfs_submit_direct, flags); |
0934856d MX |
7714 | if (rw & WRITE) { |
7715 | if (ret < 0 && ret != -EIOCBQUEUED) | |
7716 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 7717 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
7718 | btrfs_delalloc_release_space(inode, |
7719 | count - (size_t)ret); | |
172a5049 MX |
7720 | else |
7721 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 7722 | } |
38851cc1 | 7723 | out: |
2e60a51e MX |
7724 | if (wakeup) |
7725 | inode_dio_done(inode); | |
38851cc1 MX |
7726 | if (relock) |
7727 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
7728 | |
7729 | return ret; | |
16432985 CM |
7730 | } |
7731 | ||
05dadc09 TI |
7732 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
7733 | ||
1506fcc8 YS |
7734 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
7735 | __u64 start, __u64 len) | |
7736 | { | |
05dadc09 TI |
7737 | int ret; |
7738 | ||
7739 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
7740 | if (ret) | |
7741 | return ret; | |
7742 | ||
ec29ed5b | 7743 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
7744 | } |
7745 | ||
a52d9a80 | 7746 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7747 | { |
d1310b2e CM |
7748 | struct extent_io_tree *tree; |
7749 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 7750 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 7751 | } |
1832a6d5 | 7752 | |
a52d9a80 | 7753 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7754 | { |
d1310b2e | 7755 | struct extent_io_tree *tree; |
b888db2b CM |
7756 | |
7757 | ||
7758 | if (current->flags & PF_MEMALLOC) { | |
7759 | redirty_page_for_writepage(wbc, page); | |
7760 | unlock_page(page); | |
7761 | return 0; | |
7762 | } | |
d1310b2e | 7763 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 7764 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
7765 | } |
7766 | ||
48a3b636 ES |
7767 | static int btrfs_writepages(struct address_space *mapping, |
7768 | struct writeback_control *wbc) | |
b293f02e | 7769 | { |
d1310b2e | 7770 | struct extent_io_tree *tree; |
771ed689 | 7771 | |
d1310b2e | 7772 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
7773 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
7774 | } | |
7775 | ||
3ab2fb5a CM |
7776 | static int |
7777 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
7778 | struct list_head *pages, unsigned nr_pages) | |
7779 | { | |
d1310b2e CM |
7780 | struct extent_io_tree *tree; |
7781 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
7782 | return extent_readpages(tree, mapping, pages, nr_pages, |
7783 | btrfs_get_extent); | |
7784 | } | |
e6dcd2dc | 7785 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 7786 | { |
d1310b2e CM |
7787 | struct extent_io_tree *tree; |
7788 | struct extent_map_tree *map; | |
a52d9a80 | 7789 | int ret; |
8c2383c3 | 7790 | |
d1310b2e CM |
7791 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
7792 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 7793 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
7794 | if (ret == 1) { |
7795 | ClearPagePrivate(page); | |
7796 | set_page_private(page, 0); | |
7797 | page_cache_release(page); | |
39279cc3 | 7798 | } |
a52d9a80 | 7799 | return ret; |
39279cc3 CM |
7800 | } |
7801 | ||
e6dcd2dc CM |
7802 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
7803 | { | |
98509cfc CM |
7804 | if (PageWriteback(page) || PageDirty(page)) |
7805 | return 0; | |
b335b003 | 7806 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
7807 | } |
7808 | ||
d47992f8 LC |
7809 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
7810 | unsigned int length) | |
39279cc3 | 7811 | { |
5fd02043 | 7812 | struct inode *inode = page->mapping->host; |
d1310b2e | 7813 | struct extent_io_tree *tree; |
e6dcd2dc | 7814 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 7815 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7816 | u64 page_start = page_offset(page); |
7817 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
131e404a | 7818 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 7819 | |
8b62b72b CM |
7820 | /* |
7821 | * we have the page locked, so new writeback can't start, | |
7822 | * and the dirty bit won't be cleared while we are here. | |
7823 | * | |
7824 | * Wait for IO on this page so that we can safely clear | |
7825 | * the PagePrivate2 bit and do ordered accounting | |
7826 | */ | |
e6dcd2dc | 7827 | wait_on_page_writeback(page); |
8b62b72b | 7828 | |
5fd02043 | 7829 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
7830 | if (offset) { |
7831 | btrfs_releasepage(page, GFP_NOFS); | |
7832 | return; | |
7833 | } | |
131e404a FDBM |
7834 | |
7835 | if (!inode_evicting) | |
7836 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); | |
7837 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
e6dcd2dc | 7838 | if (ordered) { |
eb84ae03 CM |
7839 | /* |
7840 | * IO on this page will never be started, so we need | |
7841 | * to account for any ordered extents now | |
7842 | */ | |
131e404a FDBM |
7843 | if (!inode_evicting) |
7844 | clear_extent_bit(tree, page_start, page_end, | |
7845 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
7846 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
7847 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
7848 | GFP_NOFS); | |
8b62b72b CM |
7849 | /* |
7850 | * whoever cleared the private bit is responsible | |
7851 | * for the finish_ordered_io | |
7852 | */ | |
77cef2ec JB |
7853 | if (TestClearPagePrivate2(page)) { |
7854 | struct btrfs_ordered_inode_tree *tree; | |
7855 | u64 new_len; | |
7856 | ||
7857 | tree = &BTRFS_I(inode)->ordered_tree; | |
7858 | ||
7859 | spin_lock_irq(&tree->lock); | |
7860 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
7861 | new_len = page_start - ordered->file_offset; | |
7862 | if (new_len < ordered->truncated_len) | |
7863 | ordered->truncated_len = new_len; | |
7864 | spin_unlock_irq(&tree->lock); | |
7865 | ||
7866 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
7867 | page_start, | |
7868 | PAGE_CACHE_SIZE, 1)) | |
7869 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 7870 | } |
e6dcd2dc | 7871 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
7872 | if (!inode_evicting) { |
7873 | cached_state = NULL; | |
7874 | lock_extent_bits(tree, page_start, page_end, 0, | |
7875 | &cached_state); | |
7876 | } | |
7877 | } | |
7878 | ||
7879 | if (!inode_evicting) { | |
7880 | clear_extent_bit(tree, page_start, page_end, | |
7881 | EXTENT_LOCKED | EXTENT_DIRTY | | |
7882 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
7883 | EXTENT_DEFRAG, 1, 1, | |
7884 | &cached_state, GFP_NOFS); | |
7885 | ||
7886 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 7887 | } |
e6dcd2dc | 7888 | |
4a096752 | 7889 | ClearPageChecked(page); |
9ad6b7bc | 7890 | if (PagePrivate(page)) { |
9ad6b7bc CM |
7891 | ClearPagePrivate(page); |
7892 | set_page_private(page, 0); | |
7893 | page_cache_release(page); | |
7894 | } | |
39279cc3 CM |
7895 | } |
7896 | ||
9ebefb18 CM |
7897 | /* |
7898 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
7899 | * called from a page fault handler when a page is first dirtied. Hence we must | |
7900 | * be careful to check for EOF conditions here. We set the page up correctly | |
7901 | * for a written page which means we get ENOSPC checking when writing into | |
7902 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
7903 | * support these features. | |
7904 | * | |
7905 | * We are not allowed to take the i_mutex here so we have to play games to | |
7906 | * protect against truncate races as the page could now be beyond EOF. Because | |
7907 | * vmtruncate() writes the inode size before removing pages, once we have the | |
7908 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
7909 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
7910 | * unlock the page. | |
7911 | */ | |
c2ec175c | 7912 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 7913 | { |
c2ec175c | 7914 | struct page *page = vmf->page; |
496ad9aa | 7915 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 7916 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
7917 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7918 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 7919 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7920 | char *kaddr; |
7921 | unsigned long zero_start; | |
9ebefb18 | 7922 | loff_t size; |
1832a6d5 | 7923 | int ret; |
9998eb70 | 7924 | int reserved = 0; |
a52d9a80 | 7925 | u64 page_start; |
e6dcd2dc | 7926 | u64 page_end; |
9ebefb18 | 7927 | |
b2b5ef5c | 7928 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 7929 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7930 | if (!ret) { |
e41f941a | 7931 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
7932 | reserved = 1; |
7933 | } | |
56a76f82 NP |
7934 | if (ret) { |
7935 | if (ret == -ENOMEM) | |
7936 | ret = VM_FAULT_OOM; | |
7937 | else /* -ENOSPC, -EIO, etc */ | |
7938 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
7939 | if (reserved) |
7940 | goto out; | |
7941 | goto out_noreserve; | |
56a76f82 | 7942 | } |
1832a6d5 | 7943 | |
56a76f82 | 7944 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 7945 | again: |
9ebefb18 | 7946 | lock_page(page); |
9ebefb18 | 7947 | size = i_size_read(inode); |
e6dcd2dc CM |
7948 | page_start = page_offset(page); |
7949 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 7950 | |
9ebefb18 | 7951 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 7952 | (page_start >= size)) { |
9ebefb18 CM |
7953 | /* page got truncated out from underneath us */ |
7954 | goto out_unlock; | |
7955 | } | |
e6dcd2dc CM |
7956 | wait_on_page_writeback(page); |
7957 | ||
d0082371 | 7958 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7959 | set_page_extent_mapped(page); |
7960 | ||
eb84ae03 CM |
7961 | /* |
7962 | * we can't set the delalloc bits if there are pending ordered | |
7963 | * extents. Drop our locks and wait for them to finish | |
7964 | */ | |
e6dcd2dc CM |
7965 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
7966 | if (ordered) { | |
2ac55d41 JB |
7967 | unlock_extent_cached(io_tree, page_start, page_end, |
7968 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 7969 | unlock_page(page); |
eb84ae03 | 7970 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
7971 | btrfs_put_ordered_extent(ordered); |
7972 | goto again; | |
7973 | } | |
7974 | ||
fbf19087 JB |
7975 | /* |
7976 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
7977 | * if it was already dirty, so for space accounting reasons we need to | |
7978 | * clear any delalloc bits for the range we are fixing to save. There | |
7979 | * is probably a better way to do this, but for now keep consistent with | |
7980 | * prepare_pages in the normal write path. | |
7981 | */ | |
2ac55d41 | 7982 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
7983 | EXTENT_DIRTY | EXTENT_DELALLOC | |
7984 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 7985 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 7986 | |
2ac55d41 JB |
7987 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
7988 | &cached_state); | |
9ed74f2d | 7989 | if (ret) { |
2ac55d41 JB |
7990 | unlock_extent_cached(io_tree, page_start, page_end, |
7991 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
7992 | ret = VM_FAULT_SIGBUS; |
7993 | goto out_unlock; | |
7994 | } | |
e6dcd2dc | 7995 | ret = 0; |
9ebefb18 CM |
7996 | |
7997 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 7998 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 7999 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 8000 | else |
e6dcd2dc | 8001 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 8002 | |
e6dcd2dc CM |
8003 | if (zero_start != PAGE_CACHE_SIZE) { |
8004 | kaddr = kmap(page); | |
8005 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
8006 | flush_dcache_page(page); | |
8007 | kunmap(page); | |
8008 | } | |
247e743c | 8009 | ClearPageChecked(page); |
e6dcd2dc | 8010 | set_page_dirty(page); |
50a9b214 | 8011 | SetPageUptodate(page); |
5a3f23d5 | 8012 | |
257c62e1 CM |
8013 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
8014 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 8015 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8016 | |
2ac55d41 | 8017 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
8018 | |
8019 | out_unlock: | |
b2b5ef5c JK |
8020 | if (!ret) { |
8021 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 8022 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8023 | } |
9ebefb18 | 8024 | unlock_page(page); |
1832a6d5 | 8025 | out: |
ec39e180 | 8026 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8027 | out_noreserve: |
b2b5ef5c | 8028 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
8029 | return ret; |
8030 | } | |
8031 | ||
a41ad394 | 8032 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
8033 | { |
8034 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 8035 | struct btrfs_block_rsv *rsv; |
a71754fc | 8036 | int ret = 0; |
3893e33b | 8037 | int err = 0; |
39279cc3 | 8038 | struct btrfs_trans_handle *trans; |
dbe674a9 | 8039 | u64 mask = root->sectorsize - 1; |
07127184 | 8040 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 8041 | |
0ef8b726 JB |
8042 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
8043 | (u64)-1); | |
8044 | if (ret) | |
8045 | return ret; | |
39279cc3 | 8046 | |
fcb80c2a JB |
8047 | /* |
8048 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
8049 | * 3 things going on here | |
8050 | * | |
8051 | * 1) We need to reserve space for our orphan item and the space to | |
8052 | * delete our orphan item. Lord knows we don't want to have a dangling | |
8053 | * orphan item because we didn't reserve space to remove it. | |
8054 | * | |
8055 | * 2) We need to reserve space to update our inode. | |
8056 | * | |
8057 | * 3) We need to have something to cache all the space that is going to | |
8058 | * be free'd up by the truncate operation, but also have some slack | |
8059 | * space reserved in case it uses space during the truncate (thank you | |
8060 | * very much snapshotting). | |
8061 | * | |
8062 | * And we need these to all be seperate. The fact is we can use alot of | |
8063 | * space doing the truncate, and we have no earthly idea how much space | |
8064 | * we will use, so we need the truncate reservation to be seperate so it | |
8065 | * doesn't end up using space reserved for updating the inode or | |
8066 | * removing the orphan item. We also need to be able to stop the | |
8067 | * transaction and start a new one, which means we need to be able to | |
8068 | * update the inode several times, and we have no idea of knowing how | |
8069 | * many times that will be, so we can't just reserve 1 item for the | |
8070 | * entirety of the opration, so that has to be done seperately as well. | |
8071 | * Then there is the orphan item, which does indeed need to be held on | |
8072 | * to for the whole operation, and we need nobody to touch this reserved | |
8073 | * space except the orphan code. | |
8074 | * | |
8075 | * So that leaves us with | |
8076 | * | |
8077 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
8078 | * 2) rsv - for the truncate reservation, which we will steal from the | |
8079 | * transaction reservation. | |
8080 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
8081 | * updating the inode. | |
8082 | */ | |
66d8f3dd | 8083 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8084 | if (!rsv) |
8085 | return -ENOMEM; | |
4a338542 | 8086 | rsv->size = min_size; |
ca7e70f5 | 8087 | rsv->failfast = 1; |
f0cd846e | 8088 | |
907cbceb | 8089 | /* |
07127184 | 8090 | * 1 for the truncate slack space |
907cbceb JB |
8091 | * 1 for updating the inode. |
8092 | */ | |
f3fe820c | 8093 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
8094 | if (IS_ERR(trans)) { |
8095 | err = PTR_ERR(trans); | |
8096 | goto out; | |
8097 | } | |
f0cd846e | 8098 | |
907cbceb JB |
8099 | /* Migrate the slack space for the truncate to our reserve */ |
8100 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
8101 | min_size); | |
fcb80c2a | 8102 | BUG_ON(ret); |
f0cd846e | 8103 | |
5dc562c5 JB |
8104 | /* |
8105 | * So if we truncate and then write and fsync we normally would just | |
8106 | * write the extents that changed, which is a problem if we need to | |
8107 | * first truncate that entire inode. So set this flag so we write out | |
8108 | * all of the extents in the inode to the sync log so we're completely | |
8109 | * safe. | |
8110 | */ | |
8111 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8112 | trans->block_rsv = rsv; |
907cbceb | 8113 | |
8082510e YZ |
8114 | while (1) { |
8115 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
8116 | inode->i_size, | |
8117 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 8118 | if (ret != -ENOSPC) { |
3893e33b | 8119 | err = ret; |
8082510e | 8120 | break; |
3893e33b | 8121 | } |
39279cc3 | 8122 | |
fcb80c2a | 8123 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 8124 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
8125 | if (ret) { |
8126 | err = ret; | |
8127 | break; | |
8128 | } | |
ca7e70f5 | 8129 | |
8082510e | 8130 | btrfs_end_transaction(trans, root); |
b53d3f5d | 8131 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
8132 | |
8133 | trans = btrfs_start_transaction(root, 2); | |
8134 | if (IS_ERR(trans)) { | |
8135 | ret = err = PTR_ERR(trans); | |
8136 | trans = NULL; | |
8137 | break; | |
8138 | } | |
8139 | ||
8140 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
8141 | rsv, min_size); | |
8142 | BUG_ON(ret); /* shouldn't happen */ | |
8143 | trans->block_rsv = rsv; | |
8082510e YZ |
8144 | } |
8145 | ||
8146 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 8147 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 8148 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
8149 | if (ret) |
8150 | err = ret; | |
8082510e YZ |
8151 | } |
8152 | ||
917c16b2 CM |
8153 | if (trans) { |
8154 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
8155 | ret = btrfs_update_inode(trans, root, inode); | |
8156 | if (ret && !err) | |
8157 | err = ret; | |
7b128766 | 8158 | |
7ad85bb7 | 8159 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 8160 | btrfs_btree_balance_dirty(root); |
917c16b2 | 8161 | } |
fcb80c2a JB |
8162 | |
8163 | out: | |
8164 | btrfs_free_block_rsv(root, rsv); | |
8165 | ||
3893e33b JB |
8166 | if (ret && !err) |
8167 | err = ret; | |
a41ad394 | 8168 | |
3893e33b | 8169 | return err; |
39279cc3 CM |
8170 | } |
8171 | ||
d352ac68 CM |
8172 | /* |
8173 | * create a new subvolume directory/inode (helper for the ioctl). | |
8174 | */ | |
d2fb3437 | 8175 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8176 | struct btrfs_root *new_root, |
8177 | struct btrfs_root *parent_root, | |
8178 | u64 new_dirid) | |
39279cc3 | 8179 | { |
39279cc3 | 8180 | struct inode *inode; |
76dda93c | 8181 | int err; |
00e4e6b3 | 8182 | u64 index = 0; |
39279cc3 | 8183 | |
12fc9d09 FA |
8184 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8185 | new_dirid, new_dirid, | |
8186 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8187 | &index); | |
54aa1f4d | 8188 | if (IS_ERR(inode)) |
f46b5a66 | 8189 | return PTR_ERR(inode); |
39279cc3 CM |
8190 | inode->i_op = &btrfs_dir_inode_operations; |
8191 | inode->i_fop = &btrfs_dir_file_operations; | |
8192 | ||
bfe86848 | 8193 | set_nlink(inode, 1); |
dbe674a9 | 8194 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 8195 | unlock_new_inode(inode); |
3b96362c | 8196 | |
63541927 FDBM |
8197 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8198 | if (err) | |
8199 | btrfs_err(new_root->fs_info, | |
351fd353 | 8200 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8201 | new_root->root_key.objectid, err); |
8202 | ||
76dda93c | 8203 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 8204 | |
76dda93c | 8205 | iput(inode); |
ce598979 | 8206 | return err; |
39279cc3 CM |
8207 | } |
8208 | ||
39279cc3 CM |
8209 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8210 | { | |
8211 | struct btrfs_inode *ei; | |
2ead6ae7 | 8212 | struct inode *inode; |
39279cc3 CM |
8213 | |
8214 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
8215 | if (!ei) | |
8216 | return NULL; | |
2ead6ae7 YZ |
8217 | |
8218 | ei->root = NULL; | |
2ead6ae7 | 8219 | ei->generation = 0; |
15ee9bc7 | 8220 | ei->last_trans = 0; |
257c62e1 | 8221 | ei->last_sub_trans = 0; |
e02119d5 | 8222 | ei->logged_trans = 0; |
2ead6ae7 | 8223 | ei->delalloc_bytes = 0; |
47059d93 | 8224 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8225 | ei->disk_i_size = 0; |
8226 | ei->flags = 0; | |
7709cde3 | 8227 | ei->csum_bytes = 0; |
2ead6ae7 | 8228 | ei->index_cnt = (u64)-1; |
67de1176 | 8229 | ei->dir_index = 0; |
2ead6ae7 | 8230 | ei->last_unlink_trans = 0; |
46d8bc34 | 8231 | ei->last_log_commit = 0; |
2ead6ae7 | 8232 | |
9e0baf60 JB |
8233 | spin_lock_init(&ei->lock); |
8234 | ei->outstanding_extents = 0; | |
8235 | ei->reserved_extents = 0; | |
2ead6ae7 | 8236 | |
72ac3c0d | 8237 | ei->runtime_flags = 0; |
261507a0 | 8238 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8239 | |
16cdcec7 MX |
8240 | ei->delayed_node = NULL; |
8241 | ||
2ead6ae7 | 8242 | inode = &ei->vfs_inode; |
a8067e02 | 8243 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
8244 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
8245 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
8246 | ei->io_tree.track_uptodate = 1; |
8247 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 8248 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8249 | mutex_init(&ei->log_mutex); |
f248679e | 8250 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 8251 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8252 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
2ead6ae7 YZ |
8253 | RB_CLEAR_NODE(&ei->rb_node); |
8254 | ||
8255 | return inode; | |
39279cc3 CM |
8256 | } |
8257 | ||
aaedb55b JB |
8258 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8259 | void btrfs_test_destroy_inode(struct inode *inode) | |
8260 | { | |
8261 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8262 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
8263 | } | |
8264 | #endif | |
8265 | ||
fa0d7e3d NP |
8266 | static void btrfs_i_callback(struct rcu_head *head) |
8267 | { | |
8268 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
8269 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8270 | } | |
8271 | ||
39279cc3 CM |
8272 | void btrfs_destroy_inode(struct inode *inode) |
8273 | { | |
e6dcd2dc | 8274 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
8275 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8276 | ||
b3d9b7a3 | 8277 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 8278 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
8279 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
8280 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
8281 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
8282 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 8283 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 8284 | |
a6dbd429 JB |
8285 | /* |
8286 | * This can happen where we create an inode, but somebody else also | |
8287 | * created the same inode and we need to destroy the one we already | |
8288 | * created. | |
8289 | */ | |
8290 | if (!root) | |
8291 | goto free; | |
8292 | ||
8a35d95f JB |
8293 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8294 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 8295 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 8296 | btrfs_ino(inode)); |
8a35d95f | 8297 | atomic_dec(&root->orphan_inodes); |
7b128766 | 8298 | } |
7b128766 | 8299 | |
d397712b | 8300 | while (1) { |
e6dcd2dc CM |
8301 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
8302 | if (!ordered) | |
8303 | break; | |
8304 | else { | |
c2cf52eb | 8305 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 8306 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
8307 | btrfs_remove_ordered_extent(inode, ordered); |
8308 | btrfs_put_ordered_extent(ordered); | |
8309 | btrfs_put_ordered_extent(ordered); | |
8310 | } | |
8311 | } | |
5d4f98a2 | 8312 | inode_tree_del(inode); |
5b21f2ed | 8313 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 8314 | free: |
fa0d7e3d | 8315 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
8316 | } |
8317 | ||
45321ac5 | 8318 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8319 | { |
8320 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8321 | |
6379ef9f NA |
8322 | if (root == NULL) |
8323 | return 1; | |
8324 | ||
fa6ac876 | 8325 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8326 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8327 | return 1; |
76dda93c | 8328 | else |
45321ac5 | 8329 | return generic_drop_inode(inode); |
76dda93c YZ |
8330 | } |
8331 | ||
0ee0fda0 | 8332 | static void init_once(void *foo) |
39279cc3 CM |
8333 | { |
8334 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8335 | ||
8336 | inode_init_once(&ei->vfs_inode); | |
8337 | } | |
8338 | ||
8339 | void btrfs_destroy_cachep(void) | |
8340 | { | |
8c0a8537 KS |
8341 | /* |
8342 | * Make sure all delayed rcu free inodes are flushed before we | |
8343 | * destroy cache. | |
8344 | */ | |
8345 | rcu_barrier(); | |
39279cc3 CM |
8346 | if (btrfs_inode_cachep) |
8347 | kmem_cache_destroy(btrfs_inode_cachep); | |
8348 | if (btrfs_trans_handle_cachep) | |
8349 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
8350 | if (btrfs_transaction_cachep) | |
8351 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
8352 | if (btrfs_path_cachep) |
8353 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
8354 | if (btrfs_free_space_cachep) |
8355 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
8356 | if (btrfs_delalloc_work_cachep) |
8357 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
8358 | } |
8359 | ||
8360 | int btrfs_init_cachep(void) | |
8361 | { | |
837e1972 | 8362 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
8363 | sizeof(struct btrfs_inode), 0, |
8364 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
8365 | if (!btrfs_inode_cachep) |
8366 | goto fail; | |
9601e3f6 | 8367 | |
837e1972 | 8368 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
8369 | sizeof(struct btrfs_trans_handle), 0, |
8370 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8371 | if (!btrfs_trans_handle_cachep) |
8372 | goto fail; | |
9601e3f6 | 8373 | |
837e1972 | 8374 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
8375 | sizeof(struct btrfs_transaction), 0, |
8376 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8377 | if (!btrfs_transaction_cachep) |
8378 | goto fail; | |
9601e3f6 | 8379 | |
837e1972 | 8380 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
8381 | sizeof(struct btrfs_path), 0, |
8382 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8383 | if (!btrfs_path_cachep) |
8384 | goto fail; | |
9601e3f6 | 8385 | |
837e1972 | 8386 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
8387 | sizeof(struct btrfs_free_space), 0, |
8388 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
8389 | if (!btrfs_free_space_cachep) | |
8390 | goto fail; | |
8391 | ||
8ccf6f19 MX |
8392 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
8393 | sizeof(struct btrfs_delalloc_work), 0, | |
8394 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
8395 | NULL); | |
8396 | if (!btrfs_delalloc_work_cachep) | |
8397 | goto fail; | |
8398 | ||
39279cc3 CM |
8399 | return 0; |
8400 | fail: | |
8401 | btrfs_destroy_cachep(); | |
8402 | return -ENOMEM; | |
8403 | } | |
8404 | ||
8405 | static int btrfs_getattr(struct vfsmount *mnt, | |
8406 | struct dentry *dentry, struct kstat *stat) | |
8407 | { | |
df0af1a5 | 8408 | u64 delalloc_bytes; |
39279cc3 | 8409 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
8410 | u32 blocksize = inode->i_sb->s_blocksize; |
8411 | ||
39279cc3 | 8412 | generic_fillattr(inode, stat); |
0ee5dc67 | 8413 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 8414 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
8415 | |
8416 | spin_lock(&BTRFS_I(inode)->lock); | |
8417 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
8418 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 8419 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 8420 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8421 | return 0; |
8422 | } | |
8423 | ||
d397712b CM |
8424 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8425 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
8426 | { |
8427 | struct btrfs_trans_handle *trans; | |
8428 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 8429 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
8430 | struct inode *new_inode = new_dentry->d_inode; |
8431 | struct inode *old_inode = old_dentry->d_inode; | |
8432 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 8433 | u64 index = 0; |
4df27c4d | 8434 | u64 root_objectid; |
39279cc3 | 8435 | int ret; |
33345d01 | 8436 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 8437 | |
33345d01 | 8438 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
8439 | return -EPERM; |
8440 | ||
4df27c4d | 8441 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 8442 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8443 | return -EXDEV; |
8444 | ||
33345d01 LZ |
8445 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
8446 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 8447 | return -ENOTEMPTY; |
5f39d397 | 8448 | |
4df27c4d YZ |
8449 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8450 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8451 | return -ENOTEMPTY; | |
9c52057c CM |
8452 | |
8453 | ||
8454 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 8455 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
8456 | new_dentry->d_name.name, |
8457 | new_dentry->d_name.len); | |
8458 | ||
8459 | if (ret) { | |
8460 | if (ret == -EEXIST) { | |
8461 | /* we shouldn't get | |
8462 | * eexist without a new_inode */ | |
fae7f21c | 8463 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
8464 | return ret; |
8465 | } | |
8466 | } else { | |
8467 | /* maybe -EOVERFLOW */ | |
8468 | return ret; | |
8469 | } | |
8470 | } | |
8471 | ret = 0; | |
8472 | ||
5a3f23d5 | 8473 | /* |
8d875f95 CM |
8474 | * we're using rename to replace one file with another. Start IO on it |
8475 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 8476 | */ |
8d875f95 | 8477 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
8478 | filemap_flush(old_inode->i_mapping); |
8479 | ||
76dda93c | 8480 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8481 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8482 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8483 | /* |
8484 | * We want to reserve the absolute worst case amount of items. So if | |
8485 | * both inodes are subvols and we need to unlink them then that would | |
8486 | * require 4 item modifications, but if they are both normal inodes it | |
8487 | * would require 5 item modifications, so we'll assume their normal | |
8488 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8489 | * should cover the worst case number of items we'll modify. | |
8490 | */ | |
6e137ed3 | 8491 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8492 | if (IS_ERR(trans)) { |
8493 | ret = PTR_ERR(trans); | |
8494 | goto out_notrans; | |
8495 | } | |
76dda93c | 8496 | |
4df27c4d YZ |
8497 | if (dest != root) |
8498 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8499 | |
a5719521 YZ |
8500 | ret = btrfs_set_inode_index(new_dir, &index); |
8501 | if (ret) | |
8502 | goto out_fail; | |
5a3f23d5 | 8503 | |
67de1176 | 8504 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 8505 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 8506 | /* force full log commit if subvolume involved. */ |
995946dd | 8507 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 8508 | } else { |
a5719521 YZ |
8509 | ret = btrfs_insert_inode_ref(trans, dest, |
8510 | new_dentry->d_name.name, | |
8511 | new_dentry->d_name.len, | |
33345d01 LZ |
8512 | old_ino, |
8513 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8514 | if (ret) |
8515 | goto out_fail; | |
4df27c4d YZ |
8516 | /* |
8517 | * this is an ugly little race, but the rename is required | |
8518 | * to make sure that if we crash, the inode is either at the | |
8519 | * old name or the new one. pinning the log transaction lets | |
8520 | * us make sure we don't allow a log commit to come in after | |
8521 | * we unlink the name but before we add the new name back in. | |
8522 | */ | |
8523 | btrfs_pin_log_trans(root); | |
8524 | } | |
5a3f23d5 | 8525 | |
0c4d2d95 JB |
8526 | inode_inc_iversion(old_dir); |
8527 | inode_inc_iversion(new_dir); | |
8528 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8529 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8530 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8531 | old_inode->i_ctime = ctime; | |
5f39d397 | 8532 | |
12fcfd22 CM |
8533 | if (old_dentry->d_parent != new_dentry->d_parent) |
8534 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8535 | ||
33345d01 | 8536 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8537 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8538 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8539 | old_dentry->d_name.name, | |
8540 | old_dentry->d_name.len); | |
8541 | } else { | |
92986796 AV |
8542 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8543 | old_dentry->d_inode, | |
8544 | old_dentry->d_name.name, | |
8545 | old_dentry->d_name.len); | |
8546 | if (!ret) | |
8547 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8548 | } |
79787eaa JM |
8549 | if (ret) { |
8550 | btrfs_abort_transaction(trans, root, ret); | |
8551 | goto out_fail; | |
8552 | } | |
39279cc3 CM |
8553 | |
8554 | if (new_inode) { | |
0c4d2d95 | 8555 | inode_inc_iversion(new_inode); |
39279cc3 | 8556 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8557 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8558 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8559 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8560 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8561 | root_objectid, | |
8562 | new_dentry->d_name.name, | |
8563 | new_dentry->d_name.len); | |
8564 | BUG_ON(new_inode->i_nlink == 0); | |
8565 | } else { | |
8566 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8567 | new_dentry->d_inode, | |
8568 | new_dentry->d_name.name, | |
8569 | new_dentry->d_name.len); | |
8570 | } | |
4ef31a45 | 8571 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8572 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8573 | if (ret) { |
8574 | btrfs_abort_transaction(trans, root, ret); | |
8575 | goto out_fail; | |
8576 | } | |
39279cc3 | 8577 | } |
aec7477b | 8578 | |
4df27c4d YZ |
8579 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8580 | new_dentry->d_name.name, | |
a5719521 | 8581 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8582 | if (ret) { |
8583 | btrfs_abort_transaction(trans, root, ret); | |
8584 | goto out_fail; | |
8585 | } | |
39279cc3 | 8586 | |
67de1176 MX |
8587 | if (old_inode->i_nlink == 1) |
8588 | BTRFS_I(old_inode)->dir_index = index; | |
8589 | ||
33345d01 | 8590 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8591 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8592 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8593 | btrfs_end_log_trans(root); |
8594 | } | |
39279cc3 | 8595 | out_fail: |
7ad85bb7 | 8596 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8597 | out_notrans: |
33345d01 | 8598 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8599 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8600 | |
39279cc3 CM |
8601 | return ret; |
8602 | } | |
8603 | ||
80ace85c MS |
8604 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
8605 | struct inode *new_dir, struct dentry *new_dentry, | |
8606 | unsigned int flags) | |
8607 | { | |
8608 | if (flags & ~RENAME_NOREPLACE) | |
8609 | return -EINVAL; | |
8610 | ||
8611 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry); | |
8612 | } | |
8613 | ||
8ccf6f19 MX |
8614 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8615 | { | |
8616 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 8617 | struct inode *inode; |
8ccf6f19 MX |
8618 | |
8619 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8620 | work); | |
9f23e289 JB |
8621 | inode = delalloc_work->inode; |
8622 | if (delalloc_work->wait) { | |
8623 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
8624 | } else { | |
8625 | filemap_flush(inode->i_mapping); | |
8626 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
8627 | &BTRFS_I(inode)->runtime_flags)) | |
8628 | filemap_flush(inode->i_mapping); | |
8629 | } | |
8ccf6f19 MX |
8630 | |
8631 | if (delalloc_work->delay_iput) | |
9f23e289 | 8632 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 8633 | else |
9f23e289 | 8634 | iput(inode); |
8ccf6f19 MX |
8635 | complete(&delalloc_work->completion); |
8636 | } | |
8637 | ||
8638 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8639 | int wait, int delay_iput) | |
8640 | { | |
8641 | struct btrfs_delalloc_work *work; | |
8642 | ||
8643 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8644 | if (!work) | |
8645 | return NULL; | |
8646 | ||
8647 | init_completion(&work->completion); | |
8648 | INIT_LIST_HEAD(&work->list); | |
8649 | work->inode = inode; | |
8650 | work->wait = wait; | |
8651 | work->delay_iput = delay_iput; | |
9e0af237 LB |
8652 | WARN_ON_ONCE(!inode); |
8653 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
8654 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
8655 | |
8656 | return work; | |
8657 | } | |
8658 | ||
8659 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8660 | { | |
8661 | wait_for_completion(&work->completion); | |
8662 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8663 | } | |
8664 | ||
d352ac68 CM |
8665 | /* |
8666 | * some fairly slow code that needs optimization. This walks the list | |
8667 | * of all the inodes with pending delalloc and forces them to disk. | |
8668 | */ | |
6c255e67 MX |
8669 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
8670 | int nr) | |
ea8c2819 | 8671 | { |
ea8c2819 | 8672 | struct btrfs_inode *binode; |
5b21f2ed | 8673 | struct inode *inode; |
8ccf6f19 MX |
8674 | struct btrfs_delalloc_work *work, *next; |
8675 | struct list_head works; | |
1eafa6c7 | 8676 | struct list_head splice; |
8ccf6f19 | 8677 | int ret = 0; |
ea8c2819 | 8678 | |
8ccf6f19 | 8679 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8680 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8681 | |
573bfb72 | 8682 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
8683 | spin_lock(&root->delalloc_lock); |
8684 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8685 | while (!list_empty(&splice)) { |
8686 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8687 | delalloc_inodes); |
1eafa6c7 | 8688 | |
eb73c1b7 MX |
8689 | list_move_tail(&binode->delalloc_inodes, |
8690 | &root->delalloc_inodes); | |
5b21f2ed | 8691 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8692 | if (!inode) { |
eb73c1b7 | 8693 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8694 | continue; |
df0af1a5 | 8695 | } |
eb73c1b7 | 8696 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8697 | |
8698 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
8699 | if (unlikely(!work)) { | |
f4ab9ea7 JB |
8700 | if (delay_iput) |
8701 | btrfs_add_delayed_iput(inode); | |
8702 | else | |
8703 | iput(inode); | |
1eafa6c7 | 8704 | ret = -ENOMEM; |
a1ecaabb | 8705 | goto out; |
5b21f2ed | 8706 | } |
1eafa6c7 | 8707 | list_add_tail(&work->list, &works); |
a44903ab QW |
8708 | btrfs_queue_work(root->fs_info->flush_workers, |
8709 | &work->work); | |
6c255e67 MX |
8710 | ret++; |
8711 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 8712 | goto out; |
5b21f2ed | 8713 | cond_resched(); |
eb73c1b7 | 8714 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 8715 | } |
eb73c1b7 | 8716 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 8717 | |
a1ecaabb | 8718 | out: |
eb73c1b7 MX |
8719 | list_for_each_entry_safe(work, next, &works, list) { |
8720 | list_del_init(&work->list); | |
8721 | btrfs_wait_and_free_delalloc_work(work); | |
8722 | } | |
8723 | ||
8724 | if (!list_empty_careful(&splice)) { | |
8725 | spin_lock(&root->delalloc_lock); | |
8726 | list_splice_tail(&splice, &root->delalloc_inodes); | |
8727 | spin_unlock(&root->delalloc_lock); | |
8728 | } | |
573bfb72 | 8729 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
8730 | return ret; |
8731 | } | |
1eafa6c7 | 8732 | |
eb73c1b7 MX |
8733 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
8734 | { | |
8735 | int ret; | |
1eafa6c7 | 8736 | |
2c21b4d7 | 8737 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
8738 | return -EROFS; |
8739 | ||
6c255e67 MX |
8740 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
8741 | if (ret > 0) | |
8742 | ret = 0; | |
eb73c1b7 MX |
8743 | /* |
8744 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
8745 | * we have to make sure the IO is actually started and that |
8746 | * ordered extents get created before we return | |
8747 | */ | |
8748 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 8749 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 8750 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 8751 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
8752 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
8753 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
8754 | } |
8755 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
8756 | return ret; |
8757 | } | |
8758 | ||
6c255e67 MX |
8759 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
8760 | int nr) | |
eb73c1b7 MX |
8761 | { |
8762 | struct btrfs_root *root; | |
8763 | struct list_head splice; | |
8764 | int ret; | |
8765 | ||
2c21b4d7 | 8766 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
8767 | return -EROFS; |
8768 | ||
8769 | INIT_LIST_HEAD(&splice); | |
8770 | ||
573bfb72 | 8771 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
8772 | spin_lock(&fs_info->delalloc_root_lock); |
8773 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 8774 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
8775 | root = list_first_entry(&splice, struct btrfs_root, |
8776 | delalloc_root); | |
8777 | root = btrfs_grab_fs_root(root); | |
8778 | BUG_ON(!root); | |
8779 | list_move_tail(&root->delalloc_root, | |
8780 | &fs_info->delalloc_roots); | |
8781 | spin_unlock(&fs_info->delalloc_root_lock); | |
8782 | ||
6c255e67 | 8783 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 8784 | btrfs_put_fs_root(root); |
6c255e67 | 8785 | if (ret < 0) |
eb73c1b7 MX |
8786 | goto out; |
8787 | ||
6c255e67 MX |
8788 | if (nr != -1) { |
8789 | nr -= ret; | |
8790 | WARN_ON(nr < 0); | |
8791 | } | |
eb73c1b7 | 8792 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 8793 | } |
eb73c1b7 | 8794 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 8795 | |
6c255e67 | 8796 | ret = 0; |
eb73c1b7 MX |
8797 | atomic_inc(&fs_info->async_submit_draining); |
8798 | while (atomic_read(&fs_info->nr_async_submits) || | |
8799 | atomic_read(&fs_info->async_delalloc_pages)) { | |
8800 | wait_event(fs_info->async_submit_wait, | |
8801 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
8802 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
8803 | } | |
8804 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 8805 | out: |
1eafa6c7 | 8806 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
8807 | spin_lock(&fs_info->delalloc_root_lock); |
8808 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
8809 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 8810 | } |
573bfb72 | 8811 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 8812 | return ret; |
ea8c2819 CM |
8813 | } |
8814 | ||
39279cc3 CM |
8815 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
8816 | const char *symname) | |
8817 | { | |
8818 | struct btrfs_trans_handle *trans; | |
8819 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
8820 | struct btrfs_path *path; | |
8821 | struct btrfs_key key; | |
1832a6d5 | 8822 | struct inode *inode = NULL; |
39279cc3 CM |
8823 | int err; |
8824 | int drop_inode = 0; | |
8825 | u64 objectid; | |
67871254 | 8826 | u64 index = 0; |
39279cc3 CM |
8827 | int name_len; |
8828 | int datasize; | |
5f39d397 | 8829 | unsigned long ptr; |
39279cc3 | 8830 | struct btrfs_file_extent_item *ei; |
5f39d397 | 8831 | struct extent_buffer *leaf; |
39279cc3 | 8832 | |
f06becc4 | 8833 | name_len = strlen(symname); |
39279cc3 CM |
8834 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
8835 | return -ENAMETOOLONG; | |
1832a6d5 | 8836 | |
9ed74f2d JB |
8837 | /* |
8838 | * 2 items for inode item and ref | |
8839 | * 2 items for dir items | |
8840 | * 1 item for xattr if selinux is on | |
8841 | */ | |
a22285a6 YZ |
8842 | trans = btrfs_start_transaction(root, 5); |
8843 | if (IS_ERR(trans)) | |
8844 | return PTR_ERR(trans); | |
1832a6d5 | 8845 | |
581bb050 LZ |
8846 | err = btrfs_find_free_ino(root, &objectid); |
8847 | if (err) | |
8848 | goto out_unlock; | |
8849 | ||
aec7477b | 8850 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 8851 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 8852 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
8853 | if (IS_ERR(inode)) { |
8854 | err = PTR_ERR(inode); | |
39279cc3 | 8855 | goto out_unlock; |
7cf96da3 | 8856 | } |
39279cc3 | 8857 | |
ad19db71 CS |
8858 | /* |
8859 | * If the active LSM wants to access the inode during | |
8860 | * d_instantiate it needs these. Smack checks to see | |
8861 | * if the filesystem supports xattrs by looking at the | |
8862 | * ops vector. | |
8863 | */ | |
8864 | inode->i_fop = &btrfs_file_operations; | |
8865 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
8866 | inode->i_mapping->a_ops = &btrfs_aops; |
8867 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
8868 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
8869 | ||
8870 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
8871 | if (err) | |
8872 | goto out_unlock_inode; | |
ad19db71 | 8873 | |
a1b075d2 | 8874 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 8875 | if (err) |
b0d5d10f | 8876 | goto out_unlock_inode; |
39279cc3 CM |
8877 | |
8878 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
8879 | if (!path) { |
8880 | err = -ENOMEM; | |
b0d5d10f | 8881 | goto out_unlock_inode; |
d8926bb3 | 8882 | } |
33345d01 | 8883 | key.objectid = btrfs_ino(inode); |
39279cc3 | 8884 | key.offset = 0; |
962a298f | 8885 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
8886 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
8887 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
8888 | datasize); | |
54aa1f4d | 8889 | if (err) { |
b0839166 | 8890 | btrfs_free_path(path); |
b0d5d10f | 8891 | goto out_unlock_inode; |
54aa1f4d | 8892 | } |
5f39d397 CM |
8893 | leaf = path->nodes[0]; |
8894 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
8895 | struct btrfs_file_extent_item); | |
8896 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
8897 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 8898 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
8899 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
8900 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
8901 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
8902 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
8903 | ||
39279cc3 | 8904 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
8905 | write_extent_buffer(leaf, symname, ptr, name_len); |
8906 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 8907 | btrfs_free_path(path); |
5f39d397 | 8908 | |
39279cc3 CM |
8909 | inode->i_op = &btrfs_symlink_inode_operations; |
8910 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 8911 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 8912 | inode_set_bytes(inode, name_len); |
f06becc4 | 8913 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 8914 | err = btrfs_update_inode(trans, root, inode); |
b0d5d10f | 8915 | if (err) { |
54aa1f4d | 8916 | drop_inode = 1; |
b0d5d10f CM |
8917 | goto out_unlock_inode; |
8918 | } | |
8919 | ||
8920 | unlock_new_inode(inode); | |
8921 | d_instantiate(dentry, inode); | |
39279cc3 CM |
8922 | |
8923 | out_unlock: | |
7ad85bb7 | 8924 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
8925 | if (drop_inode) { |
8926 | inode_dec_link_count(inode); | |
8927 | iput(inode); | |
8928 | } | |
b53d3f5d | 8929 | btrfs_btree_balance_dirty(root); |
39279cc3 | 8930 | return err; |
b0d5d10f CM |
8931 | |
8932 | out_unlock_inode: | |
8933 | drop_inode = 1; | |
8934 | unlock_new_inode(inode); | |
8935 | goto out_unlock; | |
39279cc3 | 8936 | } |
16432985 | 8937 | |
0af3d00b JB |
8938 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
8939 | u64 start, u64 num_bytes, u64 min_size, | |
8940 | loff_t actual_len, u64 *alloc_hint, | |
8941 | struct btrfs_trans_handle *trans) | |
d899e052 | 8942 | { |
5dc562c5 JB |
8943 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
8944 | struct extent_map *em; | |
d899e052 YZ |
8945 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8946 | struct btrfs_key ins; | |
d899e052 | 8947 | u64 cur_offset = start; |
55a61d1d | 8948 | u64 i_size; |
154ea289 | 8949 | u64 cur_bytes; |
d899e052 | 8950 | int ret = 0; |
0af3d00b | 8951 | bool own_trans = true; |
d899e052 | 8952 | |
0af3d00b JB |
8953 | if (trans) |
8954 | own_trans = false; | |
d899e052 | 8955 | while (num_bytes > 0) { |
0af3d00b JB |
8956 | if (own_trans) { |
8957 | trans = btrfs_start_transaction(root, 3); | |
8958 | if (IS_ERR(trans)) { | |
8959 | ret = PTR_ERR(trans); | |
8960 | break; | |
8961 | } | |
5a303d5d YZ |
8962 | } |
8963 | ||
154ea289 CM |
8964 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
8965 | cur_bytes = max(cur_bytes, min_size); | |
00361589 | 8966 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 8967 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 8968 | if (ret) { |
0af3d00b JB |
8969 | if (own_trans) |
8970 | btrfs_end_transaction(trans, root); | |
a22285a6 | 8971 | break; |
d899e052 | 8972 | } |
5a303d5d | 8973 | |
d899e052 YZ |
8974 | ret = insert_reserved_file_extent(trans, inode, |
8975 | cur_offset, ins.objectid, | |
8976 | ins.offset, ins.offset, | |
920bbbfb | 8977 | ins.offset, 0, 0, 0, |
d899e052 | 8978 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 8979 | if (ret) { |
857cc2fc | 8980 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 8981 | ins.offset, 0); |
79787eaa JM |
8982 | btrfs_abort_transaction(trans, root, ret); |
8983 | if (own_trans) | |
8984 | btrfs_end_transaction(trans, root); | |
8985 | break; | |
8986 | } | |
a1ed835e CM |
8987 | btrfs_drop_extent_cache(inode, cur_offset, |
8988 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 8989 | |
5dc562c5 JB |
8990 | em = alloc_extent_map(); |
8991 | if (!em) { | |
8992 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
8993 | &BTRFS_I(inode)->runtime_flags); | |
8994 | goto next; | |
8995 | } | |
8996 | ||
8997 | em->start = cur_offset; | |
8998 | em->orig_start = cur_offset; | |
8999 | em->len = ins.offset; | |
9000 | em->block_start = ins.objectid; | |
9001 | em->block_len = ins.offset; | |
b4939680 | 9002 | em->orig_block_len = ins.offset; |
cc95bef6 | 9003 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9004 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
9005 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
9006 | em->generation = trans->transid; | |
9007 | ||
9008 | while (1) { | |
9009 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9010 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9011 | write_unlock(&em_tree->lock); |
9012 | if (ret != -EEXIST) | |
9013 | break; | |
9014 | btrfs_drop_extent_cache(inode, cur_offset, | |
9015 | cur_offset + ins.offset - 1, | |
9016 | 0); | |
9017 | } | |
9018 | free_extent_map(em); | |
9019 | next: | |
d899e052 YZ |
9020 | num_bytes -= ins.offset; |
9021 | cur_offset += ins.offset; | |
efa56464 | 9022 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9023 | |
0c4d2d95 | 9024 | inode_inc_iversion(inode); |
d899e052 | 9025 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 9026 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9027 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9028 | (actual_len > inode->i_size) && |
9029 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9030 | if (cur_offset > actual_len) |
55a61d1d | 9031 | i_size = actual_len; |
d1ea6a61 | 9032 | else |
55a61d1d JB |
9033 | i_size = cur_offset; |
9034 | i_size_write(inode, i_size); | |
9035 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
9036 | } |
9037 | ||
d899e052 | 9038 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
9039 | |
9040 | if (ret) { | |
9041 | btrfs_abort_transaction(trans, root, ret); | |
9042 | if (own_trans) | |
9043 | btrfs_end_transaction(trans, root); | |
9044 | break; | |
9045 | } | |
d899e052 | 9046 | |
0af3d00b JB |
9047 | if (own_trans) |
9048 | btrfs_end_transaction(trans, root); | |
5a303d5d | 9049 | } |
d899e052 YZ |
9050 | return ret; |
9051 | } | |
9052 | ||
0af3d00b JB |
9053 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9054 | u64 start, u64 num_bytes, u64 min_size, | |
9055 | loff_t actual_len, u64 *alloc_hint) | |
9056 | { | |
9057 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9058 | min_size, actual_len, alloc_hint, | |
9059 | NULL); | |
9060 | } | |
9061 | ||
9062 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9063 | struct btrfs_trans_handle *trans, int mode, | |
9064 | u64 start, u64 num_bytes, u64 min_size, | |
9065 | loff_t actual_len, u64 *alloc_hint) | |
9066 | { | |
9067 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9068 | min_size, actual_len, alloc_hint, trans); | |
9069 | } | |
9070 | ||
e6dcd2dc CM |
9071 | static int btrfs_set_page_dirty(struct page *page) |
9072 | { | |
e6dcd2dc CM |
9073 | return __set_page_dirty_nobuffers(page); |
9074 | } | |
9075 | ||
10556cb2 | 9076 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9077 | { |
b83cc969 | 9078 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9079 | umode_t mode = inode->i_mode; |
b83cc969 | 9080 | |
cb6db4e5 JM |
9081 | if (mask & MAY_WRITE && |
9082 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9083 | if (btrfs_root_readonly(root)) | |
9084 | return -EROFS; | |
9085 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9086 | return -EACCES; | |
9087 | } | |
2830ba7f | 9088 | return generic_permission(inode, mask); |
fdebe2bd | 9089 | } |
39279cc3 | 9090 | |
ef3b9af5 FM |
9091 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9092 | { | |
9093 | struct btrfs_trans_handle *trans; | |
9094 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9095 | struct inode *inode = NULL; | |
9096 | u64 objectid; | |
9097 | u64 index; | |
9098 | int ret = 0; | |
9099 | ||
9100 | /* | |
9101 | * 5 units required for adding orphan entry | |
9102 | */ | |
9103 | trans = btrfs_start_transaction(root, 5); | |
9104 | if (IS_ERR(trans)) | |
9105 | return PTR_ERR(trans); | |
9106 | ||
9107 | ret = btrfs_find_free_ino(root, &objectid); | |
9108 | if (ret) | |
9109 | goto out; | |
9110 | ||
9111 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
9112 | btrfs_ino(dir), objectid, mode, &index); | |
9113 | if (IS_ERR(inode)) { | |
9114 | ret = PTR_ERR(inode); | |
9115 | inode = NULL; | |
9116 | goto out; | |
9117 | } | |
9118 | ||
ef3b9af5 FM |
9119 | inode->i_fop = &btrfs_file_operations; |
9120 | inode->i_op = &btrfs_file_inode_operations; | |
9121 | ||
9122 | inode->i_mapping->a_ops = &btrfs_aops; | |
9123 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
9124 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
9125 | ||
b0d5d10f CM |
9126 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9127 | if (ret) | |
9128 | goto out_inode; | |
9129 | ||
9130 | ret = btrfs_update_inode(trans, root, inode); | |
9131 | if (ret) | |
9132 | goto out_inode; | |
ef3b9af5 FM |
9133 | ret = btrfs_orphan_add(trans, inode); |
9134 | if (ret) | |
b0d5d10f | 9135 | goto out_inode; |
ef3b9af5 | 9136 | |
5762b5c9 FM |
9137 | /* |
9138 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9139 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9140 | * through: | |
9141 | * | |
9142 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9143 | */ | |
9144 | set_nlink(inode, 1); | |
b0d5d10f | 9145 | unlock_new_inode(inode); |
ef3b9af5 FM |
9146 | d_tmpfile(dentry, inode); |
9147 | mark_inode_dirty(inode); | |
9148 | ||
9149 | out: | |
9150 | btrfs_end_transaction(trans, root); | |
9151 | if (ret) | |
9152 | iput(inode); | |
9153 | btrfs_balance_delayed_items(root); | |
9154 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 9155 | return ret; |
b0d5d10f CM |
9156 | |
9157 | out_inode: | |
9158 | unlock_new_inode(inode); | |
9159 | goto out; | |
9160 | ||
ef3b9af5 FM |
9161 | } |
9162 | ||
6e1d5dcc | 9163 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 9164 | .getattr = btrfs_getattr, |
39279cc3 CM |
9165 | .lookup = btrfs_lookup, |
9166 | .create = btrfs_create, | |
9167 | .unlink = btrfs_unlink, | |
9168 | .link = btrfs_link, | |
9169 | .mkdir = btrfs_mkdir, | |
9170 | .rmdir = btrfs_rmdir, | |
80ace85c | 9171 | .rename2 = btrfs_rename2, |
39279cc3 CM |
9172 | .symlink = btrfs_symlink, |
9173 | .setattr = btrfs_setattr, | |
618e21d5 | 9174 | .mknod = btrfs_mknod, |
95819c05 CH |
9175 | .setxattr = btrfs_setxattr, |
9176 | .getxattr = btrfs_getxattr, | |
5103e947 | 9177 | .listxattr = btrfs_listxattr, |
95819c05 | 9178 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9179 | .permission = btrfs_permission, |
4e34e719 | 9180 | .get_acl = btrfs_get_acl, |
996a710d | 9181 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9182 | .update_time = btrfs_update_time, |
ef3b9af5 | 9183 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 9184 | }; |
6e1d5dcc | 9185 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 9186 | .lookup = btrfs_lookup, |
fdebe2bd | 9187 | .permission = btrfs_permission, |
4e34e719 | 9188 | .get_acl = btrfs_get_acl, |
996a710d | 9189 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9190 | .update_time = btrfs_update_time, |
39279cc3 | 9191 | }; |
76dda93c | 9192 | |
828c0950 | 9193 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
9194 | .llseek = generic_file_llseek, |
9195 | .read = generic_read_dir, | |
9cdda8d3 | 9196 | .iterate = btrfs_real_readdir, |
34287aa3 | 9197 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 9198 | #ifdef CONFIG_COMPAT |
34287aa3 | 9199 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 9200 | #endif |
6bf13c0c | 9201 | .release = btrfs_release_file, |
e02119d5 | 9202 | .fsync = btrfs_sync_file, |
39279cc3 CM |
9203 | }; |
9204 | ||
d1310b2e | 9205 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 9206 | .fill_delalloc = run_delalloc_range, |
065631f6 | 9207 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 9208 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 9209 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 9210 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 9211 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
9212 | .set_bit_hook = btrfs_set_bit_hook, |
9213 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
9214 | .merge_extent_hook = btrfs_merge_extent_hook, |
9215 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
9216 | }; |
9217 | ||
35054394 CM |
9218 | /* |
9219 | * btrfs doesn't support the bmap operation because swapfiles | |
9220 | * use bmap to make a mapping of extents in the file. They assume | |
9221 | * these extents won't change over the life of the file and they | |
9222 | * use the bmap result to do IO directly to the drive. | |
9223 | * | |
9224 | * the btrfs bmap call would return logical addresses that aren't | |
9225 | * suitable for IO and they also will change frequently as COW | |
9226 | * operations happen. So, swapfile + btrfs == corruption. | |
9227 | * | |
9228 | * For now we're avoiding this by dropping bmap. | |
9229 | */ | |
7f09410b | 9230 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
9231 | .readpage = btrfs_readpage, |
9232 | .writepage = btrfs_writepage, | |
b293f02e | 9233 | .writepages = btrfs_writepages, |
3ab2fb5a | 9234 | .readpages = btrfs_readpages, |
16432985 | 9235 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
9236 | .invalidatepage = btrfs_invalidatepage, |
9237 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 9238 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 9239 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
9240 | }; |
9241 | ||
7f09410b | 9242 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
9243 | .readpage = btrfs_readpage, |
9244 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
9245 | .invalidatepage = btrfs_invalidatepage, |
9246 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
9247 | }; |
9248 | ||
6e1d5dcc | 9249 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
9250 | .getattr = btrfs_getattr, |
9251 | .setattr = btrfs_setattr, | |
95819c05 CH |
9252 | .setxattr = btrfs_setxattr, |
9253 | .getxattr = btrfs_getxattr, | |
5103e947 | 9254 | .listxattr = btrfs_listxattr, |
95819c05 | 9255 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9256 | .permission = btrfs_permission, |
1506fcc8 | 9257 | .fiemap = btrfs_fiemap, |
4e34e719 | 9258 | .get_acl = btrfs_get_acl, |
996a710d | 9259 | .set_acl = btrfs_set_acl, |
e41f941a | 9260 | .update_time = btrfs_update_time, |
39279cc3 | 9261 | }; |
6e1d5dcc | 9262 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
9263 | .getattr = btrfs_getattr, |
9264 | .setattr = btrfs_setattr, | |
fdebe2bd | 9265 | .permission = btrfs_permission, |
95819c05 CH |
9266 | .setxattr = btrfs_setxattr, |
9267 | .getxattr = btrfs_getxattr, | |
33268eaf | 9268 | .listxattr = btrfs_listxattr, |
95819c05 | 9269 | .removexattr = btrfs_removexattr, |
4e34e719 | 9270 | .get_acl = btrfs_get_acl, |
996a710d | 9271 | .set_acl = btrfs_set_acl, |
e41f941a | 9272 | .update_time = btrfs_update_time, |
618e21d5 | 9273 | }; |
6e1d5dcc | 9274 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
9275 | .readlink = generic_readlink, |
9276 | .follow_link = page_follow_link_light, | |
9277 | .put_link = page_put_link, | |
f209561a | 9278 | .getattr = btrfs_getattr, |
22c44fe6 | 9279 | .setattr = btrfs_setattr, |
fdebe2bd | 9280 | .permission = btrfs_permission, |
0279b4cd JO |
9281 | .setxattr = btrfs_setxattr, |
9282 | .getxattr = btrfs_getxattr, | |
9283 | .listxattr = btrfs_listxattr, | |
9284 | .removexattr = btrfs_removexattr, | |
e41f941a | 9285 | .update_time = btrfs_update_time, |
39279cc3 | 9286 | }; |
76dda93c | 9287 | |
82d339d9 | 9288 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 9289 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 9290 | .d_release = btrfs_dentry_release, |
76dda93c | 9291 | }; |