]>
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 { | |
64 | u64 ino; | |
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 | */ | |
d397712b | 128 | static noinline 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; | |
156 | btrfs_set_key_type(&key, 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 | ||
d352ac68 | 351 | /* |
771ed689 CM |
352 | * we create compressed extents in two phases. The first |
353 | * phase compresses a range of pages that have already been | |
354 | * locked (both pages and state bits are locked). | |
c8b97818 | 355 | * |
771ed689 CM |
356 | * This is done inside an ordered work queue, and the compression |
357 | * is spread across many cpus. The actual IO submission is step | |
358 | * two, and the ordered work queue takes care of making sure that | |
359 | * happens in the same order things were put onto the queue by | |
360 | * writepages and friends. | |
c8b97818 | 361 | * |
771ed689 CM |
362 | * If this code finds it can't get good compression, it puts an |
363 | * entry onto the work queue to write the uncompressed bytes. This | |
364 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
365 | * are written in the same order that the flusher thread sent them |
366 | * down. | |
d352ac68 | 367 | */ |
771ed689 CM |
368 | static noinline int compress_file_range(struct inode *inode, |
369 | struct page *locked_page, | |
370 | u64 start, u64 end, | |
371 | struct async_cow *async_cow, | |
372 | int *num_added) | |
b888db2b CM |
373 | { |
374 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 375 | u64 num_bytes; |
db94535d | 376 | u64 blocksize = root->sectorsize; |
c8b97818 | 377 | u64 actual_end; |
42dc7bab | 378 | u64 isize = i_size_read(inode); |
e6dcd2dc | 379 | int ret = 0; |
c8b97818 CM |
380 | struct page **pages = NULL; |
381 | unsigned long nr_pages; | |
382 | unsigned long nr_pages_ret = 0; | |
383 | unsigned long total_compressed = 0; | |
384 | unsigned long total_in = 0; | |
385 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 386 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
387 | int i; |
388 | int will_compress; | |
261507a0 | 389 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 390 | int redirty = 0; |
b888db2b | 391 | |
4cb13e5d LB |
392 | /* if this is a small write inside eof, kick off a defrag */ |
393 | if ((end - start + 1) < 16 * 1024 && | |
394 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
395 | btrfs_add_inode_defrag(NULL, inode); |
396 | ||
42dc7bab | 397 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
398 | again: |
399 | will_compress = 0; | |
400 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
401 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 402 | |
f03d9301 CM |
403 | /* |
404 | * we don't want to send crud past the end of i_size through | |
405 | * compression, that's just a waste of CPU time. So, if the | |
406 | * end of the file is before the start of our current | |
407 | * requested range of bytes, we bail out to the uncompressed | |
408 | * cleanup code that can deal with all of this. | |
409 | * | |
410 | * It isn't really the fastest way to fix things, but this is a | |
411 | * very uncommon corner. | |
412 | */ | |
413 | if (actual_end <= start) | |
414 | goto cleanup_and_bail_uncompressed; | |
415 | ||
c8b97818 CM |
416 | total_compressed = actual_end - start; |
417 | ||
418 | /* we want to make sure that amount of ram required to uncompress | |
419 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
420 | * of a compressed extent to 128k. This is a crucial number |
421 | * because it also controls how easily we can spread reads across | |
422 | * cpus for decompression. | |
423 | * | |
424 | * We also want to make sure the amount of IO required to do | |
425 | * a random read is reasonably small, so we limit the size of | |
426 | * a compressed extent to 128k. | |
c8b97818 CM |
427 | */ |
428 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 429 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 430 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
431 | total_in = 0; |
432 | ret = 0; | |
db94535d | 433 | |
771ed689 CM |
434 | /* |
435 | * we do compression for mount -o compress and when the | |
436 | * inode has not been flagged as nocompress. This flag can | |
437 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 438 | */ |
6cbff00f | 439 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 440 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
441 | (BTRFS_I(inode)->force_compress) || |
442 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 443 | WARN_ON(pages); |
cfbc246e | 444 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
445 | if (!pages) { |
446 | /* just bail out to the uncompressed code */ | |
447 | goto cont; | |
448 | } | |
c8b97818 | 449 | |
261507a0 LZ |
450 | if (BTRFS_I(inode)->force_compress) |
451 | compress_type = BTRFS_I(inode)->force_compress; | |
452 | ||
4adaa611 CM |
453 | /* |
454 | * we need to call clear_page_dirty_for_io on each | |
455 | * page in the range. Otherwise applications with the file | |
456 | * mmap'd can wander in and change the page contents while | |
457 | * we are compressing them. | |
458 | * | |
459 | * If the compression fails for any reason, we set the pages | |
460 | * dirty again later on. | |
461 | */ | |
462 | extent_range_clear_dirty_for_io(inode, start, end); | |
463 | redirty = 1; | |
261507a0 LZ |
464 | ret = btrfs_compress_pages(compress_type, |
465 | inode->i_mapping, start, | |
466 | total_compressed, pages, | |
467 | nr_pages, &nr_pages_ret, | |
468 | &total_in, | |
469 | &total_compressed, | |
470 | max_compressed); | |
c8b97818 CM |
471 | |
472 | if (!ret) { | |
473 | unsigned long offset = total_compressed & | |
474 | (PAGE_CACHE_SIZE - 1); | |
475 | struct page *page = pages[nr_pages_ret - 1]; | |
476 | char *kaddr; | |
477 | ||
478 | /* zero the tail end of the last page, we might be | |
479 | * sending it down to disk | |
480 | */ | |
481 | if (offset) { | |
7ac687d9 | 482 | kaddr = kmap_atomic(page); |
c8b97818 CM |
483 | memset(kaddr + offset, 0, |
484 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 485 | kunmap_atomic(kaddr); |
c8b97818 CM |
486 | } |
487 | will_compress = 1; | |
488 | } | |
489 | } | |
560f7d75 | 490 | cont: |
c8b97818 CM |
491 | if (start == 0) { |
492 | /* lets try to make an inline extent */ | |
771ed689 | 493 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 494 | /* we didn't compress the entire range, try |
771ed689 | 495 | * to make an uncompressed inline extent. |
c8b97818 | 496 | */ |
00361589 JB |
497 | ret = cow_file_range_inline(root, inode, start, end, |
498 | 0, 0, NULL); | |
c8b97818 | 499 | } else { |
771ed689 | 500 | /* try making a compressed inline extent */ |
00361589 | 501 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
502 | total_compressed, |
503 | compress_type, pages); | |
c8b97818 | 504 | } |
79787eaa | 505 | if (ret <= 0) { |
151a41bc JB |
506 | unsigned long clear_flags = EXTENT_DELALLOC | |
507 | EXTENT_DEFRAG; | |
508 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; | |
509 | ||
771ed689 | 510 | /* |
79787eaa JM |
511 | * inline extent creation worked or returned error, |
512 | * we don't need to create any more async work items. | |
513 | * Unlock and free up our temp pages. | |
771ed689 | 514 | */ |
c2790a2e | 515 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 516 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
517 | PAGE_CLEAR_DIRTY | |
518 | PAGE_SET_WRITEBACK | | |
519 | PAGE_END_WRITEBACK); | |
c8b97818 CM |
520 | goto free_pages_out; |
521 | } | |
522 | } | |
523 | ||
524 | if (will_compress) { | |
525 | /* | |
526 | * we aren't doing an inline extent round the compressed size | |
527 | * up to a block size boundary so the allocator does sane | |
528 | * things | |
529 | */ | |
fda2832f | 530 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
531 | |
532 | /* | |
533 | * one last check to make sure the compression is really a | |
534 | * win, compare the page count read with the blocks on disk | |
535 | */ | |
fda2832f | 536 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
537 | if (total_compressed >= total_in) { |
538 | will_compress = 0; | |
539 | } else { | |
c8b97818 CM |
540 | num_bytes = total_in; |
541 | } | |
542 | } | |
543 | if (!will_compress && pages) { | |
544 | /* | |
545 | * the compression code ran but failed to make things smaller, | |
546 | * free any pages it allocated and our page pointer array | |
547 | */ | |
548 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 549 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
550 | page_cache_release(pages[i]); |
551 | } | |
552 | kfree(pages); | |
553 | pages = NULL; | |
554 | total_compressed = 0; | |
555 | nr_pages_ret = 0; | |
556 | ||
557 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
558 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
559 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 560 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 561 | } |
c8b97818 | 562 | } |
771ed689 CM |
563 | if (will_compress) { |
564 | *num_added += 1; | |
c8b97818 | 565 | |
771ed689 CM |
566 | /* the async work queues will take care of doing actual |
567 | * allocation on disk for these compressed pages, | |
568 | * and will submit them to the elevator. | |
569 | */ | |
570 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
571 | total_compressed, pages, nr_pages_ret, |
572 | compress_type); | |
179e29e4 | 573 | |
24ae6365 | 574 | if (start + num_bytes < end) { |
771ed689 CM |
575 | start += num_bytes; |
576 | pages = NULL; | |
577 | cond_resched(); | |
578 | goto again; | |
579 | } | |
580 | } else { | |
f03d9301 | 581 | cleanup_and_bail_uncompressed: |
771ed689 CM |
582 | /* |
583 | * No compression, but we still need to write the pages in | |
584 | * the file we've been given so far. redirty the locked | |
585 | * page if it corresponds to our extent and set things up | |
586 | * for the async work queue to run cow_file_range to do | |
587 | * the normal delalloc dance | |
588 | */ | |
589 | if (page_offset(locked_page) >= start && | |
590 | page_offset(locked_page) <= end) { | |
591 | __set_page_dirty_nobuffers(locked_page); | |
592 | /* unlocked later on in the async handlers */ | |
593 | } | |
4adaa611 CM |
594 | if (redirty) |
595 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
596 | add_async_extent(async_cow, start, end - start + 1, |
597 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
598 | *num_added += 1; |
599 | } | |
3b951516 | 600 | |
771ed689 | 601 | out: |
79787eaa | 602 | return ret; |
771ed689 CM |
603 | |
604 | free_pages_out: | |
605 | for (i = 0; i < nr_pages_ret; i++) { | |
606 | WARN_ON(pages[i]->mapping); | |
607 | page_cache_release(pages[i]); | |
608 | } | |
d397712b | 609 | kfree(pages); |
771ed689 CM |
610 | |
611 | goto out; | |
612 | } | |
613 | ||
614 | /* | |
615 | * phase two of compressed writeback. This is the ordered portion | |
616 | * of the code, which only gets called in the order the work was | |
617 | * queued. We walk all the async extents created by compress_file_range | |
618 | * and send them down to the disk. | |
619 | */ | |
620 | static noinline int submit_compressed_extents(struct inode *inode, | |
621 | struct async_cow *async_cow) | |
622 | { | |
623 | struct async_extent *async_extent; | |
624 | u64 alloc_hint = 0; | |
771ed689 CM |
625 | struct btrfs_key ins; |
626 | struct extent_map *em; | |
627 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
628 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
629 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 630 | int ret = 0; |
771ed689 CM |
631 | |
632 | if (list_empty(&async_cow->extents)) | |
633 | return 0; | |
634 | ||
3e04e7f1 | 635 | again: |
d397712b | 636 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
637 | async_extent = list_entry(async_cow->extents.next, |
638 | struct async_extent, list); | |
639 | list_del(&async_extent->list); | |
c8b97818 | 640 | |
771ed689 CM |
641 | io_tree = &BTRFS_I(inode)->io_tree; |
642 | ||
f5a84ee3 | 643 | retry: |
771ed689 CM |
644 | /* did the compression code fall back to uncompressed IO? */ |
645 | if (!async_extent->pages) { | |
646 | int page_started = 0; | |
647 | unsigned long nr_written = 0; | |
648 | ||
649 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 650 | async_extent->start + |
d0082371 | 651 | async_extent->ram_size - 1); |
771ed689 CM |
652 | |
653 | /* allocate blocks */ | |
f5a84ee3 JB |
654 | ret = cow_file_range(inode, async_cow->locked_page, |
655 | async_extent->start, | |
656 | async_extent->start + | |
657 | async_extent->ram_size - 1, | |
658 | &page_started, &nr_written, 0); | |
771ed689 | 659 | |
79787eaa JM |
660 | /* JDM XXX */ |
661 | ||
771ed689 CM |
662 | /* |
663 | * if page_started, cow_file_range inserted an | |
664 | * inline extent and took care of all the unlocking | |
665 | * and IO for us. Otherwise, we need to submit | |
666 | * all those pages down to the drive. | |
667 | */ | |
f5a84ee3 | 668 | if (!page_started && !ret) |
771ed689 CM |
669 | extent_write_locked_range(io_tree, |
670 | inode, async_extent->start, | |
d397712b | 671 | async_extent->start + |
771ed689 CM |
672 | async_extent->ram_size - 1, |
673 | btrfs_get_extent, | |
674 | WB_SYNC_ALL); | |
3e04e7f1 JB |
675 | else if (ret) |
676 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
677 | kfree(async_extent); |
678 | cond_resched(); | |
679 | continue; | |
680 | } | |
681 | ||
682 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 683 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 684 | |
00361589 | 685 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
686 | async_extent->compressed_size, |
687 | async_extent->compressed_size, | |
81c9ad23 | 688 | 0, alloc_hint, &ins, 1); |
f5a84ee3 JB |
689 | if (ret) { |
690 | int i; | |
3e04e7f1 | 691 | |
f5a84ee3 JB |
692 | for (i = 0; i < async_extent->nr_pages; i++) { |
693 | WARN_ON(async_extent->pages[i]->mapping); | |
694 | page_cache_release(async_extent->pages[i]); | |
695 | } | |
696 | kfree(async_extent->pages); | |
697 | async_extent->nr_pages = 0; | |
698 | async_extent->pages = NULL; | |
3e04e7f1 | 699 | |
fdf8e2ea JB |
700 | if (ret == -ENOSPC) { |
701 | unlock_extent(io_tree, async_extent->start, | |
702 | async_extent->start + | |
703 | async_extent->ram_size - 1); | |
79787eaa | 704 | goto retry; |
fdf8e2ea | 705 | } |
3e04e7f1 | 706 | goto out_free; |
f5a84ee3 JB |
707 | } |
708 | ||
c2167754 YZ |
709 | /* |
710 | * here we're doing allocation and writeback of the | |
711 | * compressed pages | |
712 | */ | |
713 | btrfs_drop_extent_cache(inode, async_extent->start, | |
714 | async_extent->start + | |
715 | async_extent->ram_size - 1, 0); | |
716 | ||
172ddd60 | 717 | em = alloc_extent_map(); |
b9aa55be LB |
718 | if (!em) { |
719 | ret = -ENOMEM; | |
3e04e7f1 | 720 | goto out_free_reserve; |
b9aa55be | 721 | } |
771ed689 CM |
722 | em->start = async_extent->start; |
723 | em->len = async_extent->ram_size; | |
445a6944 | 724 | em->orig_start = em->start; |
2ab28f32 JB |
725 | em->mod_start = em->start; |
726 | em->mod_len = em->len; | |
c8b97818 | 727 | |
771ed689 CM |
728 | em->block_start = ins.objectid; |
729 | em->block_len = ins.offset; | |
b4939680 | 730 | em->orig_block_len = ins.offset; |
cc95bef6 | 731 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 732 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 733 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
734 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
735 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 736 | em->generation = -1; |
771ed689 | 737 | |
d397712b | 738 | while (1) { |
890871be | 739 | write_lock(&em_tree->lock); |
09a2a8f9 | 740 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 741 | write_unlock(&em_tree->lock); |
771ed689 CM |
742 | if (ret != -EEXIST) { |
743 | free_extent_map(em); | |
744 | break; | |
745 | } | |
746 | btrfs_drop_extent_cache(inode, async_extent->start, | |
747 | async_extent->start + | |
748 | async_extent->ram_size - 1, 0); | |
749 | } | |
750 | ||
3e04e7f1 JB |
751 | if (ret) |
752 | goto out_free_reserve; | |
753 | ||
261507a0 LZ |
754 | ret = btrfs_add_ordered_extent_compress(inode, |
755 | async_extent->start, | |
756 | ins.objectid, | |
757 | async_extent->ram_size, | |
758 | ins.offset, | |
759 | BTRFS_ORDERED_COMPRESSED, | |
760 | async_extent->compress_type); | |
3e04e7f1 JB |
761 | if (ret) |
762 | goto out_free_reserve; | |
771ed689 | 763 | |
771ed689 CM |
764 | /* |
765 | * clear dirty, set writeback and unlock the pages. | |
766 | */ | |
c2790a2e | 767 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
768 | async_extent->start + |
769 | async_extent->ram_size - 1, | |
151a41bc JB |
770 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
771 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 772 | PAGE_SET_WRITEBACK); |
771ed689 | 773 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
774 | async_extent->start, |
775 | async_extent->ram_size, | |
776 | ins.objectid, | |
777 | ins.offset, async_extent->pages, | |
778 | async_extent->nr_pages); | |
771ed689 CM |
779 | alloc_hint = ins.objectid + ins.offset; |
780 | kfree(async_extent); | |
3e04e7f1 JB |
781 | if (ret) |
782 | goto out; | |
771ed689 CM |
783 | cond_resched(); |
784 | } | |
79787eaa JM |
785 | ret = 0; |
786 | out: | |
787 | return ret; | |
3e04e7f1 JB |
788 | out_free_reserve: |
789 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 790 | out_free: |
c2790a2e | 791 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
792 | async_extent->start + |
793 | async_extent->ram_size - 1, | |
c2790a2e | 794 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
795 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
796 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
797 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 798 | kfree(async_extent); |
3e04e7f1 | 799 | goto again; |
771ed689 CM |
800 | } |
801 | ||
4b46fce2 JB |
802 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
803 | u64 num_bytes) | |
804 | { | |
805 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
806 | struct extent_map *em; | |
807 | u64 alloc_hint = 0; | |
808 | ||
809 | read_lock(&em_tree->lock); | |
810 | em = search_extent_mapping(em_tree, start, num_bytes); | |
811 | if (em) { | |
812 | /* | |
813 | * if block start isn't an actual block number then find the | |
814 | * first block in this inode and use that as a hint. If that | |
815 | * block is also bogus then just don't worry about it. | |
816 | */ | |
817 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
818 | free_extent_map(em); | |
819 | em = search_extent_mapping(em_tree, 0, 0); | |
820 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
821 | alloc_hint = em->block_start; | |
822 | if (em) | |
823 | free_extent_map(em); | |
824 | } else { | |
825 | alloc_hint = em->block_start; | |
826 | free_extent_map(em); | |
827 | } | |
828 | } | |
829 | read_unlock(&em_tree->lock); | |
830 | ||
831 | return alloc_hint; | |
832 | } | |
833 | ||
771ed689 CM |
834 | /* |
835 | * when extent_io.c finds a delayed allocation range in the file, | |
836 | * the call backs end up in this code. The basic idea is to | |
837 | * allocate extents on disk for the range, and create ordered data structs | |
838 | * in ram to track those extents. | |
839 | * | |
840 | * locked_page is the page that writepage had locked already. We use | |
841 | * it to make sure we don't do extra locks or unlocks. | |
842 | * | |
843 | * *page_started is set to one if we unlock locked_page and do everything | |
844 | * required to start IO on it. It may be clean and already done with | |
845 | * IO when we return. | |
846 | */ | |
00361589 JB |
847 | static noinline int cow_file_range(struct inode *inode, |
848 | struct page *locked_page, | |
849 | u64 start, u64 end, int *page_started, | |
850 | unsigned long *nr_written, | |
851 | int unlock) | |
771ed689 | 852 | { |
00361589 | 853 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
854 | u64 alloc_hint = 0; |
855 | u64 num_bytes; | |
856 | unsigned long ram_size; | |
857 | u64 disk_num_bytes; | |
858 | u64 cur_alloc_size; | |
859 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
860 | struct btrfs_key ins; |
861 | struct extent_map *em; | |
862 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
863 | int ret = 0; | |
864 | ||
02ecd2c2 JB |
865 | if (btrfs_is_free_space_inode(inode)) { |
866 | WARN_ON_ONCE(1); | |
867 | return -EINVAL; | |
868 | } | |
771ed689 | 869 | |
fda2832f | 870 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
871 | num_bytes = max(blocksize, num_bytes); |
872 | disk_num_bytes = num_bytes; | |
771ed689 | 873 | |
4cb5300b | 874 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
875 | if (num_bytes < 64 * 1024 && |
876 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 877 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 878 | |
771ed689 CM |
879 | if (start == 0) { |
880 | /* lets try to make an inline extent */ | |
00361589 JB |
881 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
882 | NULL); | |
771ed689 | 883 | if (ret == 0) { |
c2790a2e JB |
884 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
885 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 886 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
887 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
888 | PAGE_END_WRITEBACK); | |
c2167754 | 889 | |
771ed689 CM |
890 | *nr_written = *nr_written + |
891 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
892 | *page_started = 1; | |
771ed689 | 893 | goto out; |
79787eaa | 894 | } else if (ret < 0) { |
79787eaa | 895 | goto out_unlock; |
771ed689 CM |
896 | } |
897 | } | |
898 | ||
899 | BUG_ON(disk_num_bytes > | |
6c41761f | 900 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 901 | |
4b46fce2 | 902 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
903 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
904 | ||
d397712b | 905 | while (disk_num_bytes > 0) { |
a791e35e CM |
906 | unsigned long op; |
907 | ||
287a0ab9 | 908 | cur_alloc_size = disk_num_bytes; |
00361589 | 909 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 910 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 911 | &ins, 1); |
00361589 | 912 | if (ret < 0) |
79787eaa | 913 | goto out_unlock; |
d397712b | 914 | |
172ddd60 | 915 | em = alloc_extent_map(); |
b9aa55be LB |
916 | if (!em) { |
917 | ret = -ENOMEM; | |
ace68bac | 918 | goto out_reserve; |
b9aa55be | 919 | } |
e6dcd2dc | 920 | em->start = start; |
445a6944 | 921 | em->orig_start = em->start; |
771ed689 CM |
922 | ram_size = ins.offset; |
923 | em->len = ins.offset; | |
2ab28f32 JB |
924 | em->mod_start = em->start; |
925 | em->mod_len = em->len; | |
c8b97818 | 926 | |
e6dcd2dc | 927 | em->block_start = ins.objectid; |
c8b97818 | 928 | em->block_len = ins.offset; |
b4939680 | 929 | em->orig_block_len = ins.offset; |
cc95bef6 | 930 | em->ram_bytes = ram_size; |
e6dcd2dc | 931 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 932 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 933 | em->generation = -1; |
c8b97818 | 934 | |
d397712b | 935 | while (1) { |
890871be | 936 | write_lock(&em_tree->lock); |
09a2a8f9 | 937 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 938 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
939 | if (ret != -EEXIST) { |
940 | free_extent_map(em); | |
941 | break; | |
942 | } | |
943 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 944 | start + ram_size - 1, 0); |
e6dcd2dc | 945 | } |
ace68bac LB |
946 | if (ret) |
947 | goto out_reserve; | |
e6dcd2dc | 948 | |
98d20f67 | 949 | cur_alloc_size = ins.offset; |
e6dcd2dc | 950 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 951 | ram_size, cur_alloc_size, 0); |
ace68bac LB |
952 | if (ret) |
953 | goto out_reserve; | |
c8b97818 | 954 | |
17d217fe YZ |
955 | if (root->root_key.objectid == |
956 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
957 | ret = btrfs_reloc_clone_csums(inode, start, | |
958 | cur_alloc_size); | |
00361589 | 959 | if (ret) |
ace68bac | 960 | goto out_reserve; |
17d217fe YZ |
961 | } |
962 | ||
d397712b | 963 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 964 | break; |
d397712b | 965 | |
c8b97818 CM |
966 | /* we're not doing compressed IO, don't unlock the first |
967 | * page (which the caller expects to stay locked), don't | |
968 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
969 | * |
970 | * Do set the Private2 bit so we know this page was properly | |
971 | * setup for writepage | |
c8b97818 | 972 | */ |
c2790a2e JB |
973 | op = unlock ? PAGE_UNLOCK : 0; |
974 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 975 | |
c2790a2e JB |
976 | extent_clear_unlock_delalloc(inode, start, |
977 | start + ram_size - 1, locked_page, | |
978 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
979 | op); | |
c8b97818 | 980 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
981 | num_bytes -= cur_alloc_size; |
982 | alloc_hint = ins.objectid + ins.offset; | |
983 | start += cur_alloc_size; | |
b888db2b | 984 | } |
79787eaa | 985 | out: |
be20aa9d | 986 | return ret; |
b7d5b0a8 | 987 | |
ace68bac LB |
988 | out_reserve: |
989 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 990 | out_unlock: |
c2790a2e | 991 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
992 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
993 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
994 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
995 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 996 | goto out; |
771ed689 | 997 | } |
c8b97818 | 998 | |
771ed689 CM |
999 | /* |
1000 | * work queue call back to started compression on a file and pages | |
1001 | */ | |
1002 | static noinline void async_cow_start(struct btrfs_work *work) | |
1003 | { | |
1004 | struct async_cow *async_cow; | |
1005 | int num_added = 0; | |
1006 | async_cow = container_of(work, struct async_cow, work); | |
1007 | ||
1008 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1009 | async_cow->start, async_cow->end, async_cow, | |
1010 | &num_added); | |
8180ef88 | 1011 | if (num_added == 0) { |
cb77fcd8 | 1012 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1013 | async_cow->inode = NULL; |
8180ef88 | 1014 | } |
771ed689 CM |
1015 | } |
1016 | ||
1017 | /* | |
1018 | * work queue call back to submit previously compressed pages | |
1019 | */ | |
1020 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1021 | { | |
1022 | struct async_cow *async_cow; | |
1023 | struct btrfs_root *root; | |
1024 | unsigned long nr_pages; | |
1025 | ||
1026 | async_cow = container_of(work, struct async_cow, work); | |
1027 | ||
1028 | root = async_cow->root; | |
1029 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1030 | PAGE_CACHE_SHIFT; | |
1031 | ||
66657b31 | 1032 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1033 | 5 * 1024 * 1024 && |
771ed689 CM |
1034 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1035 | wake_up(&root->fs_info->async_submit_wait); | |
1036 | ||
d397712b | 1037 | if (async_cow->inode) |
771ed689 | 1038 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1039 | } |
c8b97818 | 1040 | |
771ed689 CM |
1041 | static noinline void async_cow_free(struct btrfs_work *work) |
1042 | { | |
1043 | struct async_cow *async_cow; | |
1044 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1045 | if (async_cow->inode) |
cb77fcd8 | 1046 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1047 | kfree(async_cow); |
1048 | } | |
1049 | ||
1050 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1051 | u64 start, u64 end, int *page_started, | |
1052 | unsigned long *nr_written) | |
1053 | { | |
1054 | struct async_cow *async_cow; | |
1055 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1056 | unsigned long nr_pages; | |
1057 | u64 cur_end; | |
287082b0 | 1058 | int limit = 10 * 1024 * 1024; |
771ed689 | 1059 | |
a3429ab7 CM |
1060 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1061 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1062 | while (start < end) { |
771ed689 | 1063 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1064 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1065 | async_cow->inode = igrab(inode); |
771ed689 CM |
1066 | async_cow->root = root; |
1067 | async_cow->locked_page = locked_page; | |
1068 | async_cow->start = start; | |
1069 | ||
6cbff00f | 1070 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1071 | cur_end = end; |
1072 | else | |
1073 | cur_end = min(end, start + 512 * 1024 - 1); | |
1074 | ||
1075 | async_cow->end = cur_end; | |
1076 | INIT_LIST_HEAD(&async_cow->extents); | |
1077 | ||
1078 | async_cow->work.func = async_cow_start; | |
1079 | async_cow->work.ordered_func = async_cow_submit; | |
1080 | async_cow->work.ordered_free = async_cow_free; | |
1081 | async_cow->work.flags = 0; | |
1082 | ||
771ed689 CM |
1083 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1084 | PAGE_CACHE_SHIFT; | |
1085 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1086 | ||
1087 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1088 | &async_cow->work); | |
1089 | ||
1090 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1091 | wait_event(root->fs_info->async_submit_wait, | |
1092 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1093 | limit)); | |
1094 | } | |
1095 | ||
d397712b | 1096 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1097 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1098 | wait_event(root->fs_info->async_submit_wait, | |
1099 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1100 | 0)); | |
1101 | } | |
1102 | ||
1103 | *nr_written += nr_pages; | |
1104 | start = cur_end + 1; | |
1105 | } | |
1106 | *page_started = 1; | |
1107 | return 0; | |
be20aa9d CM |
1108 | } |
1109 | ||
d397712b | 1110 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1111 | u64 bytenr, u64 num_bytes) |
1112 | { | |
1113 | int ret; | |
1114 | struct btrfs_ordered_sum *sums; | |
1115 | LIST_HEAD(list); | |
1116 | ||
07d400a6 | 1117 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1118 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1119 | if (ret == 0 && list_empty(&list)) |
1120 | return 0; | |
1121 | ||
1122 | while (!list_empty(&list)) { | |
1123 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1124 | list_del(&sums->list); | |
1125 | kfree(sums); | |
1126 | } | |
1127 | return 1; | |
1128 | } | |
1129 | ||
d352ac68 CM |
1130 | /* |
1131 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1132 | * of the extents that exist in the file, and COWs the file as required. | |
1133 | * | |
1134 | * If no cow copies or snapshots exist, we write directly to the existing | |
1135 | * blocks on disk | |
1136 | */ | |
7f366cfe CM |
1137 | static noinline int run_delalloc_nocow(struct inode *inode, |
1138 | struct page *locked_page, | |
771ed689 CM |
1139 | u64 start, u64 end, int *page_started, int force, |
1140 | unsigned long *nr_written) | |
be20aa9d | 1141 | { |
be20aa9d | 1142 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1143 | struct btrfs_trans_handle *trans; |
be20aa9d | 1144 | struct extent_buffer *leaf; |
be20aa9d | 1145 | struct btrfs_path *path; |
80ff3856 | 1146 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1147 | struct btrfs_key found_key; |
80ff3856 YZ |
1148 | u64 cow_start; |
1149 | u64 cur_offset; | |
1150 | u64 extent_end; | |
5d4f98a2 | 1151 | u64 extent_offset; |
80ff3856 YZ |
1152 | u64 disk_bytenr; |
1153 | u64 num_bytes; | |
b4939680 | 1154 | u64 disk_num_bytes; |
cc95bef6 | 1155 | u64 ram_bytes; |
80ff3856 | 1156 | int extent_type; |
79787eaa | 1157 | int ret, err; |
d899e052 | 1158 | int type; |
80ff3856 YZ |
1159 | int nocow; |
1160 | int check_prev = 1; | |
82d5902d | 1161 | bool nolock; |
33345d01 | 1162 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1163 | |
1164 | path = btrfs_alloc_path(); | |
17ca04af | 1165 | if (!path) { |
c2790a2e JB |
1166 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1167 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1168 | EXTENT_DO_ACCOUNTING | |
1169 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1170 | PAGE_CLEAR_DIRTY | |
1171 | PAGE_SET_WRITEBACK | | |
1172 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1173 | return -ENOMEM; |
17ca04af | 1174 | } |
82d5902d | 1175 | |
83eea1f1 | 1176 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1177 | |
1178 | if (nolock) | |
7a7eaa40 | 1179 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1180 | else |
7a7eaa40 | 1181 | trans = btrfs_join_transaction(root); |
ff5714cc | 1182 | |
79787eaa | 1183 | if (IS_ERR(trans)) { |
c2790a2e JB |
1184 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1185 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1186 | EXTENT_DO_ACCOUNTING | |
1187 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1188 | PAGE_CLEAR_DIRTY | |
1189 | PAGE_SET_WRITEBACK | | |
1190 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1191 | btrfs_free_path(path); |
1192 | return PTR_ERR(trans); | |
1193 | } | |
1194 | ||
74b21075 | 1195 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1196 | |
80ff3856 YZ |
1197 | cow_start = (u64)-1; |
1198 | cur_offset = start; | |
1199 | while (1) { | |
33345d01 | 1200 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1201 | cur_offset, 0); |
d788a349 | 1202 | if (ret < 0) |
79787eaa | 1203 | goto error; |
80ff3856 YZ |
1204 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1205 | leaf = path->nodes[0]; | |
1206 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1207 | path->slots[0] - 1); | |
33345d01 | 1208 | if (found_key.objectid == ino && |
80ff3856 YZ |
1209 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1210 | path->slots[0]--; | |
1211 | } | |
1212 | check_prev = 0; | |
1213 | next_slot: | |
1214 | leaf = path->nodes[0]; | |
1215 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1216 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1217 | if (ret < 0) |
79787eaa | 1218 | goto error; |
80ff3856 YZ |
1219 | if (ret > 0) |
1220 | break; | |
1221 | leaf = path->nodes[0]; | |
1222 | } | |
be20aa9d | 1223 | |
80ff3856 YZ |
1224 | nocow = 0; |
1225 | disk_bytenr = 0; | |
17d217fe | 1226 | num_bytes = 0; |
80ff3856 YZ |
1227 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1228 | ||
33345d01 | 1229 | if (found_key.objectid > ino || |
80ff3856 YZ |
1230 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1231 | found_key.offset > end) | |
1232 | break; | |
1233 | ||
1234 | if (found_key.offset > cur_offset) { | |
1235 | extent_end = found_key.offset; | |
e9061e21 | 1236 | extent_type = 0; |
80ff3856 YZ |
1237 | goto out_check; |
1238 | } | |
1239 | ||
1240 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1241 | struct btrfs_file_extent_item); | |
1242 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1243 | ||
cc95bef6 | 1244 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1245 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1246 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1247 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1248 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1249 | extent_end = found_key.offset + |
1250 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1251 | disk_num_bytes = |
1252 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1253 | if (extent_end <= start) { |
1254 | path->slots[0]++; | |
1255 | goto next_slot; | |
1256 | } | |
17d217fe YZ |
1257 | if (disk_bytenr == 0) |
1258 | goto out_check; | |
80ff3856 YZ |
1259 | if (btrfs_file_extent_compression(leaf, fi) || |
1260 | btrfs_file_extent_encryption(leaf, fi) || | |
1261 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1262 | goto out_check; | |
d899e052 YZ |
1263 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1264 | goto out_check; | |
d2fb3437 | 1265 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1266 | goto out_check; |
33345d01 | 1267 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1268 | found_key.offset - |
1269 | extent_offset, disk_bytenr)) | |
17d217fe | 1270 | goto out_check; |
5d4f98a2 | 1271 | disk_bytenr += extent_offset; |
17d217fe YZ |
1272 | disk_bytenr += cur_offset - found_key.offset; |
1273 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1274 | /* | |
1275 | * force cow if csum exists in the range. | |
1276 | * this ensure that csum for a given extent are | |
1277 | * either valid or do not exist. | |
1278 | */ | |
1279 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1280 | goto out_check; | |
80ff3856 YZ |
1281 | nocow = 1; |
1282 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1283 | extent_end = found_key.offset + | |
1284 | btrfs_file_extent_inline_len(leaf, fi); | |
1285 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1286 | } else { | |
1287 | BUG_ON(1); | |
1288 | } | |
1289 | out_check: | |
1290 | if (extent_end <= start) { | |
1291 | path->slots[0]++; | |
1292 | goto next_slot; | |
1293 | } | |
1294 | if (!nocow) { | |
1295 | if (cow_start == (u64)-1) | |
1296 | cow_start = cur_offset; | |
1297 | cur_offset = extent_end; | |
1298 | if (cur_offset > end) | |
1299 | break; | |
1300 | path->slots[0]++; | |
1301 | goto next_slot; | |
7ea394f1 YZ |
1302 | } |
1303 | ||
b3b4aa74 | 1304 | btrfs_release_path(path); |
80ff3856 | 1305 | if (cow_start != (u64)-1) { |
00361589 JB |
1306 | ret = cow_file_range(inode, locked_page, |
1307 | cow_start, found_key.offset - 1, | |
1308 | page_started, nr_written, 1); | |
d788a349 | 1309 | if (ret) |
79787eaa | 1310 | goto error; |
80ff3856 | 1311 | cow_start = (u64)-1; |
7ea394f1 | 1312 | } |
80ff3856 | 1313 | |
d899e052 YZ |
1314 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1315 | struct extent_map *em; | |
1316 | struct extent_map_tree *em_tree; | |
1317 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1318 | em = alloc_extent_map(); |
79787eaa | 1319 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1320 | em->start = cur_offset; |
70c8a91c | 1321 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1322 | em->len = num_bytes; |
1323 | em->block_len = num_bytes; | |
1324 | em->block_start = disk_bytenr; | |
b4939680 | 1325 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1326 | em->ram_bytes = ram_bytes; |
d899e052 | 1327 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1328 | em->mod_start = em->start; |
1329 | em->mod_len = em->len; | |
d899e052 | 1330 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1331 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1332 | em->generation = -1; |
d899e052 | 1333 | while (1) { |
890871be | 1334 | write_lock(&em_tree->lock); |
09a2a8f9 | 1335 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1336 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1337 | if (ret != -EEXIST) { |
1338 | free_extent_map(em); | |
1339 | break; | |
1340 | } | |
1341 | btrfs_drop_extent_cache(inode, em->start, | |
1342 | em->start + em->len - 1, 0); | |
1343 | } | |
1344 | type = BTRFS_ORDERED_PREALLOC; | |
1345 | } else { | |
1346 | type = BTRFS_ORDERED_NOCOW; | |
1347 | } | |
80ff3856 YZ |
1348 | |
1349 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1350 | num_bytes, num_bytes, type); |
79787eaa | 1351 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1352 | |
efa56464 YZ |
1353 | if (root->root_key.objectid == |
1354 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1355 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1356 | num_bytes); | |
d788a349 | 1357 | if (ret) |
79787eaa | 1358 | goto error; |
efa56464 YZ |
1359 | } |
1360 | ||
c2790a2e JB |
1361 | extent_clear_unlock_delalloc(inode, cur_offset, |
1362 | cur_offset + num_bytes - 1, | |
1363 | locked_page, EXTENT_LOCKED | | |
1364 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1365 | PAGE_SET_PRIVATE2); | |
80ff3856 YZ |
1366 | cur_offset = extent_end; |
1367 | if (cur_offset > end) | |
1368 | break; | |
be20aa9d | 1369 | } |
b3b4aa74 | 1370 | btrfs_release_path(path); |
80ff3856 | 1371 | |
17ca04af | 1372 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1373 | cow_start = cur_offset; |
17ca04af JB |
1374 | cur_offset = end; |
1375 | } | |
1376 | ||
80ff3856 | 1377 | if (cow_start != (u64)-1) { |
00361589 JB |
1378 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1379 | page_started, nr_written, 1); | |
d788a349 | 1380 | if (ret) |
79787eaa | 1381 | goto error; |
80ff3856 YZ |
1382 | } |
1383 | ||
79787eaa | 1384 | error: |
a698d075 | 1385 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1386 | if (!ret) |
1387 | ret = err; | |
1388 | ||
17ca04af | 1389 | if (ret && cur_offset < end) |
c2790a2e JB |
1390 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1391 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1392 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1393 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1394 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1395 | PAGE_SET_WRITEBACK | |
1396 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1397 | btrfs_free_path(path); |
79787eaa | 1398 | return ret; |
be20aa9d CM |
1399 | } |
1400 | ||
d352ac68 CM |
1401 | /* |
1402 | * extent_io.c call back to do delayed allocation processing | |
1403 | */ | |
c8b97818 | 1404 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1405 | u64 start, u64 end, int *page_started, |
1406 | unsigned long *nr_written) | |
be20aa9d | 1407 | { |
be20aa9d | 1408 | int ret; |
7f366cfe | 1409 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1410 | |
7ddf5a42 | 1411 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1412 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1413 | page_started, 1, nr_written); |
7ddf5a42 | 1414 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1415 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1416 | page_started, 0, nr_written); |
7ddf5a42 JB |
1417 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1418 | !(BTRFS_I(inode)->force_compress) && | |
1419 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1420 | ret = cow_file_range(inode, locked_page, start, end, |
1421 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1422 | } else { |
1423 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1424 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1425 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1426 | page_started, nr_written); |
7ddf5a42 | 1427 | } |
b888db2b CM |
1428 | return ret; |
1429 | } | |
1430 | ||
1bf85046 JM |
1431 | static void btrfs_split_extent_hook(struct inode *inode, |
1432 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1433 | { |
0ca1f7ce | 1434 | /* not delalloc, ignore it */ |
9ed74f2d | 1435 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1436 | return; |
9ed74f2d | 1437 | |
9e0baf60 JB |
1438 | spin_lock(&BTRFS_I(inode)->lock); |
1439 | BTRFS_I(inode)->outstanding_extents++; | |
1440 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1441 | } |
1442 | ||
1443 | /* | |
1444 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1445 | * extents so we can keep track of new extents that are just merged onto old | |
1446 | * extents, such as when we are doing sequential writes, so we can properly | |
1447 | * account for the metadata space we'll need. | |
1448 | */ | |
1bf85046 JM |
1449 | static void btrfs_merge_extent_hook(struct inode *inode, |
1450 | struct extent_state *new, | |
1451 | struct extent_state *other) | |
9ed74f2d | 1452 | { |
9ed74f2d JB |
1453 | /* not delalloc, ignore it */ |
1454 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1455 | return; |
9ed74f2d | 1456 | |
9e0baf60 JB |
1457 | spin_lock(&BTRFS_I(inode)->lock); |
1458 | BTRFS_I(inode)->outstanding_extents--; | |
1459 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1460 | } |
1461 | ||
eb73c1b7 MX |
1462 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1463 | struct inode *inode) | |
1464 | { | |
1465 | spin_lock(&root->delalloc_lock); | |
1466 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1467 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1468 | &root->delalloc_inodes); | |
1469 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1470 | &BTRFS_I(inode)->runtime_flags); | |
1471 | root->nr_delalloc_inodes++; | |
1472 | if (root->nr_delalloc_inodes == 1) { | |
1473 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1474 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1475 | list_add_tail(&root->delalloc_root, | |
1476 | &root->fs_info->delalloc_roots); | |
1477 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1478 | } | |
1479 | } | |
1480 | spin_unlock(&root->delalloc_lock); | |
1481 | } | |
1482 | ||
1483 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1484 | struct inode *inode) | |
1485 | { | |
1486 | spin_lock(&root->delalloc_lock); | |
1487 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1488 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1489 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1490 | &BTRFS_I(inode)->runtime_flags); | |
1491 | root->nr_delalloc_inodes--; | |
1492 | if (!root->nr_delalloc_inodes) { | |
1493 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1494 | BUG_ON(list_empty(&root->delalloc_root)); | |
1495 | list_del_init(&root->delalloc_root); | |
1496 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1497 | } | |
1498 | } | |
1499 | spin_unlock(&root->delalloc_lock); | |
1500 | } | |
1501 | ||
d352ac68 CM |
1502 | /* |
1503 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1504 | * bytes in this file, and to maintain the list of inodes that | |
1505 | * have pending delalloc work to be done. | |
1506 | */ | |
1bf85046 | 1507 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1508 | struct extent_state *state, unsigned long *bits) |
291d673e | 1509 | { |
9ed74f2d | 1510 | |
75eff68e CM |
1511 | /* |
1512 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1513 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1514 | * bit, which is only set or cleared with irqs on |
1515 | */ | |
0ca1f7ce | 1516 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1517 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1518 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1519 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1520 | |
9e0baf60 | 1521 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1522 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1523 | } else { |
1524 | spin_lock(&BTRFS_I(inode)->lock); | |
1525 | BTRFS_I(inode)->outstanding_extents++; | |
1526 | spin_unlock(&BTRFS_I(inode)->lock); | |
1527 | } | |
287a0ab9 | 1528 | |
963d678b MX |
1529 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1530 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1531 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1532 | BTRFS_I(inode)->delalloc_bytes += len; |
df0af1a5 | 1533 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1534 | &BTRFS_I(inode)->runtime_flags)) |
1535 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1536 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1537 | } |
291d673e CM |
1538 | } |
1539 | ||
d352ac68 CM |
1540 | /* |
1541 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1542 | */ | |
1bf85046 | 1543 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1544 | struct extent_state *state, |
1545 | unsigned long *bits) | |
291d673e | 1546 | { |
75eff68e CM |
1547 | /* |
1548 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1549 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1550 | * bit, which is only set or cleared with irqs on |
1551 | */ | |
0ca1f7ce | 1552 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1553 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1554 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1555 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1556 | |
9e0baf60 | 1557 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1558 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1559 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1560 | spin_lock(&BTRFS_I(inode)->lock); | |
1561 | BTRFS_I(inode)->outstanding_extents--; | |
1562 | spin_unlock(&BTRFS_I(inode)->lock); | |
1563 | } | |
0ca1f7ce | 1564 | |
b6d08f06 JB |
1565 | /* |
1566 | * We don't reserve metadata space for space cache inodes so we | |
1567 | * don't need to call dellalloc_release_metadata if there is an | |
1568 | * error. | |
1569 | */ | |
1570 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1571 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1572 | btrfs_delalloc_release_metadata(inode, len); |
1573 | ||
0cb59c99 | 1574 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1575 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1576 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1577 | |
963d678b MX |
1578 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1579 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1580 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1581 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1582 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1583 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1584 | &BTRFS_I(inode)->runtime_flags)) |
1585 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1586 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1587 | } |
291d673e CM |
1588 | } |
1589 | ||
d352ac68 CM |
1590 | /* |
1591 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1592 | * we don't create bios that span stripes or chunks | |
1593 | */ | |
64a16701 | 1594 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1595 | size_t size, struct bio *bio, |
1596 | unsigned long bio_flags) | |
239b14b3 CM |
1597 | { |
1598 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1599 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1600 | u64 length = 0; |
1601 | u64 map_length; | |
239b14b3 CM |
1602 | int ret; |
1603 | ||
771ed689 CM |
1604 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1605 | return 0; | |
1606 | ||
f2d8d74d | 1607 | length = bio->bi_size; |
239b14b3 | 1608 | map_length = length; |
64a16701 | 1609 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1610 | &map_length, NULL, 0); |
3ec706c8 | 1611 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1612 | BUG_ON(ret < 0); |
d397712b | 1613 | if (map_length < length + size) |
239b14b3 | 1614 | return 1; |
3444a972 | 1615 | return 0; |
239b14b3 CM |
1616 | } |
1617 | ||
d352ac68 CM |
1618 | /* |
1619 | * in order to insert checksums into the metadata in large chunks, | |
1620 | * we wait until bio submission time. All the pages in the bio are | |
1621 | * checksummed and sums are attached onto the ordered extent record. | |
1622 | * | |
1623 | * At IO completion time the cums attached on the ordered extent record | |
1624 | * are inserted into the btree | |
1625 | */ | |
d397712b CM |
1626 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1627 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1628 | unsigned long bio_flags, |
1629 | u64 bio_offset) | |
065631f6 | 1630 | { |
065631f6 | 1631 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1632 | int ret = 0; |
e015640f | 1633 | |
d20f7043 | 1634 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1635 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1636 | return 0; |
1637 | } | |
e015640f | 1638 | |
4a69a410 CM |
1639 | /* |
1640 | * in order to insert checksums into the metadata in large chunks, | |
1641 | * we wait until bio submission time. All the pages in the bio are | |
1642 | * checksummed and sums are attached onto the ordered extent record. | |
1643 | * | |
1644 | * At IO completion time the cums attached on the ordered extent record | |
1645 | * are inserted into the btree | |
1646 | */ | |
b2950863 | 1647 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1648 | int mirror_num, unsigned long bio_flags, |
1649 | u64 bio_offset) | |
4a69a410 CM |
1650 | { |
1651 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1652 | int ret; |
1653 | ||
1654 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1655 | if (ret) | |
1656 | bio_endio(bio, ret); | |
1657 | return ret; | |
44b8bd7e CM |
1658 | } |
1659 | ||
d352ac68 | 1660 | /* |
cad321ad CM |
1661 | * extent_io.c submission hook. This does the right thing for csum calculation |
1662 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1663 | */ |
b2950863 | 1664 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1665 | int mirror_num, unsigned long bio_flags, |
1666 | u64 bio_offset) | |
44b8bd7e CM |
1667 | { |
1668 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1669 | int ret = 0; | |
19b9bdb0 | 1670 | int skip_sum; |
0417341e | 1671 | int metadata = 0; |
b812ce28 | 1672 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1673 | |
6cbff00f | 1674 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1675 | |
83eea1f1 | 1676 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1677 | metadata = 2; |
1678 | ||
7b6d91da | 1679 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1680 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1681 | if (ret) | |
61891923 | 1682 | goto out; |
5fd02043 | 1683 | |
d20f7043 | 1684 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1685 | ret = btrfs_submit_compressed_read(inode, bio, |
1686 | mirror_num, | |
1687 | bio_flags); | |
1688 | goto out; | |
c2db1073 TI |
1689 | } else if (!skip_sum) { |
1690 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1691 | if (ret) | |
61891923 | 1692 | goto out; |
c2db1073 | 1693 | } |
4d1b5fb4 | 1694 | goto mapit; |
b812ce28 | 1695 | } else if (async && !skip_sum) { |
17d217fe YZ |
1696 | /* csum items have already been cloned */ |
1697 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1698 | goto mapit; | |
19b9bdb0 | 1699 | /* we're doing a write, do the async checksumming */ |
61891923 | 1700 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1701 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1702 | bio_flags, bio_offset, |
1703 | __btrfs_submit_bio_start, | |
4a69a410 | 1704 | __btrfs_submit_bio_done); |
61891923 | 1705 | goto out; |
b812ce28 JB |
1706 | } else if (!skip_sum) { |
1707 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1708 | if (ret) | |
1709 | goto out; | |
19b9bdb0 CM |
1710 | } |
1711 | ||
0b86a832 | 1712 | mapit: |
61891923 SB |
1713 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1714 | ||
1715 | out: | |
1716 | if (ret < 0) | |
1717 | bio_endio(bio, ret); | |
1718 | return ret; | |
065631f6 | 1719 | } |
6885f308 | 1720 | |
d352ac68 CM |
1721 | /* |
1722 | * given a list of ordered sums record them in the inode. This happens | |
1723 | * at IO completion time based on sums calculated at bio submission time. | |
1724 | */ | |
ba1da2f4 | 1725 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1726 | struct inode *inode, u64 file_offset, |
1727 | struct list_head *list) | |
1728 | { | |
e6dcd2dc CM |
1729 | struct btrfs_ordered_sum *sum; |
1730 | ||
c6e30871 | 1731 | list_for_each_entry(sum, list, list) { |
39847c4d | 1732 | trans->adding_csums = 1; |
d20f7043 CM |
1733 | btrfs_csum_file_blocks(trans, |
1734 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1735 | trans->adding_csums = 0; |
e6dcd2dc CM |
1736 | } |
1737 | return 0; | |
1738 | } | |
1739 | ||
2ac55d41 JB |
1740 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1741 | struct extent_state **cached_state) | |
ea8c2819 | 1742 | { |
6c1500f2 | 1743 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1744 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1745 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1746 | } |
1747 | ||
d352ac68 | 1748 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1749 | struct btrfs_writepage_fixup { |
1750 | struct page *page; | |
1751 | struct btrfs_work work; | |
1752 | }; | |
1753 | ||
b2950863 | 1754 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1755 | { |
1756 | struct btrfs_writepage_fixup *fixup; | |
1757 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1758 | struct extent_state *cached_state = NULL; |
247e743c CM |
1759 | struct page *page; |
1760 | struct inode *inode; | |
1761 | u64 page_start; | |
1762 | u64 page_end; | |
87826df0 | 1763 | int ret; |
247e743c CM |
1764 | |
1765 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1766 | page = fixup->page; | |
4a096752 | 1767 | again: |
247e743c CM |
1768 | lock_page(page); |
1769 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1770 | ClearPageChecked(page); | |
1771 | goto out_page; | |
1772 | } | |
1773 | ||
1774 | inode = page->mapping->host; | |
1775 | page_start = page_offset(page); | |
1776 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1777 | ||
2ac55d41 | 1778 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1779 | &cached_state); |
4a096752 CM |
1780 | |
1781 | /* already ordered? We're done */ | |
8b62b72b | 1782 | if (PagePrivate2(page)) |
247e743c | 1783 | goto out; |
4a096752 CM |
1784 | |
1785 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1786 | if (ordered) { | |
2ac55d41 JB |
1787 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1788 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1789 | unlock_page(page); |
1790 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1791 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1792 | goto again; |
1793 | } | |
247e743c | 1794 | |
87826df0 JM |
1795 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1796 | if (ret) { | |
1797 | mapping_set_error(page->mapping, ret); | |
1798 | end_extent_writepage(page, ret, page_start, page_end); | |
1799 | ClearPageChecked(page); | |
1800 | goto out; | |
1801 | } | |
1802 | ||
2ac55d41 | 1803 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1804 | ClearPageChecked(page); |
87826df0 | 1805 | set_page_dirty(page); |
247e743c | 1806 | out: |
2ac55d41 JB |
1807 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1808 | &cached_state, GFP_NOFS); | |
247e743c CM |
1809 | out_page: |
1810 | unlock_page(page); | |
1811 | page_cache_release(page); | |
b897abec | 1812 | kfree(fixup); |
247e743c CM |
1813 | } |
1814 | ||
1815 | /* | |
1816 | * There are a few paths in the higher layers of the kernel that directly | |
1817 | * set the page dirty bit without asking the filesystem if it is a | |
1818 | * good idea. This causes problems because we want to make sure COW | |
1819 | * properly happens and the data=ordered rules are followed. | |
1820 | * | |
c8b97818 | 1821 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1822 | * hasn't been properly setup for IO. We kick off an async process |
1823 | * to fix it up. The async helper will wait for ordered extents, set | |
1824 | * the delalloc bit and make it safe to write the page. | |
1825 | */ | |
b2950863 | 1826 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1827 | { |
1828 | struct inode *inode = page->mapping->host; | |
1829 | struct btrfs_writepage_fixup *fixup; | |
1830 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1831 | |
8b62b72b CM |
1832 | /* this page is properly in the ordered list */ |
1833 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1834 | return 0; |
1835 | ||
1836 | if (PageChecked(page)) | |
1837 | return -EAGAIN; | |
1838 | ||
1839 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1840 | if (!fixup) | |
1841 | return -EAGAIN; | |
f421950f | 1842 | |
247e743c CM |
1843 | SetPageChecked(page); |
1844 | page_cache_get(page); | |
1845 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1846 | fixup->page = page; | |
1847 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1848 | return -EBUSY; |
247e743c CM |
1849 | } |
1850 | ||
d899e052 YZ |
1851 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1852 | struct inode *inode, u64 file_pos, | |
1853 | u64 disk_bytenr, u64 disk_num_bytes, | |
1854 | u64 num_bytes, u64 ram_bytes, | |
1855 | u8 compression, u8 encryption, | |
1856 | u16 other_encoding, int extent_type) | |
1857 | { | |
1858 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1859 | struct btrfs_file_extent_item *fi; | |
1860 | struct btrfs_path *path; | |
1861 | struct extent_buffer *leaf; | |
1862 | struct btrfs_key ins; | |
1acae57b | 1863 | int extent_inserted = 0; |
d899e052 YZ |
1864 | int ret; |
1865 | ||
1866 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1867 | if (!path) |
1868 | return -ENOMEM; | |
d899e052 | 1869 | |
a1ed835e CM |
1870 | /* |
1871 | * we may be replacing one extent in the tree with another. | |
1872 | * The new extent is pinned in the extent map, and we don't want | |
1873 | * to drop it from the cache until it is completely in the btree. | |
1874 | * | |
1875 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1876 | * the caller is expected to unpin it and allow it to be merged | |
1877 | * with the others. | |
1878 | */ | |
1acae57b FDBM |
1879 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
1880 | file_pos + num_bytes, NULL, 0, | |
1881 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
1882 | if (ret) |
1883 | goto out; | |
d899e052 | 1884 | |
1acae57b FDBM |
1885 | if (!extent_inserted) { |
1886 | ins.objectid = btrfs_ino(inode); | |
1887 | ins.offset = file_pos; | |
1888 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1889 | ||
1890 | path->leave_spinning = 1; | |
1891 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
1892 | sizeof(*fi)); | |
1893 | if (ret) | |
1894 | goto out; | |
1895 | } | |
d899e052 YZ |
1896 | leaf = path->nodes[0]; |
1897 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1898 | struct btrfs_file_extent_item); | |
1899 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1900 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1901 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1902 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1903 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1904 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1905 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1906 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1907 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1908 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1909 | |
d899e052 | 1910 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1911 | btrfs_release_path(path); |
d899e052 YZ |
1912 | |
1913 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1914 | |
1915 | ins.objectid = disk_bytenr; | |
1916 | ins.offset = disk_num_bytes; | |
1917 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1918 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1919 | root->root_key.objectid, | |
33345d01 | 1920 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1921 | out: |
d899e052 | 1922 | btrfs_free_path(path); |
b9473439 | 1923 | |
79787eaa | 1924 | return ret; |
d899e052 YZ |
1925 | } |
1926 | ||
38c227d8 LB |
1927 | /* snapshot-aware defrag */ |
1928 | struct sa_defrag_extent_backref { | |
1929 | struct rb_node node; | |
1930 | struct old_sa_defrag_extent *old; | |
1931 | u64 root_id; | |
1932 | u64 inum; | |
1933 | u64 file_pos; | |
1934 | u64 extent_offset; | |
1935 | u64 num_bytes; | |
1936 | u64 generation; | |
1937 | }; | |
1938 | ||
1939 | struct old_sa_defrag_extent { | |
1940 | struct list_head list; | |
1941 | struct new_sa_defrag_extent *new; | |
1942 | ||
1943 | u64 extent_offset; | |
1944 | u64 bytenr; | |
1945 | u64 offset; | |
1946 | u64 len; | |
1947 | int count; | |
1948 | }; | |
1949 | ||
1950 | struct new_sa_defrag_extent { | |
1951 | struct rb_root root; | |
1952 | struct list_head head; | |
1953 | struct btrfs_path *path; | |
1954 | struct inode *inode; | |
1955 | u64 file_pos; | |
1956 | u64 len; | |
1957 | u64 bytenr; | |
1958 | u64 disk_len; | |
1959 | u8 compress_type; | |
1960 | }; | |
1961 | ||
1962 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
1963 | struct sa_defrag_extent_backref *b2) | |
1964 | { | |
1965 | if (b1->root_id < b2->root_id) | |
1966 | return -1; | |
1967 | else if (b1->root_id > b2->root_id) | |
1968 | return 1; | |
1969 | ||
1970 | if (b1->inum < b2->inum) | |
1971 | return -1; | |
1972 | else if (b1->inum > b2->inum) | |
1973 | return 1; | |
1974 | ||
1975 | if (b1->file_pos < b2->file_pos) | |
1976 | return -1; | |
1977 | else if (b1->file_pos > b2->file_pos) | |
1978 | return 1; | |
1979 | ||
1980 | /* | |
1981 | * [------------------------------] ===> (a range of space) | |
1982 | * |<--->| |<---->| =============> (fs/file tree A) | |
1983 | * |<---------------------------->| ===> (fs/file tree B) | |
1984 | * | |
1985 | * A range of space can refer to two file extents in one tree while | |
1986 | * refer to only one file extent in another tree. | |
1987 | * | |
1988 | * So we may process a disk offset more than one time(two extents in A) | |
1989 | * and locate at the same extent(one extent in B), then insert two same | |
1990 | * backrefs(both refer to the extent in B). | |
1991 | */ | |
1992 | return 0; | |
1993 | } | |
1994 | ||
1995 | static void backref_insert(struct rb_root *root, | |
1996 | struct sa_defrag_extent_backref *backref) | |
1997 | { | |
1998 | struct rb_node **p = &root->rb_node; | |
1999 | struct rb_node *parent = NULL; | |
2000 | struct sa_defrag_extent_backref *entry; | |
2001 | int ret; | |
2002 | ||
2003 | while (*p) { | |
2004 | parent = *p; | |
2005 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2006 | ||
2007 | ret = backref_comp(backref, entry); | |
2008 | if (ret < 0) | |
2009 | p = &(*p)->rb_left; | |
2010 | else | |
2011 | p = &(*p)->rb_right; | |
2012 | } | |
2013 | ||
2014 | rb_link_node(&backref->node, parent, p); | |
2015 | rb_insert_color(&backref->node, root); | |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * Note the backref might has changed, and in this case we just return 0. | |
2020 | */ | |
2021 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2022 | void *ctx) | |
2023 | { | |
2024 | struct btrfs_file_extent_item *extent; | |
2025 | struct btrfs_fs_info *fs_info; | |
2026 | struct old_sa_defrag_extent *old = ctx; | |
2027 | struct new_sa_defrag_extent *new = old->new; | |
2028 | struct btrfs_path *path = new->path; | |
2029 | struct btrfs_key key; | |
2030 | struct btrfs_root *root; | |
2031 | struct sa_defrag_extent_backref *backref; | |
2032 | struct extent_buffer *leaf; | |
2033 | struct inode *inode = new->inode; | |
2034 | int slot; | |
2035 | int ret; | |
2036 | u64 extent_offset; | |
2037 | u64 num_bytes; | |
2038 | ||
2039 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2040 | inum == btrfs_ino(inode)) | |
2041 | return 0; | |
2042 | ||
2043 | key.objectid = root_id; | |
2044 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2045 | key.offset = (u64)-1; | |
2046 | ||
2047 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2048 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2049 | if (IS_ERR(root)) { | |
2050 | if (PTR_ERR(root) == -ENOENT) | |
2051 | return 0; | |
2052 | WARN_ON(1); | |
2053 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2054 | inum, offset, root_id); | |
2055 | return PTR_ERR(root); | |
2056 | } | |
2057 | ||
2058 | key.objectid = inum; | |
2059 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2060 | if (offset > (u64)-1 << 32) | |
2061 | key.offset = 0; | |
2062 | else | |
2063 | key.offset = offset; | |
2064 | ||
2065 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2066 | if (WARN_ON(ret < 0)) |
38c227d8 | 2067 | return ret; |
50f1319c | 2068 | ret = 0; |
38c227d8 LB |
2069 | |
2070 | while (1) { | |
2071 | cond_resched(); | |
2072 | ||
2073 | leaf = path->nodes[0]; | |
2074 | slot = path->slots[0]; | |
2075 | ||
2076 | if (slot >= btrfs_header_nritems(leaf)) { | |
2077 | ret = btrfs_next_leaf(root, path); | |
2078 | if (ret < 0) { | |
2079 | goto out; | |
2080 | } else if (ret > 0) { | |
2081 | ret = 0; | |
2082 | goto out; | |
2083 | } | |
2084 | continue; | |
2085 | } | |
2086 | ||
2087 | path->slots[0]++; | |
2088 | ||
2089 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2090 | ||
2091 | if (key.objectid > inum) | |
2092 | goto out; | |
2093 | ||
2094 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2095 | continue; | |
2096 | ||
2097 | extent = btrfs_item_ptr(leaf, slot, | |
2098 | struct btrfs_file_extent_item); | |
2099 | ||
2100 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2101 | continue; | |
2102 | ||
e68afa49 LB |
2103 | /* |
2104 | * 'offset' refers to the exact key.offset, | |
2105 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2106 | * (key.offset - extent_offset). | |
2107 | */ | |
2108 | if (key.offset != offset) | |
38c227d8 LB |
2109 | continue; |
2110 | ||
e68afa49 | 2111 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2112 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2113 | |
38c227d8 LB |
2114 | if (extent_offset >= old->extent_offset + old->offset + |
2115 | old->len || extent_offset + num_bytes <= | |
2116 | old->extent_offset + old->offset) | |
2117 | continue; | |
38c227d8 LB |
2118 | break; |
2119 | } | |
2120 | ||
2121 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2122 | if (!backref) { | |
2123 | ret = -ENOENT; | |
2124 | goto out; | |
2125 | } | |
2126 | ||
2127 | backref->root_id = root_id; | |
2128 | backref->inum = inum; | |
e68afa49 | 2129 | backref->file_pos = offset; |
38c227d8 LB |
2130 | backref->num_bytes = num_bytes; |
2131 | backref->extent_offset = extent_offset; | |
2132 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2133 | backref->old = old; | |
2134 | backref_insert(&new->root, backref); | |
2135 | old->count++; | |
2136 | out: | |
2137 | btrfs_release_path(path); | |
2138 | WARN_ON(ret); | |
2139 | return ret; | |
2140 | } | |
2141 | ||
2142 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2143 | struct new_sa_defrag_extent *new) | |
2144 | { | |
2145 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2146 | struct old_sa_defrag_extent *old, *tmp; | |
2147 | int ret; | |
2148 | ||
2149 | new->path = path; | |
2150 | ||
2151 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2152 | ret = iterate_inodes_from_logical(old->bytenr + |
2153 | old->extent_offset, fs_info, | |
38c227d8 LB |
2154 | path, record_one_backref, |
2155 | old); | |
4724b106 JB |
2156 | if (ret < 0 && ret != -ENOENT) |
2157 | return false; | |
38c227d8 LB |
2158 | |
2159 | /* no backref to be processed for this extent */ | |
2160 | if (!old->count) { | |
2161 | list_del(&old->list); | |
2162 | kfree(old); | |
2163 | } | |
2164 | } | |
2165 | ||
2166 | if (list_empty(&new->head)) | |
2167 | return false; | |
2168 | ||
2169 | return true; | |
2170 | } | |
2171 | ||
2172 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2173 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2174 | struct new_sa_defrag_extent *new) |
38c227d8 | 2175 | { |
116e0024 | 2176 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2177 | return 0; |
2178 | ||
2179 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2180 | return 0; | |
2181 | ||
116e0024 LB |
2182 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2183 | return 0; | |
2184 | ||
2185 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2186 | btrfs_file_extent_other_encoding(leaf, fi)) |
2187 | return 0; | |
2188 | ||
2189 | return 1; | |
2190 | } | |
2191 | ||
2192 | /* | |
2193 | * Note the backref might has changed, and in this case we just return 0. | |
2194 | */ | |
2195 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2196 | struct sa_defrag_extent_backref *prev, | |
2197 | struct sa_defrag_extent_backref *backref) | |
2198 | { | |
2199 | struct btrfs_file_extent_item *extent; | |
2200 | struct btrfs_file_extent_item *item; | |
2201 | struct btrfs_ordered_extent *ordered; | |
2202 | struct btrfs_trans_handle *trans; | |
2203 | struct btrfs_fs_info *fs_info; | |
2204 | struct btrfs_root *root; | |
2205 | struct btrfs_key key; | |
2206 | struct extent_buffer *leaf; | |
2207 | struct old_sa_defrag_extent *old = backref->old; | |
2208 | struct new_sa_defrag_extent *new = old->new; | |
2209 | struct inode *src_inode = new->inode; | |
2210 | struct inode *inode; | |
2211 | struct extent_state *cached = NULL; | |
2212 | int ret = 0; | |
2213 | u64 start; | |
2214 | u64 len; | |
2215 | u64 lock_start; | |
2216 | u64 lock_end; | |
2217 | bool merge = false; | |
2218 | int index; | |
2219 | ||
2220 | if (prev && prev->root_id == backref->root_id && | |
2221 | prev->inum == backref->inum && | |
2222 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2223 | merge = true; | |
2224 | ||
2225 | /* step 1: get root */ | |
2226 | key.objectid = backref->root_id; | |
2227 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2228 | key.offset = (u64)-1; | |
2229 | ||
2230 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2231 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2232 | ||
2233 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2234 | if (IS_ERR(root)) { | |
2235 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2236 | if (PTR_ERR(root) == -ENOENT) | |
2237 | return 0; | |
2238 | return PTR_ERR(root); | |
2239 | } | |
38c227d8 LB |
2240 | |
2241 | /* step 2: get inode */ | |
2242 | key.objectid = backref->inum; | |
2243 | key.type = BTRFS_INODE_ITEM_KEY; | |
2244 | key.offset = 0; | |
2245 | ||
2246 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2247 | if (IS_ERR(inode)) { | |
2248 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2249 | return 0; | |
2250 | } | |
2251 | ||
2252 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2253 | ||
2254 | /* step 3: relink backref */ | |
2255 | lock_start = backref->file_pos; | |
2256 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2257 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2258 | 0, &cached); | |
2259 | ||
2260 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2261 | if (ordered) { | |
2262 | btrfs_put_ordered_extent(ordered); | |
2263 | goto out_unlock; | |
2264 | } | |
2265 | ||
2266 | trans = btrfs_join_transaction(root); | |
2267 | if (IS_ERR(trans)) { | |
2268 | ret = PTR_ERR(trans); | |
2269 | goto out_unlock; | |
2270 | } | |
2271 | ||
2272 | key.objectid = backref->inum; | |
2273 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2274 | key.offset = backref->file_pos; | |
2275 | ||
2276 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2277 | if (ret < 0) { | |
2278 | goto out_free_path; | |
2279 | } else if (ret > 0) { | |
2280 | ret = 0; | |
2281 | goto out_free_path; | |
2282 | } | |
2283 | ||
2284 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2285 | struct btrfs_file_extent_item); | |
2286 | ||
2287 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2288 | backref->generation) | |
2289 | goto out_free_path; | |
2290 | ||
2291 | btrfs_release_path(path); | |
2292 | ||
2293 | start = backref->file_pos; | |
2294 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2295 | start += old->extent_offset + old->offset - | |
2296 | backref->extent_offset; | |
2297 | ||
2298 | len = min(backref->extent_offset + backref->num_bytes, | |
2299 | old->extent_offset + old->offset + old->len); | |
2300 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2301 | ||
2302 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2303 | start + len, 1); | |
2304 | if (ret) | |
2305 | goto out_free_path; | |
2306 | again: | |
2307 | key.objectid = btrfs_ino(inode); | |
2308 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2309 | key.offset = start; | |
2310 | ||
a09a0a70 | 2311 | path->leave_spinning = 1; |
38c227d8 LB |
2312 | if (merge) { |
2313 | struct btrfs_file_extent_item *fi; | |
2314 | u64 extent_len; | |
2315 | struct btrfs_key found_key; | |
2316 | ||
3c9665df | 2317 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2318 | if (ret < 0) |
2319 | goto out_free_path; | |
2320 | ||
2321 | path->slots[0]--; | |
2322 | leaf = path->nodes[0]; | |
2323 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2324 | ||
2325 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2326 | struct btrfs_file_extent_item); | |
2327 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2328 | ||
116e0024 LB |
2329 | if (extent_len + found_key.offset == start && |
2330 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2331 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2332 | extent_len + len); | |
2333 | btrfs_mark_buffer_dirty(leaf); | |
2334 | inode_add_bytes(inode, len); | |
2335 | ||
2336 | ret = 1; | |
2337 | goto out_free_path; | |
2338 | } else { | |
2339 | merge = false; | |
2340 | btrfs_release_path(path); | |
2341 | goto again; | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2346 | sizeof(*extent)); | |
2347 | if (ret) { | |
2348 | btrfs_abort_transaction(trans, root, ret); | |
2349 | goto out_free_path; | |
2350 | } | |
2351 | ||
2352 | leaf = path->nodes[0]; | |
2353 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2354 | struct btrfs_file_extent_item); | |
2355 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2356 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2357 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2358 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2359 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2360 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2361 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2362 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2363 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2364 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2365 | ||
2366 | btrfs_mark_buffer_dirty(leaf); | |
2367 | inode_add_bytes(inode, len); | |
a09a0a70 | 2368 | btrfs_release_path(path); |
38c227d8 LB |
2369 | |
2370 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2371 | new->disk_len, 0, | |
2372 | backref->root_id, backref->inum, | |
2373 | new->file_pos, 0); /* start - extent_offset */ | |
2374 | if (ret) { | |
2375 | btrfs_abort_transaction(trans, root, ret); | |
2376 | goto out_free_path; | |
2377 | } | |
2378 | ||
2379 | ret = 1; | |
2380 | out_free_path: | |
2381 | btrfs_release_path(path); | |
a09a0a70 | 2382 | path->leave_spinning = 0; |
38c227d8 LB |
2383 | btrfs_end_transaction(trans, root); |
2384 | out_unlock: | |
2385 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2386 | &cached, GFP_NOFS); | |
2387 | iput(inode); | |
2388 | return ret; | |
2389 | } | |
2390 | ||
6f519564 LB |
2391 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2392 | { | |
2393 | struct old_sa_defrag_extent *old, *tmp; | |
2394 | ||
2395 | if (!new) | |
2396 | return; | |
2397 | ||
2398 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2399 | list_del(&old->list); | |
2400 | kfree(old); | |
2401 | } | |
2402 | kfree(new); | |
2403 | } | |
2404 | ||
38c227d8 LB |
2405 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2406 | { | |
2407 | struct btrfs_path *path; | |
38c227d8 LB |
2408 | struct sa_defrag_extent_backref *backref; |
2409 | struct sa_defrag_extent_backref *prev = NULL; | |
2410 | struct inode *inode; | |
2411 | struct btrfs_root *root; | |
2412 | struct rb_node *node; | |
2413 | int ret; | |
2414 | ||
2415 | inode = new->inode; | |
2416 | root = BTRFS_I(inode)->root; | |
2417 | ||
2418 | path = btrfs_alloc_path(); | |
2419 | if (!path) | |
2420 | return; | |
2421 | ||
2422 | if (!record_extent_backrefs(path, new)) { | |
2423 | btrfs_free_path(path); | |
2424 | goto out; | |
2425 | } | |
2426 | btrfs_release_path(path); | |
2427 | ||
2428 | while (1) { | |
2429 | node = rb_first(&new->root); | |
2430 | if (!node) | |
2431 | break; | |
2432 | rb_erase(node, &new->root); | |
2433 | ||
2434 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2435 | ||
2436 | ret = relink_extent_backref(path, prev, backref); | |
2437 | WARN_ON(ret < 0); | |
2438 | ||
2439 | kfree(prev); | |
2440 | ||
2441 | if (ret == 1) | |
2442 | prev = backref; | |
2443 | else | |
2444 | prev = NULL; | |
2445 | cond_resched(); | |
2446 | } | |
2447 | kfree(prev); | |
2448 | ||
2449 | btrfs_free_path(path); | |
38c227d8 | 2450 | out: |
6f519564 LB |
2451 | free_sa_defrag_extent(new); |
2452 | ||
38c227d8 LB |
2453 | atomic_dec(&root->fs_info->defrag_running); |
2454 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2455 | } |
2456 | ||
2457 | static struct new_sa_defrag_extent * | |
2458 | record_old_file_extents(struct inode *inode, | |
2459 | struct btrfs_ordered_extent *ordered) | |
2460 | { | |
2461 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2462 | struct btrfs_path *path; | |
2463 | struct btrfs_key key; | |
6f519564 | 2464 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2465 | struct new_sa_defrag_extent *new; |
2466 | int ret; | |
2467 | ||
2468 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2469 | if (!new) | |
2470 | return NULL; | |
2471 | ||
2472 | new->inode = inode; | |
2473 | new->file_pos = ordered->file_offset; | |
2474 | new->len = ordered->len; | |
2475 | new->bytenr = ordered->start; | |
2476 | new->disk_len = ordered->disk_len; | |
2477 | new->compress_type = ordered->compress_type; | |
2478 | new->root = RB_ROOT; | |
2479 | INIT_LIST_HEAD(&new->head); | |
2480 | ||
2481 | path = btrfs_alloc_path(); | |
2482 | if (!path) | |
2483 | goto out_kfree; | |
2484 | ||
2485 | key.objectid = btrfs_ino(inode); | |
2486 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2487 | key.offset = new->file_pos; | |
2488 | ||
2489 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2490 | if (ret < 0) | |
2491 | goto out_free_path; | |
2492 | if (ret > 0 && path->slots[0] > 0) | |
2493 | path->slots[0]--; | |
2494 | ||
2495 | /* find out all the old extents for the file range */ | |
2496 | while (1) { | |
2497 | struct btrfs_file_extent_item *extent; | |
2498 | struct extent_buffer *l; | |
2499 | int slot; | |
2500 | u64 num_bytes; | |
2501 | u64 offset; | |
2502 | u64 end; | |
2503 | u64 disk_bytenr; | |
2504 | u64 extent_offset; | |
2505 | ||
2506 | l = path->nodes[0]; | |
2507 | slot = path->slots[0]; | |
2508 | ||
2509 | if (slot >= btrfs_header_nritems(l)) { | |
2510 | ret = btrfs_next_leaf(root, path); | |
2511 | if (ret < 0) | |
6f519564 | 2512 | goto out_free_path; |
38c227d8 LB |
2513 | else if (ret > 0) |
2514 | break; | |
2515 | continue; | |
2516 | } | |
2517 | ||
2518 | btrfs_item_key_to_cpu(l, &key, slot); | |
2519 | ||
2520 | if (key.objectid != btrfs_ino(inode)) | |
2521 | break; | |
2522 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2523 | break; | |
2524 | if (key.offset >= new->file_pos + new->len) | |
2525 | break; | |
2526 | ||
2527 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2528 | ||
2529 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2530 | if (key.offset + num_bytes < new->file_pos) | |
2531 | goto next; | |
2532 | ||
2533 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2534 | if (!disk_bytenr) | |
2535 | goto next; | |
2536 | ||
2537 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2538 | ||
2539 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2540 | if (!old) | |
6f519564 | 2541 | goto out_free_path; |
38c227d8 LB |
2542 | |
2543 | offset = max(new->file_pos, key.offset); | |
2544 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2545 | ||
2546 | old->bytenr = disk_bytenr; | |
2547 | old->extent_offset = extent_offset; | |
2548 | old->offset = offset - key.offset; | |
2549 | old->len = end - offset; | |
2550 | old->new = new; | |
2551 | old->count = 0; | |
2552 | list_add_tail(&old->list, &new->head); | |
2553 | next: | |
2554 | path->slots[0]++; | |
2555 | cond_resched(); | |
2556 | } | |
2557 | ||
2558 | btrfs_free_path(path); | |
2559 | atomic_inc(&root->fs_info->defrag_running); | |
2560 | ||
2561 | return new; | |
2562 | ||
38c227d8 LB |
2563 | out_free_path: |
2564 | btrfs_free_path(path); | |
2565 | out_kfree: | |
6f519564 | 2566 | free_sa_defrag_extent(new); |
38c227d8 LB |
2567 | return NULL; |
2568 | } | |
2569 | ||
d352ac68 CM |
2570 | /* as ordered data IO finishes, this gets called so we can finish |
2571 | * an ordered extent if the range of bytes in the file it covers are | |
2572 | * fully written. | |
2573 | */ | |
5fd02043 | 2574 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2575 | { |
5fd02043 | 2576 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2577 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2578 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2579 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2580 | struct extent_state *cached_state = NULL; |
38c227d8 | 2581 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2582 | int compress_type = 0; |
77cef2ec JB |
2583 | int ret = 0; |
2584 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2585 | bool nolock; |
77cef2ec | 2586 | bool truncated = false; |
e6dcd2dc | 2587 | |
83eea1f1 | 2588 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2589 | |
5fd02043 JB |
2590 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2591 | ret = -EIO; | |
2592 | goto out; | |
2593 | } | |
2594 | ||
77cef2ec JB |
2595 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2596 | truncated = true; | |
2597 | logical_len = ordered_extent->truncated_len; | |
2598 | /* Truncated the entire extent, don't bother adding */ | |
2599 | if (!logical_len) | |
2600 | goto out; | |
2601 | } | |
2602 | ||
c2167754 | 2603 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2604 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2605 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2606 | if (nolock) | |
2607 | trans = btrfs_join_transaction_nolock(root); | |
2608 | else | |
2609 | trans = btrfs_join_transaction(root); | |
2610 | if (IS_ERR(trans)) { | |
2611 | ret = PTR_ERR(trans); | |
2612 | trans = NULL; | |
2613 | goto out; | |
c2167754 | 2614 | } |
6c760c07 JB |
2615 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2616 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2617 | if (ret) /* -ENOMEM or corruption */ | |
2618 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2619 | goto out; |
2620 | } | |
e6dcd2dc | 2621 | |
2ac55d41 JB |
2622 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2623 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2624 | 0, &cached_state); |
e6dcd2dc | 2625 | |
38c227d8 LB |
2626 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2627 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2628 | EXTENT_DEFRAG, 1, cached_state); | |
2629 | if (ret) { | |
2630 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
2631 | if (last_snapshot >= BTRFS_I(inode)->generation) | |
2632 | /* the inode is shared */ | |
2633 | new = record_old_file_extents(inode, ordered_extent); | |
2634 | ||
2635 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2636 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2637 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2638 | } | |
2639 | ||
0cb59c99 | 2640 | if (nolock) |
7a7eaa40 | 2641 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2642 | else |
7a7eaa40 | 2643 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2644 | if (IS_ERR(trans)) { |
2645 | ret = PTR_ERR(trans); | |
2646 | trans = NULL; | |
2647 | goto out_unlock; | |
2648 | } | |
0ca1f7ce | 2649 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2650 | |
c8b97818 | 2651 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2652 | compress_type = ordered_extent->compress_type; |
d899e052 | 2653 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2654 | BUG_ON(compress_type); |
920bbbfb | 2655 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2656 | ordered_extent->file_offset, |
2657 | ordered_extent->file_offset + | |
77cef2ec | 2658 | logical_len); |
d899e052 | 2659 | } else { |
0af3d00b | 2660 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2661 | ret = insert_reserved_file_extent(trans, inode, |
2662 | ordered_extent->file_offset, | |
2663 | ordered_extent->start, | |
2664 | ordered_extent->disk_len, | |
77cef2ec | 2665 | logical_len, logical_len, |
261507a0 | 2666 | compress_type, 0, 0, |
d899e052 | 2667 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 2668 | } |
5dc562c5 JB |
2669 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2670 | ordered_extent->file_offset, ordered_extent->len, | |
2671 | trans->transid); | |
79787eaa JM |
2672 | if (ret < 0) { |
2673 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2674 | goto out_unlock; |
79787eaa | 2675 | } |
2ac55d41 | 2676 | |
e6dcd2dc CM |
2677 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2678 | &ordered_extent->list); | |
2679 | ||
6c760c07 JB |
2680 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2681 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2682 | if (ret) { /* -ENOMEM or corruption */ | |
2683 | btrfs_abort_transaction(trans, root, ret); | |
2684 | goto out_unlock; | |
1ef30be1 JB |
2685 | } |
2686 | ret = 0; | |
5fd02043 JB |
2687 | out_unlock: |
2688 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2689 | ordered_extent->file_offset + | |
2690 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2691 | out: |
5b0e95bf | 2692 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2693 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2694 | if (trans) |
2695 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2696 | |
77cef2ec JB |
2697 | if (ret || truncated) { |
2698 | u64 start, end; | |
2699 | ||
2700 | if (truncated) | |
2701 | start = ordered_extent->file_offset + logical_len; | |
2702 | else | |
2703 | start = ordered_extent->file_offset; | |
2704 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2705 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2706 | ||
2707 | /* Drop the cache for the part of the extent we didn't write. */ | |
2708 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2709 | |
0bec9ef5 JB |
2710 | /* |
2711 | * If the ordered extent had an IOERR or something else went | |
2712 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2713 | * back to the allocator. We only free the extent in the |
2714 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2715 | */ |
77cef2ec JB |
2716 | if ((ret || !logical_len) && |
2717 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2718 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2719 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2720 | ordered_extent->disk_len); | |
2721 | } | |
2722 | ||
2723 | ||
5fd02043 | 2724 | /* |
8bad3c02 LB |
2725 | * This needs to be done to make sure anybody waiting knows we are done |
2726 | * updating everything for this ordered extent. | |
5fd02043 JB |
2727 | */ |
2728 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2729 | ||
38c227d8 | 2730 | /* for snapshot-aware defrag */ |
6f519564 LB |
2731 | if (new) { |
2732 | if (ret) { | |
2733 | free_sa_defrag_extent(new); | |
2734 | atomic_dec(&root->fs_info->defrag_running); | |
2735 | } else { | |
2736 | relink_file_extents(new); | |
2737 | } | |
2738 | } | |
38c227d8 | 2739 | |
e6dcd2dc CM |
2740 | /* once for us */ |
2741 | btrfs_put_ordered_extent(ordered_extent); | |
2742 | /* once for the tree */ | |
2743 | btrfs_put_ordered_extent(ordered_extent); | |
2744 | ||
5fd02043 JB |
2745 | return ret; |
2746 | } | |
2747 | ||
2748 | static void finish_ordered_fn(struct btrfs_work *work) | |
2749 | { | |
2750 | struct btrfs_ordered_extent *ordered_extent; | |
2751 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2752 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2753 | } |
2754 | ||
b2950863 | 2755 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2756 | struct extent_state *state, int uptodate) |
2757 | { | |
5fd02043 JB |
2758 | struct inode *inode = page->mapping->host; |
2759 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2760 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2761 | struct btrfs_workers *workers; | |
2762 | ||
1abe9b8a | 2763 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2764 | ||
8b62b72b | 2765 | ClearPagePrivate2(page); |
5fd02043 JB |
2766 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2767 | end - start + 1, uptodate)) | |
2768 | return 0; | |
2769 | ||
2770 | ordered_extent->work.func = finish_ordered_fn; | |
2771 | ordered_extent->work.flags = 0; | |
2772 | ||
83eea1f1 | 2773 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2774 | workers = &root->fs_info->endio_freespace_worker; |
2775 | else | |
2776 | workers = &root->fs_info->endio_write_workers; | |
2777 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2778 | ||
2779 | return 0; | |
211f90e6 CM |
2780 | } |
2781 | ||
d352ac68 CM |
2782 | /* |
2783 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2784 | * if there's a match, we allow the bio to finish. If not, the code in |
2785 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2786 | */ |
facc8a22 MX |
2787 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2788 | u64 phy_offset, struct page *page, | |
2789 | u64 start, u64 end, int mirror) | |
07157aac | 2790 | { |
4eee4fa4 | 2791 | size_t offset = start - page_offset(page); |
07157aac | 2792 | struct inode *inode = page->mapping->host; |
d1310b2e | 2793 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2794 | char *kaddr; |
ff79f819 | 2795 | struct btrfs_root *root = BTRFS_I(inode)->root; |
facc8a22 | 2796 | u32 csum_expected; |
ff79f819 | 2797 | u32 csum = ~(u32)0; |
c2cf52eb SK |
2798 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
2799 | DEFAULT_RATELIMIT_BURST); | |
d1310b2e | 2800 | |
d20f7043 CM |
2801 | if (PageChecked(page)) { |
2802 | ClearPageChecked(page); | |
2803 | goto good; | |
2804 | } | |
6cbff00f CH |
2805 | |
2806 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2807 | goto good; |
17d217fe YZ |
2808 | |
2809 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2810 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2811 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2812 | GFP_NOFS); | |
b6cda9bc | 2813 | return 0; |
17d217fe | 2814 | } |
d20f7043 | 2815 | |
facc8a22 MX |
2816 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
2817 | csum_expected = *(((u32 *)io_bio->csum) + phy_offset); | |
d397712b | 2818 | |
facc8a22 | 2819 | kaddr = kmap_atomic(page); |
b0496686 | 2820 | csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1); |
ff79f819 | 2821 | btrfs_csum_final(csum, (char *)&csum); |
facc8a22 | 2822 | if (csum != csum_expected) |
07157aac | 2823 | goto zeroit; |
d397712b | 2824 | |
7ac687d9 | 2825 | kunmap_atomic(kaddr); |
d20f7043 | 2826 | good: |
07157aac CM |
2827 | return 0; |
2828 | ||
2829 | zeroit: | |
c2cf52eb | 2830 | if (__ratelimit(&_rs)) |
facc8a22 | 2831 | btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", |
c1c9ff7c | 2832 | btrfs_ino(page->mapping->host), start, csum, csum_expected); |
db94535d CM |
2833 | memset(kaddr + offset, 1, end - start + 1); |
2834 | flush_dcache_page(page); | |
7ac687d9 | 2835 | kunmap_atomic(kaddr); |
facc8a22 | 2836 | if (csum_expected == 0) |
3b951516 | 2837 | return 0; |
7e38326f | 2838 | return -EIO; |
07157aac | 2839 | } |
b888db2b | 2840 | |
24bbcf04 YZ |
2841 | struct delayed_iput { |
2842 | struct list_head list; | |
2843 | struct inode *inode; | |
2844 | }; | |
2845 | ||
79787eaa JM |
2846 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2847 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2848 | void btrfs_add_delayed_iput(struct inode *inode) |
2849 | { | |
2850 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2851 | struct delayed_iput *delayed; | |
2852 | ||
2853 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2854 | return; | |
2855 | ||
2856 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2857 | delayed->inode = inode; | |
2858 | ||
2859 | spin_lock(&fs_info->delayed_iput_lock); | |
2860 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2861 | spin_unlock(&fs_info->delayed_iput_lock); | |
2862 | } | |
2863 | ||
2864 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2865 | { | |
2866 | LIST_HEAD(list); | |
2867 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2868 | struct delayed_iput *delayed; | |
2869 | int empty; | |
2870 | ||
2871 | spin_lock(&fs_info->delayed_iput_lock); | |
2872 | empty = list_empty(&fs_info->delayed_iputs); | |
2873 | spin_unlock(&fs_info->delayed_iput_lock); | |
2874 | if (empty) | |
2875 | return; | |
2876 | ||
24bbcf04 YZ |
2877 | spin_lock(&fs_info->delayed_iput_lock); |
2878 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2879 | spin_unlock(&fs_info->delayed_iput_lock); | |
2880 | ||
2881 | while (!list_empty(&list)) { | |
2882 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2883 | list_del(&delayed->list); | |
2884 | iput(delayed->inode); | |
2885 | kfree(delayed); | |
2886 | } | |
24bbcf04 YZ |
2887 | } |
2888 | ||
d68fc57b | 2889 | /* |
42b2aa86 | 2890 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2891 | * files in the subvolume, it removes orphan item and frees block_rsv |
2892 | * structure. | |
2893 | */ | |
2894 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2895 | struct btrfs_root *root) | |
2896 | { | |
90290e19 | 2897 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2898 | int ret; |
2899 | ||
8a35d95f | 2900 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2901 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2902 | return; | |
2903 | ||
90290e19 | 2904 | spin_lock(&root->orphan_lock); |
8a35d95f | 2905 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2906 | spin_unlock(&root->orphan_lock); |
2907 | return; | |
2908 | } | |
2909 | ||
2910 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2911 | spin_unlock(&root->orphan_lock); | |
2912 | return; | |
2913 | } | |
2914 | ||
2915 | block_rsv = root->orphan_block_rsv; | |
2916 | root->orphan_block_rsv = NULL; | |
2917 | spin_unlock(&root->orphan_lock); | |
2918 | ||
d68fc57b YZ |
2919 | if (root->orphan_item_inserted && |
2920 | btrfs_root_refs(&root->root_item) > 0) { | |
2921 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2922 | root->root_key.objectid); | |
4ef31a45 JB |
2923 | if (ret) |
2924 | btrfs_abort_transaction(trans, root, ret); | |
2925 | else | |
2926 | root->orphan_item_inserted = 0; | |
d68fc57b YZ |
2927 | } |
2928 | ||
90290e19 JB |
2929 | if (block_rsv) { |
2930 | WARN_ON(block_rsv->size > 0); | |
2931 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2932 | } |
2933 | } | |
2934 | ||
7b128766 JB |
2935 | /* |
2936 | * This creates an orphan entry for the given inode in case something goes | |
2937 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2938 | * |
2939 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2940 | * this function. | |
7b128766 JB |
2941 | */ |
2942 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2943 | { | |
2944 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2945 | struct btrfs_block_rsv *block_rsv = NULL; |
2946 | int reserve = 0; | |
2947 | int insert = 0; | |
2948 | int ret; | |
7b128766 | 2949 | |
d68fc57b | 2950 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2951 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2952 | if (!block_rsv) |
2953 | return -ENOMEM; | |
d68fc57b | 2954 | } |
7b128766 | 2955 | |
d68fc57b YZ |
2956 | spin_lock(&root->orphan_lock); |
2957 | if (!root->orphan_block_rsv) { | |
2958 | root->orphan_block_rsv = block_rsv; | |
2959 | } else if (block_rsv) { | |
2960 | btrfs_free_block_rsv(root, block_rsv); | |
2961 | block_rsv = NULL; | |
7b128766 | 2962 | } |
7b128766 | 2963 | |
8a35d95f JB |
2964 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2965 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2966 | #if 0 |
2967 | /* | |
2968 | * For proper ENOSPC handling, we should do orphan | |
2969 | * cleanup when mounting. But this introduces backward | |
2970 | * compatibility issue. | |
2971 | */ | |
2972 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2973 | insert = 2; | |
2974 | else | |
2975 | insert = 1; | |
2976 | #endif | |
2977 | insert = 1; | |
321f0e70 | 2978 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2979 | } |
2980 | ||
72ac3c0d JB |
2981 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2982 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2983 | reserve = 1; |
d68fc57b | 2984 | spin_unlock(&root->orphan_lock); |
7b128766 | 2985 | |
d68fc57b YZ |
2986 | /* grab metadata reservation from transaction handle */ |
2987 | if (reserve) { | |
2988 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2989 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2990 | } |
7b128766 | 2991 | |
d68fc57b YZ |
2992 | /* insert an orphan item to track this unlinked/truncated file */ |
2993 | if (insert >= 1) { | |
33345d01 | 2994 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 2995 | if (ret) { |
703c88e0 | 2996 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
2997 | if (reserve) { |
2998 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
2999 | &BTRFS_I(inode)->runtime_flags); | |
3000 | btrfs_orphan_release_metadata(inode); | |
3001 | } | |
3002 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3003 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3004 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3005 | btrfs_abort_transaction(trans, root, ret); |
3006 | return ret; | |
3007 | } | |
79787eaa JM |
3008 | } |
3009 | ret = 0; | |
d68fc57b YZ |
3010 | } |
3011 | ||
3012 | /* insert an orphan item to track subvolume contains orphan files */ | |
3013 | if (insert >= 2) { | |
3014 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3015 | root->root_key.objectid); | |
79787eaa JM |
3016 | if (ret && ret != -EEXIST) { |
3017 | btrfs_abort_transaction(trans, root, ret); | |
3018 | return ret; | |
3019 | } | |
d68fc57b YZ |
3020 | } |
3021 | return 0; | |
7b128766 JB |
3022 | } |
3023 | ||
3024 | /* | |
3025 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3026 | * item for this particular inode. | |
3027 | */ | |
48a3b636 ES |
3028 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3029 | struct inode *inode) | |
7b128766 JB |
3030 | { |
3031 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3032 | int delete_item = 0; |
3033 | int release_rsv = 0; | |
7b128766 JB |
3034 | int ret = 0; |
3035 | ||
d68fc57b | 3036 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3037 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3038 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3039 | delete_item = 1; |
7b128766 | 3040 | |
72ac3c0d JB |
3041 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3042 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3043 | release_rsv = 1; |
d68fc57b | 3044 | spin_unlock(&root->orphan_lock); |
7b128766 | 3045 | |
703c88e0 | 3046 | if (delete_item) { |
8a35d95f | 3047 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3048 | if (trans) |
3049 | ret = btrfs_del_orphan_item(trans, root, | |
3050 | btrfs_ino(inode)); | |
8a35d95f | 3051 | } |
7b128766 | 3052 | |
703c88e0 FDBM |
3053 | if (release_rsv) |
3054 | btrfs_orphan_release_metadata(inode); | |
3055 | ||
4ef31a45 | 3056 | return ret; |
7b128766 JB |
3057 | } |
3058 | ||
3059 | /* | |
3060 | * this cleans up any orphans that may be left on the list from the last use | |
3061 | * of this root. | |
3062 | */ | |
66b4ffd1 | 3063 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3064 | { |
3065 | struct btrfs_path *path; | |
3066 | struct extent_buffer *leaf; | |
7b128766 JB |
3067 | struct btrfs_key key, found_key; |
3068 | struct btrfs_trans_handle *trans; | |
3069 | struct inode *inode; | |
8f6d7f4f | 3070 | u64 last_objectid = 0; |
7b128766 JB |
3071 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3072 | ||
d68fc57b | 3073 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3074 | return 0; |
c71bf099 YZ |
3075 | |
3076 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3077 | if (!path) { |
3078 | ret = -ENOMEM; | |
3079 | goto out; | |
3080 | } | |
7b128766 JB |
3081 | path->reada = -1; |
3082 | ||
3083 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
3084 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
3085 | key.offset = (u64)-1; | |
3086 | ||
7b128766 JB |
3087 | while (1) { |
3088 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3089 | if (ret < 0) |
3090 | goto out; | |
7b128766 JB |
3091 | |
3092 | /* | |
3093 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3094 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3095 | * find the key and see if we have stuff that matches |
3096 | */ | |
3097 | if (ret > 0) { | |
66b4ffd1 | 3098 | ret = 0; |
7b128766 JB |
3099 | if (path->slots[0] == 0) |
3100 | break; | |
3101 | path->slots[0]--; | |
3102 | } | |
3103 | ||
3104 | /* pull out the item */ | |
3105 | leaf = path->nodes[0]; | |
7b128766 JB |
3106 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3107 | ||
3108 | /* make sure the item matches what we want */ | |
3109 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3110 | break; | |
3111 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
3112 | break; | |
3113 | ||
3114 | /* release the path since we're done with it */ | |
b3b4aa74 | 3115 | btrfs_release_path(path); |
7b128766 JB |
3116 | |
3117 | /* | |
3118 | * this is where we are basically btrfs_lookup, without the | |
3119 | * crossing root thing. we store the inode number in the | |
3120 | * offset of the orphan item. | |
3121 | */ | |
8f6d7f4f JB |
3122 | |
3123 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3124 | btrfs_err(root->fs_info, |
3125 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3126 | ret = -EINVAL; |
3127 | goto out; | |
3128 | } | |
3129 | ||
3130 | last_objectid = found_key.offset; | |
3131 | ||
5d4f98a2 YZ |
3132 | found_key.objectid = found_key.offset; |
3133 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3134 | found_key.offset = 0; | |
73f73415 | 3135 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3136 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3137 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3138 | goto out; |
7b128766 | 3139 | |
f8e9e0b0 AJ |
3140 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3141 | struct btrfs_root *dead_root; | |
3142 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3143 | int is_dead_root = 0; | |
3144 | ||
3145 | /* | |
3146 | * this is an orphan in the tree root. Currently these | |
3147 | * could come from 2 sources: | |
3148 | * a) a snapshot deletion in progress | |
3149 | * b) a free space cache inode | |
3150 | * We need to distinguish those two, as the snapshot | |
3151 | * orphan must not get deleted. | |
3152 | * find_dead_roots already ran before us, so if this | |
3153 | * is a snapshot deletion, we should find the root | |
3154 | * in the dead_roots list | |
3155 | */ | |
3156 | spin_lock(&fs_info->trans_lock); | |
3157 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3158 | root_list) { | |
3159 | if (dead_root->root_key.objectid == | |
3160 | found_key.objectid) { | |
3161 | is_dead_root = 1; | |
3162 | break; | |
3163 | } | |
3164 | } | |
3165 | spin_unlock(&fs_info->trans_lock); | |
3166 | if (is_dead_root) { | |
3167 | /* prevent this orphan from being found again */ | |
3168 | key.offset = found_key.objectid - 1; | |
3169 | continue; | |
3170 | } | |
3171 | } | |
7b128766 | 3172 | /* |
a8c9e576 JB |
3173 | * Inode is already gone but the orphan item is still there, |
3174 | * kill the orphan item. | |
7b128766 | 3175 | */ |
a8c9e576 JB |
3176 | if (ret == -ESTALE) { |
3177 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3178 | if (IS_ERR(trans)) { |
3179 | ret = PTR_ERR(trans); | |
3180 | goto out; | |
3181 | } | |
c2cf52eb SK |
3182 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3183 | found_key.objectid); | |
a8c9e576 JB |
3184 | ret = btrfs_del_orphan_item(trans, root, |
3185 | found_key.objectid); | |
5b21f2ed | 3186 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3187 | if (ret) |
3188 | goto out; | |
7b128766 JB |
3189 | continue; |
3190 | } | |
3191 | ||
a8c9e576 JB |
3192 | /* |
3193 | * add this inode to the orphan list so btrfs_orphan_del does | |
3194 | * the proper thing when we hit it | |
3195 | */ | |
8a35d95f JB |
3196 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3197 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3198 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3199 | |
7b128766 JB |
3200 | /* if we have links, this was a truncate, lets do that */ |
3201 | if (inode->i_nlink) { | |
fae7f21c | 3202 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3203 | iput(inode); |
3204 | continue; | |
3205 | } | |
7b128766 | 3206 | nr_truncate++; |
f3fe820c JB |
3207 | |
3208 | /* 1 for the orphan item deletion. */ | |
3209 | trans = btrfs_start_transaction(root, 1); | |
3210 | if (IS_ERR(trans)) { | |
c69b26b0 | 3211 | iput(inode); |
f3fe820c JB |
3212 | ret = PTR_ERR(trans); |
3213 | goto out; | |
3214 | } | |
3215 | ret = btrfs_orphan_add(trans, inode); | |
3216 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3217 | if (ret) { |
3218 | iput(inode); | |
f3fe820c | 3219 | goto out; |
c69b26b0 | 3220 | } |
f3fe820c | 3221 | |
66b4ffd1 | 3222 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3223 | if (ret) |
3224 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3225 | } else { |
3226 | nr_unlink++; | |
3227 | } | |
3228 | ||
3229 | /* this will do delete_inode and everything for us */ | |
3230 | iput(inode); | |
66b4ffd1 JB |
3231 | if (ret) |
3232 | goto out; | |
7b128766 | 3233 | } |
3254c876 MX |
3234 | /* release the path since we're done with it */ |
3235 | btrfs_release_path(path); | |
3236 | ||
d68fc57b YZ |
3237 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3238 | ||
3239 | if (root->orphan_block_rsv) | |
3240 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3241 | (u64)-1); | |
3242 | ||
3243 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 3244 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3245 | if (!IS_ERR(trans)) |
3246 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3247 | } |
7b128766 JB |
3248 | |
3249 | if (nr_unlink) | |
4884b476 | 3250 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3251 | if (nr_truncate) |
4884b476 | 3252 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3253 | |
3254 | out: | |
3255 | if (ret) | |
c2cf52eb SK |
3256 | btrfs_crit(root->fs_info, |
3257 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3258 | btrfs_free_path(path); |
3259 | return ret; | |
7b128766 JB |
3260 | } |
3261 | ||
46a53cca CM |
3262 | /* |
3263 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3264 | * don't find any xattrs, we know there can't be any acls. | |
3265 | * | |
3266 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3267 | */ | |
3268 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3269 | int slot, u64 objectid, |
3270 | int *first_xattr_slot) | |
46a53cca CM |
3271 | { |
3272 | u32 nritems = btrfs_header_nritems(leaf); | |
3273 | struct btrfs_key found_key; | |
f23b5a59 JB |
3274 | static u64 xattr_access = 0; |
3275 | static u64 xattr_default = 0; | |
46a53cca CM |
3276 | int scanned = 0; |
3277 | ||
f23b5a59 JB |
3278 | if (!xattr_access) { |
3279 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3280 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3281 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3282 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3283 | } | |
3284 | ||
46a53cca | 3285 | slot++; |
63541927 | 3286 | *first_xattr_slot = -1; |
46a53cca CM |
3287 | while (slot < nritems) { |
3288 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3289 | ||
3290 | /* we found a different objectid, there must not be acls */ | |
3291 | if (found_key.objectid != objectid) | |
3292 | return 0; | |
3293 | ||
3294 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3295 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3296 | if (*first_xattr_slot == -1) |
3297 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3298 | if (found_key.offset == xattr_access || |
3299 | found_key.offset == xattr_default) | |
3300 | return 1; | |
3301 | } | |
46a53cca CM |
3302 | |
3303 | /* | |
3304 | * we found a key greater than an xattr key, there can't | |
3305 | * be any acls later on | |
3306 | */ | |
3307 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3308 | return 0; | |
3309 | ||
3310 | slot++; | |
3311 | scanned++; | |
3312 | ||
3313 | /* | |
3314 | * it goes inode, inode backrefs, xattrs, extents, | |
3315 | * so if there are a ton of hard links to an inode there can | |
3316 | * be a lot of backrefs. Don't waste time searching too hard, | |
3317 | * this is just an optimization | |
3318 | */ | |
3319 | if (scanned >= 8) | |
3320 | break; | |
3321 | } | |
3322 | /* we hit the end of the leaf before we found an xattr or | |
3323 | * something larger than an xattr. We have to assume the inode | |
3324 | * has acls | |
3325 | */ | |
63541927 FDBM |
3326 | if (*first_xattr_slot == -1) |
3327 | *first_xattr_slot = slot; | |
46a53cca CM |
3328 | return 1; |
3329 | } | |
3330 | ||
d352ac68 CM |
3331 | /* |
3332 | * read an inode from the btree into the in-memory inode | |
3333 | */ | |
5d4f98a2 | 3334 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3335 | { |
3336 | struct btrfs_path *path; | |
5f39d397 | 3337 | struct extent_buffer *leaf; |
39279cc3 | 3338 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3339 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3340 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3341 | struct btrfs_key location; | |
67de1176 | 3342 | unsigned long ptr; |
46a53cca | 3343 | int maybe_acls; |
618e21d5 | 3344 | u32 rdev; |
39279cc3 | 3345 | int ret; |
2f7e33d4 | 3346 | bool filled = false; |
63541927 | 3347 | int first_xattr_slot; |
2f7e33d4 MX |
3348 | |
3349 | ret = btrfs_fill_inode(inode, &rdev); | |
3350 | if (!ret) | |
3351 | filled = true; | |
39279cc3 CM |
3352 | |
3353 | path = btrfs_alloc_path(); | |
1748f843 MF |
3354 | if (!path) |
3355 | goto make_bad; | |
3356 | ||
39279cc3 | 3357 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3358 | |
39279cc3 | 3359 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3360 | if (ret) |
39279cc3 | 3361 | goto make_bad; |
39279cc3 | 3362 | |
5f39d397 | 3363 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3364 | |
3365 | if (filled) | |
67de1176 | 3366 | goto cache_index; |
2f7e33d4 | 3367 | |
5f39d397 CM |
3368 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3369 | struct btrfs_inode_item); | |
5f39d397 | 3370 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3371 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3372 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3373 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3374 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3375 | |
3376 | tspec = btrfs_inode_atime(inode_item); | |
3377 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3378 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3379 | ||
3380 | tspec = btrfs_inode_mtime(inode_item); | |
3381 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3382 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3383 | ||
3384 | tspec = btrfs_inode_ctime(inode_item); | |
3385 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3386 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3387 | ||
a76a3cd4 | 3388 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3389 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3390 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3391 | ||
3392 | /* | |
3393 | * If we were modified in the current generation and evicted from memory | |
3394 | * and then re-read we need to do a full sync since we don't have any | |
3395 | * idea about which extents were modified before we were evicted from | |
3396 | * cache. | |
3397 | */ | |
3398 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3399 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3400 | &BTRFS_I(inode)->runtime_flags); | |
3401 | ||
0c4d2d95 | 3402 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3403 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3404 | inode->i_rdev = 0; |
5f39d397 CM |
3405 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3406 | ||
aec7477b | 3407 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3408 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
67de1176 MX |
3409 | |
3410 | cache_index: | |
3411 | path->slots[0]++; | |
3412 | if (inode->i_nlink != 1 || | |
3413 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3414 | goto cache_acl; | |
3415 | ||
3416 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3417 | if (location.objectid != btrfs_ino(inode)) | |
3418 | goto cache_acl; | |
3419 | ||
3420 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3421 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3422 | struct btrfs_inode_ref *ref; | |
3423 | ||
3424 | ref = (struct btrfs_inode_ref *)ptr; | |
3425 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3426 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3427 | struct btrfs_inode_extref *extref; | |
3428 | ||
3429 | extref = (struct btrfs_inode_extref *)ptr; | |
3430 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3431 | extref); | |
3432 | } | |
2f7e33d4 | 3433 | cache_acl: |
46a53cca CM |
3434 | /* |
3435 | * try to precache a NULL acl entry for files that don't have | |
3436 | * any xattrs or acls | |
3437 | */ | |
33345d01 | 3438 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3439 | btrfs_ino(inode), &first_xattr_slot); |
3440 | if (first_xattr_slot != -1) { | |
3441 | path->slots[0] = first_xattr_slot; | |
3442 | ret = btrfs_load_inode_props(inode, path); | |
3443 | if (ret) | |
3444 | btrfs_err(root->fs_info, | |
3445 | "error loading props for ino %llu (root %llu): %d\n", | |
3446 | btrfs_ino(inode), | |
3447 | root->root_key.objectid, ret); | |
3448 | } | |
3449 | btrfs_free_path(path); | |
3450 | ||
72c04902 AV |
3451 | if (!maybe_acls) |
3452 | cache_no_acl(inode); | |
46a53cca | 3453 | |
39279cc3 | 3454 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3455 | case S_IFREG: |
3456 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3457 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3458 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3459 | inode->i_fop = &btrfs_file_operations; |
3460 | inode->i_op = &btrfs_file_inode_operations; | |
3461 | break; | |
3462 | case S_IFDIR: | |
3463 | inode->i_fop = &btrfs_dir_file_operations; | |
3464 | if (root == root->fs_info->tree_root) | |
3465 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3466 | else | |
3467 | inode->i_op = &btrfs_dir_inode_operations; | |
3468 | break; | |
3469 | case S_IFLNK: | |
3470 | inode->i_op = &btrfs_symlink_inode_operations; | |
3471 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3472 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3473 | break; |
618e21d5 | 3474 | default: |
0279b4cd | 3475 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3476 | init_special_inode(inode, inode->i_mode, rdev); |
3477 | break; | |
39279cc3 | 3478 | } |
6cbff00f CH |
3479 | |
3480 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3481 | return; |
3482 | ||
3483 | make_bad: | |
39279cc3 | 3484 | btrfs_free_path(path); |
39279cc3 CM |
3485 | make_bad_inode(inode); |
3486 | } | |
3487 | ||
d352ac68 CM |
3488 | /* |
3489 | * given a leaf and an inode, copy the inode fields into the leaf | |
3490 | */ | |
e02119d5 CM |
3491 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3492 | struct extent_buffer *leaf, | |
5f39d397 | 3493 | struct btrfs_inode_item *item, |
39279cc3 CM |
3494 | struct inode *inode) |
3495 | { | |
51fab693 LB |
3496 | struct btrfs_map_token token; |
3497 | ||
3498 | btrfs_init_map_token(&token); | |
5f39d397 | 3499 | |
51fab693 LB |
3500 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3501 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3502 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3503 | &token); | |
3504 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3505 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3506 | |
51fab693 LB |
3507 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3508 | inode->i_atime.tv_sec, &token); | |
3509 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3510 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3511 | |
51fab693 LB |
3512 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3513 | inode->i_mtime.tv_sec, &token); | |
3514 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3515 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3516 | |
51fab693 LB |
3517 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3518 | inode->i_ctime.tv_sec, &token); | |
3519 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3520 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3521 | |
51fab693 LB |
3522 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3523 | &token); | |
3524 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3525 | &token); | |
3526 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3527 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3528 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3529 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3530 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3531 | } |
3532 | ||
d352ac68 CM |
3533 | /* |
3534 | * copy everything in the in-memory inode into the btree. | |
3535 | */ | |
2115133f | 3536 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3537 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3538 | { |
3539 | struct btrfs_inode_item *inode_item; | |
3540 | struct btrfs_path *path; | |
5f39d397 | 3541 | struct extent_buffer *leaf; |
39279cc3 CM |
3542 | int ret; |
3543 | ||
3544 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3545 | if (!path) |
3546 | return -ENOMEM; | |
3547 | ||
b9473439 | 3548 | path->leave_spinning = 1; |
16cdcec7 MX |
3549 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3550 | 1); | |
39279cc3 CM |
3551 | if (ret) { |
3552 | if (ret > 0) | |
3553 | ret = -ENOENT; | |
3554 | goto failed; | |
3555 | } | |
3556 | ||
5f39d397 CM |
3557 | leaf = path->nodes[0]; |
3558 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3559 | struct btrfs_inode_item); |
39279cc3 | 3560 | |
e02119d5 | 3561 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3562 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3563 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3564 | ret = 0; |
3565 | failed: | |
39279cc3 CM |
3566 | btrfs_free_path(path); |
3567 | return ret; | |
3568 | } | |
3569 | ||
2115133f CM |
3570 | /* |
3571 | * copy everything in the in-memory inode into the btree. | |
3572 | */ | |
3573 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3574 | struct btrfs_root *root, struct inode *inode) | |
3575 | { | |
3576 | int ret; | |
3577 | ||
3578 | /* | |
3579 | * If the inode is a free space inode, we can deadlock during commit | |
3580 | * if we put it into the delayed code. | |
3581 | * | |
3582 | * The data relocation inode should also be directly updated | |
3583 | * without delay | |
3584 | */ | |
83eea1f1 | 3585 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 3586 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
3587 | btrfs_update_root_times(trans, root); |
3588 | ||
2115133f CM |
3589 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3590 | if (!ret) | |
3591 | btrfs_set_inode_last_trans(trans, inode); | |
3592 | return ret; | |
3593 | } | |
3594 | ||
3595 | return btrfs_update_inode_item(trans, root, inode); | |
3596 | } | |
3597 | ||
be6aef60 JB |
3598 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3599 | struct btrfs_root *root, | |
3600 | struct inode *inode) | |
2115133f CM |
3601 | { |
3602 | int ret; | |
3603 | ||
3604 | ret = btrfs_update_inode(trans, root, inode); | |
3605 | if (ret == -ENOSPC) | |
3606 | return btrfs_update_inode_item(trans, root, inode); | |
3607 | return ret; | |
3608 | } | |
3609 | ||
d352ac68 CM |
3610 | /* |
3611 | * unlink helper that gets used here in inode.c and in the tree logging | |
3612 | * recovery code. It remove a link in a directory with a given name, and | |
3613 | * also drops the back refs in the inode to the directory | |
3614 | */ | |
92986796 AV |
3615 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3616 | struct btrfs_root *root, | |
3617 | struct inode *dir, struct inode *inode, | |
3618 | const char *name, int name_len) | |
39279cc3 CM |
3619 | { |
3620 | struct btrfs_path *path; | |
39279cc3 | 3621 | int ret = 0; |
5f39d397 | 3622 | struct extent_buffer *leaf; |
39279cc3 | 3623 | struct btrfs_dir_item *di; |
5f39d397 | 3624 | struct btrfs_key key; |
aec7477b | 3625 | u64 index; |
33345d01 LZ |
3626 | u64 ino = btrfs_ino(inode); |
3627 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3628 | |
3629 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3630 | if (!path) { |
3631 | ret = -ENOMEM; | |
554233a6 | 3632 | goto out; |
54aa1f4d CM |
3633 | } |
3634 | ||
b9473439 | 3635 | path->leave_spinning = 1; |
33345d01 | 3636 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3637 | name, name_len, -1); |
3638 | if (IS_ERR(di)) { | |
3639 | ret = PTR_ERR(di); | |
3640 | goto err; | |
3641 | } | |
3642 | if (!di) { | |
3643 | ret = -ENOENT; | |
3644 | goto err; | |
3645 | } | |
5f39d397 CM |
3646 | leaf = path->nodes[0]; |
3647 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3648 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3649 | if (ret) |
3650 | goto err; | |
b3b4aa74 | 3651 | btrfs_release_path(path); |
39279cc3 | 3652 | |
67de1176 MX |
3653 | /* |
3654 | * If we don't have dir index, we have to get it by looking up | |
3655 | * the inode ref, since we get the inode ref, remove it directly, | |
3656 | * it is unnecessary to do delayed deletion. | |
3657 | * | |
3658 | * But if we have dir index, needn't search inode ref to get it. | |
3659 | * Since the inode ref is close to the inode item, it is better | |
3660 | * that we delay to delete it, and just do this deletion when | |
3661 | * we update the inode item. | |
3662 | */ | |
3663 | if (BTRFS_I(inode)->dir_index) { | |
3664 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3665 | if (!ret) { | |
3666 | index = BTRFS_I(inode)->dir_index; | |
3667 | goto skip_backref; | |
3668 | } | |
3669 | } | |
3670 | ||
33345d01 LZ |
3671 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3672 | dir_ino, &index); | |
aec7477b | 3673 | if (ret) { |
c2cf52eb SK |
3674 | btrfs_info(root->fs_info, |
3675 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3676 | name_len, name, ino, dir_ino); |
79787eaa | 3677 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3678 | goto err; |
3679 | } | |
67de1176 | 3680 | skip_backref: |
16cdcec7 | 3681 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3682 | if (ret) { |
3683 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3684 | goto err; |
79787eaa | 3685 | } |
39279cc3 | 3686 | |
e02119d5 | 3687 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3688 | inode, dir_ino); |
79787eaa JM |
3689 | if (ret != 0 && ret != -ENOENT) { |
3690 | btrfs_abort_transaction(trans, root, ret); | |
3691 | goto err; | |
3692 | } | |
e02119d5 CM |
3693 | |
3694 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3695 | dir, index); | |
6418c961 CM |
3696 | if (ret == -ENOENT) |
3697 | ret = 0; | |
d4e3991b ZB |
3698 | else if (ret) |
3699 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3700 | err: |
3701 | btrfs_free_path(path); | |
e02119d5 CM |
3702 | if (ret) |
3703 | goto out; | |
3704 | ||
3705 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3706 | inode_inc_iversion(inode); |
3707 | inode_inc_iversion(dir); | |
e02119d5 | 3708 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3709 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3710 | out: |
39279cc3 CM |
3711 | return ret; |
3712 | } | |
3713 | ||
92986796 AV |
3714 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3715 | struct btrfs_root *root, | |
3716 | struct inode *dir, struct inode *inode, | |
3717 | const char *name, int name_len) | |
3718 | { | |
3719 | int ret; | |
3720 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3721 | if (!ret) { | |
8b558c5f | 3722 | drop_nlink(inode); |
92986796 AV |
3723 | ret = btrfs_update_inode(trans, root, inode); |
3724 | } | |
3725 | return ret; | |
3726 | } | |
39279cc3 | 3727 | |
a22285a6 YZ |
3728 | /* |
3729 | * helper to start transaction for unlink and rmdir. | |
3730 | * | |
d52be818 JB |
3731 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3732 | * if we cannot make our reservations the normal way try and see if there is | |
3733 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3734 | * allow the unlink to occur. | |
a22285a6 | 3735 | */ |
d52be818 | 3736 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3737 | { |
39279cc3 | 3738 | struct btrfs_trans_handle *trans; |
a22285a6 | 3739 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3740 | int ret; |
3741 | ||
e70bea5f JB |
3742 | /* |
3743 | * 1 for the possible orphan item | |
3744 | * 1 for the dir item | |
3745 | * 1 for the dir index | |
3746 | * 1 for the inode ref | |
e70bea5f JB |
3747 | * 1 for the inode |
3748 | */ | |
6e137ed3 | 3749 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3750 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3751 | return trans; | |
4df27c4d | 3752 | |
d52be818 JB |
3753 | if (PTR_ERR(trans) == -ENOSPC) { |
3754 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3755 | |
d52be818 JB |
3756 | trans = btrfs_start_transaction(root, 0); |
3757 | if (IS_ERR(trans)) | |
3758 | return trans; | |
3759 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3760 | &root->fs_info->trans_block_rsv, | |
3761 | num_bytes, 5); | |
3762 | if (ret) { | |
3763 | btrfs_end_transaction(trans, root); | |
3764 | return ERR_PTR(ret); | |
a22285a6 | 3765 | } |
5a77d76c | 3766 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3767 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3768 | } |
d52be818 | 3769 | return trans; |
a22285a6 YZ |
3770 | } |
3771 | ||
3772 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3773 | { | |
3774 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3775 | struct btrfs_trans_handle *trans; | |
3776 | struct inode *inode = dentry->d_inode; | |
3777 | int ret; | |
a22285a6 | 3778 | |
d52be818 | 3779 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3780 | if (IS_ERR(trans)) |
3781 | return PTR_ERR(trans); | |
5f39d397 | 3782 | |
12fcfd22 CM |
3783 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3784 | ||
e02119d5 CM |
3785 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3786 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3787 | if (ret) |
3788 | goto out; | |
7b128766 | 3789 | |
a22285a6 | 3790 | if (inode->i_nlink == 0) { |
7b128766 | 3791 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3792 | if (ret) |
3793 | goto out; | |
a22285a6 | 3794 | } |
7b128766 | 3795 | |
b532402e | 3796 | out: |
d52be818 | 3797 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3798 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3799 | return ret; |
3800 | } | |
3801 | ||
4df27c4d YZ |
3802 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3803 | struct btrfs_root *root, | |
3804 | struct inode *dir, u64 objectid, | |
3805 | const char *name, int name_len) | |
3806 | { | |
3807 | struct btrfs_path *path; | |
3808 | struct extent_buffer *leaf; | |
3809 | struct btrfs_dir_item *di; | |
3810 | struct btrfs_key key; | |
3811 | u64 index; | |
3812 | int ret; | |
33345d01 | 3813 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3814 | |
3815 | path = btrfs_alloc_path(); | |
3816 | if (!path) | |
3817 | return -ENOMEM; | |
3818 | ||
33345d01 | 3819 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3820 | name, name_len, -1); |
79787eaa JM |
3821 | if (IS_ERR_OR_NULL(di)) { |
3822 | if (!di) | |
3823 | ret = -ENOENT; | |
3824 | else | |
3825 | ret = PTR_ERR(di); | |
3826 | goto out; | |
3827 | } | |
4df27c4d YZ |
3828 | |
3829 | leaf = path->nodes[0]; | |
3830 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3831 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3832 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3833 | if (ret) { |
3834 | btrfs_abort_transaction(trans, root, ret); | |
3835 | goto out; | |
3836 | } | |
b3b4aa74 | 3837 | btrfs_release_path(path); |
4df27c4d YZ |
3838 | |
3839 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3840 | objectid, root->root_key.objectid, | |
33345d01 | 3841 | dir_ino, &index, name, name_len); |
4df27c4d | 3842 | if (ret < 0) { |
79787eaa JM |
3843 | if (ret != -ENOENT) { |
3844 | btrfs_abort_transaction(trans, root, ret); | |
3845 | goto out; | |
3846 | } | |
33345d01 | 3847 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3848 | name, name_len); |
79787eaa JM |
3849 | if (IS_ERR_OR_NULL(di)) { |
3850 | if (!di) | |
3851 | ret = -ENOENT; | |
3852 | else | |
3853 | ret = PTR_ERR(di); | |
3854 | btrfs_abort_transaction(trans, root, ret); | |
3855 | goto out; | |
3856 | } | |
4df27c4d YZ |
3857 | |
3858 | leaf = path->nodes[0]; | |
3859 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3860 | btrfs_release_path(path); |
4df27c4d YZ |
3861 | index = key.offset; |
3862 | } | |
945d8962 | 3863 | btrfs_release_path(path); |
4df27c4d | 3864 | |
16cdcec7 | 3865 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3866 | if (ret) { |
3867 | btrfs_abort_transaction(trans, root, ret); | |
3868 | goto out; | |
3869 | } | |
4df27c4d YZ |
3870 | |
3871 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3872 | inode_inc_iversion(dir); |
4df27c4d | 3873 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3874 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3875 | if (ret) |
3876 | btrfs_abort_transaction(trans, root, ret); | |
3877 | out: | |
71d7aed0 | 3878 | btrfs_free_path(path); |
79787eaa | 3879 | return ret; |
4df27c4d YZ |
3880 | } |
3881 | ||
39279cc3 CM |
3882 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3883 | { | |
3884 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3885 | int err = 0; |
39279cc3 | 3886 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3887 | struct btrfs_trans_handle *trans; |
39279cc3 | 3888 | |
b3ae244e | 3889 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3890 | return -ENOTEMPTY; |
b3ae244e DS |
3891 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3892 | return -EPERM; | |
134d4512 | 3893 | |
d52be818 | 3894 | trans = __unlink_start_trans(dir); |
a22285a6 | 3895 | if (IS_ERR(trans)) |
5df6a9f6 | 3896 | return PTR_ERR(trans); |
5df6a9f6 | 3897 | |
33345d01 | 3898 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3899 | err = btrfs_unlink_subvol(trans, root, dir, |
3900 | BTRFS_I(inode)->location.objectid, | |
3901 | dentry->d_name.name, | |
3902 | dentry->d_name.len); | |
3903 | goto out; | |
3904 | } | |
3905 | ||
7b128766 JB |
3906 | err = btrfs_orphan_add(trans, inode); |
3907 | if (err) | |
4df27c4d | 3908 | goto out; |
7b128766 | 3909 | |
39279cc3 | 3910 | /* now the directory is empty */ |
e02119d5 CM |
3911 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3912 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3913 | if (!err) |
dbe674a9 | 3914 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3915 | out: |
d52be818 | 3916 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3917 | btrfs_btree_balance_dirty(root); |
3954401f | 3918 | |
39279cc3 CM |
3919 | return err; |
3920 | } | |
3921 | ||
39279cc3 CM |
3922 | /* |
3923 | * this can truncate away extent items, csum items and directory items. | |
3924 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3925 | * any higher than new_size |
39279cc3 CM |
3926 | * |
3927 | * csum items that cross the new i_size are truncated to the new size | |
3928 | * as well. | |
7b128766 JB |
3929 | * |
3930 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3931 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3932 | */ |
8082510e YZ |
3933 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3934 | struct btrfs_root *root, | |
3935 | struct inode *inode, | |
3936 | u64 new_size, u32 min_type) | |
39279cc3 | 3937 | { |
39279cc3 | 3938 | struct btrfs_path *path; |
5f39d397 | 3939 | struct extent_buffer *leaf; |
39279cc3 | 3940 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3941 | struct btrfs_key key; |
3942 | struct btrfs_key found_key; | |
39279cc3 | 3943 | u64 extent_start = 0; |
db94535d | 3944 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3945 | u64 extent_offset = 0; |
39279cc3 | 3946 | u64 item_end = 0; |
7f4f6e0a | 3947 | u64 last_size = (u64)-1; |
8082510e | 3948 | u32 found_type = (u8)-1; |
39279cc3 CM |
3949 | int found_extent; |
3950 | int del_item; | |
85e21bac CM |
3951 | int pending_del_nr = 0; |
3952 | int pending_del_slot = 0; | |
179e29e4 | 3953 | int extent_type = -1; |
8082510e YZ |
3954 | int ret; |
3955 | int err = 0; | |
33345d01 | 3956 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3957 | |
3958 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3959 | |
0eb0e19c MF |
3960 | path = btrfs_alloc_path(); |
3961 | if (!path) | |
3962 | return -ENOMEM; | |
3963 | path->reada = -1; | |
3964 | ||
5dc562c5 JB |
3965 | /* |
3966 | * We want to drop from the next block forward in case this new size is | |
3967 | * not block aligned since we will be keeping the last block of the | |
3968 | * extent just the way it is. | |
3969 | */ | |
0af3d00b | 3970 | if (root->ref_cows || root == root->fs_info->tree_root) |
fda2832f QW |
3971 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
3972 | root->sectorsize), (u64)-1, 0); | |
8082510e | 3973 | |
16cdcec7 MX |
3974 | /* |
3975 | * This function is also used to drop the items in the log tree before | |
3976 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3977 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3978 | * items. | |
3979 | */ | |
3980 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3981 | btrfs_kill_delayed_inode_items(inode); | |
3982 | ||
33345d01 | 3983 | key.objectid = ino; |
39279cc3 | 3984 | key.offset = (u64)-1; |
5f39d397 CM |
3985 | key.type = (u8)-1; |
3986 | ||
85e21bac | 3987 | search_again: |
b9473439 | 3988 | path->leave_spinning = 1; |
85e21bac | 3989 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3990 | if (ret < 0) { |
3991 | err = ret; | |
3992 | goto out; | |
3993 | } | |
d397712b | 3994 | |
85e21bac | 3995 | if (ret > 0) { |
e02119d5 CM |
3996 | /* there are no items in the tree for us to truncate, we're |
3997 | * done | |
3998 | */ | |
8082510e YZ |
3999 | if (path->slots[0] == 0) |
4000 | goto out; | |
85e21bac CM |
4001 | path->slots[0]--; |
4002 | } | |
4003 | ||
d397712b | 4004 | while (1) { |
39279cc3 | 4005 | fi = NULL; |
5f39d397 CM |
4006 | leaf = path->nodes[0]; |
4007 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4008 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 4009 | |
33345d01 | 4010 | if (found_key.objectid != ino) |
39279cc3 | 4011 | break; |
5f39d397 | 4012 | |
85e21bac | 4013 | if (found_type < min_type) |
39279cc3 CM |
4014 | break; |
4015 | ||
5f39d397 | 4016 | item_end = found_key.offset; |
39279cc3 | 4017 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4018 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4019 | struct btrfs_file_extent_item); |
179e29e4 CM |
4020 | extent_type = btrfs_file_extent_type(leaf, fi); |
4021 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4022 | item_end += |
db94535d | 4023 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4024 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4025 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 4026 | fi); |
39279cc3 | 4027 | } |
008630c1 | 4028 | item_end--; |
39279cc3 | 4029 | } |
8082510e YZ |
4030 | if (found_type > min_type) { |
4031 | del_item = 1; | |
4032 | } else { | |
4033 | if (item_end < new_size) | |
b888db2b | 4034 | break; |
8082510e YZ |
4035 | if (found_key.offset >= new_size) |
4036 | del_item = 1; | |
4037 | else | |
4038 | del_item = 0; | |
39279cc3 | 4039 | } |
39279cc3 | 4040 | found_extent = 0; |
39279cc3 | 4041 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4042 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4043 | goto delete; | |
4044 | ||
7f4f6e0a JB |
4045 | if (del_item) |
4046 | last_size = found_key.offset; | |
4047 | else | |
4048 | last_size = new_size; | |
4049 | ||
179e29e4 | 4050 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4051 | u64 num_dec; |
db94535d | 4052 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4053 | if (!del_item) { |
db94535d CM |
4054 | u64 orig_num_bytes = |
4055 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4056 | extent_num_bytes = ALIGN(new_size - |
4057 | found_key.offset, | |
4058 | root->sectorsize); | |
db94535d CM |
4059 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4060 | extent_num_bytes); | |
4061 | num_dec = (orig_num_bytes - | |
9069218d | 4062 | extent_num_bytes); |
e02119d5 | 4063 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 4064 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4065 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4066 | } else { |
db94535d CM |
4067 | extent_num_bytes = |
4068 | btrfs_file_extent_disk_num_bytes(leaf, | |
4069 | fi); | |
5d4f98a2 YZ |
4070 | extent_offset = found_key.offset - |
4071 | btrfs_file_extent_offset(leaf, fi); | |
4072 | ||
39279cc3 | 4073 | /* FIXME blocksize != 4096 */ |
9069218d | 4074 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4075 | if (extent_start != 0) { |
4076 | found_extent = 1; | |
e02119d5 | 4077 | if (root->ref_cows) |
a76a3cd4 | 4078 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4079 | } |
39279cc3 | 4080 | } |
9069218d | 4081 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4082 | /* |
4083 | * we can't truncate inline items that have had | |
4084 | * special encodings | |
4085 | */ | |
4086 | if (!del_item && | |
4087 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4088 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4089 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4090 | u32 size = new_size - found_key.offset; |
4091 | ||
4092 | if (root->ref_cows) { | |
a76a3cd4 YZ |
4093 | inode_sub_bytes(inode, item_end + 1 - |
4094 | new_size); | |
e02119d5 CM |
4095 | } |
4096 | size = | |
4097 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4098 | btrfs_truncate_item(root, path, size, 1); |
e02119d5 | 4099 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
4100 | inode_sub_bytes(inode, item_end + 1 - |
4101 | found_key.offset); | |
9069218d | 4102 | } |
39279cc3 | 4103 | } |
179e29e4 | 4104 | delete: |
39279cc3 | 4105 | if (del_item) { |
85e21bac CM |
4106 | if (!pending_del_nr) { |
4107 | /* no pending yet, add ourselves */ | |
4108 | pending_del_slot = path->slots[0]; | |
4109 | pending_del_nr = 1; | |
4110 | } else if (pending_del_nr && | |
4111 | path->slots[0] + 1 == pending_del_slot) { | |
4112 | /* hop on the pending chunk */ | |
4113 | pending_del_nr++; | |
4114 | pending_del_slot = path->slots[0]; | |
4115 | } else { | |
d397712b | 4116 | BUG(); |
85e21bac | 4117 | } |
39279cc3 CM |
4118 | } else { |
4119 | break; | |
4120 | } | |
0af3d00b JB |
4121 | if (found_extent && (root->ref_cows || |
4122 | root == root->fs_info->tree_root)) { | |
b9473439 | 4123 | btrfs_set_path_blocking(path); |
39279cc3 | 4124 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4125 | extent_num_bytes, 0, |
4126 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4127 | ino, extent_offset, 0); |
39279cc3 CM |
4128 | BUG_ON(ret); |
4129 | } | |
85e21bac | 4130 | |
8082510e YZ |
4131 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4132 | break; | |
4133 | ||
4134 | if (path->slots[0] == 0 || | |
4135 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4136 | if (pending_del_nr) { |
4137 | ret = btrfs_del_items(trans, root, path, | |
4138 | pending_del_slot, | |
4139 | pending_del_nr); | |
79787eaa JM |
4140 | if (ret) { |
4141 | btrfs_abort_transaction(trans, | |
4142 | root, ret); | |
4143 | goto error; | |
4144 | } | |
8082510e YZ |
4145 | pending_del_nr = 0; |
4146 | } | |
b3b4aa74 | 4147 | btrfs_release_path(path); |
85e21bac | 4148 | goto search_again; |
8082510e YZ |
4149 | } else { |
4150 | path->slots[0]--; | |
85e21bac | 4151 | } |
39279cc3 | 4152 | } |
8082510e | 4153 | out: |
85e21bac CM |
4154 | if (pending_del_nr) { |
4155 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4156 | pending_del_nr); | |
79787eaa JM |
4157 | if (ret) |
4158 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4159 | } |
79787eaa | 4160 | error: |
7f4f6e0a JB |
4161 | if (last_size != (u64)-1) |
4162 | btrfs_ordered_update_i_size(inode, last_size, NULL); | |
39279cc3 | 4163 | btrfs_free_path(path); |
8082510e | 4164 | return err; |
39279cc3 CM |
4165 | } |
4166 | ||
4167 | /* | |
2aaa6655 JB |
4168 | * btrfs_truncate_page - read, zero a chunk and write a page |
4169 | * @inode - inode that we're zeroing | |
4170 | * @from - the offset to start zeroing | |
4171 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4172 | * offset | |
4173 | * @front - zero up to the offset instead of from the offset on | |
4174 | * | |
4175 | * This will find the page for the "from" offset and cow the page and zero the | |
4176 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4177 | */ |
2aaa6655 JB |
4178 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4179 | int front) | |
39279cc3 | 4180 | { |
2aaa6655 | 4181 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4182 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4183 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4184 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4185 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4186 | char *kaddr; |
db94535d | 4187 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4188 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4189 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4190 | struct page *page; | |
3b16a4e3 | 4191 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4192 | int ret = 0; |
a52d9a80 | 4193 | u64 page_start; |
e6dcd2dc | 4194 | u64 page_end; |
39279cc3 | 4195 | |
2aaa6655 JB |
4196 | if ((offset & (blocksize - 1)) == 0 && |
4197 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4198 | goto out; |
0ca1f7ce | 4199 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4200 | if (ret) |
4201 | goto out; | |
39279cc3 | 4202 | |
211c17f5 | 4203 | again: |
3b16a4e3 | 4204 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4205 | if (!page) { |
0ca1f7ce | 4206 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4207 | ret = -ENOMEM; |
39279cc3 | 4208 | goto out; |
5d5e103a | 4209 | } |
e6dcd2dc CM |
4210 | |
4211 | page_start = page_offset(page); | |
4212 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4213 | ||
39279cc3 | 4214 | if (!PageUptodate(page)) { |
9ebefb18 | 4215 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4216 | lock_page(page); |
211c17f5 CM |
4217 | if (page->mapping != mapping) { |
4218 | unlock_page(page); | |
4219 | page_cache_release(page); | |
4220 | goto again; | |
4221 | } | |
39279cc3 CM |
4222 | if (!PageUptodate(page)) { |
4223 | ret = -EIO; | |
89642229 | 4224 | goto out_unlock; |
39279cc3 CM |
4225 | } |
4226 | } | |
211c17f5 | 4227 | wait_on_page_writeback(page); |
e6dcd2dc | 4228 | |
d0082371 | 4229 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4230 | set_page_extent_mapped(page); |
4231 | ||
4232 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4233 | if (ordered) { | |
2ac55d41 JB |
4234 | unlock_extent_cached(io_tree, page_start, page_end, |
4235 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4236 | unlock_page(page); |
4237 | page_cache_release(page); | |
eb84ae03 | 4238 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4239 | btrfs_put_ordered_extent(ordered); |
4240 | goto again; | |
4241 | } | |
4242 | ||
2ac55d41 | 4243 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4244 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4245 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4246 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4247 | |
2ac55d41 JB |
4248 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4249 | &cached_state); | |
9ed74f2d | 4250 | if (ret) { |
2ac55d41 JB |
4251 | unlock_extent_cached(io_tree, page_start, page_end, |
4252 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4253 | goto out_unlock; |
4254 | } | |
4255 | ||
e6dcd2dc | 4256 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4257 | if (!len) |
4258 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4259 | kaddr = kmap(page); |
2aaa6655 JB |
4260 | if (front) |
4261 | memset(kaddr, 0, offset); | |
4262 | else | |
4263 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4264 | flush_dcache_page(page); |
4265 | kunmap(page); | |
4266 | } | |
247e743c | 4267 | ClearPageChecked(page); |
e6dcd2dc | 4268 | set_page_dirty(page); |
2ac55d41 JB |
4269 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4270 | GFP_NOFS); | |
39279cc3 | 4271 | |
89642229 | 4272 | out_unlock: |
5d5e103a | 4273 | if (ret) |
0ca1f7ce | 4274 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4275 | unlock_page(page); |
4276 | page_cache_release(page); | |
4277 | out: | |
4278 | return ret; | |
4279 | } | |
4280 | ||
16e7549f JB |
4281 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4282 | u64 offset, u64 len) | |
4283 | { | |
4284 | struct btrfs_trans_handle *trans; | |
4285 | int ret; | |
4286 | ||
4287 | /* | |
4288 | * Still need to make sure the inode looks like it's been updated so | |
4289 | * that any holes get logged if we fsync. | |
4290 | */ | |
4291 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4292 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4293 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4294 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4295 | return 0; | |
4296 | } | |
4297 | ||
4298 | /* | |
4299 | * 1 - for the one we're dropping | |
4300 | * 1 - for the one we're adding | |
4301 | * 1 - for updating the inode. | |
4302 | */ | |
4303 | trans = btrfs_start_transaction(root, 3); | |
4304 | if (IS_ERR(trans)) | |
4305 | return PTR_ERR(trans); | |
4306 | ||
4307 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4308 | if (ret) { | |
4309 | btrfs_abort_transaction(trans, root, ret); | |
4310 | btrfs_end_transaction(trans, root); | |
4311 | return ret; | |
4312 | } | |
4313 | ||
4314 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4315 | 0, 0, len, 0, len, 0, 0, 0); | |
4316 | if (ret) | |
4317 | btrfs_abort_transaction(trans, root, ret); | |
4318 | else | |
4319 | btrfs_update_inode(trans, root, inode); | |
4320 | btrfs_end_transaction(trans, root); | |
4321 | return ret; | |
4322 | } | |
4323 | ||
695a0d0d JB |
4324 | /* |
4325 | * This function puts in dummy file extents for the area we're creating a hole | |
4326 | * for. So if we are truncating this file to a larger size we need to insert | |
4327 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4328 | * the range between oldsize and size | |
4329 | */ | |
a41ad394 | 4330 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4331 | { |
9036c102 YZ |
4332 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4333 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4334 | struct extent_map *em = NULL; |
2ac55d41 | 4335 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4336 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4337 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4338 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4339 | u64 last_byte; |
4340 | u64 cur_offset; | |
4341 | u64 hole_size; | |
9ed74f2d | 4342 | int err = 0; |
39279cc3 | 4343 | |
a71754fc JB |
4344 | /* |
4345 | * If our size started in the middle of a page we need to zero out the | |
4346 | * rest of the page before we expand the i_size, otherwise we could | |
4347 | * expose stale data. | |
4348 | */ | |
4349 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4350 | if (err) | |
4351 | return err; | |
4352 | ||
9036c102 YZ |
4353 | if (size <= hole_start) |
4354 | return 0; | |
4355 | ||
9036c102 YZ |
4356 | while (1) { |
4357 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4358 | |
2ac55d41 | 4359 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4360 | &cached_state); |
fa7c1494 MX |
4361 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4362 | block_end - hole_start); | |
9036c102 YZ |
4363 | if (!ordered) |
4364 | break; | |
2ac55d41 JB |
4365 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4366 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4367 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4368 | btrfs_put_ordered_extent(ordered); |
4369 | } | |
39279cc3 | 4370 | |
9036c102 YZ |
4371 | cur_offset = hole_start; |
4372 | while (1) { | |
4373 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4374 | block_end - cur_offset, 0); | |
79787eaa JM |
4375 | if (IS_ERR(em)) { |
4376 | err = PTR_ERR(em); | |
f2767956 | 4377 | em = NULL; |
79787eaa JM |
4378 | break; |
4379 | } | |
9036c102 | 4380 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4381 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4382 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4383 | struct extent_map *hole_em; |
9036c102 | 4384 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4385 | |
16e7549f JB |
4386 | err = maybe_insert_hole(root, inode, cur_offset, |
4387 | hole_size); | |
4388 | if (err) | |
3893e33b | 4389 | break; |
5dc562c5 JB |
4390 | btrfs_drop_extent_cache(inode, cur_offset, |
4391 | cur_offset + hole_size - 1, 0); | |
4392 | hole_em = alloc_extent_map(); | |
4393 | if (!hole_em) { | |
4394 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4395 | &BTRFS_I(inode)->runtime_flags); | |
4396 | goto next; | |
4397 | } | |
4398 | hole_em->start = cur_offset; | |
4399 | hole_em->len = hole_size; | |
4400 | hole_em->orig_start = cur_offset; | |
8082510e | 4401 | |
5dc562c5 JB |
4402 | hole_em->block_start = EXTENT_MAP_HOLE; |
4403 | hole_em->block_len = 0; | |
b4939680 | 4404 | hole_em->orig_block_len = 0; |
cc95bef6 | 4405 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4406 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4407 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4408 | hole_em->generation = root->fs_info->generation; |
8082510e | 4409 | |
5dc562c5 JB |
4410 | while (1) { |
4411 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4412 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4413 | write_unlock(&em_tree->lock); |
4414 | if (err != -EEXIST) | |
4415 | break; | |
4416 | btrfs_drop_extent_cache(inode, cur_offset, | |
4417 | cur_offset + | |
4418 | hole_size - 1, 0); | |
4419 | } | |
4420 | free_extent_map(hole_em); | |
9036c102 | 4421 | } |
16e7549f | 4422 | next: |
9036c102 | 4423 | free_extent_map(em); |
a22285a6 | 4424 | em = NULL; |
9036c102 | 4425 | cur_offset = last_byte; |
8082510e | 4426 | if (cur_offset >= block_end) |
9036c102 YZ |
4427 | break; |
4428 | } | |
a22285a6 | 4429 | free_extent_map(em); |
2ac55d41 JB |
4430 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4431 | GFP_NOFS); | |
9036c102 YZ |
4432 | return err; |
4433 | } | |
39279cc3 | 4434 | |
3972f260 | 4435 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4436 | { |
f4a2f4c5 MX |
4437 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4438 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4439 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4440 | loff_t newsize = attr->ia_size; |
4441 | int mask = attr->ia_valid; | |
8082510e YZ |
4442 | int ret; |
4443 | ||
3972f260 ES |
4444 | /* |
4445 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4446 | * special case where we need to update the times despite not having | |
4447 | * these flags set. For all other operations the VFS set these flags | |
4448 | * explicitly if it wants a timestamp update. | |
4449 | */ | |
4450 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
4451 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
4452 | ||
a41ad394 | 4453 | if (newsize > oldsize) { |
7caef267 | 4454 | truncate_pagecache(inode, newsize); |
a41ad394 | 4455 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
f4a2f4c5 | 4456 | if (ret) |
8082510e | 4457 | return ret; |
8082510e | 4458 | |
f4a2f4c5 MX |
4459 | trans = btrfs_start_transaction(root, 1); |
4460 | if (IS_ERR(trans)) | |
4461 | return PTR_ERR(trans); | |
4462 | ||
4463 | i_size_write(inode, newsize); | |
4464 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4465 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4466 | btrfs_end_transaction(trans, root); |
a41ad394 | 4467 | } else { |
8082510e | 4468 | |
a41ad394 JB |
4469 | /* |
4470 | * We're truncating a file that used to have good data down to | |
4471 | * zero. Make sure it gets into the ordered flush list so that | |
4472 | * any new writes get down to disk quickly. | |
4473 | */ | |
4474 | if (newsize == 0) | |
72ac3c0d JB |
4475 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4476 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4477 | |
f3fe820c JB |
4478 | /* |
4479 | * 1 for the orphan item we're going to add | |
4480 | * 1 for the orphan item deletion. | |
4481 | */ | |
4482 | trans = btrfs_start_transaction(root, 2); | |
4483 | if (IS_ERR(trans)) | |
4484 | return PTR_ERR(trans); | |
4485 | ||
4486 | /* | |
4487 | * We need to do this in case we fail at _any_ point during the | |
4488 | * actual truncate. Once we do the truncate_setsize we could | |
4489 | * invalidate pages which forces any outstanding ordered io to | |
4490 | * be instantly completed which will give us extents that need | |
4491 | * to be truncated. If we fail to get an orphan inode down we | |
4492 | * could have left over extents that were never meant to live, | |
4493 | * so we need to garuntee from this point on that everything | |
4494 | * will be consistent. | |
4495 | */ | |
4496 | ret = btrfs_orphan_add(trans, inode); | |
4497 | btrfs_end_transaction(trans, root); | |
4498 | if (ret) | |
4499 | return ret; | |
4500 | ||
a41ad394 JB |
4501 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4502 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4503 | |
4504 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4505 | btrfs_inode_block_unlocked_dio(inode); | |
4506 | inode_dio_wait(inode); | |
4507 | btrfs_inode_resume_unlocked_dio(inode); | |
4508 | ||
a41ad394 | 4509 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4510 | if (ret && inode->i_nlink) { |
4511 | int err; | |
4512 | ||
4513 | /* | |
4514 | * failed to truncate, disk_i_size is only adjusted down | |
4515 | * as we remove extents, so it should represent the true | |
4516 | * size of the inode, so reset the in memory size and | |
4517 | * delete our orphan entry. | |
4518 | */ | |
4519 | trans = btrfs_join_transaction(root); | |
4520 | if (IS_ERR(trans)) { | |
4521 | btrfs_orphan_del(NULL, inode); | |
4522 | return ret; | |
4523 | } | |
4524 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4525 | err = btrfs_orphan_del(trans, inode); | |
4526 | if (err) | |
4527 | btrfs_abort_transaction(trans, root, err); | |
4528 | btrfs_end_transaction(trans, root); | |
4529 | } | |
8082510e YZ |
4530 | } |
4531 | ||
a41ad394 | 4532 | return ret; |
8082510e YZ |
4533 | } |
4534 | ||
9036c102 YZ |
4535 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4536 | { | |
4537 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4538 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4539 | int err; |
39279cc3 | 4540 | |
b83cc969 LZ |
4541 | if (btrfs_root_readonly(root)) |
4542 | return -EROFS; | |
4543 | ||
9036c102 YZ |
4544 | err = inode_change_ok(inode, attr); |
4545 | if (err) | |
4546 | return err; | |
2bf5a725 | 4547 | |
5a3f23d5 | 4548 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4549 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4550 | if (err) |
4551 | return err; | |
39279cc3 | 4552 | } |
9036c102 | 4553 | |
1025774c CH |
4554 | if (attr->ia_valid) { |
4555 | setattr_copy(inode, attr); | |
0c4d2d95 | 4556 | inode_inc_iversion(inode); |
22c44fe6 | 4557 | err = btrfs_dirty_inode(inode); |
1025774c | 4558 | |
22c44fe6 | 4559 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
4560 | err = btrfs_acl_chmod(inode); |
4561 | } | |
33268eaf | 4562 | |
39279cc3 CM |
4563 | return err; |
4564 | } | |
61295eb8 | 4565 | |
131e404a FDBM |
4566 | /* |
4567 | * While truncating the inode pages during eviction, we get the VFS calling | |
4568 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
4569 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
4570 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
4571 | * extent_state structures over and over, wasting lots of time. | |
4572 | * | |
4573 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
4574 | * those expensive operations on a per page basis and do only the ordered io | |
4575 | * finishing, while we release here the extent_map and extent_state structures, | |
4576 | * without the excessive merging and splitting. | |
4577 | */ | |
4578 | static void evict_inode_truncate_pages(struct inode *inode) | |
4579 | { | |
4580 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
4581 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
4582 | struct rb_node *node; | |
4583 | ||
4584 | ASSERT(inode->i_state & I_FREEING); | |
4585 | truncate_inode_pages(&inode->i_data, 0); | |
4586 | ||
4587 | write_lock(&map_tree->lock); | |
4588 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
4589 | struct extent_map *em; | |
4590 | ||
4591 | node = rb_first(&map_tree->map); | |
4592 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
4593 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
4594 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
4595 | remove_extent_mapping(map_tree, em); |
4596 | free_extent_map(em); | |
4597 | } | |
4598 | write_unlock(&map_tree->lock); | |
4599 | ||
4600 | spin_lock(&io_tree->lock); | |
4601 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
4602 | struct extent_state *state; | |
4603 | struct extent_state *cached_state = NULL; | |
4604 | ||
4605 | node = rb_first(&io_tree->state); | |
4606 | state = rb_entry(node, struct extent_state, rb_node); | |
4607 | atomic_inc(&state->refs); | |
4608 | spin_unlock(&io_tree->lock); | |
4609 | ||
4610 | lock_extent_bits(io_tree, state->start, state->end, | |
4611 | 0, &cached_state); | |
4612 | clear_extent_bit(io_tree, state->start, state->end, | |
4613 | EXTENT_LOCKED | EXTENT_DIRTY | | |
4614 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
4615 | EXTENT_DEFRAG, 1, 1, | |
4616 | &cached_state, GFP_NOFS); | |
4617 | free_extent_state(state); | |
4618 | ||
4619 | spin_lock(&io_tree->lock); | |
4620 | } | |
4621 | spin_unlock(&io_tree->lock); | |
4622 | } | |
4623 | ||
bd555975 | 4624 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4625 | { |
4626 | struct btrfs_trans_handle *trans; | |
4627 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4628 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4629 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4630 | int ret; |
4631 | ||
1abe9b8a | 4632 | trace_btrfs_inode_evict(inode); |
4633 | ||
131e404a FDBM |
4634 | evict_inode_truncate_pages(inode); |
4635 | ||
69e9c6c6 SB |
4636 | if (inode->i_nlink && |
4637 | ((btrfs_root_refs(&root->root_item) != 0 && | |
4638 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
4639 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
4640 | goto no_delete; |
4641 | ||
39279cc3 | 4642 | if (is_bad_inode(inode)) { |
7b128766 | 4643 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4644 | goto no_delete; |
4645 | } | |
bd555975 | 4646 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4647 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4648 | |
c71bf099 | 4649 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4650 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4651 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4652 | goto no_delete; |
4653 | } | |
4654 | ||
76dda93c | 4655 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
4656 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
4657 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
4658 | goto no_delete; |
4659 | } | |
4660 | ||
0e8c36a9 MX |
4661 | ret = btrfs_commit_inode_delayed_inode(inode); |
4662 | if (ret) { | |
4663 | btrfs_orphan_del(NULL, inode); | |
4664 | goto no_delete; | |
4665 | } | |
4666 | ||
66d8f3dd | 4667 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4668 | if (!rsv) { |
4669 | btrfs_orphan_del(NULL, inode); | |
4670 | goto no_delete; | |
4671 | } | |
4a338542 | 4672 | rsv->size = min_size; |
ca7e70f5 | 4673 | rsv->failfast = 1; |
726c35fa | 4674 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4675 | |
dbe674a9 | 4676 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4677 | |
4289a667 | 4678 | /* |
8407aa46 MX |
4679 | * This is a bit simpler than btrfs_truncate since we've already |
4680 | * reserved our space for our orphan item in the unlink, so we just | |
4681 | * need to reserve some slack space in case we add bytes and update | |
4682 | * inode item when doing the truncate. | |
4289a667 | 4683 | */ |
8082510e | 4684 | while (1) { |
08e007d2 MX |
4685 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4686 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4687 | |
4688 | /* | |
4689 | * Try and steal from the global reserve since we will | |
4690 | * likely not use this space anyway, we want to try as | |
4691 | * hard as possible to get this to work. | |
4692 | */ | |
4693 | if (ret) | |
4694 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4695 | |
d68fc57b | 4696 | if (ret) { |
c2cf52eb SK |
4697 | btrfs_warn(root->fs_info, |
4698 | "Could not get space for a delete, will truncate on mount %d", | |
4699 | ret); | |
4289a667 JB |
4700 | btrfs_orphan_del(NULL, inode); |
4701 | btrfs_free_block_rsv(root, rsv); | |
4702 | goto no_delete; | |
d68fc57b | 4703 | } |
7b128766 | 4704 | |
0e8c36a9 | 4705 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4706 | if (IS_ERR(trans)) { |
4707 | btrfs_orphan_del(NULL, inode); | |
4708 | btrfs_free_block_rsv(root, rsv); | |
4709 | goto no_delete; | |
d68fc57b | 4710 | } |
7b128766 | 4711 | |
4289a667 JB |
4712 | trans->block_rsv = rsv; |
4713 | ||
d68fc57b | 4714 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4715 | if (ret != -ENOSPC) |
8082510e | 4716 | break; |
85e21bac | 4717 | |
8407aa46 | 4718 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4719 | btrfs_end_transaction(trans, root); |
4720 | trans = NULL; | |
b53d3f5d | 4721 | btrfs_btree_balance_dirty(root); |
8082510e | 4722 | } |
5f39d397 | 4723 | |
4289a667 JB |
4724 | btrfs_free_block_rsv(root, rsv); |
4725 | ||
4ef31a45 JB |
4726 | /* |
4727 | * Errors here aren't a big deal, it just means we leave orphan items | |
4728 | * in the tree. They will be cleaned up on the next mount. | |
4729 | */ | |
8082510e | 4730 | if (ret == 0) { |
4289a667 | 4731 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
4732 | btrfs_orphan_del(trans, inode); |
4733 | } else { | |
4734 | btrfs_orphan_del(NULL, inode); | |
8082510e | 4735 | } |
54aa1f4d | 4736 | |
4289a667 | 4737 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4738 | if (!(root == root->fs_info->tree_root || |
4739 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4740 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4741 | |
54aa1f4d | 4742 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4743 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4744 | no_delete: |
89042e5a | 4745 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4746 | clear_inode(inode); |
8082510e | 4747 | return; |
39279cc3 CM |
4748 | } |
4749 | ||
4750 | /* | |
4751 | * this returns the key found in the dir entry in the location pointer. | |
4752 | * If no dir entries were found, location->objectid is 0. | |
4753 | */ | |
4754 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4755 | struct btrfs_key *location) | |
4756 | { | |
4757 | const char *name = dentry->d_name.name; | |
4758 | int namelen = dentry->d_name.len; | |
4759 | struct btrfs_dir_item *di; | |
4760 | struct btrfs_path *path; | |
4761 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4762 | int ret = 0; |
39279cc3 CM |
4763 | |
4764 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4765 | if (!path) |
4766 | return -ENOMEM; | |
3954401f | 4767 | |
33345d01 | 4768 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4769 | namelen, 0); |
0d9f7f3e Y |
4770 | if (IS_ERR(di)) |
4771 | ret = PTR_ERR(di); | |
d397712b | 4772 | |
c704005d | 4773 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4774 | goto out_err; |
d397712b | 4775 | |
5f39d397 | 4776 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4777 | out: |
39279cc3 CM |
4778 | btrfs_free_path(path); |
4779 | return ret; | |
3954401f CM |
4780 | out_err: |
4781 | location->objectid = 0; | |
4782 | goto out; | |
39279cc3 CM |
4783 | } |
4784 | ||
4785 | /* | |
4786 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4787 | * needs to be changed to reflect the root directory of the tree root. This | |
4788 | * is kind of like crossing a mount point. | |
4789 | */ | |
4790 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4791 | struct inode *dir, |
4792 | struct dentry *dentry, | |
4793 | struct btrfs_key *location, | |
4794 | struct btrfs_root **sub_root) | |
39279cc3 | 4795 | { |
4df27c4d YZ |
4796 | struct btrfs_path *path; |
4797 | struct btrfs_root *new_root; | |
4798 | struct btrfs_root_ref *ref; | |
4799 | struct extent_buffer *leaf; | |
4800 | int ret; | |
4801 | int err = 0; | |
39279cc3 | 4802 | |
4df27c4d YZ |
4803 | path = btrfs_alloc_path(); |
4804 | if (!path) { | |
4805 | err = -ENOMEM; | |
4806 | goto out; | |
4807 | } | |
39279cc3 | 4808 | |
4df27c4d | 4809 | err = -ENOENT; |
75ac2dd9 KN |
4810 | ret = btrfs_find_item(root->fs_info->tree_root, path, |
4811 | BTRFS_I(dir)->root->root_key.objectid, | |
4812 | location->objectid, BTRFS_ROOT_REF_KEY, NULL); | |
4df27c4d YZ |
4813 | if (ret) { |
4814 | if (ret < 0) | |
4815 | err = ret; | |
4816 | goto out; | |
4817 | } | |
39279cc3 | 4818 | |
4df27c4d YZ |
4819 | leaf = path->nodes[0]; |
4820 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4821 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4822 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4823 | goto out; | |
39279cc3 | 4824 | |
4df27c4d YZ |
4825 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4826 | (unsigned long)(ref + 1), | |
4827 | dentry->d_name.len); | |
4828 | if (ret) | |
4829 | goto out; | |
4830 | ||
b3b4aa74 | 4831 | btrfs_release_path(path); |
4df27c4d YZ |
4832 | |
4833 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4834 | if (IS_ERR(new_root)) { | |
4835 | err = PTR_ERR(new_root); | |
4836 | goto out; | |
4837 | } | |
4838 | ||
4df27c4d YZ |
4839 | *sub_root = new_root; |
4840 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4841 | location->type = BTRFS_INODE_ITEM_KEY; | |
4842 | location->offset = 0; | |
4843 | err = 0; | |
4844 | out: | |
4845 | btrfs_free_path(path); | |
4846 | return err; | |
39279cc3 CM |
4847 | } |
4848 | ||
5d4f98a2 YZ |
4849 | static void inode_tree_add(struct inode *inode) |
4850 | { | |
4851 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4852 | struct btrfs_inode *entry; | |
03e860bd FNP |
4853 | struct rb_node **p; |
4854 | struct rb_node *parent; | |
cef21937 | 4855 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 4856 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 4857 | |
1d3382cb | 4858 | if (inode_unhashed(inode)) |
76dda93c | 4859 | return; |
e1409cef | 4860 | parent = NULL; |
5d4f98a2 | 4861 | spin_lock(&root->inode_lock); |
e1409cef | 4862 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
4863 | while (*p) { |
4864 | parent = *p; | |
4865 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4866 | ||
33345d01 | 4867 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4868 | p = &parent->rb_left; |
33345d01 | 4869 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4870 | p = &parent->rb_right; |
5d4f98a2 YZ |
4871 | else { |
4872 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4873 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 4874 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
4875 | RB_CLEAR_NODE(parent); |
4876 | spin_unlock(&root->inode_lock); | |
cef21937 | 4877 | return; |
5d4f98a2 YZ |
4878 | } |
4879 | } | |
cef21937 FDBM |
4880 | rb_link_node(new, parent, p); |
4881 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
4882 | spin_unlock(&root->inode_lock); |
4883 | } | |
4884 | ||
4885 | static void inode_tree_del(struct inode *inode) | |
4886 | { | |
4887 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4888 | int empty = 0; |
5d4f98a2 | 4889 | |
03e860bd | 4890 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4891 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4892 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4893 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4894 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4895 | } |
03e860bd | 4896 | spin_unlock(&root->inode_lock); |
76dda93c | 4897 | |
69e9c6c6 | 4898 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
4899 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4900 | spin_lock(&root->inode_lock); | |
4901 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4902 | spin_unlock(&root->inode_lock); | |
4903 | if (empty) | |
4904 | btrfs_add_dead_root(root); | |
4905 | } | |
4906 | } | |
4907 | ||
143bede5 | 4908 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4909 | { |
4910 | struct rb_node *node; | |
4911 | struct rb_node *prev; | |
4912 | struct btrfs_inode *entry; | |
4913 | struct inode *inode; | |
4914 | u64 objectid = 0; | |
4915 | ||
4916 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4917 | ||
4918 | spin_lock(&root->inode_lock); | |
4919 | again: | |
4920 | node = root->inode_tree.rb_node; | |
4921 | prev = NULL; | |
4922 | while (node) { | |
4923 | prev = node; | |
4924 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4925 | ||
33345d01 | 4926 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4927 | node = node->rb_left; |
33345d01 | 4928 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4929 | node = node->rb_right; |
4930 | else | |
4931 | break; | |
4932 | } | |
4933 | if (!node) { | |
4934 | while (prev) { | |
4935 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4936 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4937 | node = prev; |
4938 | break; | |
4939 | } | |
4940 | prev = rb_next(prev); | |
4941 | } | |
4942 | } | |
4943 | while (node) { | |
4944 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4945 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4946 | inode = igrab(&entry->vfs_inode); |
4947 | if (inode) { | |
4948 | spin_unlock(&root->inode_lock); | |
4949 | if (atomic_read(&inode->i_count) > 1) | |
4950 | d_prune_aliases(inode); | |
4951 | /* | |
45321ac5 | 4952 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4953 | * the inode cache when its usage count |
4954 | * hits zero. | |
4955 | */ | |
4956 | iput(inode); | |
4957 | cond_resched(); | |
4958 | spin_lock(&root->inode_lock); | |
4959 | goto again; | |
4960 | } | |
4961 | ||
4962 | if (cond_resched_lock(&root->inode_lock)) | |
4963 | goto again; | |
4964 | ||
4965 | node = rb_next(node); | |
4966 | } | |
4967 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4968 | } |
4969 | ||
e02119d5 CM |
4970 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4971 | { | |
4972 | struct btrfs_iget_args *args = p; | |
4973 | inode->i_ino = args->ino; | |
e02119d5 | 4974 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4975 | return 0; |
4976 | } | |
4977 | ||
4978 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4979 | { | |
4980 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4981 | return args->ino == btrfs_ino(inode) && |
d397712b | 4982 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4983 | } |
4984 | ||
5d4f98a2 YZ |
4985 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4986 | u64 objectid, | |
4987 | struct btrfs_root *root) | |
39279cc3 CM |
4988 | { |
4989 | struct inode *inode; | |
4990 | struct btrfs_iget_args args; | |
778ba82b FDBM |
4991 | unsigned long hashval = btrfs_inode_hash(objectid, root); |
4992 | ||
39279cc3 CM |
4993 | args.ino = objectid; |
4994 | args.root = root; | |
4995 | ||
778ba82b | 4996 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
4997 | btrfs_init_locked_inode, |
4998 | (void *)&args); | |
4999 | return inode; | |
5000 | } | |
5001 | ||
1a54ef8c BR |
5002 | /* Get an inode object given its location and corresponding root. |
5003 | * Returns in *is_new if the inode was read from disk | |
5004 | */ | |
5005 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5006 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5007 | { |
5008 | struct inode *inode; | |
5009 | ||
5010 | inode = btrfs_iget_locked(s, location->objectid, root); | |
5011 | if (!inode) | |
5d4f98a2 | 5012 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5013 | |
5014 | if (inode->i_state & I_NEW) { | |
5015 | BTRFS_I(inode)->root = root; | |
5016 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
5017 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
5018 | if (!is_bad_inode(inode)) { |
5019 | inode_tree_add(inode); | |
5020 | unlock_new_inode(inode); | |
5021 | if (new) | |
5022 | *new = 1; | |
5023 | } else { | |
e0b6d65b ST |
5024 | unlock_new_inode(inode); |
5025 | iput(inode); | |
5026 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5027 | } |
5028 | } | |
5029 | ||
1a54ef8c BR |
5030 | return inode; |
5031 | } | |
5032 | ||
4df27c4d YZ |
5033 | static struct inode *new_simple_dir(struct super_block *s, |
5034 | struct btrfs_key *key, | |
5035 | struct btrfs_root *root) | |
5036 | { | |
5037 | struct inode *inode = new_inode(s); | |
5038 | ||
5039 | if (!inode) | |
5040 | return ERR_PTR(-ENOMEM); | |
5041 | ||
4df27c4d YZ |
5042 | BTRFS_I(inode)->root = root; |
5043 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5044 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5045 | |
5046 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5047 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5048 | inode->i_fop = &simple_dir_operations; |
5049 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
5050 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
5051 | ||
5052 | return inode; | |
5053 | } | |
5054 | ||
3de4586c | 5055 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5056 | { |
d397712b | 5057 | struct inode *inode; |
4df27c4d | 5058 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5059 | struct btrfs_root *sub_root = root; |
5060 | struct btrfs_key location; | |
76dda93c | 5061 | int index; |
b4aff1f8 | 5062 | int ret = 0; |
39279cc3 CM |
5063 | |
5064 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5065 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5066 | |
39e3c955 | 5067 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5068 | if (ret < 0) |
5069 | return ERR_PTR(ret); | |
5f39d397 | 5070 | |
4df27c4d | 5071 | if (location.objectid == 0) |
5662344b | 5072 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5073 | |
5074 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5075 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5076 | return inode; |
5077 | } | |
5078 | ||
5079 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5080 | ||
76dda93c | 5081 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5082 | ret = fixup_tree_root_location(root, dir, dentry, |
5083 | &location, &sub_root); | |
5084 | if (ret < 0) { | |
5085 | if (ret != -ENOENT) | |
5086 | inode = ERR_PTR(ret); | |
5087 | else | |
5088 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5089 | } else { | |
73f73415 | 5090 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5091 | } |
76dda93c YZ |
5092 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5093 | ||
34d19bad | 5094 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5095 | down_read(&root->fs_info->cleanup_work_sem); |
5096 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5097 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5098 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5099 | if (ret) { |
5100 | iput(inode); | |
66b4ffd1 | 5101 | inode = ERR_PTR(ret); |
01cd3367 | 5102 | } |
c71bf099 YZ |
5103 | } |
5104 | ||
3de4586c CM |
5105 | return inode; |
5106 | } | |
5107 | ||
fe15ce44 | 5108 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5109 | { |
5110 | struct btrfs_root *root; | |
848cce0d | 5111 | struct inode *inode = dentry->d_inode; |
76dda93c | 5112 | |
848cce0d LZ |
5113 | if (!inode && !IS_ROOT(dentry)) |
5114 | inode = dentry->d_parent->d_inode; | |
76dda93c | 5115 | |
848cce0d LZ |
5116 | if (inode) { |
5117 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5118 | if (btrfs_root_refs(&root->root_item) == 0) |
5119 | return 1; | |
848cce0d LZ |
5120 | |
5121 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5122 | return 1; | |
efefb143 | 5123 | } |
76dda93c YZ |
5124 | return 0; |
5125 | } | |
5126 | ||
b4aff1f8 JB |
5127 | static void btrfs_dentry_release(struct dentry *dentry) |
5128 | { | |
5129 | if (dentry->d_fsdata) | |
5130 | kfree(dentry->d_fsdata); | |
5131 | } | |
5132 | ||
3de4586c | 5133 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5134 | unsigned int flags) |
3de4586c | 5135 | { |
5662344b | 5136 | struct inode *inode; |
a66e7cc6 | 5137 | |
5662344b TI |
5138 | inode = btrfs_lookup_dentry(dir, dentry); |
5139 | if (IS_ERR(inode)) { | |
5140 | if (PTR_ERR(inode) == -ENOENT) | |
5141 | inode = NULL; | |
5142 | else | |
5143 | return ERR_CAST(inode); | |
5144 | } | |
5145 | ||
5146 | return d_splice_alias(inode, dentry); | |
39279cc3 CM |
5147 | } |
5148 | ||
16cdcec7 | 5149 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5150 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5151 | }; | |
5152 | ||
9cdda8d3 | 5153 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5154 | { |
9cdda8d3 | 5155 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5156 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5157 | struct btrfs_item *item; | |
5158 | struct btrfs_dir_item *di; | |
5159 | struct btrfs_key key; | |
5f39d397 | 5160 | struct btrfs_key found_key; |
39279cc3 | 5161 | struct btrfs_path *path; |
16cdcec7 MX |
5162 | struct list_head ins_list; |
5163 | struct list_head del_list; | |
39279cc3 | 5164 | int ret; |
5f39d397 | 5165 | struct extent_buffer *leaf; |
39279cc3 | 5166 | int slot; |
39279cc3 CM |
5167 | unsigned char d_type; |
5168 | int over = 0; | |
5169 | u32 di_cur; | |
5170 | u32 di_total; | |
5171 | u32 di_len; | |
5172 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5173 | char tmp_name[32]; |
5174 | char *name_ptr; | |
5175 | int name_len; | |
9cdda8d3 | 5176 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5177 | |
5178 | /* FIXME, use a real flag for deciding about the key type */ | |
5179 | if (root->fs_info->tree_root == root) | |
5180 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5181 | |
9cdda8d3 AV |
5182 | if (!dir_emit_dots(file, ctx)) |
5183 | return 0; | |
5184 | ||
49593bfa | 5185 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5186 | if (!path) |
5187 | return -ENOMEM; | |
ff5714cc | 5188 | |
026fd317 | 5189 | path->reada = 1; |
49593bfa | 5190 | |
16cdcec7 MX |
5191 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5192 | INIT_LIST_HEAD(&ins_list); | |
5193 | INIT_LIST_HEAD(&del_list); | |
5194 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5195 | } | |
5196 | ||
39279cc3 | 5197 | btrfs_set_key_type(&key, key_type); |
9cdda8d3 | 5198 | key.offset = ctx->pos; |
33345d01 | 5199 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5200 | |
39279cc3 CM |
5201 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5202 | if (ret < 0) | |
5203 | goto err; | |
49593bfa DW |
5204 | |
5205 | while (1) { | |
5f39d397 | 5206 | leaf = path->nodes[0]; |
39279cc3 | 5207 | slot = path->slots[0]; |
b9e03af0 LZ |
5208 | if (slot >= btrfs_header_nritems(leaf)) { |
5209 | ret = btrfs_next_leaf(root, path); | |
5210 | if (ret < 0) | |
5211 | goto err; | |
5212 | else if (ret > 0) | |
5213 | break; | |
5214 | continue; | |
39279cc3 | 5215 | } |
3de4586c | 5216 | |
dd3cc16b | 5217 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5218 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5219 | ||
5220 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5221 | break; |
5f39d397 | 5222 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 5223 | break; |
9cdda8d3 | 5224 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5225 | goto next; |
16cdcec7 MX |
5226 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5227 | btrfs_should_delete_dir_index(&del_list, | |
5228 | found_key.offset)) | |
5229 | goto next; | |
5f39d397 | 5230 | |
9cdda8d3 | 5231 | ctx->pos = found_key.offset; |
16cdcec7 | 5232 | is_curr = 1; |
49593bfa | 5233 | |
39279cc3 CM |
5234 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5235 | di_cur = 0; | |
5f39d397 | 5236 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5237 | |
5238 | while (di_cur < di_total) { | |
5f39d397 CM |
5239 | struct btrfs_key location; |
5240 | ||
22a94d44 JB |
5241 | if (verify_dir_item(root, leaf, di)) |
5242 | break; | |
5243 | ||
5f39d397 | 5244 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5245 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5246 | name_ptr = tmp_name; |
5247 | } else { | |
5248 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5249 | if (!name_ptr) { |
5250 | ret = -ENOMEM; | |
5251 | goto err; | |
5252 | } | |
5f39d397 CM |
5253 | } |
5254 | read_extent_buffer(leaf, name_ptr, | |
5255 | (unsigned long)(di + 1), name_len); | |
5256 | ||
5257 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5258 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5259 | |
fede766f | 5260 | |
3de4586c | 5261 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5262 | * skip it. |
5263 | * | |
5264 | * In contrast to old kernels, we insert the snapshot's | |
5265 | * dir item and dir index after it has been created, so | |
5266 | * we won't find a reference to our own snapshot. We | |
5267 | * still keep the following code for backward | |
5268 | * compatibility. | |
3de4586c CM |
5269 | */ |
5270 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5271 | location.objectid == root->root_key.objectid) { | |
5272 | over = 0; | |
5273 | goto skip; | |
5274 | } | |
9cdda8d3 AV |
5275 | over = !dir_emit(ctx, name_ptr, name_len, |
5276 | location.objectid, d_type); | |
5f39d397 | 5277 | |
3de4586c | 5278 | skip: |
5f39d397 CM |
5279 | if (name_ptr != tmp_name) |
5280 | kfree(name_ptr); | |
5281 | ||
39279cc3 CM |
5282 | if (over) |
5283 | goto nopos; | |
5103e947 | 5284 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5285 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5286 | di_cur += di_len; |
5287 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5288 | } | |
b9e03af0 LZ |
5289 | next: |
5290 | path->slots[0]++; | |
39279cc3 | 5291 | } |
49593bfa | 5292 | |
16cdcec7 MX |
5293 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5294 | if (is_curr) | |
9cdda8d3 AV |
5295 | ctx->pos++; |
5296 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5297 | if (ret) |
5298 | goto nopos; | |
5299 | } | |
5300 | ||
49593bfa | 5301 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5302 | ctx->pos++; |
5303 | ||
5304 | /* | |
5305 | * Stop new entries from being returned after we return the last | |
5306 | * entry. | |
5307 | * | |
5308 | * New directory entries are assigned a strictly increasing | |
5309 | * offset. This means that new entries created during readdir | |
5310 | * are *guaranteed* to be seen in the future by that readdir. | |
5311 | * This has broken buggy programs which operate on names as | |
5312 | * they're returned by readdir. Until we re-use freed offsets | |
5313 | * we have this hack to stop new entries from being returned | |
5314 | * under the assumption that they'll never reach this huge | |
5315 | * offset. | |
5316 | * | |
5317 | * This is being careful not to overflow 32bit loff_t unless the | |
5318 | * last entry requires it because doing so has broken 32bit apps | |
5319 | * in the past. | |
5320 | */ | |
5321 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5322 | if (ctx->pos >= INT_MAX) | |
5323 | ctx->pos = LLONG_MAX; | |
5324 | else | |
5325 | ctx->pos = INT_MAX; | |
5326 | } | |
39279cc3 CM |
5327 | nopos: |
5328 | ret = 0; | |
5329 | err: | |
16cdcec7 MX |
5330 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5331 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5332 | btrfs_free_path(path); |
39279cc3 CM |
5333 | return ret; |
5334 | } | |
5335 | ||
a9185b41 | 5336 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5337 | { |
5338 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5339 | struct btrfs_trans_handle *trans; | |
5340 | int ret = 0; | |
0af3d00b | 5341 | bool nolock = false; |
39279cc3 | 5342 | |
72ac3c0d | 5343 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5344 | return 0; |
5345 | ||
83eea1f1 | 5346 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5347 | nolock = true; |
0af3d00b | 5348 | |
a9185b41 | 5349 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5350 | if (nolock) |
7a7eaa40 | 5351 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5352 | else |
7a7eaa40 | 5353 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5354 | if (IS_ERR(trans)) |
5355 | return PTR_ERR(trans); | |
a698d075 | 5356 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5357 | } |
5358 | return ret; | |
5359 | } | |
5360 | ||
5361 | /* | |
54aa1f4d | 5362 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5363 | * inode changes. But, it is most likely to find the inode in cache. |
5364 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5365 | * to keep or drop this code. | |
5366 | */ | |
48a3b636 | 5367 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5368 | { |
5369 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5370 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5371 | int ret; |
5372 | ||
72ac3c0d | 5373 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5374 | return 0; |
39279cc3 | 5375 | |
7a7eaa40 | 5376 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5377 | if (IS_ERR(trans)) |
5378 | return PTR_ERR(trans); | |
8929ecfa YZ |
5379 | |
5380 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5381 | if (ret && ret == -ENOSPC) { |
5382 | /* whoops, lets try again with the full transaction */ | |
5383 | btrfs_end_transaction(trans, root); | |
5384 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5385 | if (IS_ERR(trans)) |
5386 | return PTR_ERR(trans); | |
8929ecfa | 5387 | |
94b60442 | 5388 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5389 | } |
39279cc3 | 5390 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5391 | if (BTRFS_I(inode)->delayed_node) |
5392 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5393 | |
5394 | return ret; | |
5395 | } | |
5396 | ||
5397 | /* | |
5398 | * This is a copy of file_update_time. We need this so we can return error on | |
5399 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5400 | */ | |
e41f941a JB |
5401 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5402 | int flags) | |
22c44fe6 | 5403 | { |
2bc55652 AB |
5404 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5405 | ||
5406 | if (btrfs_root_readonly(root)) | |
5407 | return -EROFS; | |
5408 | ||
e41f941a | 5409 | if (flags & S_VERSION) |
22c44fe6 | 5410 | inode_inc_iversion(inode); |
e41f941a JB |
5411 | if (flags & S_CTIME) |
5412 | inode->i_ctime = *now; | |
5413 | if (flags & S_MTIME) | |
5414 | inode->i_mtime = *now; | |
5415 | if (flags & S_ATIME) | |
5416 | inode->i_atime = *now; | |
5417 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5418 | } |
5419 | ||
d352ac68 CM |
5420 | /* |
5421 | * find the highest existing sequence number in a directory | |
5422 | * and then set the in-memory index_cnt variable to reflect | |
5423 | * free sequence numbers | |
5424 | */ | |
aec7477b JB |
5425 | static int btrfs_set_inode_index_count(struct inode *inode) |
5426 | { | |
5427 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5428 | struct btrfs_key key, found_key; | |
5429 | struct btrfs_path *path; | |
5430 | struct extent_buffer *leaf; | |
5431 | int ret; | |
5432 | ||
33345d01 | 5433 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
5434 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
5435 | key.offset = (u64)-1; | |
5436 | ||
5437 | path = btrfs_alloc_path(); | |
5438 | if (!path) | |
5439 | return -ENOMEM; | |
5440 | ||
5441 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5442 | if (ret < 0) | |
5443 | goto out; | |
5444 | /* FIXME: we should be able to handle this */ | |
5445 | if (ret == 0) | |
5446 | goto out; | |
5447 | ret = 0; | |
5448 | ||
5449 | /* | |
5450 | * MAGIC NUMBER EXPLANATION: | |
5451 | * since we search a directory based on f_pos we have to start at 2 | |
5452 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5453 | * else has to start at 2 | |
5454 | */ | |
5455 | if (path->slots[0] == 0) { | |
5456 | BTRFS_I(inode)->index_cnt = 2; | |
5457 | goto out; | |
5458 | } | |
5459 | ||
5460 | path->slots[0]--; | |
5461 | ||
5462 | leaf = path->nodes[0]; | |
5463 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5464 | ||
33345d01 | 5465 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
5466 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
5467 | BTRFS_I(inode)->index_cnt = 2; | |
5468 | goto out; | |
5469 | } | |
5470 | ||
5471 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5472 | out: | |
5473 | btrfs_free_path(path); | |
5474 | return ret; | |
5475 | } | |
5476 | ||
d352ac68 CM |
5477 | /* |
5478 | * helper to find a free sequence number in a given directory. This current | |
5479 | * code is very simple, later versions will do smarter things in the btree | |
5480 | */ | |
3de4586c | 5481 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5482 | { |
5483 | int ret = 0; | |
5484 | ||
5485 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5486 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5487 | if (ret) { | |
5488 | ret = btrfs_set_inode_index_count(dir); | |
5489 | if (ret) | |
5490 | return ret; | |
5491 | } | |
aec7477b JB |
5492 | } |
5493 | ||
00e4e6b3 | 5494 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5495 | BTRFS_I(dir)->index_cnt++; |
5496 | ||
5497 | return ret; | |
5498 | } | |
5499 | ||
39279cc3 CM |
5500 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5501 | struct btrfs_root *root, | |
aec7477b | 5502 | struct inode *dir, |
9c58309d | 5503 | const char *name, int name_len, |
175a4eb7 AV |
5504 | u64 ref_objectid, u64 objectid, |
5505 | umode_t mode, u64 *index) | |
39279cc3 CM |
5506 | { |
5507 | struct inode *inode; | |
5f39d397 | 5508 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5509 | struct btrfs_key *location; |
5f39d397 | 5510 | struct btrfs_path *path; |
9c58309d CM |
5511 | struct btrfs_inode_ref *ref; |
5512 | struct btrfs_key key[2]; | |
5513 | u32 sizes[2]; | |
5514 | unsigned long ptr; | |
39279cc3 | 5515 | int ret; |
39279cc3 | 5516 | |
5f39d397 | 5517 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5518 | if (!path) |
5519 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5520 | |
39279cc3 | 5521 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5522 | if (!inode) { |
5523 | btrfs_free_path(path); | |
39279cc3 | 5524 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5525 | } |
39279cc3 | 5526 | |
581bb050 LZ |
5527 | /* |
5528 | * we have to initialize this early, so we can reclaim the inode | |
5529 | * number if we fail afterwards in this function. | |
5530 | */ | |
5531 | inode->i_ino = objectid; | |
5532 | ||
aec7477b | 5533 | if (dir) { |
1abe9b8a | 5534 | trace_btrfs_inode_request(dir); |
5535 | ||
3de4586c | 5536 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5537 | if (ret) { |
8fb27640 | 5538 | btrfs_free_path(path); |
09771430 | 5539 | iput(inode); |
aec7477b | 5540 | return ERR_PTR(ret); |
09771430 | 5541 | } |
aec7477b JB |
5542 | } |
5543 | /* | |
5544 | * index_cnt is ignored for everything but a dir, | |
5545 | * btrfs_get_inode_index_count has an explanation for the magic | |
5546 | * number | |
5547 | */ | |
5548 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 5549 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 5550 | BTRFS_I(inode)->root = root; |
e02119d5 | 5551 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5552 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5553 | |
5dc562c5 JB |
5554 | /* |
5555 | * We could have gotten an inode number from somebody who was fsynced | |
5556 | * and then removed in this same transaction, so let's just set full | |
5557 | * sync since it will be a full sync anyway and this will blow away the | |
5558 | * old info in the log. | |
5559 | */ | |
5560 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5561 | ||
9c58309d CM |
5562 | key[0].objectid = objectid; |
5563 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
5564 | key[0].offset = 0; | |
5565 | ||
f186373f MF |
5566 | /* |
5567 | * Start new inodes with an inode_ref. This is slightly more | |
5568 | * efficient for small numbers of hard links since they will | |
5569 | * be packed into one item. Extended refs will kick in if we | |
5570 | * add more hard links than can fit in the ref item. | |
5571 | */ | |
9c58309d CM |
5572 | key[1].objectid = objectid; |
5573 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
5574 | key[1].offset = ref_objectid; | |
5575 | ||
5576 | sizes[0] = sizeof(struct btrfs_inode_item); | |
5577 | sizes[1] = name_len + sizeof(*ref); | |
5578 | ||
b9473439 | 5579 | path->leave_spinning = 1; |
9c58309d CM |
5580 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
5581 | if (ret != 0) | |
5f39d397 CM |
5582 | goto fail; |
5583 | ||
ecc11fab | 5584 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5585 | inode_set_bytes(inode, 0); |
39279cc3 | 5586 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5587 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5588 | struct btrfs_inode_item); | |
293f7e07 LZ |
5589 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5590 | sizeof(*inode_item)); | |
e02119d5 | 5591 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
5592 | |
5593 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5594 | struct btrfs_inode_ref); | |
5595 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 5596 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
5597 | ptr = (unsigned long)(ref + 1); |
5598 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5599 | ||
5f39d397 CM |
5600 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5601 | btrfs_free_path(path); | |
5602 | ||
39279cc3 CM |
5603 | location = &BTRFS_I(inode)->location; |
5604 | location->objectid = objectid; | |
39279cc3 CM |
5605 | location->offset = 0; |
5606 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
5607 | ||
6cbff00f CH |
5608 | btrfs_inherit_iflags(inode, dir); |
5609 | ||
569254b0 | 5610 | if (S_ISREG(mode)) { |
94272164 CM |
5611 | if (btrfs_test_opt(root, NODATASUM)) |
5612 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5613 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5614 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5615 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5616 | } |
5617 | ||
778ba82b | 5618 | btrfs_insert_inode_hash(inode); |
5d4f98a2 | 5619 | inode_tree_add(inode); |
1abe9b8a | 5620 | |
5621 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5622 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5623 | |
8ea05e3a AB |
5624 | btrfs_update_root_times(trans, root); |
5625 | ||
63541927 FDBM |
5626 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
5627 | if (ret) | |
5628 | btrfs_err(root->fs_info, | |
5629 | "error inheriting props for ino %llu (root %llu): %d", | |
5630 | btrfs_ino(inode), root->root_key.objectid, ret); | |
5631 | ||
39279cc3 | 5632 | return inode; |
5f39d397 | 5633 | fail: |
aec7477b JB |
5634 | if (dir) |
5635 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 5636 | btrfs_free_path(path); |
09771430 | 5637 | iput(inode); |
5f39d397 | 5638 | return ERR_PTR(ret); |
39279cc3 CM |
5639 | } |
5640 | ||
5641 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5642 | { | |
5643 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5644 | } | |
5645 | ||
d352ac68 CM |
5646 | /* |
5647 | * utility function to add 'inode' into 'parent_inode' with | |
5648 | * a give name and a given sequence number. | |
5649 | * if 'add_backref' is true, also insert a backref from the | |
5650 | * inode to the parent directory. | |
5651 | */ | |
e02119d5 CM |
5652 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5653 | struct inode *parent_inode, struct inode *inode, | |
5654 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5655 | { |
4df27c4d | 5656 | int ret = 0; |
39279cc3 | 5657 | struct btrfs_key key; |
e02119d5 | 5658 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5659 | u64 ino = btrfs_ino(inode); |
5660 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5661 | |
33345d01 | 5662 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5663 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5664 | } else { | |
33345d01 | 5665 | key.objectid = ino; |
4df27c4d YZ |
5666 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
5667 | key.offset = 0; | |
5668 | } | |
5669 | ||
33345d01 | 5670 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5671 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5672 | key.objectid, root->root_key.objectid, | |
33345d01 | 5673 | parent_ino, index, name, name_len); |
4df27c4d | 5674 | } else if (add_backref) { |
33345d01 LZ |
5675 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5676 | parent_ino, index); | |
4df27c4d | 5677 | } |
39279cc3 | 5678 | |
79787eaa JM |
5679 | /* Nothing to clean up yet */ |
5680 | if (ret) | |
5681 | return ret; | |
4df27c4d | 5682 | |
79787eaa JM |
5683 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5684 | parent_inode, &key, | |
5685 | btrfs_inode_type(inode), index); | |
9c52057c | 5686 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5687 | goto fail_dir_item; |
5688 | else if (ret) { | |
5689 | btrfs_abort_transaction(trans, root, ret); | |
5690 | return ret; | |
39279cc3 | 5691 | } |
79787eaa JM |
5692 | |
5693 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5694 | name_len * 2); | |
0c4d2d95 | 5695 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5696 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5697 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5698 | if (ret) | |
5699 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5700 | return ret; |
fe66a05a CM |
5701 | |
5702 | fail_dir_item: | |
5703 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5704 | u64 local_index; | |
5705 | int err; | |
5706 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5707 | key.objectid, root->root_key.objectid, | |
5708 | parent_ino, &local_index, name, name_len); | |
5709 | ||
5710 | } else if (add_backref) { | |
5711 | u64 local_index; | |
5712 | int err; | |
5713 | ||
5714 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5715 | ino, parent_ino, &local_index); | |
5716 | } | |
5717 | return ret; | |
39279cc3 CM |
5718 | } |
5719 | ||
5720 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5721 | struct inode *dir, struct dentry *dentry, |
5722 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5723 | { |
a1b075d2 JB |
5724 | int err = btrfs_add_link(trans, dir, inode, |
5725 | dentry->d_name.name, dentry->d_name.len, | |
5726 | backref, index); | |
39279cc3 CM |
5727 | if (err > 0) |
5728 | err = -EEXIST; | |
5729 | return err; | |
5730 | } | |
5731 | ||
618e21d5 | 5732 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5733 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5734 | { |
5735 | struct btrfs_trans_handle *trans; | |
5736 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5737 | struct inode *inode = NULL; |
618e21d5 JB |
5738 | int err; |
5739 | int drop_inode = 0; | |
5740 | u64 objectid; | |
00e4e6b3 | 5741 | u64 index = 0; |
618e21d5 JB |
5742 | |
5743 | if (!new_valid_dev(rdev)) | |
5744 | return -EINVAL; | |
5745 | ||
9ed74f2d JB |
5746 | /* |
5747 | * 2 for inode item and ref | |
5748 | * 2 for dir items | |
5749 | * 1 for xattr if selinux is on | |
5750 | */ | |
a22285a6 YZ |
5751 | trans = btrfs_start_transaction(root, 5); |
5752 | if (IS_ERR(trans)) | |
5753 | return PTR_ERR(trans); | |
1832a6d5 | 5754 | |
581bb050 LZ |
5755 | err = btrfs_find_free_ino(root, &objectid); |
5756 | if (err) | |
5757 | goto out_unlock; | |
5758 | ||
aec7477b | 5759 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5760 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5761 | mode, &index); |
7cf96da3 TI |
5762 | if (IS_ERR(inode)) { |
5763 | err = PTR_ERR(inode); | |
618e21d5 | 5764 | goto out_unlock; |
7cf96da3 | 5765 | } |
618e21d5 | 5766 | |
2a7dba39 | 5767 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5768 | if (err) { |
5769 | drop_inode = 1; | |
5770 | goto out_unlock; | |
5771 | } | |
5772 | ||
ad19db71 CS |
5773 | /* |
5774 | * If the active LSM wants to access the inode during | |
5775 | * d_instantiate it needs these. Smack checks to see | |
5776 | * if the filesystem supports xattrs by looking at the | |
5777 | * ops vector. | |
5778 | */ | |
5779 | ||
5780 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5781 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5782 | if (err) |
5783 | drop_inode = 1; | |
5784 | else { | |
618e21d5 | 5785 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5786 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5787 | d_instantiate(dentry, inode); |
618e21d5 | 5788 | } |
618e21d5 | 5789 | out_unlock: |
7ad85bb7 | 5790 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5791 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5792 | if (drop_inode) { |
5793 | inode_dec_link_count(inode); | |
5794 | iput(inode); | |
5795 | } | |
618e21d5 JB |
5796 | return err; |
5797 | } | |
5798 | ||
39279cc3 | 5799 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5800 | umode_t mode, bool excl) |
39279cc3 CM |
5801 | { |
5802 | struct btrfs_trans_handle *trans; | |
5803 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5804 | struct inode *inode = NULL; |
43baa579 | 5805 | int drop_inode_on_err = 0; |
a22285a6 | 5806 | int err; |
39279cc3 | 5807 | u64 objectid; |
00e4e6b3 | 5808 | u64 index = 0; |
39279cc3 | 5809 | |
9ed74f2d JB |
5810 | /* |
5811 | * 2 for inode item and ref | |
5812 | * 2 for dir items | |
5813 | * 1 for xattr if selinux is on | |
5814 | */ | |
a22285a6 YZ |
5815 | trans = btrfs_start_transaction(root, 5); |
5816 | if (IS_ERR(trans)) | |
5817 | return PTR_ERR(trans); | |
9ed74f2d | 5818 | |
581bb050 LZ |
5819 | err = btrfs_find_free_ino(root, &objectid); |
5820 | if (err) | |
5821 | goto out_unlock; | |
5822 | ||
aec7477b | 5823 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5824 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5825 | mode, &index); |
7cf96da3 TI |
5826 | if (IS_ERR(inode)) { |
5827 | err = PTR_ERR(inode); | |
39279cc3 | 5828 | goto out_unlock; |
7cf96da3 | 5829 | } |
43baa579 | 5830 | drop_inode_on_err = 1; |
39279cc3 | 5831 | |
2a7dba39 | 5832 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5833 | if (err) |
33268eaf | 5834 | goto out_unlock; |
33268eaf | 5835 | |
9185aa58 FB |
5836 | err = btrfs_update_inode(trans, root, inode); |
5837 | if (err) | |
5838 | goto out_unlock; | |
5839 | ||
ad19db71 CS |
5840 | /* |
5841 | * If the active LSM wants to access the inode during | |
5842 | * d_instantiate it needs these. Smack checks to see | |
5843 | * if the filesystem supports xattrs by looking at the | |
5844 | * ops vector. | |
5845 | */ | |
5846 | inode->i_fop = &btrfs_file_operations; | |
5847 | inode->i_op = &btrfs_file_inode_operations; | |
5848 | ||
a1b075d2 | 5849 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5850 | if (err) |
43baa579 FB |
5851 | goto out_unlock; |
5852 | ||
5853 | inode->i_mapping->a_ops = &btrfs_aops; | |
5854 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5855 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5856 | d_instantiate(dentry, inode); | |
5857 | ||
39279cc3 | 5858 | out_unlock: |
7ad85bb7 | 5859 | btrfs_end_transaction(trans, root); |
43baa579 | 5860 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5861 | inode_dec_link_count(inode); |
5862 | iput(inode); | |
5863 | } | |
b53d3f5d | 5864 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5865 | return err; |
5866 | } | |
5867 | ||
5868 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5869 | struct dentry *dentry) | |
5870 | { | |
5871 | struct btrfs_trans_handle *trans; | |
5872 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5873 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5874 | u64 index; |
39279cc3 CM |
5875 | int err; |
5876 | int drop_inode = 0; | |
5877 | ||
4a8be425 TH |
5878 | /* do not allow sys_link's with other subvols of the same device */ |
5879 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5880 | return -EXDEV; |
4a8be425 | 5881 | |
f186373f | 5882 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5883 | return -EMLINK; |
4a8be425 | 5884 | |
3de4586c | 5885 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5886 | if (err) |
5887 | goto fail; | |
5888 | ||
a22285a6 | 5889 | /* |
7e6b6465 | 5890 | * 2 items for inode and inode ref |
a22285a6 | 5891 | * 2 items for dir items |
7e6b6465 | 5892 | * 1 item for parent inode |
a22285a6 | 5893 | */ |
7e6b6465 | 5894 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5895 | if (IS_ERR(trans)) { |
5896 | err = PTR_ERR(trans); | |
5897 | goto fail; | |
5898 | } | |
5f39d397 | 5899 | |
67de1176 MX |
5900 | /* There are several dir indexes for this inode, clear the cache. */ |
5901 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 5902 | inc_nlink(inode); |
0c4d2d95 | 5903 | inode_inc_iversion(inode); |
3153495d | 5904 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5905 | ihold(inode); |
e9976151 | 5906 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5907 | |
a1b075d2 | 5908 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5909 | |
a5719521 | 5910 | if (err) { |
54aa1f4d | 5911 | drop_inode = 1; |
a5719521 | 5912 | } else { |
10d9f309 | 5913 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5914 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5915 | if (err) |
5916 | goto fail; | |
08c422c2 | 5917 | d_instantiate(dentry, inode); |
6a912213 | 5918 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5919 | } |
39279cc3 | 5920 | |
7ad85bb7 | 5921 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5922 | fail: |
39279cc3 CM |
5923 | if (drop_inode) { |
5924 | inode_dec_link_count(inode); | |
5925 | iput(inode); | |
5926 | } | |
b53d3f5d | 5927 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5928 | return err; |
5929 | } | |
5930 | ||
18bb1db3 | 5931 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5932 | { |
b9d86667 | 5933 | struct inode *inode = NULL; |
39279cc3 CM |
5934 | struct btrfs_trans_handle *trans; |
5935 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5936 | int err = 0; | |
5937 | int drop_on_err = 0; | |
b9d86667 | 5938 | u64 objectid = 0; |
00e4e6b3 | 5939 | u64 index = 0; |
39279cc3 | 5940 | |
9ed74f2d JB |
5941 | /* |
5942 | * 2 items for inode and ref | |
5943 | * 2 items for dir items | |
5944 | * 1 for xattr if selinux is on | |
5945 | */ | |
a22285a6 YZ |
5946 | trans = btrfs_start_transaction(root, 5); |
5947 | if (IS_ERR(trans)) | |
5948 | return PTR_ERR(trans); | |
39279cc3 | 5949 | |
581bb050 LZ |
5950 | err = btrfs_find_free_ino(root, &objectid); |
5951 | if (err) | |
5952 | goto out_fail; | |
5953 | ||
aec7477b | 5954 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5955 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5956 | S_IFDIR | mode, &index); |
39279cc3 CM |
5957 | if (IS_ERR(inode)) { |
5958 | err = PTR_ERR(inode); | |
5959 | goto out_fail; | |
5960 | } | |
5f39d397 | 5961 | |
39279cc3 | 5962 | drop_on_err = 1; |
33268eaf | 5963 | |
2a7dba39 | 5964 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5965 | if (err) |
5966 | goto out_fail; | |
5967 | ||
39279cc3 CM |
5968 | inode->i_op = &btrfs_dir_inode_operations; |
5969 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5970 | |
dbe674a9 | 5971 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5972 | err = btrfs_update_inode(trans, root, inode); |
5973 | if (err) | |
5974 | goto out_fail; | |
5f39d397 | 5975 | |
a1b075d2 JB |
5976 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5977 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5978 | if (err) |
5979 | goto out_fail; | |
5f39d397 | 5980 | |
39279cc3 CM |
5981 | d_instantiate(dentry, inode); |
5982 | drop_on_err = 0; | |
39279cc3 CM |
5983 | |
5984 | out_fail: | |
7ad85bb7 | 5985 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5986 | if (drop_on_err) |
5987 | iput(inode); | |
b53d3f5d | 5988 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5989 | return err; |
5990 | } | |
5991 | ||
d352ac68 CM |
5992 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5993 | * and an extent that you want to insert, deal with overlap and insert | |
5994 | * the new extent into the tree. | |
5995 | */ | |
3b951516 CM |
5996 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5997 | struct extent_map *existing, | |
e6dcd2dc CM |
5998 | struct extent_map *em, |
5999 | u64 map_start, u64 map_len) | |
3b951516 CM |
6000 | { |
6001 | u64 start_diff; | |
3b951516 | 6002 | |
e6dcd2dc CM |
6003 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
6004 | start_diff = map_start - em->start; | |
6005 | em->start = map_start; | |
6006 | em->len = map_len; | |
c8b97818 CM |
6007 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6008 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6009 | em->block_start += start_diff; |
c8b97818 CM |
6010 | em->block_len -= start_diff; |
6011 | } | |
09a2a8f9 | 6012 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6013 | } |
6014 | ||
c8b97818 CM |
6015 | static noinline int uncompress_inline(struct btrfs_path *path, |
6016 | struct inode *inode, struct page *page, | |
6017 | size_t pg_offset, u64 extent_offset, | |
6018 | struct btrfs_file_extent_item *item) | |
6019 | { | |
6020 | int ret; | |
6021 | struct extent_buffer *leaf = path->nodes[0]; | |
6022 | char *tmp; | |
6023 | size_t max_size; | |
6024 | unsigned long inline_size; | |
6025 | unsigned long ptr; | |
261507a0 | 6026 | int compress_type; |
c8b97818 CM |
6027 | |
6028 | WARN_ON(pg_offset != 0); | |
261507a0 | 6029 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6030 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6031 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6032 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6033 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6034 | if (!tmp) |
6035 | return -ENOMEM; | |
c8b97818 CM |
6036 | ptr = btrfs_file_extent_inline_start(item); |
6037 | ||
6038 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6039 | ||
5b050f04 | 6040 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
6041 | ret = btrfs_decompress(compress_type, tmp, page, |
6042 | extent_offset, inline_size, max_size); | |
c8b97818 | 6043 | if (ret) { |
7ac687d9 | 6044 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
6045 | unsigned long copy_size = min_t(u64, |
6046 | PAGE_CACHE_SIZE - pg_offset, | |
6047 | max_size - extent_offset); | |
6048 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 6049 | kunmap_atomic(kaddr); |
c8b97818 CM |
6050 | } |
6051 | kfree(tmp); | |
6052 | return 0; | |
6053 | } | |
6054 | ||
d352ac68 CM |
6055 | /* |
6056 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6057 | * the ugly parts come from merging extents from the disk with the in-ram |
6058 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6059 | * where the in-ram extents might be locked pending data=ordered completion. |
6060 | * | |
6061 | * This also copies inline extents directly into the page. | |
6062 | */ | |
d397712b | 6063 | |
a52d9a80 | 6064 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6065 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6066 | int create) |
6067 | { | |
6068 | int ret; | |
6069 | int err = 0; | |
db94535d | 6070 | u64 bytenr; |
a52d9a80 CM |
6071 | u64 extent_start = 0; |
6072 | u64 extent_end = 0; | |
33345d01 | 6073 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6074 | u32 found_type; |
f421950f | 6075 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6076 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6077 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6078 | struct extent_buffer *leaf; |
6079 | struct btrfs_key found_key; | |
a52d9a80 CM |
6080 | struct extent_map *em = NULL; |
6081 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6082 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6083 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 6084 | int compress_type; |
a52d9a80 | 6085 | |
a52d9a80 | 6086 | again: |
890871be | 6087 | read_lock(&em_tree->lock); |
d1310b2e | 6088 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6089 | if (em) |
6090 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6091 | read_unlock(&em_tree->lock); |
d1310b2e | 6092 | |
a52d9a80 | 6093 | if (em) { |
e1c4b745 CM |
6094 | if (em->start > start || em->start + em->len <= start) |
6095 | free_extent_map(em); | |
6096 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6097 | free_extent_map(em); |
6098 | else | |
6099 | goto out; | |
a52d9a80 | 6100 | } |
172ddd60 | 6101 | em = alloc_extent_map(); |
a52d9a80 | 6102 | if (!em) { |
d1310b2e CM |
6103 | err = -ENOMEM; |
6104 | goto out; | |
a52d9a80 | 6105 | } |
e6dcd2dc | 6106 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6107 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6108 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6109 | em->len = (u64)-1; |
c8b97818 | 6110 | em->block_len = (u64)-1; |
f421950f CM |
6111 | |
6112 | if (!path) { | |
6113 | path = btrfs_alloc_path(); | |
026fd317 JB |
6114 | if (!path) { |
6115 | err = -ENOMEM; | |
6116 | goto out; | |
6117 | } | |
6118 | /* | |
6119 | * Chances are we'll be called again, so go ahead and do | |
6120 | * readahead | |
6121 | */ | |
6122 | path->reada = 1; | |
f421950f CM |
6123 | } |
6124 | ||
179e29e4 CM |
6125 | ret = btrfs_lookup_file_extent(trans, root, path, |
6126 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6127 | if (ret < 0) { |
6128 | err = ret; | |
6129 | goto out; | |
6130 | } | |
6131 | ||
6132 | if (ret != 0) { | |
6133 | if (path->slots[0] == 0) | |
6134 | goto not_found; | |
6135 | path->slots[0]--; | |
6136 | } | |
6137 | ||
5f39d397 CM |
6138 | leaf = path->nodes[0]; |
6139 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6140 | struct btrfs_file_extent_item); |
a52d9a80 | 6141 | /* are we inside the extent that was found? */ |
5f39d397 CM |
6142 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6143 | found_type = btrfs_key_type(&found_key); | |
6144 | if (found_key.objectid != objectid || | |
a52d9a80 | 6145 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6146 | /* |
6147 | * If we backup past the first extent we want to move forward | |
6148 | * and see if there is an extent in front of us, otherwise we'll | |
6149 | * say there is a hole for our whole search range which can | |
6150 | * cause problems. | |
6151 | */ | |
6152 | extent_end = start; | |
6153 | goto next; | |
a52d9a80 CM |
6154 | } |
6155 | ||
5f39d397 CM |
6156 | found_type = btrfs_file_extent_type(leaf, item); |
6157 | extent_start = found_key.offset; | |
261507a0 | 6158 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
6159 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6160 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6161 | extent_end = extent_start + |
db94535d | 6162 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6163 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6164 | size_t size; | |
6165 | size = btrfs_file_extent_inline_len(leaf, item); | |
fda2832f | 6166 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6167 | } |
25a50341 | 6168 | next: |
9036c102 YZ |
6169 | if (start >= extent_end) { |
6170 | path->slots[0]++; | |
6171 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6172 | ret = btrfs_next_leaf(root, path); | |
6173 | if (ret < 0) { | |
6174 | err = ret; | |
6175 | goto out; | |
a52d9a80 | 6176 | } |
9036c102 YZ |
6177 | if (ret > 0) |
6178 | goto not_found; | |
6179 | leaf = path->nodes[0]; | |
a52d9a80 | 6180 | } |
9036c102 YZ |
6181 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6182 | if (found_key.objectid != objectid || | |
6183 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6184 | goto not_found; | |
6185 | if (start + len <= found_key.offset) | |
6186 | goto not_found; | |
6187 | em->start = start; | |
70c8a91c | 6188 | em->orig_start = start; |
9036c102 YZ |
6189 | em->len = found_key.offset - start; |
6190 | goto not_found_em; | |
6191 | } | |
6192 | ||
cc95bef6 | 6193 | em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); |
d899e052 YZ |
6194 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6195 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
6196 | em->start = extent_start; |
6197 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
6198 | em->orig_start = extent_start - |
6199 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
6200 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
6201 | item); | |
db94535d CM |
6202 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
6203 | if (bytenr == 0) { | |
5f39d397 | 6204 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
6205 | goto insert; |
6206 | } | |
261507a0 | 6207 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 6208 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 6209 | em->compress_type = compress_type; |
c8b97818 | 6210 | em->block_start = bytenr; |
b4939680 | 6211 | em->block_len = em->orig_block_len; |
c8b97818 CM |
6212 | } else { |
6213 | bytenr += btrfs_file_extent_offset(leaf, item); | |
6214 | em->block_start = bytenr; | |
6215 | em->block_len = em->len; | |
d899e052 YZ |
6216 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
6217 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 6218 | } |
a52d9a80 CM |
6219 | goto insert; |
6220 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6221 | unsigned long ptr; |
a52d9a80 | 6222 | char *map; |
3326d1b0 CM |
6223 | size_t size; |
6224 | size_t extent_offset; | |
6225 | size_t copy_size; | |
a52d9a80 | 6226 | |
689f9346 | 6227 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 6228 | if (!page || create) { |
689f9346 | 6229 | em->start = extent_start; |
9036c102 | 6230 | em->len = extent_end - extent_start; |
689f9346 Y |
6231 | goto out; |
6232 | } | |
5f39d397 | 6233 | |
9036c102 YZ |
6234 | size = btrfs_file_extent_inline_len(leaf, item); |
6235 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 6236 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6237 | size - extent_offset); |
3326d1b0 | 6238 | em->start = extent_start + extent_offset; |
fda2832f | 6239 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6240 | em->orig_block_len = em->len; |
70c8a91c | 6241 | em->orig_start = em->start; |
261507a0 | 6242 | if (compress_type) { |
c8b97818 | 6243 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
6244 | em->compress_type = compress_type; |
6245 | } | |
689f9346 | 6246 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6247 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6248 | if (btrfs_file_extent_compression(leaf, item) != |
6249 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6250 | ret = uncompress_inline(path, inode, page, |
6251 | pg_offset, | |
6252 | extent_offset, item); | |
79787eaa | 6253 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
6254 | } else { |
6255 | map = kmap(page); | |
6256 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6257 | copy_size); | |
93c82d57 CM |
6258 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6259 | memset(map + pg_offset + copy_size, 0, | |
6260 | PAGE_CACHE_SIZE - pg_offset - | |
6261 | copy_size); | |
6262 | } | |
c8b97818 CM |
6263 | kunmap(page); |
6264 | } | |
179e29e4 CM |
6265 | flush_dcache_page(page); |
6266 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6267 | BUG(); |
179e29e4 CM |
6268 | if (!trans) { |
6269 | kunmap(page); | |
6270 | free_extent_map(em); | |
6271 | em = NULL; | |
ff5714cc | 6272 | |
b3b4aa74 | 6273 | btrfs_release_path(path); |
7a7eaa40 | 6274 | trans = btrfs_join_transaction(root); |
ff5714cc | 6275 | |
3612b495 TI |
6276 | if (IS_ERR(trans)) |
6277 | return ERR_CAST(trans); | |
179e29e4 CM |
6278 | goto again; |
6279 | } | |
c8b97818 | 6280 | map = kmap(page); |
70dec807 | 6281 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6282 | copy_size); |
c8b97818 | 6283 | kunmap(page); |
179e29e4 | 6284 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6285 | } |
d1310b2e | 6286 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6287 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
6288 | goto insert; |
6289 | } else { | |
31b1a2bd | 6290 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
6291 | } |
6292 | not_found: | |
6293 | em->start = start; | |
70c8a91c | 6294 | em->orig_start = start; |
d1310b2e | 6295 | em->len = len; |
a52d9a80 | 6296 | not_found_em: |
5f39d397 | 6297 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6298 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6299 | insert: |
b3b4aa74 | 6300 | btrfs_release_path(path); |
d1310b2e | 6301 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6302 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6303 | em->start, em->len, start, len); |
a52d9a80 CM |
6304 | err = -EIO; |
6305 | goto out; | |
6306 | } | |
d1310b2e CM |
6307 | |
6308 | err = 0; | |
890871be | 6309 | write_lock(&em_tree->lock); |
09a2a8f9 | 6310 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6311 | /* it is possible that someone inserted the extent into the tree |
6312 | * while we had the lock dropped. It is also possible that | |
6313 | * an overlapping map exists in the tree | |
6314 | */ | |
a52d9a80 | 6315 | if (ret == -EEXIST) { |
3b951516 | 6316 | struct extent_map *existing; |
e6dcd2dc CM |
6317 | |
6318 | ret = 0; | |
6319 | ||
3b951516 | 6320 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
6321 | if (existing && (existing->start > start || |
6322 | existing->start + existing->len <= start)) { | |
6323 | free_extent_map(existing); | |
6324 | existing = NULL; | |
6325 | } | |
3b951516 CM |
6326 | if (!existing) { |
6327 | existing = lookup_extent_mapping(em_tree, em->start, | |
6328 | em->len); | |
6329 | if (existing) { | |
6330 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
6331 | em, start, |
6332 | root->sectorsize); | |
3b951516 CM |
6333 | free_extent_map(existing); |
6334 | if (err) { | |
6335 | free_extent_map(em); | |
6336 | em = NULL; | |
6337 | } | |
6338 | } else { | |
6339 | err = -EIO; | |
3b951516 CM |
6340 | free_extent_map(em); |
6341 | em = NULL; | |
6342 | } | |
6343 | } else { | |
6344 | free_extent_map(em); | |
6345 | em = existing; | |
e6dcd2dc | 6346 | err = 0; |
a52d9a80 | 6347 | } |
a52d9a80 | 6348 | } |
890871be | 6349 | write_unlock(&em_tree->lock); |
a52d9a80 | 6350 | out: |
1abe9b8a | 6351 | |
4cd8587c | 6352 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6353 | |
f421950f CM |
6354 | if (path) |
6355 | btrfs_free_path(path); | |
a52d9a80 CM |
6356 | if (trans) { |
6357 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6358 | if (!err) |
a52d9a80 CM |
6359 | err = ret; |
6360 | } | |
a52d9a80 CM |
6361 | if (err) { |
6362 | free_extent_map(em); | |
a52d9a80 CM |
6363 | return ERR_PTR(err); |
6364 | } | |
79787eaa | 6365 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6366 | return em; |
6367 | } | |
6368 | ||
ec29ed5b CM |
6369 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6370 | size_t pg_offset, u64 start, u64 len, | |
6371 | int create) | |
6372 | { | |
6373 | struct extent_map *em; | |
6374 | struct extent_map *hole_em = NULL; | |
6375 | u64 range_start = start; | |
6376 | u64 end; | |
6377 | u64 found; | |
6378 | u64 found_end; | |
6379 | int err = 0; | |
6380 | ||
6381 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6382 | if (IS_ERR(em)) | |
6383 | return em; | |
6384 | if (em) { | |
6385 | /* | |
f9e4fb53 LB |
6386 | * if our em maps to |
6387 | * - a hole or | |
6388 | * - a pre-alloc extent, | |
6389 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6390 | */ |
f9e4fb53 LB |
6391 | if (em->block_start != EXTENT_MAP_HOLE && |
6392 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6393 | return em; |
6394 | else | |
6395 | hole_em = em; | |
6396 | } | |
6397 | ||
6398 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6399 | end = start + len; | |
6400 | if (end < start) | |
6401 | end = (u64)-1; | |
6402 | else | |
6403 | end -= 1; | |
6404 | ||
6405 | em = NULL; | |
6406 | ||
6407 | /* ok, we didn't find anything, lets look for delalloc */ | |
6408 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6409 | end, len, EXTENT_DELALLOC, 1); | |
6410 | found_end = range_start + found; | |
6411 | if (found_end < range_start) | |
6412 | found_end = (u64)-1; | |
6413 | ||
6414 | /* | |
6415 | * we didn't find anything useful, return | |
6416 | * the original results from get_extent() | |
6417 | */ | |
6418 | if (range_start > end || found_end <= start) { | |
6419 | em = hole_em; | |
6420 | hole_em = NULL; | |
6421 | goto out; | |
6422 | } | |
6423 | ||
6424 | /* adjust the range_start to make sure it doesn't | |
6425 | * go backwards from the start they passed in | |
6426 | */ | |
67871254 | 6427 | range_start = max(start, range_start); |
ec29ed5b CM |
6428 | found = found_end - range_start; |
6429 | ||
6430 | if (found > 0) { | |
6431 | u64 hole_start = start; | |
6432 | u64 hole_len = len; | |
6433 | ||
172ddd60 | 6434 | em = alloc_extent_map(); |
ec29ed5b CM |
6435 | if (!em) { |
6436 | err = -ENOMEM; | |
6437 | goto out; | |
6438 | } | |
6439 | /* | |
6440 | * when btrfs_get_extent can't find anything it | |
6441 | * returns one huge hole | |
6442 | * | |
6443 | * make sure what it found really fits our range, and | |
6444 | * adjust to make sure it is based on the start from | |
6445 | * the caller | |
6446 | */ | |
6447 | if (hole_em) { | |
6448 | u64 calc_end = extent_map_end(hole_em); | |
6449 | ||
6450 | if (calc_end <= start || (hole_em->start > end)) { | |
6451 | free_extent_map(hole_em); | |
6452 | hole_em = NULL; | |
6453 | } else { | |
6454 | hole_start = max(hole_em->start, start); | |
6455 | hole_len = calc_end - hole_start; | |
6456 | } | |
6457 | } | |
6458 | em->bdev = NULL; | |
6459 | if (hole_em && range_start > hole_start) { | |
6460 | /* our hole starts before our delalloc, so we | |
6461 | * have to return just the parts of the hole | |
6462 | * that go until the delalloc starts | |
6463 | */ | |
6464 | em->len = min(hole_len, | |
6465 | range_start - hole_start); | |
6466 | em->start = hole_start; | |
6467 | em->orig_start = hole_start; | |
6468 | /* | |
6469 | * don't adjust block start at all, | |
6470 | * it is fixed at EXTENT_MAP_HOLE | |
6471 | */ | |
6472 | em->block_start = hole_em->block_start; | |
6473 | em->block_len = hole_len; | |
f9e4fb53 LB |
6474 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6475 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6476 | } else { |
6477 | em->start = range_start; | |
6478 | em->len = found; | |
6479 | em->orig_start = range_start; | |
6480 | em->block_start = EXTENT_MAP_DELALLOC; | |
6481 | em->block_len = found; | |
6482 | } | |
6483 | } else if (hole_em) { | |
6484 | return hole_em; | |
6485 | } | |
6486 | out: | |
6487 | ||
6488 | free_extent_map(hole_em); | |
6489 | if (err) { | |
6490 | free_extent_map(em); | |
6491 | return ERR_PTR(err); | |
6492 | } | |
6493 | return em; | |
6494 | } | |
6495 | ||
4b46fce2 JB |
6496 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6497 | u64 start, u64 len) | |
6498 | { | |
6499 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 6500 | struct extent_map *em; |
4b46fce2 JB |
6501 | struct btrfs_key ins; |
6502 | u64 alloc_hint; | |
6503 | int ret; | |
4b46fce2 | 6504 | |
4b46fce2 | 6505 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 6506 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
81c9ad23 | 6507 | alloc_hint, &ins, 1); |
00361589 JB |
6508 | if (ret) |
6509 | return ERR_PTR(ret); | |
4b46fce2 | 6510 | |
70c8a91c | 6511 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6512 | ins.offset, ins.offset, ins.offset, 0); |
00361589 JB |
6513 | if (IS_ERR(em)) { |
6514 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
6515 | return em; | |
6516 | } | |
4b46fce2 JB |
6517 | |
6518 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6519 | ins.offset, ins.offset, 0); | |
6520 | if (ret) { | |
6521 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
00361589 JB |
6522 | free_extent_map(em); |
6523 | return ERR_PTR(ret); | |
4b46fce2 | 6524 | } |
00361589 | 6525 | |
4b46fce2 JB |
6526 | return em; |
6527 | } | |
6528 | ||
46bfbb5c CM |
6529 | /* |
6530 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6531 | * block must be cow'd | |
6532 | */ | |
00361589 | 6533 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6534 | u64 *orig_start, u64 *orig_block_len, |
6535 | u64 *ram_bytes) | |
46bfbb5c | 6536 | { |
00361589 | 6537 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
6538 | struct btrfs_path *path; |
6539 | int ret; | |
6540 | struct extent_buffer *leaf; | |
6541 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6542 | struct btrfs_file_extent_item *fi; | |
6543 | struct btrfs_key key; | |
6544 | u64 disk_bytenr; | |
6545 | u64 backref_offset; | |
6546 | u64 extent_end; | |
6547 | u64 num_bytes; | |
6548 | int slot; | |
6549 | int found_type; | |
7ee9e440 | 6550 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 6551 | |
46bfbb5c CM |
6552 | path = btrfs_alloc_path(); |
6553 | if (!path) | |
6554 | return -ENOMEM; | |
6555 | ||
00361589 | 6556 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6557 | offset, 0); |
6558 | if (ret < 0) | |
6559 | goto out; | |
6560 | ||
6561 | slot = path->slots[0]; | |
6562 | if (ret == 1) { | |
6563 | if (slot == 0) { | |
6564 | /* can't find the item, must cow */ | |
6565 | ret = 0; | |
6566 | goto out; | |
6567 | } | |
6568 | slot--; | |
6569 | } | |
6570 | ret = 0; | |
6571 | leaf = path->nodes[0]; | |
6572 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6573 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6574 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6575 | /* not our file or wrong item type, must cow */ | |
6576 | goto out; | |
6577 | } | |
6578 | ||
6579 | if (key.offset > offset) { | |
6580 | /* Wrong offset, must cow */ | |
6581 | goto out; | |
6582 | } | |
6583 | ||
6584 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6585 | found_type = btrfs_file_extent_type(leaf, fi); | |
6586 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6587 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6588 | /* not a regular extent, must cow */ | |
6589 | goto out; | |
6590 | } | |
7ee9e440 JB |
6591 | |
6592 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6593 | goto out; | |
6594 | ||
e77751aa MX |
6595 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
6596 | if (extent_end <= offset) | |
6597 | goto out; | |
6598 | ||
46bfbb5c | 6599 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6600 | if (disk_bytenr == 0) |
6601 | goto out; | |
6602 | ||
6603 | if (btrfs_file_extent_compression(leaf, fi) || | |
6604 | btrfs_file_extent_encryption(leaf, fi) || | |
6605 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6606 | goto out; | |
6607 | ||
46bfbb5c CM |
6608 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6609 | ||
7ee9e440 JB |
6610 | if (orig_start) { |
6611 | *orig_start = key.offset - backref_offset; | |
6612 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6613 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6614 | } | |
eb384b55 | 6615 | |
46bfbb5c CM |
6616 | if (btrfs_extent_readonly(root, disk_bytenr)) |
6617 | goto out; | |
1bda19eb | 6618 | btrfs_release_path(path); |
46bfbb5c CM |
6619 | |
6620 | /* | |
6621 | * look for other files referencing this extent, if we | |
6622 | * find any we must cow | |
6623 | */ | |
00361589 JB |
6624 | trans = btrfs_join_transaction(root); |
6625 | if (IS_ERR(trans)) { | |
6626 | ret = 0; | |
46bfbb5c | 6627 | goto out; |
00361589 JB |
6628 | } |
6629 | ||
6630 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
6631 | key.offset - backref_offset, disk_bytenr); | |
6632 | btrfs_end_transaction(trans, root); | |
6633 | if (ret) { | |
6634 | ret = 0; | |
6635 | goto out; | |
6636 | } | |
46bfbb5c CM |
6637 | |
6638 | /* | |
6639 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6640 | * in this extent we are about to write. If there | |
6641 | * are any csums in that range we have to cow in order | |
6642 | * to keep the csums correct | |
6643 | */ | |
6644 | disk_bytenr += backref_offset; | |
6645 | disk_bytenr += offset - key.offset; | |
eb384b55 | 6646 | num_bytes = min(offset + *len, extent_end) - offset; |
46bfbb5c CM |
6647 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6648 | goto out; | |
6649 | /* | |
6650 | * all of the above have passed, it is safe to overwrite this extent | |
6651 | * without cow | |
6652 | */ | |
eb384b55 | 6653 | *len = num_bytes; |
46bfbb5c CM |
6654 | ret = 1; |
6655 | out: | |
6656 | btrfs_free_path(path); | |
6657 | return ret; | |
6658 | } | |
6659 | ||
eb838e73 JB |
6660 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6661 | struct extent_state **cached_state, int writing) | |
6662 | { | |
6663 | struct btrfs_ordered_extent *ordered; | |
6664 | int ret = 0; | |
6665 | ||
6666 | while (1) { | |
6667 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6668 | 0, cached_state); | |
6669 | /* | |
6670 | * We're concerned with the entire range that we're going to be | |
6671 | * doing DIO to, so we need to make sure theres no ordered | |
6672 | * extents in this range. | |
6673 | */ | |
6674 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6675 | lockend - lockstart + 1); | |
6676 | ||
6677 | /* | |
6678 | * We need to make sure there are no buffered pages in this | |
6679 | * range either, we could have raced between the invalidate in | |
6680 | * generic_file_direct_write and locking the extent. The | |
6681 | * invalidate needs to happen so that reads after a write do not | |
6682 | * get stale data. | |
6683 | */ | |
6684 | if (!ordered && (!writing || | |
6685 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
6686 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
6687 | *cached_state))) | |
6688 | break; | |
6689 | ||
6690 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6691 | cached_state, GFP_NOFS); | |
6692 | ||
6693 | if (ordered) { | |
6694 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6695 | btrfs_put_ordered_extent(ordered); | |
6696 | } else { | |
6697 | /* Screw you mmap */ | |
6698 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
6699 | lockstart, | |
6700 | lockend); | |
6701 | if (ret) | |
6702 | break; | |
6703 | ||
6704 | /* | |
6705 | * If we found a page that couldn't be invalidated just | |
6706 | * fall back to buffered. | |
6707 | */ | |
6708 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6709 | lockstart >> PAGE_CACHE_SHIFT, | |
6710 | lockend >> PAGE_CACHE_SHIFT); | |
6711 | if (ret) | |
6712 | break; | |
6713 | } | |
6714 | ||
6715 | cond_resched(); | |
6716 | } | |
6717 | ||
6718 | return ret; | |
6719 | } | |
6720 | ||
69ffb543 JB |
6721 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
6722 | u64 len, u64 orig_start, | |
6723 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
6724 | u64 orig_block_len, u64 ram_bytes, |
6725 | int type) | |
69ffb543 JB |
6726 | { |
6727 | struct extent_map_tree *em_tree; | |
6728 | struct extent_map *em; | |
6729 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6730 | int ret; | |
6731 | ||
6732 | em_tree = &BTRFS_I(inode)->extent_tree; | |
6733 | em = alloc_extent_map(); | |
6734 | if (!em) | |
6735 | return ERR_PTR(-ENOMEM); | |
6736 | ||
6737 | em->start = start; | |
6738 | em->orig_start = orig_start; | |
2ab28f32 JB |
6739 | em->mod_start = start; |
6740 | em->mod_len = len; | |
69ffb543 JB |
6741 | em->len = len; |
6742 | em->block_len = block_len; | |
6743 | em->block_start = block_start; | |
6744 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6745 | em->orig_block_len = orig_block_len; |
cc95bef6 | 6746 | em->ram_bytes = ram_bytes; |
70c8a91c | 6747 | em->generation = -1; |
69ffb543 JB |
6748 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6749 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6750 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6751 | |
6752 | do { | |
6753 | btrfs_drop_extent_cache(inode, em->start, | |
6754 | em->start + em->len - 1, 0); | |
6755 | write_lock(&em_tree->lock); | |
09a2a8f9 | 6756 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
6757 | write_unlock(&em_tree->lock); |
6758 | } while (ret == -EEXIST); | |
6759 | ||
6760 | if (ret) { | |
6761 | free_extent_map(em); | |
6762 | return ERR_PTR(ret); | |
6763 | } | |
6764 | ||
6765 | return em; | |
6766 | } | |
6767 | ||
6768 | ||
4b46fce2 JB |
6769 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6770 | struct buffer_head *bh_result, int create) | |
6771 | { | |
6772 | struct extent_map *em; | |
6773 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6774 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6775 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6776 | u64 lockstart, lockend; |
4b46fce2 | 6777 | u64 len = bh_result->b_size; |
eb838e73 | 6778 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 6779 | int ret = 0; |
eb838e73 | 6780 | |
172a5049 | 6781 | if (create) |
eb838e73 | 6782 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 6783 | else |
c329861d | 6784 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 6785 | |
c329861d JB |
6786 | lockstart = start; |
6787 | lockend = start + len - 1; | |
6788 | ||
eb838e73 JB |
6789 | /* |
6790 | * If this errors out it's because we couldn't invalidate pagecache for | |
6791 | * this range and we need to fallback to buffered. | |
6792 | */ | |
6793 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6794 | return -ENOTBLK; | |
6795 | ||
4b46fce2 | 6796 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
6797 | if (IS_ERR(em)) { |
6798 | ret = PTR_ERR(em); | |
6799 | goto unlock_err; | |
6800 | } | |
4b46fce2 JB |
6801 | |
6802 | /* | |
6803 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6804 | * io. INLINE is special, and we could probably kludge it in here, but | |
6805 | * it's still buffered so for safety lets just fall back to the generic | |
6806 | * buffered path. | |
6807 | * | |
6808 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6809 | * decompress it, so there will be buffering required no matter what we | |
6810 | * do, so go ahead and fallback to buffered. | |
6811 | * | |
6812 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6813 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6814 | * the generic code. | |
6815 | */ | |
6816 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6817 | em->block_start == EXTENT_MAP_INLINE) { | |
6818 | free_extent_map(em); | |
eb838e73 JB |
6819 | ret = -ENOTBLK; |
6820 | goto unlock_err; | |
4b46fce2 JB |
6821 | } |
6822 | ||
6823 | /* Just a good old fashioned hole, return */ | |
6824 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6825 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6826 | free_extent_map(em); | |
eb838e73 | 6827 | goto unlock_err; |
4b46fce2 JB |
6828 | } |
6829 | ||
6830 | /* | |
6831 | * We don't allocate a new extent in the following cases | |
6832 | * | |
6833 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6834 | * existing extent. | |
6835 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6836 | * just use the extent. | |
6837 | * | |
6838 | */ | |
46bfbb5c | 6839 | if (!create) { |
eb838e73 JB |
6840 | len = min(len, em->len - (start - em->start)); |
6841 | lockstart = start + len; | |
6842 | goto unlock; | |
46bfbb5c | 6843 | } |
4b46fce2 JB |
6844 | |
6845 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6846 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6847 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6848 | int type; |
6849 | int ret; | |
eb384b55 | 6850 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
6851 | |
6852 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6853 | type = BTRFS_ORDERED_PREALLOC; | |
6854 | else | |
6855 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6856 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6857 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 6858 | |
00361589 | 6859 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 6860 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
6861 | if (type == BTRFS_ORDERED_PREALLOC) { |
6862 | free_extent_map(em); | |
6863 | em = create_pinned_em(inode, start, len, | |
6864 | orig_start, | |
b4939680 | 6865 | block_start, len, |
cc95bef6 JB |
6866 | orig_block_len, |
6867 | ram_bytes, type); | |
00361589 | 6868 | if (IS_ERR(em)) |
69ffb543 | 6869 | goto unlock_err; |
69ffb543 JB |
6870 | } |
6871 | ||
46bfbb5c CM |
6872 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6873 | block_start, len, len, type); | |
46bfbb5c CM |
6874 | if (ret) { |
6875 | free_extent_map(em); | |
eb838e73 | 6876 | goto unlock_err; |
46bfbb5c CM |
6877 | } |
6878 | goto unlock; | |
4b46fce2 | 6879 | } |
4b46fce2 | 6880 | } |
00361589 | 6881 | |
46bfbb5c CM |
6882 | /* |
6883 | * this will cow the extent, reset the len in case we changed | |
6884 | * it above | |
6885 | */ | |
6886 | len = bh_result->b_size; | |
70c8a91c JB |
6887 | free_extent_map(em); |
6888 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6889 | if (IS_ERR(em)) { |
6890 | ret = PTR_ERR(em); | |
6891 | goto unlock_err; | |
6892 | } | |
46bfbb5c CM |
6893 | len = min(len, em->len - (start - em->start)); |
6894 | unlock: | |
4b46fce2 JB |
6895 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6896 | inode->i_blkbits; | |
46bfbb5c | 6897 | bh_result->b_size = len; |
4b46fce2 JB |
6898 | bh_result->b_bdev = em->bdev; |
6899 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6900 | if (create) { |
6901 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6902 | set_buffer_new(bh_result); | |
6903 | ||
6904 | /* | |
6905 | * Need to update the i_size under the extent lock so buffered | |
6906 | * readers will get the updated i_size when we unlock. | |
6907 | */ | |
6908 | if (start + len > i_size_read(inode)) | |
6909 | i_size_write(inode, start + len); | |
0934856d | 6910 | |
172a5049 MX |
6911 | spin_lock(&BTRFS_I(inode)->lock); |
6912 | BTRFS_I(inode)->outstanding_extents++; | |
6913 | spin_unlock(&BTRFS_I(inode)->lock); | |
6914 | ||
0934856d MX |
6915 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6916 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
6917 | &cached_state, GFP_NOFS); | |
6918 | BUG_ON(ret); | |
c3473e83 | 6919 | } |
4b46fce2 | 6920 | |
eb838e73 JB |
6921 | /* |
6922 | * In the case of write we need to clear and unlock the entire range, | |
6923 | * in the case of read we need to unlock only the end area that we | |
6924 | * aren't using if there is any left over space. | |
6925 | */ | |
24c03fa5 | 6926 | if (lockstart < lockend) { |
0934856d MX |
6927 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6928 | lockend, unlock_bits, 1, 0, | |
6929 | &cached_state, GFP_NOFS); | |
24c03fa5 | 6930 | } else { |
eb838e73 | 6931 | free_extent_state(cached_state); |
24c03fa5 | 6932 | } |
eb838e73 | 6933 | |
4b46fce2 JB |
6934 | free_extent_map(em); |
6935 | ||
6936 | return 0; | |
eb838e73 JB |
6937 | |
6938 | unlock_err: | |
eb838e73 JB |
6939 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6940 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6941 | return ret; | |
4b46fce2 JB |
6942 | } |
6943 | ||
4b46fce2 JB |
6944 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
6945 | { | |
e65e1535 | 6946 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6947 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6948 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6949 | struct inode *inode = dip->inode; |
6950 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9be3395b | 6951 | struct bio *dio_bio; |
facc8a22 MX |
6952 | u32 *csums = (u32 *)dip->csum; |
6953 | int index = 0; | |
4b46fce2 | 6954 | u64 start; |
4b46fce2 JB |
6955 | |
6956 | start = dip->logical_offset; | |
6957 | do { | |
6958 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6959 | struct page *page = bvec->bv_page; | |
6960 | char *kaddr; | |
6961 | u32 csum = ~(u32)0; | |
6962 | unsigned long flags; | |
6963 | ||
6964 | local_irq_save(flags); | |
7ac687d9 | 6965 | kaddr = kmap_atomic(page); |
b0496686 | 6966 | csum = btrfs_csum_data(kaddr + bvec->bv_offset, |
4b46fce2 JB |
6967 | csum, bvec->bv_len); |
6968 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6969 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6970 | local_irq_restore(flags); |
6971 | ||
6972 | flush_dcache_page(bvec->bv_page); | |
facc8a22 MX |
6973 | if (csum != csums[index]) { |
6974 | btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", | |
c1c9ff7c GU |
6975 | btrfs_ino(inode), start, csum, |
6976 | csums[index]); | |
4b46fce2 JB |
6977 | err = -EIO; |
6978 | } | |
6979 | } | |
6980 | ||
6981 | start += bvec->bv_len; | |
4b46fce2 | 6982 | bvec++; |
facc8a22 | 6983 | index++; |
4b46fce2 JB |
6984 | } while (bvec <= bvec_end); |
6985 | ||
6986 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6987 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 6988 | dio_bio = dip->dio_bio; |
4b46fce2 | 6989 | |
4b46fce2 | 6990 | kfree(dip); |
c0da7aa1 JB |
6991 | |
6992 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6993 | if (err) | |
9be3395b CM |
6994 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6995 | dio_end_io(dio_bio, err); | |
6996 | bio_put(bio); | |
4b46fce2 JB |
6997 | } |
6998 | ||
6999 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
7000 | { | |
7001 | struct btrfs_dio_private *dip = bio->bi_private; | |
7002 | struct inode *inode = dip->inode; | |
7003 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 7004 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
7005 | u64 ordered_offset = dip->logical_offset; |
7006 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 7007 | struct bio *dio_bio; |
4b46fce2 JB |
7008 | int ret; |
7009 | ||
7010 | if (err) | |
7011 | goto out_done; | |
163cf09c CM |
7012 | again: |
7013 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
7014 | &ordered_offset, | |
5fd02043 | 7015 | ordered_bytes, !err); |
4b46fce2 | 7016 | if (!ret) |
163cf09c | 7017 | goto out_test; |
4b46fce2 | 7018 | |
5fd02043 JB |
7019 | ordered->work.func = finish_ordered_fn; |
7020 | ordered->work.flags = 0; | |
7021 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
7022 | &ordered->work); | |
163cf09c CM |
7023 | out_test: |
7024 | /* | |
7025 | * our bio might span multiple ordered extents. If we haven't | |
7026 | * completed the accounting for the whole dio, go back and try again | |
7027 | */ | |
7028 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
7029 | ordered_bytes = dip->logical_offset + dip->bytes - | |
7030 | ordered_offset; | |
5fd02043 | 7031 | ordered = NULL; |
163cf09c CM |
7032 | goto again; |
7033 | } | |
4b46fce2 | 7034 | out_done: |
9be3395b | 7035 | dio_bio = dip->dio_bio; |
4b46fce2 | 7036 | |
4b46fce2 | 7037 | kfree(dip); |
c0da7aa1 JB |
7038 | |
7039 | /* If we had an error make sure to clear the uptodate flag */ | |
7040 | if (err) | |
9be3395b CM |
7041 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7042 | dio_end_io(dio_bio, err); | |
7043 | bio_put(bio); | |
4b46fce2 JB |
7044 | } |
7045 | ||
eaf25d93 CM |
7046 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
7047 | struct bio *bio, int mirror_num, | |
7048 | unsigned long bio_flags, u64 offset) | |
7049 | { | |
7050 | int ret; | |
7051 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7052 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 7053 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
7054 | return 0; |
7055 | } | |
7056 | ||
e65e1535 MX |
7057 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
7058 | { | |
7059 | struct btrfs_dio_private *dip = bio->bi_private; | |
7060 | ||
7061 | if (err) { | |
efe120a0 FH |
7062 | btrfs_err(BTRFS_I(dip->inode)->root->fs_info, |
7063 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
c1c9ff7c | 7064 | btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 7065 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
7066 | dip->errors = 1; |
7067 | ||
7068 | /* | |
7069 | * before atomic variable goto zero, we must make sure | |
7070 | * dip->errors is perceived to be set. | |
7071 | */ | |
7072 | smp_mb__before_atomic_dec(); | |
7073 | } | |
7074 | ||
7075 | /* if there are more bios still pending for this dio, just exit */ | |
7076 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
7077 | goto out; | |
7078 | ||
9be3395b | 7079 | if (dip->errors) { |
e65e1535 | 7080 | bio_io_error(dip->orig_bio); |
9be3395b CM |
7081 | } else { |
7082 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
7083 | bio_endio(dip->orig_bio, 0); |
7084 | } | |
7085 | out: | |
7086 | bio_put(bio); | |
7087 | } | |
7088 | ||
7089 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
7090 | u64 first_sector, gfp_t gfp_flags) | |
7091 | { | |
7092 | int nr_vecs = bio_get_nr_vecs(bdev); | |
7093 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
7094 | } | |
7095 | ||
7096 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
7097 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 7098 | int async_submit) |
e65e1535 | 7099 | { |
facc8a22 | 7100 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
7101 | int write = rw & REQ_WRITE; |
7102 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7103 | int ret; | |
7104 | ||
b812ce28 JB |
7105 | if (async_submit) |
7106 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7107 | ||
e65e1535 | 7108 | bio_get(bio); |
5fd02043 JB |
7109 | |
7110 | if (!write) { | |
7111 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
7112 | if (ret) | |
7113 | goto err; | |
7114 | } | |
e65e1535 | 7115 | |
1ae39938 JB |
7116 | if (skip_sum) |
7117 | goto map; | |
7118 | ||
7119 | if (write && async_submit) { | |
e65e1535 MX |
7120 | ret = btrfs_wq_submit_bio(root->fs_info, |
7121 | inode, rw, bio, 0, 0, | |
7122 | file_offset, | |
7123 | __btrfs_submit_bio_start_direct_io, | |
7124 | __btrfs_submit_bio_done); | |
7125 | goto err; | |
1ae39938 JB |
7126 | } else if (write) { |
7127 | /* | |
7128 | * If we aren't doing async submit, calculate the csum of the | |
7129 | * bio now. | |
7130 | */ | |
7131 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
7132 | if (ret) | |
7133 | goto err; | |
c2db1073 | 7134 | } else if (!skip_sum) { |
facc8a22 MX |
7135 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio, |
7136 | file_offset); | |
c2db1073 TI |
7137 | if (ret) |
7138 | goto err; | |
7139 | } | |
e65e1535 | 7140 | |
1ae39938 JB |
7141 | map: |
7142 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
7143 | err: |
7144 | bio_put(bio); | |
7145 | return ret; | |
7146 | } | |
7147 | ||
7148 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
7149 | int skip_sum) | |
7150 | { | |
7151 | struct inode *inode = dip->inode; | |
7152 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7153 | struct bio *bio; |
7154 | struct bio *orig_bio = dip->orig_bio; | |
7155 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
7156 | u64 start_sector = orig_bio->bi_sector; | |
7157 | u64 file_offset = dip->logical_offset; | |
7158 | u64 submit_len = 0; | |
7159 | u64 map_length; | |
7160 | int nr_pages = 0; | |
e65e1535 | 7161 | int ret = 0; |
1ae39938 | 7162 | int async_submit = 0; |
e65e1535 | 7163 | |
e65e1535 | 7164 | map_length = orig_bio->bi_size; |
53b381b3 | 7165 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 MX |
7166 | &map_length, NULL, 0); |
7167 | if (ret) { | |
64728bbb | 7168 | bio_put(orig_bio); |
e65e1535 MX |
7169 | return -EIO; |
7170 | } | |
facc8a22 | 7171 | |
02f57c7a JB |
7172 | if (map_length >= orig_bio->bi_size) { |
7173 | bio = orig_bio; | |
7174 | goto submit; | |
7175 | } | |
7176 | ||
53b381b3 DW |
7177 | /* async crcs make it difficult to collect full stripe writes. */ |
7178 | if (btrfs_get_alloc_profile(root, 1) & | |
7179 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
7180 | async_submit = 0; | |
7181 | else | |
7182 | async_submit = 1; | |
7183 | ||
02f57c7a JB |
7184 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7185 | if (!bio) | |
7186 | return -ENOMEM; | |
7187 | bio->bi_private = dip; | |
7188 | bio->bi_end_io = btrfs_end_dio_bio; | |
7189 | atomic_inc(&dip->pending_bios); | |
7190 | ||
e65e1535 MX |
7191 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
7192 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
7193 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
7194 | bvec->bv_offset) < bvec->bv_len)) { | |
7195 | /* | |
7196 | * inc the count before we submit the bio so | |
7197 | * we know the end IO handler won't happen before | |
7198 | * we inc the count. Otherwise, the dip might get freed | |
7199 | * before we're done setting it up | |
7200 | */ | |
7201 | atomic_inc(&dip->pending_bios); | |
7202 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7203 | file_offset, skip_sum, | |
c329861d | 7204 | async_submit); |
e65e1535 MX |
7205 | if (ret) { |
7206 | bio_put(bio); | |
7207 | atomic_dec(&dip->pending_bios); | |
7208 | goto out_err; | |
7209 | } | |
7210 | ||
e65e1535 MX |
7211 | start_sector += submit_len >> 9; |
7212 | file_offset += submit_len; | |
7213 | ||
7214 | submit_len = 0; | |
7215 | nr_pages = 0; | |
7216 | ||
7217 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7218 | start_sector, GFP_NOFS); | |
7219 | if (!bio) | |
7220 | goto out_err; | |
7221 | bio->bi_private = dip; | |
7222 | bio->bi_end_io = btrfs_end_dio_bio; | |
7223 | ||
7224 | map_length = orig_bio->bi_size; | |
53b381b3 | 7225 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7226 | start_sector << 9, |
e65e1535 MX |
7227 | &map_length, NULL, 0); |
7228 | if (ret) { | |
7229 | bio_put(bio); | |
7230 | goto out_err; | |
7231 | } | |
7232 | } else { | |
7233 | submit_len += bvec->bv_len; | |
67871254 | 7234 | nr_pages++; |
e65e1535 MX |
7235 | bvec++; |
7236 | } | |
7237 | } | |
7238 | ||
02f57c7a | 7239 | submit: |
e65e1535 | 7240 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7241 | async_submit); |
e65e1535 MX |
7242 | if (!ret) |
7243 | return 0; | |
7244 | ||
7245 | bio_put(bio); | |
7246 | out_err: | |
7247 | dip->errors = 1; | |
7248 | /* | |
7249 | * before atomic variable goto zero, we must | |
7250 | * make sure dip->errors is perceived to be set. | |
7251 | */ | |
7252 | smp_mb__before_atomic_dec(); | |
7253 | if (atomic_dec_and_test(&dip->pending_bios)) | |
7254 | bio_io_error(dip->orig_bio); | |
7255 | ||
7256 | /* bio_end_io() will handle error, so we needn't return it */ | |
7257 | return 0; | |
7258 | } | |
7259 | ||
9be3395b CM |
7260 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7261 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7262 | { |
7263 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7264 | struct btrfs_dio_private *dip; | |
9be3395b | 7265 | struct bio *io_bio; |
4b46fce2 | 7266 | int skip_sum; |
facc8a22 | 7267 | int sum_len; |
7b6d91da | 7268 | int write = rw & REQ_WRITE; |
4b46fce2 | 7269 | int ret = 0; |
facc8a22 | 7270 | u16 csum_size; |
4b46fce2 JB |
7271 | |
7272 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7273 | ||
9be3395b | 7274 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7275 | if (!io_bio) { |
7276 | ret = -ENOMEM; | |
7277 | goto free_ordered; | |
7278 | } | |
7279 | ||
facc8a22 MX |
7280 | if (!skip_sum && !write) { |
7281 | csum_size = btrfs_super_csum_size(root->fs_info->super_copy); | |
7282 | sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits; | |
7283 | sum_len *= csum_size; | |
7284 | } else { | |
7285 | sum_len = 0; | |
7286 | } | |
7287 | ||
7288 | dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS); | |
4b46fce2 JB |
7289 | if (!dip) { |
7290 | ret = -ENOMEM; | |
9be3395b | 7291 | goto free_io_bio; |
4b46fce2 | 7292 | } |
4b46fce2 | 7293 | |
9be3395b | 7294 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
7295 | dip->inode = inode; |
7296 | dip->logical_offset = file_offset; | |
e6da5d2e | 7297 | dip->bytes = dio_bio->bi_size; |
9be3395b CM |
7298 | dip->disk_bytenr = (u64)dio_bio->bi_sector << 9; |
7299 | io_bio->bi_private = dip; | |
e65e1535 | 7300 | dip->errors = 0; |
9be3395b CM |
7301 | dip->orig_bio = io_bio; |
7302 | dip->dio_bio = dio_bio; | |
e65e1535 | 7303 | atomic_set(&dip->pending_bios, 0); |
4b46fce2 JB |
7304 | |
7305 | if (write) | |
9be3395b | 7306 | io_bio->bi_end_io = btrfs_endio_direct_write; |
4b46fce2 | 7307 | else |
9be3395b | 7308 | io_bio->bi_end_io = btrfs_endio_direct_read; |
4b46fce2 | 7309 | |
e65e1535 MX |
7310 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7311 | if (!ret) | |
eaf25d93 | 7312 | return; |
9be3395b CM |
7313 | |
7314 | free_io_bio: | |
7315 | bio_put(io_bio); | |
7316 | ||
4b46fce2 JB |
7317 | free_ordered: |
7318 | /* | |
7319 | * If this is a write, we need to clean up the reserved space and kill | |
7320 | * the ordered extent. | |
7321 | */ | |
7322 | if (write) { | |
7323 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7324 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7325 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7326 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7327 | btrfs_free_reserved_extent(root, ordered->start, | |
7328 | ordered->disk_len); | |
7329 | btrfs_put_ordered_extent(ordered); | |
7330 | btrfs_put_ordered_extent(ordered); | |
7331 | } | |
9be3395b | 7332 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7333 | } |
7334 | ||
5a5f79b5 CM |
7335 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
7336 | const struct iovec *iov, loff_t offset, | |
7337 | unsigned long nr_segs) | |
7338 | { | |
7339 | int seg; | |
a1b75f7d | 7340 | int i; |
5a5f79b5 CM |
7341 | size_t size; |
7342 | unsigned long addr; | |
7343 | unsigned blocksize_mask = root->sectorsize - 1; | |
7344 | ssize_t retval = -EINVAL; | |
7345 | loff_t end = offset; | |
7346 | ||
7347 | if (offset & blocksize_mask) | |
7348 | goto out; | |
7349 | ||
7350 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
7351 | for (seg = 0; seg < nr_segs; seg++) { | |
7352 | addr = (unsigned long)iov[seg].iov_base; | |
7353 | size = iov[seg].iov_len; | |
7354 | end += size; | |
a1b75f7d | 7355 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 7356 | goto out; |
a1b75f7d JB |
7357 | |
7358 | /* If this is a write we don't need to check anymore */ | |
7359 | if (rw & WRITE) | |
7360 | continue; | |
7361 | ||
7362 | /* | |
7363 | * Check to make sure we don't have duplicate iov_base's in this | |
7364 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7365 | * when reading back. | |
7366 | */ | |
7367 | for (i = seg + 1; i < nr_segs; i++) { | |
7368 | if (iov[seg].iov_base == iov[i].iov_base) | |
7369 | goto out; | |
7370 | } | |
5a5f79b5 CM |
7371 | } |
7372 | retval = 0; | |
7373 | out: | |
7374 | return retval; | |
7375 | } | |
eb838e73 | 7376 | |
16432985 CM |
7377 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
7378 | const struct iovec *iov, loff_t offset, | |
7379 | unsigned long nr_segs) | |
7380 | { | |
4b46fce2 JB |
7381 | struct file *file = iocb->ki_filp; |
7382 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7383 | size_t count = 0; |
2e60a51e | 7384 | int flags = 0; |
38851cc1 MX |
7385 | bool wakeup = true; |
7386 | bool relock = false; | |
0934856d | 7387 | ssize_t ret; |
4b46fce2 | 7388 | |
5a5f79b5 | 7389 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 7390 | offset, nr_segs)) |
5a5f79b5 | 7391 | return 0; |
3f7c579c | 7392 | |
38851cc1 MX |
7393 | atomic_inc(&inode->i_dio_count); |
7394 | smp_mb__after_atomic_inc(); | |
7395 | ||
0e267c44 JB |
7396 | /* |
7397 | * The generic stuff only does filemap_write_and_wait_range, which isn't | |
7398 | * enough if we've written compressed pages to this area, so we need to | |
7399 | * call btrfs_wait_ordered_range to make absolutely sure that any | |
7400 | * outstanding dirty pages are on disk. | |
7401 | */ | |
7402 | count = iov_length(iov, nr_segs); | |
0ef8b726 JB |
7403 | ret = btrfs_wait_ordered_range(inode, offset, count); |
7404 | if (ret) | |
7405 | return ret; | |
0e267c44 | 7406 | |
0934856d | 7407 | if (rw & WRITE) { |
38851cc1 MX |
7408 | /* |
7409 | * If the write DIO is beyond the EOF, we need update | |
7410 | * the isize, but it is protected by i_mutex. So we can | |
7411 | * not unlock the i_mutex at this case. | |
7412 | */ | |
7413 | if (offset + count <= inode->i_size) { | |
7414 | mutex_unlock(&inode->i_mutex); | |
7415 | relock = true; | |
7416 | } | |
0934856d MX |
7417 | ret = btrfs_delalloc_reserve_space(inode, count); |
7418 | if (ret) | |
38851cc1 MX |
7419 | goto out; |
7420 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
7421 | &BTRFS_I(inode)->runtime_flags))) { | |
7422 | inode_dio_done(inode); | |
7423 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7424 | wakeup = false; | |
0934856d MX |
7425 | } |
7426 | ||
7427 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
7428 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
7429 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
2e60a51e | 7430 | btrfs_submit_direct, flags); |
0934856d MX |
7431 | if (rw & WRITE) { |
7432 | if (ret < 0 && ret != -EIOCBQUEUED) | |
7433 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 7434 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
7435 | btrfs_delalloc_release_space(inode, |
7436 | count - (size_t)ret); | |
172a5049 MX |
7437 | else |
7438 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 7439 | } |
38851cc1 | 7440 | out: |
2e60a51e MX |
7441 | if (wakeup) |
7442 | inode_dio_done(inode); | |
38851cc1 MX |
7443 | if (relock) |
7444 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
7445 | |
7446 | return ret; | |
16432985 CM |
7447 | } |
7448 | ||
05dadc09 TI |
7449 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
7450 | ||
1506fcc8 YS |
7451 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
7452 | __u64 start, __u64 len) | |
7453 | { | |
05dadc09 TI |
7454 | int ret; |
7455 | ||
7456 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
7457 | if (ret) | |
7458 | return ret; | |
7459 | ||
ec29ed5b | 7460 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
7461 | } |
7462 | ||
a52d9a80 | 7463 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7464 | { |
d1310b2e CM |
7465 | struct extent_io_tree *tree; |
7466 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 7467 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 7468 | } |
1832a6d5 | 7469 | |
a52d9a80 | 7470 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7471 | { |
d1310b2e | 7472 | struct extent_io_tree *tree; |
b888db2b CM |
7473 | |
7474 | ||
7475 | if (current->flags & PF_MEMALLOC) { | |
7476 | redirty_page_for_writepage(wbc, page); | |
7477 | unlock_page(page); | |
7478 | return 0; | |
7479 | } | |
d1310b2e | 7480 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 7481 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
7482 | } |
7483 | ||
48a3b636 ES |
7484 | static int btrfs_writepages(struct address_space *mapping, |
7485 | struct writeback_control *wbc) | |
b293f02e | 7486 | { |
d1310b2e | 7487 | struct extent_io_tree *tree; |
771ed689 | 7488 | |
d1310b2e | 7489 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
7490 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
7491 | } | |
7492 | ||
3ab2fb5a CM |
7493 | static int |
7494 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
7495 | struct list_head *pages, unsigned nr_pages) | |
7496 | { | |
d1310b2e CM |
7497 | struct extent_io_tree *tree; |
7498 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
7499 | return extent_readpages(tree, mapping, pages, nr_pages, |
7500 | btrfs_get_extent); | |
7501 | } | |
e6dcd2dc | 7502 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 7503 | { |
d1310b2e CM |
7504 | struct extent_io_tree *tree; |
7505 | struct extent_map_tree *map; | |
a52d9a80 | 7506 | int ret; |
8c2383c3 | 7507 | |
d1310b2e CM |
7508 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
7509 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 7510 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
7511 | if (ret == 1) { |
7512 | ClearPagePrivate(page); | |
7513 | set_page_private(page, 0); | |
7514 | page_cache_release(page); | |
39279cc3 | 7515 | } |
a52d9a80 | 7516 | return ret; |
39279cc3 CM |
7517 | } |
7518 | ||
e6dcd2dc CM |
7519 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
7520 | { | |
98509cfc CM |
7521 | if (PageWriteback(page) || PageDirty(page)) |
7522 | return 0; | |
b335b003 | 7523 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
7524 | } |
7525 | ||
d47992f8 LC |
7526 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
7527 | unsigned int length) | |
39279cc3 | 7528 | { |
5fd02043 | 7529 | struct inode *inode = page->mapping->host; |
d1310b2e | 7530 | struct extent_io_tree *tree; |
e6dcd2dc | 7531 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 7532 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7533 | u64 page_start = page_offset(page); |
7534 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
131e404a | 7535 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 7536 | |
8b62b72b CM |
7537 | /* |
7538 | * we have the page locked, so new writeback can't start, | |
7539 | * and the dirty bit won't be cleared while we are here. | |
7540 | * | |
7541 | * Wait for IO on this page so that we can safely clear | |
7542 | * the PagePrivate2 bit and do ordered accounting | |
7543 | */ | |
e6dcd2dc | 7544 | wait_on_page_writeback(page); |
8b62b72b | 7545 | |
5fd02043 | 7546 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
7547 | if (offset) { |
7548 | btrfs_releasepage(page, GFP_NOFS); | |
7549 | return; | |
7550 | } | |
131e404a FDBM |
7551 | |
7552 | if (!inode_evicting) | |
7553 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); | |
7554 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
e6dcd2dc | 7555 | if (ordered) { |
eb84ae03 CM |
7556 | /* |
7557 | * IO on this page will never be started, so we need | |
7558 | * to account for any ordered extents now | |
7559 | */ | |
131e404a FDBM |
7560 | if (!inode_evicting) |
7561 | clear_extent_bit(tree, page_start, page_end, | |
7562 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
7563 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
7564 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
7565 | GFP_NOFS); | |
8b62b72b CM |
7566 | /* |
7567 | * whoever cleared the private bit is responsible | |
7568 | * for the finish_ordered_io | |
7569 | */ | |
77cef2ec JB |
7570 | if (TestClearPagePrivate2(page)) { |
7571 | struct btrfs_ordered_inode_tree *tree; | |
7572 | u64 new_len; | |
7573 | ||
7574 | tree = &BTRFS_I(inode)->ordered_tree; | |
7575 | ||
7576 | spin_lock_irq(&tree->lock); | |
7577 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
7578 | new_len = page_start - ordered->file_offset; | |
7579 | if (new_len < ordered->truncated_len) | |
7580 | ordered->truncated_len = new_len; | |
7581 | spin_unlock_irq(&tree->lock); | |
7582 | ||
7583 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
7584 | page_start, | |
7585 | PAGE_CACHE_SIZE, 1)) | |
7586 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 7587 | } |
e6dcd2dc | 7588 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
7589 | if (!inode_evicting) { |
7590 | cached_state = NULL; | |
7591 | lock_extent_bits(tree, page_start, page_end, 0, | |
7592 | &cached_state); | |
7593 | } | |
7594 | } | |
7595 | ||
7596 | if (!inode_evicting) { | |
7597 | clear_extent_bit(tree, page_start, page_end, | |
7598 | EXTENT_LOCKED | EXTENT_DIRTY | | |
7599 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
7600 | EXTENT_DEFRAG, 1, 1, | |
7601 | &cached_state, GFP_NOFS); | |
7602 | ||
7603 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 7604 | } |
e6dcd2dc | 7605 | |
4a096752 | 7606 | ClearPageChecked(page); |
9ad6b7bc | 7607 | if (PagePrivate(page)) { |
9ad6b7bc CM |
7608 | ClearPagePrivate(page); |
7609 | set_page_private(page, 0); | |
7610 | page_cache_release(page); | |
7611 | } | |
39279cc3 CM |
7612 | } |
7613 | ||
9ebefb18 CM |
7614 | /* |
7615 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
7616 | * called from a page fault handler when a page is first dirtied. Hence we must | |
7617 | * be careful to check for EOF conditions here. We set the page up correctly | |
7618 | * for a written page which means we get ENOSPC checking when writing into | |
7619 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
7620 | * support these features. | |
7621 | * | |
7622 | * We are not allowed to take the i_mutex here so we have to play games to | |
7623 | * protect against truncate races as the page could now be beyond EOF. Because | |
7624 | * vmtruncate() writes the inode size before removing pages, once we have the | |
7625 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
7626 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
7627 | * unlock the page. | |
7628 | */ | |
c2ec175c | 7629 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 7630 | { |
c2ec175c | 7631 | struct page *page = vmf->page; |
496ad9aa | 7632 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 7633 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
7634 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7635 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 7636 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7637 | char *kaddr; |
7638 | unsigned long zero_start; | |
9ebefb18 | 7639 | loff_t size; |
1832a6d5 | 7640 | int ret; |
9998eb70 | 7641 | int reserved = 0; |
a52d9a80 | 7642 | u64 page_start; |
e6dcd2dc | 7643 | u64 page_end; |
9ebefb18 | 7644 | |
b2b5ef5c | 7645 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 7646 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7647 | if (!ret) { |
e41f941a | 7648 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
7649 | reserved = 1; |
7650 | } | |
56a76f82 NP |
7651 | if (ret) { |
7652 | if (ret == -ENOMEM) | |
7653 | ret = VM_FAULT_OOM; | |
7654 | else /* -ENOSPC, -EIO, etc */ | |
7655 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
7656 | if (reserved) |
7657 | goto out; | |
7658 | goto out_noreserve; | |
56a76f82 | 7659 | } |
1832a6d5 | 7660 | |
56a76f82 | 7661 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 7662 | again: |
9ebefb18 | 7663 | lock_page(page); |
9ebefb18 | 7664 | size = i_size_read(inode); |
e6dcd2dc CM |
7665 | page_start = page_offset(page); |
7666 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 7667 | |
9ebefb18 | 7668 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 7669 | (page_start >= size)) { |
9ebefb18 CM |
7670 | /* page got truncated out from underneath us */ |
7671 | goto out_unlock; | |
7672 | } | |
e6dcd2dc CM |
7673 | wait_on_page_writeback(page); |
7674 | ||
d0082371 | 7675 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7676 | set_page_extent_mapped(page); |
7677 | ||
eb84ae03 CM |
7678 | /* |
7679 | * we can't set the delalloc bits if there are pending ordered | |
7680 | * extents. Drop our locks and wait for them to finish | |
7681 | */ | |
e6dcd2dc CM |
7682 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
7683 | if (ordered) { | |
2ac55d41 JB |
7684 | unlock_extent_cached(io_tree, page_start, page_end, |
7685 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 7686 | unlock_page(page); |
eb84ae03 | 7687 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
7688 | btrfs_put_ordered_extent(ordered); |
7689 | goto again; | |
7690 | } | |
7691 | ||
fbf19087 JB |
7692 | /* |
7693 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
7694 | * if it was already dirty, so for space accounting reasons we need to | |
7695 | * clear any delalloc bits for the range we are fixing to save. There | |
7696 | * is probably a better way to do this, but for now keep consistent with | |
7697 | * prepare_pages in the normal write path. | |
7698 | */ | |
2ac55d41 | 7699 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
7700 | EXTENT_DIRTY | EXTENT_DELALLOC | |
7701 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 7702 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 7703 | |
2ac55d41 JB |
7704 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
7705 | &cached_state); | |
9ed74f2d | 7706 | if (ret) { |
2ac55d41 JB |
7707 | unlock_extent_cached(io_tree, page_start, page_end, |
7708 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
7709 | ret = VM_FAULT_SIGBUS; |
7710 | goto out_unlock; | |
7711 | } | |
e6dcd2dc | 7712 | ret = 0; |
9ebefb18 CM |
7713 | |
7714 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 7715 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 7716 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 7717 | else |
e6dcd2dc | 7718 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 7719 | |
e6dcd2dc CM |
7720 | if (zero_start != PAGE_CACHE_SIZE) { |
7721 | kaddr = kmap(page); | |
7722 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
7723 | flush_dcache_page(page); | |
7724 | kunmap(page); | |
7725 | } | |
247e743c | 7726 | ClearPageChecked(page); |
e6dcd2dc | 7727 | set_page_dirty(page); |
50a9b214 | 7728 | SetPageUptodate(page); |
5a3f23d5 | 7729 | |
257c62e1 CM |
7730 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
7731 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 7732 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 7733 | |
2ac55d41 | 7734 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
7735 | |
7736 | out_unlock: | |
b2b5ef5c JK |
7737 | if (!ret) { |
7738 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 7739 | return VM_FAULT_LOCKED; |
b2b5ef5c | 7740 | } |
9ebefb18 | 7741 | unlock_page(page); |
1832a6d5 | 7742 | out: |
ec39e180 | 7743 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7744 | out_noreserve: |
b2b5ef5c | 7745 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
7746 | return ret; |
7747 | } | |
7748 | ||
a41ad394 | 7749 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
7750 | { |
7751 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 7752 | struct btrfs_block_rsv *rsv; |
a71754fc | 7753 | int ret = 0; |
3893e33b | 7754 | int err = 0; |
39279cc3 | 7755 | struct btrfs_trans_handle *trans; |
dbe674a9 | 7756 | u64 mask = root->sectorsize - 1; |
07127184 | 7757 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 7758 | |
0ef8b726 JB |
7759 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
7760 | (u64)-1); | |
7761 | if (ret) | |
7762 | return ret; | |
39279cc3 | 7763 | |
fcb80c2a JB |
7764 | /* |
7765 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
7766 | * 3 things going on here | |
7767 | * | |
7768 | * 1) We need to reserve space for our orphan item and the space to | |
7769 | * delete our orphan item. Lord knows we don't want to have a dangling | |
7770 | * orphan item because we didn't reserve space to remove it. | |
7771 | * | |
7772 | * 2) We need to reserve space to update our inode. | |
7773 | * | |
7774 | * 3) We need to have something to cache all the space that is going to | |
7775 | * be free'd up by the truncate operation, but also have some slack | |
7776 | * space reserved in case it uses space during the truncate (thank you | |
7777 | * very much snapshotting). | |
7778 | * | |
7779 | * And we need these to all be seperate. The fact is we can use alot of | |
7780 | * space doing the truncate, and we have no earthly idea how much space | |
7781 | * we will use, so we need the truncate reservation to be seperate so it | |
7782 | * doesn't end up using space reserved for updating the inode or | |
7783 | * removing the orphan item. We also need to be able to stop the | |
7784 | * transaction and start a new one, which means we need to be able to | |
7785 | * update the inode several times, and we have no idea of knowing how | |
7786 | * many times that will be, so we can't just reserve 1 item for the | |
7787 | * entirety of the opration, so that has to be done seperately as well. | |
7788 | * Then there is the orphan item, which does indeed need to be held on | |
7789 | * to for the whole operation, and we need nobody to touch this reserved | |
7790 | * space except the orphan code. | |
7791 | * | |
7792 | * So that leaves us with | |
7793 | * | |
7794 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7795 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7796 | * transaction reservation. | |
7797 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7798 | * updating the inode. | |
7799 | */ | |
66d8f3dd | 7800 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7801 | if (!rsv) |
7802 | return -ENOMEM; | |
4a338542 | 7803 | rsv->size = min_size; |
ca7e70f5 | 7804 | rsv->failfast = 1; |
f0cd846e | 7805 | |
907cbceb | 7806 | /* |
07127184 | 7807 | * 1 for the truncate slack space |
907cbceb JB |
7808 | * 1 for updating the inode. |
7809 | */ | |
f3fe820c | 7810 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7811 | if (IS_ERR(trans)) { |
7812 | err = PTR_ERR(trans); | |
7813 | goto out; | |
7814 | } | |
f0cd846e | 7815 | |
907cbceb JB |
7816 | /* Migrate the slack space for the truncate to our reserve */ |
7817 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7818 | min_size); | |
fcb80c2a | 7819 | BUG_ON(ret); |
f0cd846e | 7820 | |
5a3f23d5 CM |
7821 | /* |
7822 | * setattr is responsible for setting the ordered_data_close flag, | |
7823 | * but that is only tested during the last file release. That | |
7824 | * could happen well after the next commit, leaving a great big | |
7825 | * window where new writes may get lost if someone chooses to write | |
7826 | * to this file after truncating to zero | |
7827 | * | |
7828 | * The inode doesn't have any dirty data here, and so if we commit | |
7829 | * this is a noop. If someone immediately starts writing to the inode | |
7830 | * it is very likely we'll catch some of their writes in this | |
7831 | * transaction, and the commit will find this file on the ordered | |
7832 | * data list with good things to send down. | |
7833 | * | |
7834 | * This is a best effort solution, there is still a window where | |
7835 | * using truncate to replace the contents of the file will | |
7836 | * end up with a zero length file after a crash. | |
7837 | */ | |
72ac3c0d JB |
7838 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7839 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7840 | btrfs_add_ordered_operation(trans, root, inode); |
7841 | ||
5dc562c5 JB |
7842 | /* |
7843 | * So if we truncate and then write and fsync we normally would just | |
7844 | * write the extents that changed, which is a problem if we need to | |
7845 | * first truncate that entire inode. So set this flag so we write out | |
7846 | * all of the extents in the inode to the sync log so we're completely | |
7847 | * safe. | |
7848 | */ | |
7849 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7850 | trans->block_rsv = rsv; |
907cbceb | 7851 | |
8082510e YZ |
7852 | while (1) { |
7853 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7854 | inode->i_size, | |
7855 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7856 | if (ret != -ENOSPC) { |
3893e33b | 7857 | err = ret; |
8082510e | 7858 | break; |
3893e33b | 7859 | } |
39279cc3 | 7860 | |
fcb80c2a | 7861 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7862 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7863 | if (ret) { |
7864 | err = ret; | |
7865 | break; | |
7866 | } | |
ca7e70f5 | 7867 | |
8082510e | 7868 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7869 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7870 | |
7871 | trans = btrfs_start_transaction(root, 2); | |
7872 | if (IS_ERR(trans)) { | |
7873 | ret = err = PTR_ERR(trans); | |
7874 | trans = NULL; | |
7875 | break; | |
7876 | } | |
7877 | ||
7878 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7879 | rsv, min_size); | |
7880 | BUG_ON(ret); /* shouldn't happen */ | |
7881 | trans->block_rsv = rsv; | |
8082510e YZ |
7882 | } |
7883 | ||
7884 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7885 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7886 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7887 | if (ret) |
7888 | err = ret; | |
8082510e YZ |
7889 | } |
7890 | ||
917c16b2 CM |
7891 | if (trans) { |
7892 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7893 | ret = btrfs_update_inode(trans, root, inode); | |
7894 | if (ret && !err) | |
7895 | err = ret; | |
7b128766 | 7896 | |
7ad85bb7 | 7897 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7898 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7899 | } |
fcb80c2a JB |
7900 | |
7901 | out: | |
7902 | btrfs_free_block_rsv(root, rsv); | |
7903 | ||
3893e33b JB |
7904 | if (ret && !err) |
7905 | err = ret; | |
a41ad394 | 7906 | |
3893e33b | 7907 | return err; |
39279cc3 CM |
7908 | } |
7909 | ||
d352ac68 CM |
7910 | /* |
7911 | * create a new subvolume directory/inode (helper for the ioctl). | |
7912 | */ | |
d2fb3437 | 7913 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
7914 | struct btrfs_root *new_root, |
7915 | struct btrfs_root *parent_root, | |
7916 | u64 new_dirid) | |
39279cc3 | 7917 | { |
39279cc3 | 7918 | struct inode *inode; |
76dda93c | 7919 | int err; |
00e4e6b3 | 7920 | u64 index = 0; |
39279cc3 | 7921 | |
12fc9d09 FA |
7922 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7923 | new_dirid, new_dirid, | |
7924 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7925 | &index); | |
54aa1f4d | 7926 | if (IS_ERR(inode)) |
f46b5a66 | 7927 | return PTR_ERR(inode); |
39279cc3 CM |
7928 | inode->i_op = &btrfs_dir_inode_operations; |
7929 | inode->i_fop = &btrfs_dir_file_operations; | |
7930 | ||
bfe86848 | 7931 | set_nlink(inode, 1); |
dbe674a9 | 7932 | btrfs_i_size_write(inode, 0); |
3b96362c | 7933 | |
63541927 FDBM |
7934 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
7935 | if (err) | |
7936 | btrfs_err(new_root->fs_info, | |
7937 | "error inheriting subvolume %llu properties: %d\n", | |
7938 | new_root->root_key.objectid, err); | |
7939 | ||
76dda93c | 7940 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7941 | |
76dda93c | 7942 | iput(inode); |
ce598979 | 7943 | return err; |
39279cc3 CM |
7944 | } |
7945 | ||
39279cc3 CM |
7946 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7947 | { | |
7948 | struct btrfs_inode *ei; | |
2ead6ae7 | 7949 | struct inode *inode; |
39279cc3 CM |
7950 | |
7951 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7952 | if (!ei) | |
7953 | return NULL; | |
2ead6ae7 YZ |
7954 | |
7955 | ei->root = NULL; | |
2ead6ae7 | 7956 | ei->generation = 0; |
15ee9bc7 | 7957 | ei->last_trans = 0; |
257c62e1 | 7958 | ei->last_sub_trans = 0; |
e02119d5 | 7959 | ei->logged_trans = 0; |
2ead6ae7 | 7960 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7961 | ei->disk_i_size = 0; |
7962 | ei->flags = 0; | |
7709cde3 | 7963 | ei->csum_bytes = 0; |
2ead6ae7 | 7964 | ei->index_cnt = (u64)-1; |
67de1176 | 7965 | ei->dir_index = 0; |
2ead6ae7 | 7966 | ei->last_unlink_trans = 0; |
46d8bc34 | 7967 | ei->last_log_commit = 0; |
2ead6ae7 | 7968 | |
9e0baf60 JB |
7969 | spin_lock_init(&ei->lock); |
7970 | ei->outstanding_extents = 0; | |
7971 | ei->reserved_extents = 0; | |
2ead6ae7 | 7972 | |
72ac3c0d | 7973 | ei->runtime_flags = 0; |
261507a0 | 7974 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7975 | |
16cdcec7 MX |
7976 | ei->delayed_node = NULL; |
7977 | ||
2ead6ae7 | 7978 | inode = &ei->vfs_inode; |
a8067e02 | 7979 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7980 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7981 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7982 | ei->io_tree.track_uptodate = 1; |
7983 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7984 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7985 | mutex_init(&ei->log_mutex); |
f248679e | 7986 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7987 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7988 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7989 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7990 | RB_CLEAR_NODE(&ei->rb_node); |
7991 | ||
7992 | return inode; | |
39279cc3 CM |
7993 | } |
7994 | ||
aaedb55b JB |
7995 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
7996 | void btrfs_test_destroy_inode(struct inode *inode) | |
7997 | { | |
7998 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
7999 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
8000 | } | |
8001 | #endif | |
8002 | ||
fa0d7e3d NP |
8003 | static void btrfs_i_callback(struct rcu_head *head) |
8004 | { | |
8005 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
8006 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8007 | } | |
8008 | ||
39279cc3 CM |
8009 | void btrfs_destroy_inode(struct inode *inode) |
8010 | { | |
e6dcd2dc | 8011 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
8012 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8013 | ||
b3d9b7a3 | 8014 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 8015 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
8016 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
8017 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
8018 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
8019 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 8020 | |
a6dbd429 JB |
8021 | /* |
8022 | * This can happen where we create an inode, but somebody else also | |
8023 | * created the same inode and we need to destroy the one we already | |
8024 | * created. | |
8025 | */ | |
8026 | if (!root) | |
8027 | goto free; | |
8028 | ||
5a3f23d5 CM |
8029 | /* |
8030 | * Make sure we're properly removed from the ordered operation | |
8031 | * lists. | |
8032 | */ | |
8033 | smp_mb(); | |
8034 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
199c2a9c | 8035 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 8036 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
199c2a9c | 8037 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
8038 | } |
8039 | ||
8a35d95f JB |
8040 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8041 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 8042 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 8043 | btrfs_ino(inode)); |
8a35d95f | 8044 | atomic_dec(&root->orphan_inodes); |
7b128766 | 8045 | } |
7b128766 | 8046 | |
d397712b | 8047 | while (1) { |
e6dcd2dc CM |
8048 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
8049 | if (!ordered) | |
8050 | break; | |
8051 | else { | |
c2cf52eb | 8052 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 8053 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
8054 | btrfs_remove_ordered_extent(inode, ordered); |
8055 | btrfs_put_ordered_extent(ordered); | |
8056 | btrfs_put_ordered_extent(ordered); | |
8057 | } | |
8058 | } | |
5d4f98a2 | 8059 | inode_tree_del(inode); |
5b21f2ed | 8060 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 8061 | free: |
fa0d7e3d | 8062 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
8063 | } |
8064 | ||
45321ac5 | 8065 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8066 | { |
8067 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8068 | |
6379ef9f NA |
8069 | if (root == NULL) |
8070 | return 1; | |
8071 | ||
fa6ac876 | 8072 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8073 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8074 | return 1; |
76dda93c | 8075 | else |
45321ac5 | 8076 | return generic_drop_inode(inode); |
76dda93c YZ |
8077 | } |
8078 | ||
0ee0fda0 | 8079 | static void init_once(void *foo) |
39279cc3 CM |
8080 | { |
8081 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8082 | ||
8083 | inode_init_once(&ei->vfs_inode); | |
8084 | } | |
8085 | ||
8086 | void btrfs_destroy_cachep(void) | |
8087 | { | |
8c0a8537 KS |
8088 | /* |
8089 | * Make sure all delayed rcu free inodes are flushed before we | |
8090 | * destroy cache. | |
8091 | */ | |
8092 | rcu_barrier(); | |
39279cc3 CM |
8093 | if (btrfs_inode_cachep) |
8094 | kmem_cache_destroy(btrfs_inode_cachep); | |
8095 | if (btrfs_trans_handle_cachep) | |
8096 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
8097 | if (btrfs_transaction_cachep) | |
8098 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
8099 | if (btrfs_path_cachep) |
8100 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
8101 | if (btrfs_free_space_cachep) |
8102 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
8103 | if (btrfs_delalloc_work_cachep) |
8104 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
8105 | } |
8106 | ||
8107 | int btrfs_init_cachep(void) | |
8108 | { | |
837e1972 | 8109 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
8110 | sizeof(struct btrfs_inode), 0, |
8111 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
8112 | if (!btrfs_inode_cachep) |
8113 | goto fail; | |
9601e3f6 | 8114 | |
837e1972 | 8115 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
8116 | sizeof(struct btrfs_trans_handle), 0, |
8117 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8118 | if (!btrfs_trans_handle_cachep) |
8119 | goto fail; | |
9601e3f6 | 8120 | |
837e1972 | 8121 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
8122 | sizeof(struct btrfs_transaction), 0, |
8123 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8124 | if (!btrfs_transaction_cachep) |
8125 | goto fail; | |
9601e3f6 | 8126 | |
837e1972 | 8127 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
8128 | sizeof(struct btrfs_path), 0, |
8129 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8130 | if (!btrfs_path_cachep) |
8131 | goto fail; | |
9601e3f6 | 8132 | |
837e1972 | 8133 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
8134 | sizeof(struct btrfs_free_space), 0, |
8135 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
8136 | if (!btrfs_free_space_cachep) | |
8137 | goto fail; | |
8138 | ||
8ccf6f19 MX |
8139 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
8140 | sizeof(struct btrfs_delalloc_work), 0, | |
8141 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
8142 | NULL); | |
8143 | if (!btrfs_delalloc_work_cachep) | |
8144 | goto fail; | |
8145 | ||
39279cc3 CM |
8146 | return 0; |
8147 | fail: | |
8148 | btrfs_destroy_cachep(); | |
8149 | return -ENOMEM; | |
8150 | } | |
8151 | ||
8152 | static int btrfs_getattr(struct vfsmount *mnt, | |
8153 | struct dentry *dentry, struct kstat *stat) | |
8154 | { | |
df0af1a5 | 8155 | u64 delalloc_bytes; |
39279cc3 | 8156 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
8157 | u32 blocksize = inode->i_sb->s_blocksize; |
8158 | ||
39279cc3 | 8159 | generic_fillattr(inode, stat); |
0ee5dc67 | 8160 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 8161 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
8162 | |
8163 | spin_lock(&BTRFS_I(inode)->lock); | |
8164 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
8165 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 8166 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 8167 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8168 | return 0; |
8169 | } | |
8170 | ||
d397712b CM |
8171 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8172 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
8173 | { |
8174 | struct btrfs_trans_handle *trans; | |
8175 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 8176 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
8177 | struct inode *new_inode = new_dentry->d_inode; |
8178 | struct inode *old_inode = old_dentry->d_inode; | |
8179 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 8180 | u64 index = 0; |
4df27c4d | 8181 | u64 root_objectid; |
39279cc3 | 8182 | int ret; |
33345d01 | 8183 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 8184 | |
33345d01 | 8185 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
8186 | return -EPERM; |
8187 | ||
4df27c4d | 8188 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 8189 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8190 | return -EXDEV; |
8191 | ||
33345d01 LZ |
8192 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
8193 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 8194 | return -ENOTEMPTY; |
5f39d397 | 8195 | |
4df27c4d YZ |
8196 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8197 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8198 | return -ENOTEMPTY; | |
9c52057c CM |
8199 | |
8200 | ||
8201 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 8202 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
8203 | new_dentry->d_name.name, |
8204 | new_dentry->d_name.len); | |
8205 | ||
8206 | if (ret) { | |
8207 | if (ret == -EEXIST) { | |
8208 | /* we shouldn't get | |
8209 | * eexist without a new_inode */ | |
fae7f21c | 8210 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
8211 | return ret; |
8212 | } | |
8213 | } else { | |
8214 | /* maybe -EOVERFLOW */ | |
8215 | return ret; | |
8216 | } | |
8217 | } | |
8218 | ret = 0; | |
8219 | ||
5a3f23d5 CM |
8220 | /* |
8221 | * we're using rename to replace one file with another. | |
8222 | * and the replacement file is large. Start IO on it now so | |
8223 | * we don't add too much work to the end of the transaction | |
8224 | */ | |
4baf8c92 | 8225 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
8226 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
8227 | filemap_flush(old_inode->i_mapping); | |
8228 | ||
76dda93c | 8229 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8230 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8231 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8232 | /* |
8233 | * We want to reserve the absolute worst case amount of items. So if | |
8234 | * both inodes are subvols and we need to unlink them then that would | |
8235 | * require 4 item modifications, but if they are both normal inodes it | |
8236 | * would require 5 item modifications, so we'll assume their normal | |
8237 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8238 | * should cover the worst case number of items we'll modify. | |
8239 | */ | |
6e137ed3 | 8240 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8241 | if (IS_ERR(trans)) { |
8242 | ret = PTR_ERR(trans); | |
8243 | goto out_notrans; | |
8244 | } | |
76dda93c | 8245 | |
4df27c4d YZ |
8246 | if (dest != root) |
8247 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8248 | |
a5719521 YZ |
8249 | ret = btrfs_set_inode_index(new_dir, &index); |
8250 | if (ret) | |
8251 | goto out_fail; | |
5a3f23d5 | 8252 | |
67de1176 | 8253 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 8254 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8255 | /* force full log commit if subvolume involved. */ |
8256 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
8257 | } else { | |
a5719521 YZ |
8258 | ret = btrfs_insert_inode_ref(trans, dest, |
8259 | new_dentry->d_name.name, | |
8260 | new_dentry->d_name.len, | |
33345d01 LZ |
8261 | old_ino, |
8262 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8263 | if (ret) |
8264 | goto out_fail; | |
4df27c4d YZ |
8265 | /* |
8266 | * this is an ugly little race, but the rename is required | |
8267 | * to make sure that if we crash, the inode is either at the | |
8268 | * old name or the new one. pinning the log transaction lets | |
8269 | * us make sure we don't allow a log commit to come in after | |
8270 | * we unlink the name but before we add the new name back in. | |
8271 | */ | |
8272 | btrfs_pin_log_trans(root); | |
8273 | } | |
5a3f23d5 CM |
8274 | /* |
8275 | * make sure the inode gets flushed if it is replacing | |
8276 | * something. | |
8277 | */ | |
33345d01 | 8278 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 8279 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 8280 | |
0c4d2d95 JB |
8281 | inode_inc_iversion(old_dir); |
8282 | inode_inc_iversion(new_dir); | |
8283 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8284 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8285 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8286 | old_inode->i_ctime = ctime; | |
5f39d397 | 8287 | |
12fcfd22 CM |
8288 | if (old_dentry->d_parent != new_dentry->d_parent) |
8289 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8290 | ||
33345d01 | 8291 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8292 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8293 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8294 | old_dentry->d_name.name, | |
8295 | old_dentry->d_name.len); | |
8296 | } else { | |
92986796 AV |
8297 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8298 | old_dentry->d_inode, | |
8299 | old_dentry->d_name.name, | |
8300 | old_dentry->d_name.len); | |
8301 | if (!ret) | |
8302 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8303 | } |
79787eaa JM |
8304 | if (ret) { |
8305 | btrfs_abort_transaction(trans, root, ret); | |
8306 | goto out_fail; | |
8307 | } | |
39279cc3 CM |
8308 | |
8309 | if (new_inode) { | |
0c4d2d95 | 8310 | inode_inc_iversion(new_inode); |
39279cc3 | 8311 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8312 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8313 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8314 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8315 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8316 | root_objectid, | |
8317 | new_dentry->d_name.name, | |
8318 | new_dentry->d_name.len); | |
8319 | BUG_ON(new_inode->i_nlink == 0); | |
8320 | } else { | |
8321 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8322 | new_dentry->d_inode, | |
8323 | new_dentry->d_name.name, | |
8324 | new_dentry->d_name.len); | |
8325 | } | |
4ef31a45 | 8326 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8327 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8328 | if (ret) { |
8329 | btrfs_abort_transaction(trans, root, ret); | |
8330 | goto out_fail; | |
8331 | } | |
39279cc3 | 8332 | } |
aec7477b | 8333 | |
4df27c4d YZ |
8334 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8335 | new_dentry->d_name.name, | |
a5719521 | 8336 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8337 | if (ret) { |
8338 | btrfs_abort_transaction(trans, root, ret); | |
8339 | goto out_fail; | |
8340 | } | |
39279cc3 | 8341 | |
67de1176 MX |
8342 | if (old_inode->i_nlink == 1) |
8343 | BTRFS_I(old_inode)->dir_index = index; | |
8344 | ||
33345d01 | 8345 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8346 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8347 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8348 | btrfs_end_log_trans(root); |
8349 | } | |
39279cc3 | 8350 | out_fail: |
7ad85bb7 | 8351 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8352 | out_notrans: |
33345d01 | 8353 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8354 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8355 | |
39279cc3 CM |
8356 | return ret; |
8357 | } | |
8358 | ||
8ccf6f19 MX |
8359 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8360 | { | |
8361 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 8362 | struct inode *inode; |
8ccf6f19 MX |
8363 | |
8364 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8365 | work); | |
9f23e289 JB |
8366 | inode = delalloc_work->inode; |
8367 | if (delalloc_work->wait) { | |
8368 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
8369 | } else { | |
8370 | filemap_flush(inode->i_mapping); | |
8371 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
8372 | &BTRFS_I(inode)->runtime_flags)) | |
8373 | filemap_flush(inode->i_mapping); | |
8374 | } | |
8ccf6f19 MX |
8375 | |
8376 | if (delalloc_work->delay_iput) | |
9f23e289 | 8377 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 8378 | else |
9f23e289 | 8379 | iput(inode); |
8ccf6f19 MX |
8380 | complete(&delalloc_work->completion); |
8381 | } | |
8382 | ||
8383 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8384 | int wait, int delay_iput) | |
8385 | { | |
8386 | struct btrfs_delalloc_work *work; | |
8387 | ||
8388 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8389 | if (!work) | |
8390 | return NULL; | |
8391 | ||
8392 | init_completion(&work->completion); | |
8393 | INIT_LIST_HEAD(&work->list); | |
8394 | work->inode = inode; | |
8395 | work->wait = wait; | |
8396 | work->delay_iput = delay_iput; | |
8397 | work->work.func = btrfs_run_delalloc_work; | |
8398 | ||
8399 | return work; | |
8400 | } | |
8401 | ||
8402 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8403 | { | |
8404 | wait_for_completion(&work->completion); | |
8405 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8406 | } | |
8407 | ||
d352ac68 CM |
8408 | /* |
8409 | * some fairly slow code that needs optimization. This walks the list | |
8410 | * of all the inodes with pending delalloc and forces them to disk. | |
8411 | */ | |
eb73c1b7 | 8412 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 8413 | { |
ea8c2819 | 8414 | struct btrfs_inode *binode; |
5b21f2ed | 8415 | struct inode *inode; |
8ccf6f19 MX |
8416 | struct btrfs_delalloc_work *work, *next; |
8417 | struct list_head works; | |
1eafa6c7 | 8418 | struct list_head splice; |
8ccf6f19 | 8419 | int ret = 0; |
ea8c2819 | 8420 | |
8ccf6f19 | 8421 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8422 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8423 | |
eb73c1b7 MX |
8424 | spin_lock(&root->delalloc_lock); |
8425 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8426 | while (!list_empty(&splice)) { |
8427 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8428 | delalloc_inodes); |
1eafa6c7 | 8429 | |
eb73c1b7 MX |
8430 | list_move_tail(&binode->delalloc_inodes, |
8431 | &root->delalloc_inodes); | |
5b21f2ed | 8432 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8433 | if (!inode) { |
eb73c1b7 | 8434 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8435 | continue; |
df0af1a5 | 8436 | } |
eb73c1b7 | 8437 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8438 | |
8439 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
8440 | if (unlikely(!work)) { | |
f4ab9ea7 JB |
8441 | if (delay_iput) |
8442 | btrfs_add_delayed_iput(inode); | |
8443 | else | |
8444 | iput(inode); | |
1eafa6c7 MX |
8445 | ret = -ENOMEM; |
8446 | goto out; | |
5b21f2ed | 8447 | } |
1eafa6c7 MX |
8448 | list_add_tail(&work->list, &works); |
8449 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
8450 | &work->work); | |
8451 | ||
5b21f2ed | 8452 | cond_resched(); |
eb73c1b7 | 8453 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 8454 | } |
eb73c1b7 | 8455 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 8456 | |
1eafa6c7 MX |
8457 | list_for_each_entry_safe(work, next, &works, list) { |
8458 | list_del_init(&work->list); | |
8459 | btrfs_wait_and_free_delalloc_work(work); | |
8460 | } | |
eb73c1b7 MX |
8461 | return 0; |
8462 | out: | |
8463 | list_for_each_entry_safe(work, next, &works, list) { | |
8464 | list_del_init(&work->list); | |
8465 | btrfs_wait_and_free_delalloc_work(work); | |
8466 | } | |
8467 | ||
8468 | if (!list_empty_careful(&splice)) { | |
8469 | spin_lock(&root->delalloc_lock); | |
8470 | list_splice_tail(&splice, &root->delalloc_inodes); | |
8471 | spin_unlock(&root->delalloc_lock); | |
8472 | } | |
8473 | return ret; | |
8474 | } | |
1eafa6c7 | 8475 | |
eb73c1b7 MX |
8476 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
8477 | { | |
8478 | int ret; | |
1eafa6c7 | 8479 | |
2c21b4d7 | 8480 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
8481 | return -EROFS; |
8482 | ||
8483 | ret = __start_delalloc_inodes(root, delay_iput); | |
8484 | /* | |
8485 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
8486 | * we have to make sure the IO is actually started and that |
8487 | * ordered extents get created before we return | |
8488 | */ | |
8489 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 8490 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 8491 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 8492 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
8493 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
8494 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
8495 | } |
8496 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
8497 | return ret; |
8498 | } | |
8499 | ||
91aef86f | 8500 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput) |
eb73c1b7 MX |
8501 | { |
8502 | struct btrfs_root *root; | |
8503 | struct list_head splice; | |
8504 | int ret; | |
8505 | ||
2c21b4d7 | 8506 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
8507 | return -EROFS; |
8508 | ||
8509 | INIT_LIST_HEAD(&splice); | |
8510 | ||
8511 | spin_lock(&fs_info->delalloc_root_lock); | |
8512 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
8513 | while (!list_empty(&splice)) { | |
8514 | root = list_first_entry(&splice, struct btrfs_root, | |
8515 | delalloc_root); | |
8516 | root = btrfs_grab_fs_root(root); | |
8517 | BUG_ON(!root); | |
8518 | list_move_tail(&root->delalloc_root, | |
8519 | &fs_info->delalloc_roots); | |
8520 | spin_unlock(&fs_info->delalloc_root_lock); | |
8521 | ||
8522 | ret = __start_delalloc_inodes(root, delay_iput); | |
8523 | btrfs_put_fs_root(root); | |
8524 | if (ret) | |
8525 | goto out; | |
8526 | ||
8527 | spin_lock(&fs_info->delalloc_root_lock); | |
8ccf6f19 | 8528 | } |
eb73c1b7 | 8529 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 8530 | |
eb73c1b7 MX |
8531 | atomic_inc(&fs_info->async_submit_draining); |
8532 | while (atomic_read(&fs_info->nr_async_submits) || | |
8533 | atomic_read(&fs_info->async_delalloc_pages)) { | |
8534 | wait_event(fs_info->async_submit_wait, | |
8535 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
8536 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
8537 | } | |
8538 | atomic_dec(&fs_info->async_submit_draining); | |
8539 | return 0; | |
8540 | out: | |
1eafa6c7 | 8541 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
8542 | spin_lock(&fs_info->delalloc_root_lock); |
8543 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
8544 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 8545 | } |
8ccf6f19 | 8546 | return ret; |
ea8c2819 CM |
8547 | } |
8548 | ||
39279cc3 CM |
8549 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
8550 | const char *symname) | |
8551 | { | |
8552 | struct btrfs_trans_handle *trans; | |
8553 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
8554 | struct btrfs_path *path; | |
8555 | struct btrfs_key key; | |
1832a6d5 | 8556 | struct inode *inode = NULL; |
39279cc3 CM |
8557 | int err; |
8558 | int drop_inode = 0; | |
8559 | u64 objectid; | |
67871254 | 8560 | u64 index = 0; |
39279cc3 CM |
8561 | int name_len; |
8562 | int datasize; | |
5f39d397 | 8563 | unsigned long ptr; |
39279cc3 | 8564 | struct btrfs_file_extent_item *ei; |
5f39d397 | 8565 | struct extent_buffer *leaf; |
39279cc3 | 8566 | |
f06becc4 | 8567 | name_len = strlen(symname); |
39279cc3 CM |
8568 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
8569 | return -ENAMETOOLONG; | |
1832a6d5 | 8570 | |
9ed74f2d JB |
8571 | /* |
8572 | * 2 items for inode item and ref | |
8573 | * 2 items for dir items | |
8574 | * 1 item for xattr if selinux is on | |
8575 | */ | |
a22285a6 YZ |
8576 | trans = btrfs_start_transaction(root, 5); |
8577 | if (IS_ERR(trans)) | |
8578 | return PTR_ERR(trans); | |
1832a6d5 | 8579 | |
581bb050 LZ |
8580 | err = btrfs_find_free_ino(root, &objectid); |
8581 | if (err) | |
8582 | goto out_unlock; | |
8583 | ||
aec7477b | 8584 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 8585 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 8586 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
8587 | if (IS_ERR(inode)) { |
8588 | err = PTR_ERR(inode); | |
39279cc3 | 8589 | goto out_unlock; |
7cf96da3 | 8590 | } |
39279cc3 | 8591 | |
2a7dba39 | 8592 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
8593 | if (err) { |
8594 | drop_inode = 1; | |
8595 | goto out_unlock; | |
8596 | } | |
8597 | ||
ad19db71 CS |
8598 | /* |
8599 | * If the active LSM wants to access the inode during | |
8600 | * d_instantiate it needs these. Smack checks to see | |
8601 | * if the filesystem supports xattrs by looking at the | |
8602 | * ops vector. | |
8603 | */ | |
8604 | inode->i_fop = &btrfs_file_operations; | |
8605 | inode->i_op = &btrfs_file_inode_operations; | |
8606 | ||
a1b075d2 | 8607 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
8608 | if (err) |
8609 | drop_inode = 1; | |
8610 | else { | |
8611 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 8612 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 8613 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 8614 | } |
39279cc3 CM |
8615 | if (drop_inode) |
8616 | goto out_unlock; | |
8617 | ||
8618 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
8619 | if (!path) { |
8620 | err = -ENOMEM; | |
8621 | drop_inode = 1; | |
8622 | goto out_unlock; | |
8623 | } | |
33345d01 | 8624 | key.objectid = btrfs_ino(inode); |
39279cc3 | 8625 | key.offset = 0; |
39279cc3 CM |
8626 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
8627 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
8628 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
8629 | datasize); | |
54aa1f4d CM |
8630 | if (err) { |
8631 | drop_inode = 1; | |
b0839166 | 8632 | btrfs_free_path(path); |
54aa1f4d CM |
8633 | goto out_unlock; |
8634 | } | |
5f39d397 CM |
8635 | leaf = path->nodes[0]; |
8636 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
8637 | struct btrfs_file_extent_item); | |
8638 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
8639 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 8640 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
8641 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
8642 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
8643 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
8644 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
8645 | ||
39279cc3 | 8646 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
8647 | write_extent_buffer(leaf, symname, ptr, name_len); |
8648 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 8649 | btrfs_free_path(path); |
5f39d397 | 8650 | |
39279cc3 CM |
8651 | inode->i_op = &btrfs_symlink_inode_operations; |
8652 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 8653 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 8654 | inode_set_bytes(inode, name_len); |
f06becc4 | 8655 | btrfs_i_size_write(inode, name_len); |
54aa1f4d CM |
8656 | err = btrfs_update_inode(trans, root, inode); |
8657 | if (err) | |
8658 | drop_inode = 1; | |
39279cc3 CM |
8659 | |
8660 | out_unlock: | |
08c422c2 AV |
8661 | if (!err) |
8662 | d_instantiate(dentry, inode); | |
7ad85bb7 | 8663 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
8664 | if (drop_inode) { |
8665 | inode_dec_link_count(inode); | |
8666 | iput(inode); | |
8667 | } | |
b53d3f5d | 8668 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
8669 | return err; |
8670 | } | |
16432985 | 8671 | |
0af3d00b JB |
8672 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
8673 | u64 start, u64 num_bytes, u64 min_size, | |
8674 | loff_t actual_len, u64 *alloc_hint, | |
8675 | struct btrfs_trans_handle *trans) | |
d899e052 | 8676 | { |
5dc562c5 JB |
8677 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
8678 | struct extent_map *em; | |
d899e052 YZ |
8679 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8680 | struct btrfs_key ins; | |
d899e052 | 8681 | u64 cur_offset = start; |
55a61d1d | 8682 | u64 i_size; |
154ea289 | 8683 | u64 cur_bytes; |
d899e052 | 8684 | int ret = 0; |
0af3d00b | 8685 | bool own_trans = true; |
d899e052 | 8686 | |
0af3d00b JB |
8687 | if (trans) |
8688 | own_trans = false; | |
d899e052 | 8689 | while (num_bytes > 0) { |
0af3d00b JB |
8690 | if (own_trans) { |
8691 | trans = btrfs_start_transaction(root, 3); | |
8692 | if (IS_ERR(trans)) { | |
8693 | ret = PTR_ERR(trans); | |
8694 | break; | |
8695 | } | |
5a303d5d YZ |
8696 | } |
8697 | ||
154ea289 CM |
8698 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
8699 | cur_bytes = max(cur_bytes, min_size); | |
00361589 JB |
8700 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
8701 | *alloc_hint, &ins, 1); | |
5a303d5d | 8702 | if (ret) { |
0af3d00b JB |
8703 | if (own_trans) |
8704 | btrfs_end_transaction(trans, root); | |
a22285a6 | 8705 | break; |
d899e052 | 8706 | } |
5a303d5d | 8707 | |
d899e052 YZ |
8708 | ret = insert_reserved_file_extent(trans, inode, |
8709 | cur_offset, ins.objectid, | |
8710 | ins.offset, ins.offset, | |
920bbbfb | 8711 | ins.offset, 0, 0, 0, |
d899e052 | 8712 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 8713 | if (ret) { |
857cc2fc JB |
8714 | btrfs_free_reserved_extent(root, ins.objectid, |
8715 | ins.offset); | |
79787eaa JM |
8716 | btrfs_abort_transaction(trans, root, ret); |
8717 | if (own_trans) | |
8718 | btrfs_end_transaction(trans, root); | |
8719 | break; | |
8720 | } | |
a1ed835e CM |
8721 | btrfs_drop_extent_cache(inode, cur_offset, |
8722 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 8723 | |
5dc562c5 JB |
8724 | em = alloc_extent_map(); |
8725 | if (!em) { | |
8726 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
8727 | &BTRFS_I(inode)->runtime_flags); | |
8728 | goto next; | |
8729 | } | |
8730 | ||
8731 | em->start = cur_offset; | |
8732 | em->orig_start = cur_offset; | |
8733 | em->len = ins.offset; | |
8734 | em->block_start = ins.objectid; | |
8735 | em->block_len = ins.offset; | |
b4939680 | 8736 | em->orig_block_len = ins.offset; |
cc95bef6 | 8737 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
8738 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
8739 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
8740 | em->generation = trans->transid; | |
8741 | ||
8742 | while (1) { | |
8743 | write_lock(&em_tree->lock); | |
09a2a8f9 | 8744 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
8745 | write_unlock(&em_tree->lock); |
8746 | if (ret != -EEXIST) | |
8747 | break; | |
8748 | btrfs_drop_extent_cache(inode, cur_offset, | |
8749 | cur_offset + ins.offset - 1, | |
8750 | 0); | |
8751 | } | |
8752 | free_extent_map(em); | |
8753 | next: | |
d899e052 YZ |
8754 | num_bytes -= ins.offset; |
8755 | cur_offset += ins.offset; | |
efa56464 | 8756 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 8757 | |
0c4d2d95 | 8758 | inode_inc_iversion(inode); |
d899e052 | 8759 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 8760 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 8761 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
8762 | (actual_len > inode->i_size) && |
8763 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 8764 | if (cur_offset > actual_len) |
55a61d1d | 8765 | i_size = actual_len; |
d1ea6a61 | 8766 | else |
55a61d1d JB |
8767 | i_size = cur_offset; |
8768 | i_size_write(inode, i_size); | |
8769 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
8770 | } |
8771 | ||
d899e052 | 8772 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
8773 | |
8774 | if (ret) { | |
8775 | btrfs_abort_transaction(trans, root, ret); | |
8776 | if (own_trans) | |
8777 | btrfs_end_transaction(trans, root); | |
8778 | break; | |
8779 | } | |
d899e052 | 8780 | |
0af3d00b JB |
8781 | if (own_trans) |
8782 | btrfs_end_transaction(trans, root); | |
5a303d5d | 8783 | } |
d899e052 YZ |
8784 | return ret; |
8785 | } | |
8786 | ||
0af3d00b JB |
8787 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
8788 | u64 start, u64 num_bytes, u64 min_size, | |
8789 | loff_t actual_len, u64 *alloc_hint) | |
8790 | { | |
8791 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8792 | min_size, actual_len, alloc_hint, | |
8793 | NULL); | |
8794 | } | |
8795 | ||
8796 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
8797 | struct btrfs_trans_handle *trans, int mode, | |
8798 | u64 start, u64 num_bytes, u64 min_size, | |
8799 | loff_t actual_len, u64 *alloc_hint) | |
8800 | { | |
8801 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8802 | min_size, actual_len, alloc_hint, trans); | |
8803 | } | |
8804 | ||
e6dcd2dc CM |
8805 | static int btrfs_set_page_dirty(struct page *page) |
8806 | { | |
e6dcd2dc CM |
8807 | return __set_page_dirty_nobuffers(page); |
8808 | } | |
8809 | ||
10556cb2 | 8810 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 8811 | { |
b83cc969 | 8812 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 8813 | umode_t mode = inode->i_mode; |
b83cc969 | 8814 | |
cb6db4e5 JM |
8815 | if (mask & MAY_WRITE && |
8816 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
8817 | if (btrfs_root_readonly(root)) | |
8818 | return -EROFS; | |
8819 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
8820 | return -EACCES; | |
8821 | } | |
2830ba7f | 8822 | return generic_permission(inode, mask); |
fdebe2bd | 8823 | } |
39279cc3 | 8824 | |
6e1d5dcc | 8825 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 8826 | .getattr = btrfs_getattr, |
39279cc3 CM |
8827 | .lookup = btrfs_lookup, |
8828 | .create = btrfs_create, | |
8829 | .unlink = btrfs_unlink, | |
8830 | .link = btrfs_link, | |
8831 | .mkdir = btrfs_mkdir, | |
8832 | .rmdir = btrfs_rmdir, | |
8833 | .rename = btrfs_rename, | |
8834 | .symlink = btrfs_symlink, | |
8835 | .setattr = btrfs_setattr, | |
618e21d5 | 8836 | .mknod = btrfs_mknod, |
95819c05 CH |
8837 | .setxattr = btrfs_setxattr, |
8838 | .getxattr = btrfs_getxattr, | |
5103e947 | 8839 | .listxattr = btrfs_listxattr, |
95819c05 | 8840 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8841 | .permission = btrfs_permission, |
4e34e719 | 8842 | .get_acl = btrfs_get_acl, |
93fd63c2 | 8843 | .update_time = btrfs_update_time, |
39279cc3 | 8844 | }; |
6e1d5dcc | 8845 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8846 | .lookup = btrfs_lookup, |
fdebe2bd | 8847 | .permission = btrfs_permission, |
4e34e719 | 8848 | .get_acl = btrfs_get_acl, |
93fd63c2 | 8849 | .update_time = btrfs_update_time, |
39279cc3 | 8850 | }; |
76dda93c | 8851 | |
828c0950 | 8852 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8853 | .llseek = generic_file_llseek, |
8854 | .read = generic_read_dir, | |
9cdda8d3 | 8855 | .iterate = btrfs_real_readdir, |
34287aa3 | 8856 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8857 | #ifdef CONFIG_COMPAT |
34287aa3 | 8858 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8859 | #endif |
6bf13c0c | 8860 | .release = btrfs_release_file, |
e02119d5 | 8861 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8862 | }; |
8863 | ||
d1310b2e | 8864 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8865 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8866 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8867 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8868 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8869 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8870 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8871 | .set_bit_hook = btrfs_set_bit_hook, |
8872 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8873 | .merge_extent_hook = btrfs_merge_extent_hook, |
8874 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8875 | }; |
8876 | ||
35054394 CM |
8877 | /* |
8878 | * btrfs doesn't support the bmap operation because swapfiles | |
8879 | * use bmap to make a mapping of extents in the file. They assume | |
8880 | * these extents won't change over the life of the file and they | |
8881 | * use the bmap result to do IO directly to the drive. | |
8882 | * | |
8883 | * the btrfs bmap call would return logical addresses that aren't | |
8884 | * suitable for IO and they also will change frequently as COW | |
8885 | * operations happen. So, swapfile + btrfs == corruption. | |
8886 | * | |
8887 | * For now we're avoiding this by dropping bmap. | |
8888 | */ | |
7f09410b | 8889 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8890 | .readpage = btrfs_readpage, |
8891 | .writepage = btrfs_writepage, | |
b293f02e | 8892 | .writepages = btrfs_writepages, |
3ab2fb5a | 8893 | .readpages = btrfs_readpages, |
16432985 | 8894 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8895 | .invalidatepage = btrfs_invalidatepage, |
8896 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8897 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8898 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8899 | }; |
8900 | ||
7f09410b | 8901 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8902 | .readpage = btrfs_readpage, |
8903 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8904 | .invalidatepage = btrfs_invalidatepage, |
8905 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8906 | }; |
8907 | ||
6e1d5dcc | 8908 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8909 | .getattr = btrfs_getattr, |
8910 | .setattr = btrfs_setattr, | |
95819c05 CH |
8911 | .setxattr = btrfs_setxattr, |
8912 | .getxattr = btrfs_getxattr, | |
5103e947 | 8913 | .listxattr = btrfs_listxattr, |
95819c05 | 8914 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8915 | .permission = btrfs_permission, |
1506fcc8 | 8916 | .fiemap = btrfs_fiemap, |
4e34e719 | 8917 | .get_acl = btrfs_get_acl, |
e41f941a | 8918 | .update_time = btrfs_update_time, |
39279cc3 | 8919 | }; |
6e1d5dcc | 8920 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8921 | .getattr = btrfs_getattr, |
8922 | .setattr = btrfs_setattr, | |
fdebe2bd | 8923 | .permission = btrfs_permission, |
95819c05 CH |
8924 | .setxattr = btrfs_setxattr, |
8925 | .getxattr = btrfs_getxattr, | |
33268eaf | 8926 | .listxattr = btrfs_listxattr, |
95819c05 | 8927 | .removexattr = btrfs_removexattr, |
4e34e719 | 8928 | .get_acl = btrfs_get_acl, |
e41f941a | 8929 | .update_time = btrfs_update_time, |
618e21d5 | 8930 | }; |
6e1d5dcc | 8931 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8932 | .readlink = generic_readlink, |
8933 | .follow_link = page_follow_link_light, | |
8934 | .put_link = page_put_link, | |
f209561a | 8935 | .getattr = btrfs_getattr, |
22c44fe6 | 8936 | .setattr = btrfs_setattr, |
fdebe2bd | 8937 | .permission = btrfs_permission, |
0279b4cd JO |
8938 | .setxattr = btrfs_setxattr, |
8939 | .getxattr = btrfs_getxattr, | |
8940 | .listxattr = btrfs_listxattr, | |
8941 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8942 | .get_acl = btrfs_get_acl, |
e41f941a | 8943 | .update_time = btrfs_update_time, |
39279cc3 | 8944 | }; |
76dda93c | 8945 | |
82d339d9 | 8946 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8947 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8948 | .d_release = btrfs_dentry_release, |
76dda93c | 8949 | }; |