Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
4b4e25f2 | 39 | #include "compat.h" |
39279cc3 CM |
40 | #include "ctree.h" |
41 | #include "disk-io.h" | |
42 | #include "transaction.h" | |
43 | #include "btrfs_inode.h" | |
44 | #include "ioctl.h" | |
45 | #include "print-tree.h" | |
0b86a832 | 46 | #include "volumes.h" |
e6dcd2dc | 47 | #include "ordered-data.h" |
95819c05 | 48 | #include "xattr.h" |
e02119d5 | 49 | #include "tree-log.h" |
c8b97818 | 50 | #include "compression.h" |
b4ce94de | 51 | #include "locking.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
54 | u64 ino; | |
55 | struct btrfs_root *root; | |
56 | }; | |
57 | ||
58 | static struct inode_operations btrfs_dir_inode_operations; | |
59 | static struct inode_operations btrfs_symlink_inode_operations; | |
60 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 61 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
62 | static struct inode_operations btrfs_file_inode_operations; |
63 | static struct address_space_operations btrfs_aops; | |
64 | static struct address_space_operations btrfs_symlink_aops; | |
65 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 66 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
67 | |
68 | static struct kmem_cache *btrfs_inode_cachep; | |
69 | struct kmem_cache *btrfs_trans_handle_cachep; | |
70 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
71 | struct kmem_cache *btrfs_path_cachep; |
72 | ||
73 | #define S_SHIFT 12 | |
74 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
75 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
76 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
77 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
78 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
79 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
80 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
81 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
82 | }; | |
83 | ||
7b128766 | 84 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 85 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
86 | static noinline int cow_file_range(struct inode *inode, |
87 | struct page *locked_page, | |
88 | u64 start, u64 end, int *page_started, | |
89 | unsigned long *nr_written, int unlock); | |
7b128766 | 90 | |
0279b4cd JO |
91 | static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) |
92 | { | |
93 | int err; | |
94 | ||
95 | err = btrfs_init_acl(inode, dir); | |
96 | if (!err) | |
97 | err = btrfs_xattr_security_init(inode, dir); | |
98 | return err; | |
99 | } | |
100 | ||
c8b97818 CM |
101 | /* |
102 | * this does all the hard work for inserting an inline extent into | |
103 | * the btree. The caller should have done a btrfs_drop_extents so that | |
104 | * no overlapping inline items exist in the btree | |
105 | */ | |
d397712b | 106 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
107 | struct btrfs_root *root, struct inode *inode, |
108 | u64 start, size_t size, size_t compressed_size, | |
109 | struct page **compressed_pages) | |
110 | { | |
111 | struct btrfs_key key; | |
112 | struct btrfs_path *path; | |
113 | struct extent_buffer *leaf; | |
114 | struct page *page = NULL; | |
115 | char *kaddr; | |
116 | unsigned long ptr; | |
117 | struct btrfs_file_extent_item *ei; | |
118 | int err = 0; | |
119 | int ret; | |
120 | size_t cur_size = size; | |
121 | size_t datasize; | |
122 | unsigned long offset; | |
123 | int use_compress = 0; | |
124 | ||
125 | if (compressed_size && compressed_pages) { | |
126 | use_compress = 1; | |
127 | cur_size = compressed_size; | |
128 | } | |
129 | ||
d397712b CM |
130 | path = btrfs_alloc_path(); |
131 | if (!path) | |
c8b97818 CM |
132 | return -ENOMEM; |
133 | ||
b9473439 | 134 | path->leave_spinning = 1; |
c8b97818 CM |
135 | btrfs_set_trans_block_group(trans, inode); |
136 | ||
137 | key.objectid = inode->i_ino; | |
138 | key.offset = start; | |
139 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
140 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
141 | ||
142 | inode_add_bytes(inode, size); | |
143 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
144 | datasize); | |
145 | BUG_ON(ret); | |
146 | if (ret) { | |
147 | err = ret; | |
c8b97818 CM |
148 | goto fail; |
149 | } | |
150 | leaf = path->nodes[0]; | |
151 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
152 | struct btrfs_file_extent_item); | |
153 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
154 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
155 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
156 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
157 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
158 | ptr = btrfs_file_extent_inline_start(ei); | |
159 | ||
160 | if (use_compress) { | |
161 | struct page *cpage; | |
162 | int i = 0; | |
d397712b | 163 | while (compressed_size > 0) { |
c8b97818 | 164 | cpage = compressed_pages[i]; |
5b050f04 | 165 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
166 | PAGE_CACHE_SIZE); |
167 | ||
b9473439 | 168 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 169 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 170 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
171 | |
172 | i++; | |
173 | ptr += cur_size; | |
174 | compressed_size -= cur_size; | |
175 | } | |
176 | btrfs_set_file_extent_compression(leaf, ei, | |
177 | BTRFS_COMPRESS_ZLIB); | |
178 | } else { | |
179 | page = find_get_page(inode->i_mapping, | |
180 | start >> PAGE_CACHE_SHIFT); | |
181 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
182 | kaddr = kmap_atomic(page, KM_USER0); | |
183 | offset = start & (PAGE_CACHE_SIZE - 1); | |
184 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
185 | kunmap_atomic(kaddr, KM_USER0); | |
186 | page_cache_release(page); | |
187 | } | |
188 | btrfs_mark_buffer_dirty(leaf); | |
189 | btrfs_free_path(path); | |
190 | ||
191 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
192 | btrfs_update_inode(trans, root, inode); | |
193 | return 0; | |
194 | fail: | |
195 | btrfs_free_path(path); | |
196 | return err; | |
197 | } | |
198 | ||
199 | ||
200 | /* | |
201 | * conditionally insert an inline extent into the file. This | |
202 | * does the checks required to make sure the data is small enough | |
203 | * to fit as an inline extent. | |
204 | */ | |
7f366cfe | 205 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
206 | struct btrfs_root *root, |
207 | struct inode *inode, u64 start, u64 end, | |
208 | size_t compressed_size, | |
209 | struct page **compressed_pages) | |
210 | { | |
211 | u64 isize = i_size_read(inode); | |
212 | u64 actual_end = min(end + 1, isize); | |
213 | u64 inline_len = actual_end - start; | |
214 | u64 aligned_end = (end + root->sectorsize - 1) & | |
215 | ~((u64)root->sectorsize - 1); | |
216 | u64 hint_byte; | |
217 | u64 data_len = inline_len; | |
218 | int ret; | |
219 | ||
220 | if (compressed_size) | |
221 | data_len = compressed_size; | |
222 | ||
223 | if (start > 0 || | |
70b99e69 | 224 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
225 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
226 | (!compressed_size && | |
227 | (actual_end & (root->sectorsize - 1)) == 0) || | |
228 | end + 1 < isize || | |
229 | data_len > root->fs_info->max_inline) { | |
230 | return 1; | |
231 | } | |
232 | ||
c8b97818 | 233 | ret = btrfs_drop_extents(trans, root, inode, start, |
a1ed835e CM |
234 | aligned_end, aligned_end, start, |
235 | &hint_byte, 1); | |
c8b97818 CM |
236 | BUG_ON(ret); |
237 | ||
238 | if (isize > actual_end) | |
239 | inline_len = min_t(u64, isize, actual_end); | |
240 | ret = insert_inline_extent(trans, root, inode, start, | |
241 | inline_len, compressed_size, | |
242 | compressed_pages); | |
243 | BUG_ON(ret); | |
a1ed835e | 244 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
245 | return 0; |
246 | } | |
247 | ||
771ed689 CM |
248 | struct async_extent { |
249 | u64 start; | |
250 | u64 ram_size; | |
251 | u64 compressed_size; | |
252 | struct page **pages; | |
253 | unsigned long nr_pages; | |
254 | struct list_head list; | |
255 | }; | |
256 | ||
257 | struct async_cow { | |
258 | struct inode *inode; | |
259 | struct btrfs_root *root; | |
260 | struct page *locked_page; | |
261 | u64 start; | |
262 | u64 end; | |
263 | struct list_head extents; | |
264 | struct btrfs_work work; | |
265 | }; | |
266 | ||
267 | static noinline int add_async_extent(struct async_cow *cow, | |
268 | u64 start, u64 ram_size, | |
269 | u64 compressed_size, | |
270 | struct page **pages, | |
271 | unsigned long nr_pages) | |
272 | { | |
273 | struct async_extent *async_extent; | |
274 | ||
275 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
276 | async_extent->start = start; | |
277 | async_extent->ram_size = ram_size; | |
278 | async_extent->compressed_size = compressed_size; | |
279 | async_extent->pages = pages; | |
280 | async_extent->nr_pages = nr_pages; | |
281 | list_add_tail(&async_extent->list, &cow->extents); | |
282 | return 0; | |
283 | } | |
284 | ||
d352ac68 | 285 | /* |
771ed689 CM |
286 | * we create compressed extents in two phases. The first |
287 | * phase compresses a range of pages that have already been | |
288 | * locked (both pages and state bits are locked). | |
c8b97818 | 289 | * |
771ed689 CM |
290 | * This is done inside an ordered work queue, and the compression |
291 | * is spread across many cpus. The actual IO submission is step | |
292 | * two, and the ordered work queue takes care of making sure that | |
293 | * happens in the same order things were put onto the queue by | |
294 | * writepages and friends. | |
c8b97818 | 295 | * |
771ed689 CM |
296 | * If this code finds it can't get good compression, it puts an |
297 | * entry onto the work queue to write the uncompressed bytes. This | |
298 | * makes sure that both compressed inodes and uncompressed inodes | |
299 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 300 | */ |
771ed689 CM |
301 | static noinline int compress_file_range(struct inode *inode, |
302 | struct page *locked_page, | |
303 | u64 start, u64 end, | |
304 | struct async_cow *async_cow, | |
305 | int *num_added) | |
b888db2b CM |
306 | { |
307 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
308 | struct btrfs_trans_handle *trans; | |
db94535d | 309 | u64 num_bytes; |
c8b97818 CM |
310 | u64 orig_start; |
311 | u64 disk_num_bytes; | |
db94535d | 312 | u64 blocksize = root->sectorsize; |
c8b97818 | 313 | u64 actual_end; |
42dc7bab | 314 | u64 isize = i_size_read(inode); |
e6dcd2dc | 315 | int ret = 0; |
c8b97818 CM |
316 | struct page **pages = NULL; |
317 | unsigned long nr_pages; | |
318 | unsigned long nr_pages_ret = 0; | |
319 | unsigned long total_compressed = 0; | |
320 | unsigned long total_in = 0; | |
321 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 322 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
323 | int i; |
324 | int will_compress; | |
b888db2b | 325 | |
c8b97818 CM |
326 | orig_start = start; |
327 | ||
42dc7bab | 328 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
329 | again: |
330 | will_compress = 0; | |
331 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
332 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 333 | |
f03d9301 CM |
334 | /* |
335 | * we don't want to send crud past the end of i_size through | |
336 | * compression, that's just a waste of CPU time. So, if the | |
337 | * end of the file is before the start of our current | |
338 | * requested range of bytes, we bail out to the uncompressed | |
339 | * cleanup code that can deal with all of this. | |
340 | * | |
341 | * It isn't really the fastest way to fix things, but this is a | |
342 | * very uncommon corner. | |
343 | */ | |
344 | if (actual_end <= start) | |
345 | goto cleanup_and_bail_uncompressed; | |
346 | ||
c8b97818 CM |
347 | total_compressed = actual_end - start; |
348 | ||
349 | /* we want to make sure that amount of ram required to uncompress | |
350 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
351 | * of a compressed extent to 128k. This is a crucial number |
352 | * because it also controls how easily we can spread reads across | |
353 | * cpus for decompression. | |
354 | * | |
355 | * We also want to make sure the amount of IO required to do | |
356 | * a random read is reasonably small, so we limit the size of | |
357 | * a compressed extent to 128k. | |
c8b97818 CM |
358 | */ |
359 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 360 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 361 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
362 | disk_num_bytes = num_bytes; |
363 | total_in = 0; | |
364 | ret = 0; | |
db94535d | 365 | |
771ed689 CM |
366 | /* |
367 | * we do compression for mount -o compress and when the | |
368 | * inode has not been flagged as nocompress. This flag can | |
369 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 370 | */ |
6cbff00f | 371 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
372 | btrfs_test_opt(root, COMPRESS)) { |
373 | WARN_ON(pages); | |
cfbc246e | 374 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 375 | |
c8b97818 CM |
376 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
377 | total_compressed, pages, | |
378 | nr_pages, &nr_pages_ret, | |
379 | &total_in, | |
380 | &total_compressed, | |
381 | max_compressed); | |
382 | ||
383 | if (!ret) { | |
384 | unsigned long offset = total_compressed & | |
385 | (PAGE_CACHE_SIZE - 1); | |
386 | struct page *page = pages[nr_pages_ret - 1]; | |
387 | char *kaddr; | |
388 | ||
389 | /* zero the tail end of the last page, we might be | |
390 | * sending it down to disk | |
391 | */ | |
392 | if (offset) { | |
393 | kaddr = kmap_atomic(page, KM_USER0); | |
394 | memset(kaddr + offset, 0, | |
395 | PAGE_CACHE_SIZE - offset); | |
396 | kunmap_atomic(kaddr, KM_USER0); | |
397 | } | |
398 | will_compress = 1; | |
399 | } | |
400 | } | |
401 | if (start == 0) { | |
771ed689 CM |
402 | trans = btrfs_join_transaction(root, 1); |
403 | BUG_ON(!trans); | |
404 | btrfs_set_trans_block_group(trans, inode); | |
405 | ||
c8b97818 | 406 | /* lets try to make an inline extent */ |
771ed689 | 407 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 408 | /* we didn't compress the entire range, try |
771ed689 | 409 | * to make an uncompressed inline extent. |
c8b97818 CM |
410 | */ |
411 | ret = cow_file_range_inline(trans, root, inode, | |
412 | start, end, 0, NULL); | |
413 | } else { | |
771ed689 | 414 | /* try making a compressed inline extent */ |
c8b97818 CM |
415 | ret = cow_file_range_inline(trans, root, inode, |
416 | start, end, | |
417 | total_compressed, pages); | |
418 | } | |
771ed689 | 419 | btrfs_end_transaction(trans, root); |
c8b97818 | 420 | if (ret == 0) { |
771ed689 CM |
421 | /* |
422 | * inline extent creation worked, we don't need | |
423 | * to create any more async work items. Unlock | |
424 | * and free up our temp pages. | |
425 | */ | |
c8b97818 | 426 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
427 | &BTRFS_I(inode)->io_tree, |
428 | start, end, NULL, | |
429 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
430 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); | |
c8b97818 CM |
431 | ret = 0; |
432 | goto free_pages_out; | |
433 | } | |
434 | } | |
435 | ||
436 | if (will_compress) { | |
437 | /* | |
438 | * we aren't doing an inline extent round the compressed size | |
439 | * up to a block size boundary so the allocator does sane | |
440 | * things | |
441 | */ | |
442 | total_compressed = (total_compressed + blocksize - 1) & | |
443 | ~(blocksize - 1); | |
444 | ||
445 | /* | |
446 | * one last check to make sure the compression is really a | |
447 | * win, compare the page count read with the blocks on disk | |
448 | */ | |
449 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
450 | ~(PAGE_CACHE_SIZE - 1); | |
451 | if (total_compressed >= total_in) { | |
452 | will_compress = 0; | |
453 | } else { | |
454 | disk_num_bytes = total_compressed; | |
455 | num_bytes = total_in; | |
456 | } | |
457 | } | |
458 | if (!will_compress && pages) { | |
459 | /* | |
460 | * the compression code ran but failed to make things smaller, | |
461 | * free any pages it allocated and our page pointer array | |
462 | */ | |
463 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 464 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
465 | page_cache_release(pages[i]); |
466 | } | |
467 | kfree(pages); | |
468 | pages = NULL; | |
469 | total_compressed = 0; | |
470 | nr_pages_ret = 0; | |
471 | ||
472 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 473 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 474 | } |
771ed689 CM |
475 | if (will_compress) { |
476 | *num_added += 1; | |
c8b97818 | 477 | |
771ed689 CM |
478 | /* the async work queues will take care of doing actual |
479 | * allocation on disk for these compressed pages, | |
480 | * and will submit them to the elevator. | |
481 | */ | |
482 | add_async_extent(async_cow, start, num_bytes, | |
483 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 484 | |
42dc7bab | 485 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
486 | start += num_bytes; |
487 | pages = NULL; | |
488 | cond_resched(); | |
489 | goto again; | |
490 | } | |
491 | } else { | |
f03d9301 | 492 | cleanup_and_bail_uncompressed: |
771ed689 CM |
493 | /* |
494 | * No compression, but we still need to write the pages in | |
495 | * the file we've been given so far. redirty the locked | |
496 | * page if it corresponds to our extent and set things up | |
497 | * for the async work queue to run cow_file_range to do | |
498 | * the normal delalloc dance | |
499 | */ | |
500 | if (page_offset(locked_page) >= start && | |
501 | page_offset(locked_page) <= end) { | |
502 | __set_page_dirty_nobuffers(locked_page); | |
503 | /* unlocked later on in the async handlers */ | |
504 | } | |
505 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
506 | *num_added += 1; | |
507 | } | |
3b951516 | 508 | |
771ed689 CM |
509 | out: |
510 | return 0; | |
511 | ||
512 | free_pages_out: | |
513 | for (i = 0; i < nr_pages_ret; i++) { | |
514 | WARN_ON(pages[i]->mapping); | |
515 | page_cache_release(pages[i]); | |
516 | } | |
d397712b | 517 | kfree(pages); |
771ed689 CM |
518 | |
519 | goto out; | |
520 | } | |
521 | ||
522 | /* | |
523 | * phase two of compressed writeback. This is the ordered portion | |
524 | * of the code, which only gets called in the order the work was | |
525 | * queued. We walk all the async extents created by compress_file_range | |
526 | * and send them down to the disk. | |
527 | */ | |
528 | static noinline int submit_compressed_extents(struct inode *inode, | |
529 | struct async_cow *async_cow) | |
530 | { | |
531 | struct async_extent *async_extent; | |
532 | u64 alloc_hint = 0; | |
533 | struct btrfs_trans_handle *trans; | |
534 | struct btrfs_key ins; | |
535 | struct extent_map *em; | |
536 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
537 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
538 | struct extent_io_tree *io_tree; | |
539 | int ret; | |
540 | ||
541 | if (list_empty(&async_cow->extents)) | |
542 | return 0; | |
543 | ||
544 | trans = btrfs_join_transaction(root, 1); | |
545 | ||
d397712b | 546 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
547 | async_extent = list_entry(async_cow->extents.next, |
548 | struct async_extent, list); | |
549 | list_del(&async_extent->list); | |
c8b97818 | 550 | |
771ed689 CM |
551 | io_tree = &BTRFS_I(inode)->io_tree; |
552 | ||
553 | /* did the compression code fall back to uncompressed IO? */ | |
554 | if (!async_extent->pages) { | |
555 | int page_started = 0; | |
556 | unsigned long nr_written = 0; | |
557 | ||
558 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
559 | async_extent->start + |
560 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
561 | |
562 | /* allocate blocks */ | |
563 | cow_file_range(inode, async_cow->locked_page, | |
564 | async_extent->start, | |
565 | async_extent->start + | |
566 | async_extent->ram_size - 1, | |
567 | &page_started, &nr_written, 0); | |
568 | ||
569 | /* | |
570 | * if page_started, cow_file_range inserted an | |
571 | * inline extent and took care of all the unlocking | |
572 | * and IO for us. Otherwise, we need to submit | |
573 | * all those pages down to the drive. | |
574 | */ | |
575 | if (!page_started) | |
576 | extent_write_locked_range(io_tree, | |
577 | inode, async_extent->start, | |
d397712b | 578 | async_extent->start + |
771ed689 CM |
579 | async_extent->ram_size - 1, |
580 | btrfs_get_extent, | |
581 | WB_SYNC_ALL); | |
582 | kfree(async_extent); | |
583 | cond_resched(); | |
584 | continue; | |
585 | } | |
586 | ||
587 | lock_extent(io_tree, async_extent->start, | |
588 | async_extent->start + async_extent->ram_size - 1, | |
589 | GFP_NOFS); | |
c8b97818 | 590 | /* |
771ed689 CM |
591 | * here we're doing allocation and writeback of the |
592 | * compressed pages | |
c8b97818 | 593 | */ |
771ed689 CM |
594 | btrfs_drop_extent_cache(inode, async_extent->start, |
595 | async_extent->start + | |
596 | async_extent->ram_size - 1, 0); | |
597 | ||
598 | ret = btrfs_reserve_extent(trans, root, | |
599 | async_extent->compressed_size, | |
600 | async_extent->compressed_size, | |
601 | 0, alloc_hint, | |
602 | (u64)-1, &ins, 1); | |
603 | BUG_ON(ret); | |
604 | em = alloc_extent_map(GFP_NOFS); | |
605 | em->start = async_extent->start; | |
606 | em->len = async_extent->ram_size; | |
445a6944 | 607 | em->orig_start = em->start; |
c8b97818 | 608 | |
771ed689 CM |
609 | em->block_start = ins.objectid; |
610 | em->block_len = ins.offset; | |
611 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
612 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
613 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
614 | ||
d397712b | 615 | while (1) { |
890871be | 616 | write_lock(&em_tree->lock); |
771ed689 | 617 | ret = add_extent_mapping(em_tree, em); |
890871be | 618 | write_unlock(&em_tree->lock); |
771ed689 CM |
619 | if (ret != -EEXIST) { |
620 | free_extent_map(em); | |
621 | break; | |
622 | } | |
623 | btrfs_drop_extent_cache(inode, async_extent->start, | |
624 | async_extent->start + | |
625 | async_extent->ram_size - 1, 0); | |
626 | } | |
627 | ||
628 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
629 | ins.objectid, | |
630 | async_extent->ram_size, | |
631 | ins.offset, | |
632 | BTRFS_ORDERED_COMPRESSED); | |
633 | BUG_ON(ret); | |
634 | ||
635 | btrfs_end_transaction(trans, root); | |
636 | ||
637 | /* | |
638 | * clear dirty, set writeback and unlock the pages. | |
639 | */ | |
640 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
641 | &BTRFS_I(inode)->io_tree, |
642 | async_extent->start, | |
643 | async_extent->start + | |
644 | async_extent->ram_size - 1, | |
645 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
646 | EXTENT_CLEAR_UNLOCK | | |
647 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); | |
771ed689 CM |
648 | |
649 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
650 | async_extent->start, |
651 | async_extent->ram_size, | |
652 | ins.objectid, | |
653 | ins.offset, async_extent->pages, | |
654 | async_extent->nr_pages); | |
771ed689 CM |
655 | |
656 | BUG_ON(ret); | |
657 | trans = btrfs_join_transaction(root, 1); | |
658 | alloc_hint = ins.objectid + ins.offset; | |
659 | kfree(async_extent); | |
660 | cond_resched(); | |
661 | } | |
662 | ||
663 | btrfs_end_transaction(trans, root); | |
664 | return 0; | |
665 | } | |
666 | ||
667 | /* | |
668 | * when extent_io.c finds a delayed allocation range in the file, | |
669 | * the call backs end up in this code. The basic idea is to | |
670 | * allocate extents on disk for the range, and create ordered data structs | |
671 | * in ram to track those extents. | |
672 | * | |
673 | * locked_page is the page that writepage had locked already. We use | |
674 | * it to make sure we don't do extra locks or unlocks. | |
675 | * | |
676 | * *page_started is set to one if we unlock locked_page and do everything | |
677 | * required to start IO on it. It may be clean and already done with | |
678 | * IO when we return. | |
679 | */ | |
680 | static noinline int cow_file_range(struct inode *inode, | |
681 | struct page *locked_page, | |
682 | u64 start, u64 end, int *page_started, | |
683 | unsigned long *nr_written, | |
684 | int unlock) | |
685 | { | |
686 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
687 | struct btrfs_trans_handle *trans; | |
688 | u64 alloc_hint = 0; | |
689 | u64 num_bytes; | |
690 | unsigned long ram_size; | |
691 | u64 disk_num_bytes; | |
692 | u64 cur_alloc_size; | |
693 | u64 blocksize = root->sectorsize; | |
694 | u64 actual_end; | |
42dc7bab | 695 | u64 isize = i_size_read(inode); |
771ed689 CM |
696 | struct btrfs_key ins; |
697 | struct extent_map *em; | |
698 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
699 | int ret = 0; | |
700 | ||
701 | trans = btrfs_join_transaction(root, 1); | |
702 | BUG_ON(!trans); | |
703 | btrfs_set_trans_block_group(trans, inode); | |
704 | ||
42dc7bab | 705 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
706 | |
707 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
708 | num_bytes = max(blocksize, num_bytes); | |
709 | disk_num_bytes = num_bytes; | |
710 | ret = 0; | |
711 | ||
712 | if (start == 0) { | |
713 | /* lets try to make an inline extent */ | |
714 | ret = cow_file_range_inline(trans, root, inode, | |
715 | start, end, 0, NULL); | |
716 | if (ret == 0) { | |
717 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
718 | &BTRFS_I(inode)->io_tree, |
719 | start, end, NULL, | |
720 | EXTENT_CLEAR_UNLOCK_PAGE | | |
721 | EXTENT_CLEAR_UNLOCK | | |
722 | EXTENT_CLEAR_DELALLOC | | |
723 | EXTENT_CLEAR_DIRTY | | |
724 | EXTENT_SET_WRITEBACK | | |
725 | EXTENT_END_WRITEBACK); | |
771ed689 CM |
726 | *nr_written = *nr_written + |
727 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
728 | *page_started = 1; | |
729 | ret = 0; | |
730 | goto out; | |
731 | } | |
732 | } | |
733 | ||
734 | BUG_ON(disk_num_bytes > | |
735 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
736 | ||
b917b7c3 CM |
737 | |
738 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
739 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
740 | start, num_bytes); | |
741 | if (em) { | |
742 | alloc_hint = em->block_start; | |
743 | free_extent_map(em); | |
744 | } | |
745 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
746 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
747 | ||
d397712b | 748 | while (disk_num_bytes > 0) { |
a791e35e CM |
749 | unsigned long op; |
750 | ||
c8b97818 | 751 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 752 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 753 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 754 | (u64)-1, &ins, 1); |
d397712b CM |
755 | BUG_ON(ret); |
756 | ||
e6dcd2dc CM |
757 | em = alloc_extent_map(GFP_NOFS); |
758 | em->start = start; | |
445a6944 | 759 | em->orig_start = em->start; |
771ed689 CM |
760 | ram_size = ins.offset; |
761 | em->len = ins.offset; | |
c8b97818 | 762 | |
e6dcd2dc | 763 | em->block_start = ins.objectid; |
c8b97818 | 764 | em->block_len = ins.offset; |
e6dcd2dc | 765 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 766 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 767 | |
d397712b | 768 | while (1) { |
890871be | 769 | write_lock(&em_tree->lock); |
e6dcd2dc | 770 | ret = add_extent_mapping(em_tree, em); |
890871be | 771 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
772 | if (ret != -EEXIST) { |
773 | free_extent_map(em); | |
774 | break; | |
775 | } | |
776 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 777 | start + ram_size - 1, 0); |
e6dcd2dc CM |
778 | } |
779 | ||
98d20f67 | 780 | cur_alloc_size = ins.offset; |
e6dcd2dc | 781 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 782 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 783 | BUG_ON(ret); |
c8b97818 | 784 | |
17d217fe YZ |
785 | if (root->root_key.objectid == |
786 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
787 | ret = btrfs_reloc_clone_csums(inode, start, | |
788 | cur_alloc_size); | |
789 | BUG_ON(ret); | |
790 | } | |
791 | ||
d397712b | 792 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 793 | break; |
d397712b | 794 | |
c8b97818 CM |
795 | /* we're not doing compressed IO, don't unlock the first |
796 | * page (which the caller expects to stay locked), don't | |
797 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
798 | * |
799 | * Do set the Private2 bit so we know this page was properly | |
800 | * setup for writepage | |
c8b97818 | 801 | */ |
a791e35e CM |
802 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
803 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
804 | EXTENT_SET_PRIVATE2; | |
805 | ||
c8b97818 CM |
806 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
807 | start, start + ram_size - 1, | |
a791e35e | 808 | locked_page, op); |
c8b97818 | 809 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
810 | num_bytes -= cur_alloc_size; |
811 | alloc_hint = ins.objectid + ins.offset; | |
812 | start += cur_alloc_size; | |
b888db2b | 813 | } |
b888db2b | 814 | out: |
771ed689 | 815 | ret = 0; |
b888db2b | 816 | btrfs_end_transaction(trans, root); |
c8b97818 | 817 | |
be20aa9d | 818 | return ret; |
771ed689 | 819 | } |
c8b97818 | 820 | |
771ed689 CM |
821 | /* |
822 | * work queue call back to started compression on a file and pages | |
823 | */ | |
824 | static noinline void async_cow_start(struct btrfs_work *work) | |
825 | { | |
826 | struct async_cow *async_cow; | |
827 | int num_added = 0; | |
828 | async_cow = container_of(work, struct async_cow, work); | |
829 | ||
830 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
831 | async_cow->start, async_cow->end, async_cow, | |
832 | &num_added); | |
833 | if (num_added == 0) | |
834 | async_cow->inode = NULL; | |
835 | } | |
836 | ||
837 | /* | |
838 | * work queue call back to submit previously compressed pages | |
839 | */ | |
840 | static noinline void async_cow_submit(struct btrfs_work *work) | |
841 | { | |
842 | struct async_cow *async_cow; | |
843 | struct btrfs_root *root; | |
844 | unsigned long nr_pages; | |
845 | ||
846 | async_cow = container_of(work, struct async_cow, work); | |
847 | ||
848 | root = async_cow->root; | |
849 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
850 | PAGE_CACHE_SHIFT; | |
851 | ||
852 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
853 | ||
854 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
855 | 5 * 1042 * 1024 && | |
856 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
857 | wake_up(&root->fs_info->async_submit_wait); | |
858 | ||
d397712b | 859 | if (async_cow->inode) |
771ed689 | 860 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 861 | } |
c8b97818 | 862 | |
771ed689 CM |
863 | static noinline void async_cow_free(struct btrfs_work *work) |
864 | { | |
865 | struct async_cow *async_cow; | |
866 | async_cow = container_of(work, struct async_cow, work); | |
867 | kfree(async_cow); | |
868 | } | |
869 | ||
870 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
871 | u64 start, u64 end, int *page_started, | |
872 | unsigned long *nr_written) | |
873 | { | |
874 | struct async_cow *async_cow; | |
875 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
876 | unsigned long nr_pages; | |
877 | u64 cur_end; | |
878 | int limit = 10 * 1024 * 1042; | |
879 | ||
771ed689 | 880 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | |
2c64c53d | 881 | EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS); |
d397712b | 882 | while (start < end) { |
771ed689 CM |
883 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
884 | async_cow->inode = inode; | |
885 | async_cow->root = root; | |
886 | async_cow->locked_page = locked_page; | |
887 | async_cow->start = start; | |
888 | ||
6cbff00f | 889 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
890 | cur_end = end; |
891 | else | |
892 | cur_end = min(end, start + 512 * 1024 - 1); | |
893 | ||
894 | async_cow->end = cur_end; | |
895 | INIT_LIST_HEAD(&async_cow->extents); | |
896 | ||
897 | async_cow->work.func = async_cow_start; | |
898 | async_cow->work.ordered_func = async_cow_submit; | |
899 | async_cow->work.ordered_free = async_cow_free; | |
900 | async_cow->work.flags = 0; | |
901 | ||
771ed689 CM |
902 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
903 | PAGE_CACHE_SHIFT; | |
904 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
905 | ||
906 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
907 | &async_cow->work); | |
908 | ||
909 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
910 | wait_event(root->fs_info->async_submit_wait, | |
911 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
912 | limit)); | |
913 | } | |
914 | ||
d397712b | 915 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
916 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
917 | wait_event(root->fs_info->async_submit_wait, | |
918 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
919 | 0)); | |
920 | } | |
921 | ||
922 | *nr_written += nr_pages; | |
923 | start = cur_end + 1; | |
924 | } | |
925 | *page_started = 1; | |
926 | return 0; | |
be20aa9d CM |
927 | } |
928 | ||
d397712b | 929 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
930 | u64 bytenr, u64 num_bytes) |
931 | { | |
932 | int ret; | |
933 | struct btrfs_ordered_sum *sums; | |
934 | LIST_HEAD(list); | |
935 | ||
07d400a6 YZ |
936 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
937 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
938 | if (ret == 0 && list_empty(&list)) |
939 | return 0; | |
940 | ||
941 | while (!list_empty(&list)) { | |
942 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
943 | list_del(&sums->list); | |
944 | kfree(sums); | |
945 | } | |
946 | return 1; | |
947 | } | |
948 | ||
d352ac68 CM |
949 | /* |
950 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
951 | * of the extents that exist in the file, and COWs the file as required. | |
952 | * | |
953 | * If no cow copies or snapshots exist, we write directly to the existing | |
954 | * blocks on disk | |
955 | */ | |
7f366cfe CM |
956 | static noinline int run_delalloc_nocow(struct inode *inode, |
957 | struct page *locked_page, | |
771ed689 CM |
958 | u64 start, u64 end, int *page_started, int force, |
959 | unsigned long *nr_written) | |
be20aa9d | 960 | { |
be20aa9d | 961 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 962 | struct btrfs_trans_handle *trans; |
be20aa9d | 963 | struct extent_buffer *leaf; |
be20aa9d | 964 | struct btrfs_path *path; |
80ff3856 | 965 | struct btrfs_file_extent_item *fi; |
be20aa9d | 966 | struct btrfs_key found_key; |
80ff3856 YZ |
967 | u64 cow_start; |
968 | u64 cur_offset; | |
969 | u64 extent_end; | |
5d4f98a2 | 970 | u64 extent_offset; |
80ff3856 YZ |
971 | u64 disk_bytenr; |
972 | u64 num_bytes; | |
973 | int extent_type; | |
974 | int ret; | |
d899e052 | 975 | int type; |
80ff3856 YZ |
976 | int nocow; |
977 | int check_prev = 1; | |
be20aa9d CM |
978 | |
979 | path = btrfs_alloc_path(); | |
980 | BUG_ON(!path); | |
7ea394f1 YZ |
981 | trans = btrfs_join_transaction(root, 1); |
982 | BUG_ON(!trans); | |
be20aa9d | 983 | |
80ff3856 YZ |
984 | cow_start = (u64)-1; |
985 | cur_offset = start; | |
986 | while (1) { | |
987 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
988 | cur_offset, 0); | |
989 | BUG_ON(ret < 0); | |
990 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
991 | leaf = path->nodes[0]; | |
992 | btrfs_item_key_to_cpu(leaf, &found_key, | |
993 | path->slots[0] - 1); | |
994 | if (found_key.objectid == inode->i_ino && | |
995 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
996 | path->slots[0]--; | |
997 | } | |
998 | check_prev = 0; | |
999 | next_slot: | |
1000 | leaf = path->nodes[0]; | |
1001 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1002 | ret = btrfs_next_leaf(root, path); | |
1003 | if (ret < 0) | |
1004 | BUG_ON(1); | |
1005 | if (ret > 0) | |
1006 | break; | |
1007 | leaf = path->nodes[0]; | |
1008 | } | |
be20aa9d | 1009 | |
80ff3856 YZ |
1010 | nocow = 0; |
1011 | disk_bytenr = 0; | |
17d217fe | 1012 | num_bytes = 0; |
80ff3856 YZ |
1013 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1014 | ||
1015 | if (found_key.objectid > inode->i_ino || | |
1016 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1017 | found_key.offset > end) | |
1018 | break; | |
1019 | ||
1020 | if (found_key.offset > cur_offset) { | |
1021 | extent_end = found_key.offset; | |
1022 | goto out_check; | |
1023 | } | |
1024 | ||
1025 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1026 | struct btrfs_file_extent_item); | |
1027 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1028 | ||
d899e052 YZ |
1029 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1030 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1031 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1032 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1033 | extent_end = found_key.offset + |
1034 | btrfs_file_extent_num_bytes(leaf, fi); | |
1035 | if (extent_end <= start) { | |
1036 | path->slots[0]++; | |
1037 | goto next_slot; | |
1038 | } | |
17d217fe YZ |
1039 | if (disk_bytenr == 0) |
1040 | goto out_check; | |
80ff3856 YZ |
1041 | if (btrfs_file_extent_compression(leaf, fi) || |
1042 | btrfs_file_extent_encryption(leaf, fi) || | |
1043 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1044 | goto out_check; | |
d899e052 YZ |
1045 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1046 | goto out_check; | |
d2fb3437 | 1047 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1048 | goto out_check; |
17d217fe | 1049 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1050 | found_key.offset - |
1051 | extent_offset, disk_bytenr)) | |
17d217fe | 1052 | goto out_check; |
5d4f98a2 | 1053 | disk_bytenr += extent_offset; |
17d217fe YZ |
1054 | disk_bytenr += cur_offset - found_key.offset; |
1055 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1056 | /* | |
1057 | * force cow if csum exists in the range. | |
1058 | * this ensure that csum for a given extent are | |
1059 | * either valid or do not exist. | |
1060 | */ | |
1061 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1062 | goto out_check; | |
80ff3856 YZ |
1063 | nocow = 1; |
1064 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1065 | extent_end = found_key.offset + | |
1066 | btrfs_file_extent_inline_len(leaf, fi); | |
1067 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1068 | } else { | |
1069 | BUG_ON(1); | |
1070 | } | |
1071 | out_check: | |
1072 | if (extent_end <= start) { | |
1073 | path->slots[0]++; | |
1074 | goto next_slot; | |
1075 | } | |
1076 | if (!nocow) { | |
1077 | if (cow_start == (u64)-1) | |
1078 | cow_start = cur_offset; | |
1079 | cur_offset = extent_end; | |
1080 | if (cur_offset > end) | |
1081 | break; | |
1082 | path->slots[0]++; | |
1083 | goto next_slot; | |
7ea394f1 YZ |
1084 | } |
1085 | ||
1086 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1087 | if (cow_start != (u64)-1) { |
1088 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1089 | found_key.offset - 1, page_started, |
1090 | nr_written, 1); | |
80ff3856 YZ |
1091 | BUG_ON(ret); |
1092 | cow_start = (u64)-1; | |
7ea394f1 | 1093 | } |
80ff3856 | 1094 | |
d899e052 YZ |
1095 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1096 | struct extent_map *em; | |
1097 | struct extent_map_tree *em_tree; | |
1098 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1099 | em = alloc_extent_map(GFP_NOFS); | |
1100 | em->start = cur_offset; | |
445a6944 | 1101 | em->orig_start = em->start; |
d899e052 YZ |
1102 | em->len = num_bytes; |
1103 | em->block_len = num_bytes; | |
1104 | em->block_start = disk_bytenr; | |
1105 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1106 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1107 | while (1) { | |
890871be | 1108 | write_lock(&em_tree->lock); |
d899e052 | 1109 | ret = add_extent_mapping(em_tree, em); |
890871be | 1110 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1111 | if (ret != -EEXIST) { |
1112 | free_extent_map(em); | |
1113 | break; | |
1114 | } | |
1115 | btrfs_drop_extent_cache(inode, em->start, | |
1116 | em->start + em->len - 1, 0); | |
1117 | } | |
1118 | type = BTRFS_ORDERED_PREALLOC; | |
1119 | } else { | |
1120 | type = BTRFS_ORDERED_NOCOW; | |
1121 | } | |
80ff3856 YZ |
1122 | |
1123 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1124 | num_bytes, num_bytes, type); |
1125 | BUG_ON(ret); | |
771ed689 | 1126 | |
d899e052 | 1127 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1128 | cur_offset, cur_offset + num_bytes - 1, |
1129 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1130 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1131 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1132 | cur_offset = extent_end; |
1133 | if (cur_offset > end) | |
1134 | break; | |
be20aa9d | 1135 | } |
80ff3856 YZ |
1136 | btrfs_release_path(root, path); |
1137 | ||
1138 | if (cur_offset <= end && cow_start == (u64)-1) | |
1139 | cow_start = cur_offset; | |
1140 | if (cow_start != (u64)-1) { | |
1141 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1142 | page_started, nr_written, 1); |
80ff3856 YZ |
1143 | BUG_ON(ret); |
1144 | } | |
1145 | ||
1146 | ret = btrfs_end_transaction(trans, root); | |
1147 | BUG_ON(ret); | |
7ea394f1 | 1148 | btrfs_free_path(path); |
80ff3856 | 1149 | return 0; |
be20aa9d CM |
1150 | } |
1151 | ||
d352ac68 CM |
1152 | /* |
1153 | * extent_io.c call back to do delayed allocation processing | |
1154 | */ | |
c8b97818 | 1155 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1156 | u64 start, u64 end, int *page_started, |
1157 | unsigned long *nr_written) | |
be20aa9d | 1158 | { |
be20aa9d | 1159 | int ret; |
7f366cfe | 1160 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1161 | |
6cbff00f | 1162 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1163 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1164 | page_started, 1, nr_written); |
6cbff00f | 1165 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1166 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1167 | page_started, 0, nr_written); |
7f366cfe CM |
1168 | else if (!btrfs_test_opt(root, COMPRESS)) |
1169 | ret = cow_file_range(inode, locked_page, start, end, | |
1170 | page_started, nr_written, 1); | |
be20aa9d | 1171 | else |
771ed689 | 1172 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1173 | page_started, nr_written); |
b888db2b CM |
1174 | return ret; |
1175 | } | |
1176 | ||
9ed74f2d JB |
1177 | static int btrfs_split_extent_hook(struct inode *inode, |
1178 | struct extent_state *orig, u64 split) | |
1179 | { | |
1180 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1181 | u64 size; | |
1182 | ||
1183 | if (!(orig->state & EXTENT_DELALLOC)) | |
1184 | return 0; | |
1185 | ||
1186 | size = orig->end - orig->start + 1; | |
1187 | if (size > root->fs_info->max_extent) { | |
1188 | u64 num_extents; | |
1189 | u64 new_size; | |
1190 | ||
1191 | new_size = orig->end - split + 1; | |
1192 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1193 | root->fs_info->max_extent); | |
1194 | ||
1195 | /* | |
1196 | * if we break a large extent up then leave delalloc_extents be, | |
1197 | * since we've already accounted for the large extent. | |
1198 | */ | |
1199 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1200 | root->fs_info->max_extent) < num_extents) | |
1201 | return 0; | |
1202 | } | |
1203 | ||
1204 | BTRFS_I(inode)->delalloc_extents++; | |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1211 | * extents so we can keep track of new extents that are just merged onto old | |
1212 | * extents, such as when we are doing sequential writes, so we can properly | |
1213 | * account for the metadata space we'll need. | |
1214 | */ | |
1215 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1216 | struct extent_state *new, | |
1217 | struct extent_state *other) | |
1218 | { | |
1219 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1220 | u64 new_size, old_size; | |
1221 | u64 num_extents; | |
1222 | ||
1223 | /* not delalloc, ignore it */ | |
1224 | if (!(other->state & EXTENT_DELALLOC)) | |
1225 | return 0; | |
1226 | ||
1227 | old_size = other->end - other->start + 1; | |
1228 | if (new->start < other->start) | |
1229 | new_size = other->end - new->start + 1; | |
1230 | else | |
1231 | new_size = new->end - other->start + 1; | |
1232 | ||
1233 | /* we're not bigger than the max, unreserve the space and go */ | |
1234 | if (new_size <= root->fs_info->max_extent) { | |
1235 | BTRFS_I(inode)->delalloc_extents--; | |
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * If we grew by another max_extent, just return, we want to keep that | |
1241 | * reserved amount. | |
1242 | */ | |
1243 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1244 | root->fs_info->max_extent); | |
1245 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1246 | root->fs_info->max_extent) > num_extents) | |
1247 | return 0; | |
1248 | ||
1249 | BTRFS_I(inode)->delalloc_extents--; | |
1250 | ||
1251 | return 0; | |
1252 | } | |
1253 | ||
d352ac68 CM |
1254 | /* |
1255 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1256 | * bytes in this file, and to maintain the list of inodes that | |
1257 | * have pending delalloc work to be done. | |
1258 | */ | |
b2950863 | 1259 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1260 | unsigned long old, unsigned long bits) |
291d673e | 1261 | { |
9ed74f2d | 1262 | |
75eff68e CM |
1263 | /* |
1264 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1265 | * but in this case, we are only testeing for the DELALLOC | |
1266 | * bit, which is only set or cleared with irqs on | |
1267 | */ | |
b0c68f8b | 1268 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1269 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d JB |
1270 | |
1271 | BTRFS_I(inode)->delalloc_extents++; | |
6a63209f | 1272 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1273 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1274 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1275 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1276 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1277 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1278 | &root->fs_info->delalloc_inodes); | |
1279 | } | |
75eff68e | 1280 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1281 | } |
1282 | return 0; | |
1283 | } | |
1284 | ||
d352ac68 CM |
1285 | /* |
1286 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1287 | */ | |
9ed74f2d JB |
1288 | static int btrfs_clear_bit_hook(struct inode *inode, |
1289 | struct extent_state *state, unsigned long bits) | |
291d673e | 1290 | { |
75eff68e CM |
1291 | /* |
1292 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1293 | * but in this case, we are only testeing for the DELALLOC | |
1294 | * bit, which is only set or cleared with irqs on | |
1295 | */ | |
9ed74f2d | 1296 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1297 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1298 | |
9ed74f2d JB |
1299 | BTRFS_I(inode)->delalloc_extents--; |
1300 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1301 | ||
75eff68e | 1302 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1303 | if (state->end - state->start + 1 > |
1304 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1305 | printk(KERN_INFO "btrfs warning: delalloc account " |
1306 | "%llu %llu\n", | |
9ed74f2d JB |
1307 | (unsigned long long) |
1308 | state->end - state->start + 1, | |
d397712b CM |
1309 | (unsigned long long) |
1310 | root->fs_info->delalloc_bytes); | |
6a63209f | 1311 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1312 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1313 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1314 | } else { |
6a63209f | 1315 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1316 | state->end - |
1317 | state->start + 1); | |
1318 | root->fs_info->delalloc_bytes -= state->end - | |
1319 | state->start + 1; | |
1320 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1321 | state->start + 1; | |
b0c68f8b | 1322 | } |
ea8c2819 CM |
1323 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1324 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1325 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1326 | } | |
75eff68e | 1327 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1328 | } |
1329 | return 0; | |
1330 | } | |
1331 | ||
d352ac68 CM |
1332 | /* |
1333 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1334 | * we don't create bios that span stripes or chunks | |
1335 | */ | |
239b14b3 | 1336 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1337 | size_t size, struct bio *bio, |
1338 | unsigned long bio_flags) | |
239b14b3 CM |
1339 | { |
1340 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1341 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1342 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1343 | u64 length = 0; |
1344 | u64 map_length; | |
239b14b3 CM |
1345 | int ret; |
1346 | ||
771ed689 CM |
1347 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1348 | return 0; | |
1349 | ||
f2d8d74d | 1350 | length = bio->bi_size; |
239b14b3 CM |
1351 | map_tree = &root->fs_info->mapping_tree; |
1352 | map_length = length; | |
cea9e445 | 1353 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1354 | &map_length, NULL, 0); |
cea9e445 | 1355 | |
d397712b | 1356 | if (map_length < length + size) |
239b14b3 | 1357 | return 1; |
239b14b3 CM |
1358 | return 0; |
1359 | } | |
1360 | ||
d352ac68 CM |
1361 | /* |
1362 | * in order to insert checksums into the metadata in large chunks, | |
1363 | * we wait until bio submission time. All the pages in the bio are | |
1364 | * checksummed and sums are attached onto the ordered extent record. | |
1365 | * | |
1366 | * At IO completion time the cums attached on the ordered extent record | |
1367 | * are inserted into the btree | |
1368 | */ | |
d397712b CM |
1369 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1370 | struct bio *bio, int mirror_num, | |
1371 | unsigned long bio_flags) | |
065631f6 | 1372 | { |
065631f6 | 1373 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1374 | int ret = 0; |
e015640f | 1375 | |
d20f7043 | 1376 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1377 | BUG_ON(ret); |
4a69a410 CM |
1378 | return 0; |
1379 | } | |
e015640f | 1380 | |
4a69a410 CM |
1381 | /* |
1382 | * in order to insert checksums into the metadata in large chunks, | |
1383 | * we wait until bio submission time. All the pages in the bio are | |
1384 | * checksummed and sums are attached onto the ordered extent record. | |
1385 | * | |
1386 | * At IO completion time the cums attached on the ordered extent record | |
1387 | * are inserted into the btree | |
1388 | */ | |
b2950863 | 1389 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1390 | int mirror_num, unsigned long bio_flags) |
1391 | { | |
1392 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1393 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1394 | } |
1395 | ||
d352ac68 | 1396 | /* |
cad321ad CM |
1397 | * extent_io.c submission hook. This does the right thing for csum calculation |
1398 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1399 | */ |
b2950863 | 1400 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1401 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1402 | { |
1403 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1404 | int ret = 0; | |
19b9bdb0 | 1405 | int skip_sum; |
44b8bd7e | 1406 | |
6cbff00f | 1407 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1408 | |
e6dcd2dc CM |
1409 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1410 | BUG_ON(ret); | |
065631f6 | 1411 | |
4d1b5fb4 | 1412 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1413 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1414 | return btrfs_submit_compressed_read(inode, bio, |
1415 | mirror_num, bio_flags); | |
d20f7043 CM |
1416 | } else if (!skip_sum) |
1417 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1418 | goto mapit; |
19b9bdb0 | 1419 | } else if (!skip_sum) { |
17d217fe YZ |
1420 | /* csum items have already been cloned */ |
1421 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1422 | goto mapit; | |
19b9bdb0 CM |
1423 | /* we're doing a write, do the async checksumming */ |
1424 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1425 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1426 | bio_flags, __btrfs_submit_bio_start, |
1427 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1428 | } |
1429 | ||
0b86a832 | 1430 | mapit: |
8b712842 | 1431 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1432 | } |
6885f308 | 1433 | |
d352ac68 CM |
1434 | /* |
1435 | * given a list of ordered sums record them in the inode. This happens | |
1436 | * at IO completion time based on sums calculated at bio submission time. | |
1437 | */ | |
ba1da2f4 | 1438 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1439 | struct inode *inode, u64 file_offset, |
1440 | struct list_head *list) | |
1441 | { | |
e6dcd2dc CM |
1442 | struct btrfs_ordered_sum *sum; |
1443 | ||
1444 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1445 | |
1446 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1447 | btrfs_csum_file_blocks(trans, |
1448 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1449 | } |
1450 | return 0; | |
1451 | } | |
1452 | ||
ea8c2819 CM |
1453 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1454 | { | |
d397712b | 1455 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1456 | WARN_ON(1); |
ea8c2819 CM |
1457 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1458 | GFP_NOFS); | |
1459 | } | |
1460 | ||
d352ac68 | 1461 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1462 | struct btrfs_writepage_fixup { |
1463 | struct page *page; | |
1464 | struct btrfs_work work; | |
1465 | }; | |
1466 | ||
b2950863 | 1467 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1468 | { |
1469 | struct btrfs_writepage_fixup *fixup; | |
1470 | struct btrfs_ordered_extent *ordered; | |
1471 | struct page *page; | |
1472 | struct inode *inode; | |
1473 | u64 page_start; | |
1474 | u64 page_end; | |
1475 | ||
1476 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1477 | page = fixup->page; | |
4a096752 | 1478 | again: |
247e743c CM |
1479 | lock_page(page); |
1480 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1481 | ClearPageChecked(page); | |
1482 | goto out_page; | |
1483 | } | |
1484 | ||
1485 | inode = page->mapping->host; | |
1486 | page_start = page_offset(page); | |
1487 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1488 | ||
1489 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1490 | |
1491 | /* already ordered? We're done */ | |
8b62b72b | 1492 | if (PagePrivate2(page)) |
247e743c | 1493 | goto out; |
4a096752 CM |
1494 | |
1495 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1496 | if (ordered) { | |
1497 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1498 | page_end, GFP_NOFS); | |
1499 | unlock_page(page); | |
1500 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1501 | goto again; | |
1502 | } | |
247e743c | 1503 | |
ea8c2819 | 1504 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1505 | ClearPageChecked(page); |
1506 | out: | |
1507 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1508 | out_page: | |
1509 | unlock_page(page); | |
1510 | page_cache_release(page); | |
1511 | } | |
1512 | ||
1513 | /* | |
1514 | * There are a few paths in the higher layers of the kernel that directly | |
1515 | * set the page dirty bit without asking the filesystem if it is a | |
1516 | * good idea. This causes problems because we want to make sure COW | |
1517 | * properly happens and the data=ordered rules are followed. | |
1518 | * | |
c8b97818 | 1519 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1520 | * hasn't been properly setup for IO. We kick off an async process |
1521 | * to fix it up. The async helper will wait for ordered extents, set | |
1522 | * the delalloc bit and make it safe to write the page. | |
1523 | */ | |
b2950863 | 1524 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1525 | { |
1526 | struct inode *inode = page->mapping->host; | |
1527 | struct btrfs_writepage_fixup *fixup; | |
1528 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1529 | |
8b62b72b CM |
1530 | /* this page is properly in the ordered list */ |
1531 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1532 | return 0; |
1533 | ||
1534 | if (PageChecked(page)) | |
1535 | return -EAGAIN; | |
1536 | ||
1537 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1538 | if (!fixup) | |
1539 | return -EAGAIN; | |
f421950f | 1540 | |
247e743c CM |
1541 | SetPageChecked(page); |
1542 | page_cache_get(page); | |
1543 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1544 | fixup->page = page; | |
1545 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1546 | return -EAGAIN; | |
1547 | } | |
1548 | ||
d899e052 YZ |
1549 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1550 | struct inode *inode, u64 file_pos, | |
1551 | u64 disk_bytenr, u64 disk_num_bytes, | |
1552 | u64 num_bytes, u64 ram_bytes, | |
e980b50c | 1553 | u64 locked_end, |
d899e052 YZ |
1554 | u8 compression, u8 encryption, |
1555 | u16 other_encoding, int extent_type) | |
1556 | { | |
1557 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1558 | struct btrfs_file_extent_item *fi; | |
1559 | struct btrfs_path *path; | |
1560 | struct extent_buffer *leaf; | |
1561 | struct btrfs_key ins; | |
1562 | u64 hint; | |
1563 | int ret; | |
1564 | ||
1565 | path = btrfs_alloc_path(); | |
1566 | BUG_ON(!path); | |
1567 | ||
b9473439 | 1568 | path->leave_spinning = 1; |
a1ed835e CM |
1569 | |
1570 | /* | |
1571 | * we may be replacing one extent in the tree with another. | |
1572 | * The new extent is pinned in the extent map, and we don't want | |
1573 | * to drop it from the cache until it is completely in the btree. | |
1574 | * | |
1575 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1576 | * the caller is expected to unpin it and allow it to be merged | |
1577 | * with the others. | |
1578 | */ | |
d899e052 | 1579 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
e980b50c | 1580 | file_pos + num_bytes, locked_end, |
a1ed835e | 1581 | file_pos, &hint, 0); |
d899e052 YZ |
1582 | BUG_ON(ret); |
1583 | ||
1584 | ins.objectid = inode->i_ino; | |
1585 | ins.offset = file_pos; | |
1586 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1587 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1588 | BUG_ON(ret); | |
1589 | leaf = path->nodes[0]; | |
1590 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1591 | struct btrfs_file_extent_item); | |
1592 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1593 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1594 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1595 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1596 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1597 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1598 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1599 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1600 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1601 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1602 | |
1603 | btrfs_unlock_up_safe(path, 1); | |
1604 | btrfs_set_lock_blocking(leaf); | |
1605 | ||
d899e052 YZ |
1606 | btrfs_mark_buffer_dirty(leaf); |
1607 | ||
1608 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1609 | |
1610 | ins.objectid = disk_bytenr; | |
1611 | ins.offset = disk_num_bytes; | |
1612 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1613 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1614 | root->root_key.objectid, | |
1615 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1616 | BUG_ON(ret); |
d899e052 | 1617 | btrfs_free_path(path); |
b9473439 | 1618 | |
d899e052 YZ |
1619 | return 0; |
1620 | } | |
1621 | ||
5d13a98f CM |
1622 | /* |
1623 | * helper function for btrfs_finish_ordered_io, this | |
1624 | * just reads in some of the csum leaves to prime them into ram | |
1625 | * before we start the transaction. It limits the amount of btree | |
1626 | * reads required while inside the transaction. | |
1627 | */ | |
1628 | static noinline void reada_csum(struct btrfs_root *root, | |
1629 | struct btrfs_path *path, | |
1630 | struct btrfs_ordered_extent *ordered_extent) | |
1631 | { | |
1632 | struct btrfs_ordered_sum *sum; | |
1633 | u64 bytenr; | |
1634 | ||
1635 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1636 | list); | |
1637 | bytenr = sum->sums[0].bytenr; | |
1638 | ||
1639 | /* | |
1640 | * we don't care about the results, the point of this search is | |
1641 | * just to get the btree leaves into ram | |
1642 | */ | |
1643 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1644 | } | |
1645 | ||
d352ac68 CM |
1646 | /* as ordered data IO finishes, this gets called so we can finish |
1647 | * an ordered extent if the range of bytes in the file it covers are | |
1648 | * fully written. | |
1649 | */ | |
211f90e6 | 1650 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1651 | { |
e6dcd2dc CM |
1652 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1653 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1654 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1655 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1656 | struct btrfs_path *path; |
d899e052 | 1657 | int compressed = 0; |
e6dcd2dc CM |
1658 | int ret; |
1659 | ||
1660 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1661 | if (!ret) |
e6dcd2dc | 1662 | return 0; |
e6dcd2dc | 1663 | |
b7ec40d7 CM |
1664 | /* |
1665 | * before we join the transaction, try to do some of our IO. | |
1666 | * This will limit the amount of IO that we have to do with | |
1667 | * the transaction running. We're unlikely to need to do any | |
1668 | * IO if the file extents are new, the disk_i_size checks | |
1669 | * covers the most common case. | |
1670 | */ | |
1671 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1672 | path = btrfs_alloc_path(); | |
1673 | if (path) { | |
1674 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1675 | inode->i_ino, | |
1676 | start, 0); | |
5d13a98f CM |
1677 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1678 | start); | |
1679 | if (!list_empty(&ordered_extent->list)) { | |
1680 | btrfs_release_path(root, path); | |
1681 | reada_csum(root, path, ordered_extent); | |
1682 | } | |
b7ec40d7 CM |
1683 | btrfs_free_path(path); |
1684 | } | |
1685 | } | |
1686 | ||
f9295749 | 1687 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc | 1688 | |
5d13a98f CM |
1689 | if (!ordered_extent) |
1690 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1691 | BUG_ON(!ordered_extent); |
7ea394f1 YZ |
1692 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1693 | goto nocow; | |
e6dcd2dc CM |
1694 | |
1695 | lock_extent(io_tree, ordered_extent->file_offset, | |
1696 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1697 | GFP_NOFS); | |
1698 | ||
c8b97818 | 1699 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1700 | compressed = 1; |
1701 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1702 | BUG_ON(compressed); | |
1703 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1704 | ordered_extent->file_offset, | |
1705 | ordered_extent->file_offset + | |
1706 | ordered_extent->len); | |
1707 | BUG_ON(ret); | |
1708 | } else { | |
1709 | ret = insert_reserved_file_extent(trans, inode, | |
1710 | ordered_extent->file_offset, | |
1711 | ordered_extent->start, | |
1712 | ordered_extent->disk_len, | |
1713 | ordered_extent->len, | |
1714 | ordered_extent->len, | |
e980b50c CM |
1715 | ordered_extent->file_offset + |
1716 | ordered_extent->len, | |
d899e052 YZ |
1717 | compressed, 0, 0, |
1718 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1719 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1720 | ordered_extent->file_offset, | |
1721 | ordered_extent->len); | |
d899e052 YZ |
1722 | BUG_ON(ret); |
1723 | } | |
e6dcd2dc CM |
1724 | unlock_extent(io_tree, ordered_extent->file_offset, |
1725 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1726 | GFP_NOFS); | |
7ea394f1 | 1727 | nocow: |
e6dcd2dc CM |
1728 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1729 | &ordered_extent->list); | |
1730 | ||
34353029 | 1731 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1732 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1733 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1734 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1735 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1736 | |
e6dcd2dc CM |
1737 | /* once for us */ |
1738 | btrfs_put_ordered_extent(ordered_extent); | |
1739 | /* once for the tree */ | |
1740 | btrfs_put_ordered_extent(ordered_extent); | |
1741 | ||
e6dcd2dc CM |
1742 | btrfs_end_transaction(trans, root); |
1743 | return 0; | |
1744 | } | |
1745 | ||
b2950863 | 1746 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1747 | struct extent_state *state, int uptodate) |
1748 | { | |
8b62b72b | 1749 | ClearPagePrivate2(page); |
211f90e6 CM |
1750 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1751 | } | |
1752 | ||
d352ac68 CM |
1753 | /* |
1754 | * When IO fails, either with EIO or csum verification fails, we | |
1755 | * try other mirrors that might have a good copy of the data. This | |
1756 | * io_failure_record is used to record state as we go through all the | |
1757 | * mirrors. If another mirror has good data, the page is set up to date | |
1758 | * and things continue. If a good mirror can't be found, the original | |
1759 | * bio end_io callback is called to indicate things have failed. | |
1760 | */ | |
7e38326f CM |
1761 | struct io_failure_record { |
1762 | struct page *page; | |
1763 | u64 start; | |
1764 | u64 len; | |
1765 | u64 logical; | |
d20f7043 | 1766 | unsigned long bio_flags; |
7e38326f CM |
1767 | int last_mirror; |
1768 | }; | |
1769 | ||
b2950863 | 1770 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1771 | struct page *page, u64 start, u64 end, |
1772 | struct extent_state *state) | |
7e38326f CM |
1773 | { |
1774 | struct io_failure_record *failrec = NULL; | |
1775 | u64 private; | |
1776 | struct extent_map *em; | |
1777 | struct inode *inode = page->mapping->host; | |
1778 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1779 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1780 | struct bio *bio; |
1781 | int num_copies; | |
1782 | int ret; | |
1259ab75 | 1783 | int rw; |
7e38326f CM |
1784 | u64 logical; |
1785 | ||
1786 | ret = get_state_private(failure_tree, start, &private); | |
1787 | if (ret) { | |
7e38326f CM |
1788 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1789 | if (!failrec) | |
1790 | return -ENOMEM; | |
1791 | failrec->start = start; | |
1792 | failrec->len = end - start + 1; | |
1793 | failrec->last_mirror = 0; | |
d20f7043 | 1794 | failrec->bio_flags = 0; |
7e38326f | 1795 | |
890871be | 1796 | read_lock(&em_tree->lock); |
3b951516 CM |
1797 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1798 | if (em->start > start || em->start + em->len < start) { | |
1799 | free_extent_map(em); | |
1800 | em = NULL; | |
1801 | } | |
890871be | 1802 | read_unlock(&em_tree->lock); |
7e38326f CM |
1803 | |
1804 | if (!em || IS_ERR(em)) { | |
1805 | kfree(failrec); | |
1806 | return -EIO; | |
1807 | } | |
1808 | logical = start - em->start; | |
1809 | logical = em->block_start + logical; | |
d20f7043 CM |
1810 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1811 | logical = em->block_start; | |
1812 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1813 | } | |
7e38326f CM |
1814 | failrec->logical = logical; |
1815 | free_extent_map(em); | |
1816 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1817 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1818 | set_state_private(failure_tree, start, |
1819 | (u64)(unsigned long)failrec); | |
7e38326f | 1820 | } else { |
587f7704 | 1821 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1822 | } |
1823 | num_copies = btrfs_num_copies( | |
1824 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1825 | failrec->logical, failrec->len); | |
1826 | failrec->last_mirror++; | |
1827 | if (!state) { | |
cad321ad | 1828 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1829 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1830 | failrec->start, | |
1831 | EXTENT_LOCKED); | |
1832 | if (state && state->start != failrec->start) | |
1833 | state = NULL; | |
cad321ad | 1834 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1835 | } |
1836 | if (!state || failrec->last_mirror > num_copies) { | |
1837 | set_state_private(failure_tree, failrec->start, 0); | |
1838 | clear_extent_bits(failure_tree, failrec->start, | |
1839 | failrec->start + failrec->len - 1, | |
1840 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1841 | kfree(failrec); | |
1842 | return -EIO; | |
1843 | } | |
1844 | bio = bio_alloc(GFP_NOFS, 1); | |
1845 | bio->bi_private = state; | |
1846 | bio->bi_end_io = failed_bio->bi_end_io; | |
1847 | bio->bi_sector = failrec->logical >> 9; | |
1848 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1849 | bio->bi_size = 0; |
d20f7043 | 1850 | |
7e38326f | 1851 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1852 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1853 | rw = WRITE; | |
1854 | else | |
1855 | rw = READ; | |
1856 | ||
1857 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1858 | failrec->last_mirror, |
d20f7043 | 1859 | failrec->bio_flags); |
1259ab75 CM |
1860 | return 0; |
1861 | } | |
1862 | ||
d352ac68 CM |
1863 | /* |
1864 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1865 | * to see if we need to process or clean up an io_failure_record | |
1866 | */ | |
b2950863 | 1867 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1868 | { |
1869 | u64 private; | |
1870 | u64 private_failure; | |
1871 | struct io_failure_record *failure; | |
1872 | int ret; | |
1873 | ||
1874 | private = 0; | |
1875 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1876 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1877 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1878 | start, &private_failure); | |
1879 | if (ret == 0) { | |
1880 | failure = (struct io_failure_record *)(unsigned long) | |
1881 | private_failure; | |
1882 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1883 | failure->start, 0); | |
1884 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1885 | failure->start, | |
1886 | failure->start + failure->len - 1, | |
1887 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1888 | GFP_NOFS); | |
1889 | kfree(failure); | |
1890 | } | |
1891 | } | |
7e38326f CM |
1892 | return 0; |
1893 | } | |
1894 | ||
d352ac68 CM |
1895 | /* |
1896 | * when reads are done, we need to check csums to verify the data is correct | |
1897 | * if there's a match, we allow the bio to finish. If not, we go through | |
1898 | * the io_failure_record routines to find good copies | |
1899 | */ | |
b2950863 | 1900 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1901 | struct extent_state *state) |
07157aac | 1902 | { |
35ebb934 | 1903 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1904 | struct inode *inode = page->mapping->host; |
d1310b2e | 1905 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1906 | char *kaddr; |
aadfeb6e | 1907 | u64 private = ~(u32)0; |
07157aac | 1908 | int ret; |
ff79f819 CM |
1909 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1910 | u32 csum = ~(u32)0; | |
d1310b2e | 1911 | |
d20f7043 CM |
1912 | if (PageChecked(page)) { |
1913 | ClearPageChecked(page); | |
1914 | goto good; | |
1915 | } | |
6cbff00f CH |
1916 | |
1917 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1918 | return 0; |
1919 | ||
1920 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1921 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1922 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1923 | GFP_NOFS); | |
b6cda9bc | 1924 | return 0; |
17d217fe | 1925 | } |
d20f7043 | 1926 | |
c2e639f0 | 1927 | if (state && state->start == start) { |
70dec807 CM |
1928 | private = state->private; |
1929 | ret = 0; | |
1930 | } else { | |
1931 | ret = get_state_private(io_tree, start, &private); | |
1932 | } | |
9ab86c8e | 1933 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1934 | if (ret) |
07157aac | 1935 | goto zeroit; |
d397712b | 1936 | |
ff79f819 CM |
1937 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1938 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1939 | if (csum != private) |
07157aac | 1940 | goto zeroit; |
d397712b | 1941 | |
9ab86c8e | 1942 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1943 | good: |
7e38326f CM |
1944 | /* if the io failure tree for this inode is non-empty, |
1945 | * check to see if we've recovered from a failed IO | |
1946 | */ | |
1259ab75 | 1947 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1948 | return 0; |
1949 | ||
1950 | zeroit: | |
193f284d CM |
1951 | if (printk_ratelimit()) { |
1952 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1953 | "private %llu\n", page->mapping->host->i_ino, | |
1954 | (unsigned long long)start, csum, | |
1955 | (unsigned long long)private); | |
1956 | } | |
db94535d CM |
1957 | memset(kaddr + offset, 1, end - start + 1); |
1958 | flush_dcache_page(page); | |
9ab86c8e | 1959 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1960 | if (private == 0) |
1961 | return 0; | |
7e38326f | 1962 | return -EIO; |
07157aac | 1963 | } |
b888db2b | 1964 | |
7b128766 JB |
1965 | /* |
1966 | * This creates an orphan entry for the given inode in case something goes | |
1967 | * wrong in the middle of an unlink/truncate. | |
1968 | */ | |
1969 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1970 | { | |
1971 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1972 | int ret = 0; | |
1973 | ||
bcc63abb | 1974 | spin_lock(&root->list_lock); |
7b128766 JB |
1975 | |
1976 | /* already on the orphan list, we're good */ | |
1977 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1978 | spin_unlock(&root->list_lock); |
7b128766 JB |
1979 | return 0; |
1980 | } | |
1981 | ||
1982 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
1983 | ||
bcc63abb | 1984 | spin_unlock(&root->list_lock); |
7b128766 JB |
1985 | |
1986 | /* | |
1987 | * insert an orphan item to track this unlinked/truncated file | |
1988 | */ | |
1989 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
1990 | ||
1991 | return ret; | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
1996 | * item for this particular inode. | |
1997 | */ | |
1998 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
1999 | { | |
2000 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2001 | int ret = 0; | |
2002 | ||
bcc63abb | 2003 | spin_lock(&root->list_lock); |
7b128766 JB |
2004 | |
2005 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2006 | spin_unlock(&root->list_lock); |
7b128766 JB |
2007 | return 0; |
2008 | } | |
2009 | ||
2010 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2011 | if (!trans) { | |
bcc63abb | 2012 | spin_unlock(&root->list_lock); |
7b128766 JB |
2013 | return 0; |
2014 | } | |
2015 | ||
bcc63abb | 2016 | spin_unlock(&root->list_lock); |
7b128766 JB |
2017 | |
2018 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2019 | ||
2020 | return ret; | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * this cleans up any orphans that may be left on the list from the last use | |
2025 | * of this root. | |
2026 | */ | |
2027 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2028 | { | |
2029 | struct btrfs_path *path; | |
2030 | struct extent_buffer *leaf; | |
2031 | struct btrfs_item *item; | |
2032 | struct btrfs_key key, found_key; | |
2033 | struct btrfs_trans_handle *trans; | |
2034 | struct inode *inode; | |
2035 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2036 | ||
7b128766 JB |
2037 | path = btrfs_alloc_path(); |
2038 | if (!path) | |
2039 | return; | |
2040 | path->reada = -1; | |
2041 | ||
2042 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2043 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2044 | key.offset = (u64)-1; | |
2045 | ||
7b128766 JB |
2046 | |
2047 | while (1) { | |
2048 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2049 | if (ret < 0) { | |
2050 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2051 | "\n", ret); | |
2052 | break; | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * if ret == 0 means we found what we were searching for, which | |
2057 | * is weird, but possible, so only screw with path if we didnt | |
2058 | * find the key and see if we have stuff that matches | |
2059 | */ | |
2060 | if (ret > 0) { | |
2061 | if (path->slots[0] == 0) | |
2062 | break; | |
2063 | path->slots[0]--; | |
2064 | } | |
2065 | ||
2066 | /* pull out the item */ | |
2067 | leaf = path->nodes[0]; | |
2068 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2069 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2070 | ||
2071 | /* make sure the item matches what we want */ | |
2072 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2073 | break; | |
2074 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2075 | break; | |
2076 | ||
2077 | /* release the path since we're done with it */ | |
2078 | btrfs_release_path(root, path); | |
2079 | ||
2080 | /* | |
2081 | * this is where we are basically btrfs_lookup, without the | |
2082 | * crossing root thing. we store the inode number in the | |
2083 | * offset of the orphan item. | |
2084 | */ | |
5d4f98a2 YZ |
2085 | found_key.objectid = found_key.offset; |
2086 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2087 | found_key.offset = 0; | |
2088 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2089 | if (IS_ERR(inode)) | |
7b128766 JB |
2090 | break; |
2091 | ||
7b128766 JB |
2092 | /* |
2093 | * add this inode to the orphan list so btrfs_orphan_del does | |
2094 | * the proper thing when we hit it | |
2095 | */ | |
bcc63abb | 2096 | spin_lock(&root->list_lock); |
7b128766 | 2097 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2098 | spin_unlock(&root->list_lock); |
7b128766 JB |
2099 | |
2100 | /* | |
2101 | * if this is a bad inode, means we actually succeeded in | |
2102 | * removing the inode, but not the orphan record, which means | |
2103 | * we need to manually delete the orphan since iput will just | |
2104 | * do a destroy_inode | |
2105 | */ | |
2106 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2107 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2108 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2109 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2110 | iput(inode); |
2111 | continue; | |
2112 | } | |
2113 | ||
2114 | /* if we have links, this was a truncate, lets do that */ | |
2115 | if (inode->i_nlink) { | |
2116 | nr_truncate++; | |
2117 | btrfs_truncate(inode); | |
2118 | } else { | |
2119 | nr_unlink++; | |
2120 | } | |
2121 | ||
2122 | /* this will do delete_inode and everything for us */ | |
2123 | iput(inode); | |
2124 | } | |
2125 | ||
2126 | if (nr_unlink) | |
2127 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2128 | if (nr_truncate) | |
2129 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2130 | ||
2131 | btrfs_free_path(path); | |
7b128766 JB |
2132 | } |
2133 | ||
46a53cca CM |
2134 | /* |
2135 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2136 | * don't find any xattrs, we know there can't be any acls. | |
2137 | * | |
2138 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2139 | */ | |
2140 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2141 | int slot, u64 objectid) | |
2142 | { | |
2143 | u32 nritems = btrfs_header_nritems(leaf); | |
2144 | struct btrfs_key found_key; | |
2145 | int scanned = 0; | |
2146 | ||
2147 | slot++; | |
2148 | while (slot < nritems) { | |
2149 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2150 | ||
2151 | /* we found a different objectid, there must not be acls */ | |
2152 | if (found_key.objectid != objectid) | |
2153 | return 0; | |
2154 | ||
2155 | /* we found an xattr, assume we've got an acl */ | |
2156 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2157 | return 1; | |
2158 | ||
2159 | /* | |
2160 | * we found a key greater than an xattr key, there can't | |
2161 | * be any acls later on | |
2162 | */ | |
2163 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2164 | return 0; | |
2165 | ||
2166 | slot++; | |
2167 | scanned++; | |
2168 | ||
2169 | /* | |
2170 | * it goes inode, inode backrefs, xattrs, extents, | |
2171 | * so if there are a ton of hard links to an inode there can | |
2172 | * be a lot of backrefs. Don't waste time searching too hard, | |
2173 | * this is just an optimization | |
2174 | */ | |
2175 | if (scanned >= 8) | |
2176 | break; | |
2177 | } | |
2178 | /* we hit the end of the leaf before we found an xattr or | |
2179 | * something larger than an xattr. We have to assume the inode | |
2180 | * has acls | |
2181 | */ | |
2182 | return 1; | |
2183 | } | |
2184 | ||
d352ac68 CM |
2185 | /* |
2186 | * read an inode from the btree into the in-memory inode | |
2187 | */ | |
5d4f98a2 | 2188 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2189 | { |
2190 | struct btrfs_path *path; | |
5f39d397 | 2191 | struct extent_buffer *leaf; |
39279cc3 | 2192 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2193 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2194 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2195 | struct btrfs_key location; | |
46a53cca | 2196 | int maybe_acls; |
39279cc3 | 2197 | u64 alloc_group_block; |
618e21d5 | 2198 | u32 rdev; |
39279cc3 CM |
2199 | int ret; |
2200 | ||
2201 | path = btrfs_alloc_path(); | |
2202 | BUG_ON(!path); | |
39279cc3 | 2203 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2204 | |
39279cc3 | 2205 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2206 | if (ret) |
39279cc3 | 2207 | goto make_bad; |
39279cc3 | 2208 | |
5f39d397 CM |
2209 | leaf = path->nodes[0]; |
2210 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2211 | struct btrfs_inode_item); | |
2212 | ||
2213 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2214 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2215 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2216 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2217 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2218 | |
2219 | tspec = btrfs_inode_atime(inode_item); | |
2220 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2221 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2222 | ||
2223 | tspec = btrfs_inode_mtime(inode_item); | |
2224 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2225 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2226 | ||
2227 | tspec = btrfs_inode_ctime(inode_item); | |
2228 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2229 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2230 | ||
a76a3cd4 | 2231 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2232 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2233 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2234 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2235 | inode->i_rdev = 0; |
5f39d397 CM |
2236 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2237 | ||
aec7477b | 2238 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2239 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2240 | |
5f39d397 | 2241 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2242 | |
46a53cca CM |
2243 | /* |
2244 | * try to precache a NULL acl entry for files that don't have | |
2245 | * any xattrs or acls | |
2246 | */ | |
2247 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2248 | if (!maybe_acls) |
2249 | cache_no_acl(inode); | |
46a53cca | 2250 | |
d2fb3437 YZ |
2251 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2252 | alloc_group_block, 0); | |
39279cc3 CM |
2253 | btrfs_free_path(path); |
2254 | inode_item = NULL; | |
2255 | ||
39279cc3 | 2256 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2257 | case S_IFREG: |
2258 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2259 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2260 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2261 | inode->i_fop = &btrfs_file_operations; |
2262 | inode->i_op = &btrfs_file_inode_operations; | |
2263 | break; | |
2264 | case S_IFDIR: | |
2265 | inode->i_fop = &btrfs_dir_file_operations; | |
2266 | if (root == root->fs_info->tree_root) | |
2267 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2268 | else | |
2269 | inode->i_op = &btrfs_dir_inode_operations; | |
2270 | break; | |
2271 | case S_IFLNK: | |
2272 | inode->i_op = &btrfs_symlink_inode_operations; | |
2273 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2274 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2275 | break; |
618e21d5 | 2276 | default: |
0279b4cd | 2277 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2278 | init_special_inode(inode, inode->i_mode, rdev); |
2279 | break; | |
39279cc3 | 2280 | } |
6cbff00f CH |
2281 | |
2282 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2283 | return; |
2284 | ||
2285 | make_bad: | |
39279cc3 | 2286 | btrfs_free_path(path); |
39279cc3 CM |
2287 | make_bad_inode(inode); |
2288 | } | |
2289 | ||
d352ac68 CM |
2290 | /* |
2291 | * given a leaf and an inode, copy the inode fields into the leaf | |
2292 | */ | |
e02119d5 CM |
2293 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2294 | struct extent_buffer *leaf, | |
5f39d397 | 2295 | struct btrfs_inode_item *item, |
39279cc3 CM |
2296 | struct inode *inode) |
2297 | { | |
5f39d397 CM |
2298 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2299 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2300 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2301 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2302 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2303 | ||
2304 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2305 | inode->i_atime.tv_sec); | |
2306 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2307 | inode->i_atime.tv_nsec); | |
2308 | ||
2309 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2310 | inode->i_mtime.tv_sec); | |
2311 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2312 | inode->i_mtime.tv_nsec); | |
2313 | ||
2314 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2315 | inode->i_ctime.tv_sec); | |
2316 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2317 | inode->i_ctime.tv_nsec); | |
2318 | ||
a76a3cd4 | 2319 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2320 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2321 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2322 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2323 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2324 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2325 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2326 | } |
2327 | ||
d352ac68 CM |
2328 | /* |
2329 | * copy everything in the in-memory inode into the btree. | |
2330 | */ | |
d397712b CM |
2331 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2332 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2333 | { |
2334 | struct btrfs_inode_item *inode_item; | |
2335 | struct btrfs_path *path; | |
5f39d397 | 2336 | struct extent_buffer *leaf; |
39279cc3 CM |
2337 | int ret; |
2338 | ||
2339 | path = btrfs_alloc_path(); | |
2340 | BUG_ON(!path); | |
b9473439 | 2341 | path->leave_spinning = 1; |
39279cc3 CM |
2342 | ret = btrfs_lookup_inode(trans, root, path, |
2343 | &BTRFS_I(inode)->location, 1); | |
2344 | if (ret) { | |
2345 | if (ret > 0) | |
2346 | ret = -ENOENT; | |
2347 | goto failed; | |
2348 | } | |
2349 | ||
b4ce94de | 2350 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2351 | leaf = path->nodes[0]; |
2352 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2353 | struct btrfs_inode_item); |
2354 | ||
e02119d5 | 2355 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2356 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2357 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2358 | ret = 0; |
2359 | failed: | |
39279cc3 CM |
2360 | btrfs_free_path(path); |
2361 | return ret; | |
2362 | } | |
2363 | ||
2364 | ||
d352ac68 CM |
2365 | /* |
2366 | * unlink helper that gets used here in inode.c and in the tree logging | |
2367 | * recovery code. It remove a link in a directory with a given name, and | |
2368 | * also drops the back refs in the inode to the directory | |
2369 | */ | |
e02119d5 CM |
2370 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2371 | struct btrfs_root *root, | |
2372 | struct inode *dir, struct inode *inode, | |
2373 | const char *name, int name_len) | |
39279cc3 CM |
2374 | { |
2375 | struct btrfs_path *path; | |
39279cc3 | 2376 | int ret = 0; |
5f39d397 | 2377 | struct extent_buffer *leaf; |
39279cc3 | 2378 | struct btrfs_dir_item *di; |
5f39d397 | 2379 | struct btrfs_key key; |
aec7477b | 2380 | u64 index; |
39279cc3 CM |
2381 | |
2382 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2383 | if (!path) { |
2384 | ret = -ENOMEM; | |
2385 | goto err; | |
2386 | } | |
2387 | ||
b9473439 | 2388 | path->leave_spinning = 1; |
39279cc3 CM |
2389 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2390 | name, name_len, -1); | |
2391 | if (IS_ERR(di)) { | |
2392 | ret = PTR_ERR(di); | |
2393 | goto err; | |
2394 | } | |
2395 | if (!di) { | |
2396 | ret = -ENOENT; | |
2397 | goto err; | |
2398 | } | |
5f39d397 CM |
2399 | leaf = path->nodes[0]; |
2400 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2401 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2402 | if (ret) |
2403 | goto err; | |
39279cc3 CM |
2404 | btrfs_release_path(root, path); |
2405 | ||
aec7477b | 2406 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2407 | inode->i_ino, |
2408 | dir->i_ino, &index); | |
aec7477b | 2409 | if (ret) { |
d397712b | 2410 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2411 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2412 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2413 | goto err; |
2414 | } | |
2415 | ||
39279cc3 | 2416 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2417 | index, name, name_len, -1); |
39279cc3 CM |
2418 | if (IS_ERR(di)) { |
2419 | ret = PTR_ERR(di); | |
2420 | goto err; | |
2421 | } | |
2422 | if (!di) { | |
2423 | ret = -ENOENT; | |
2424 | goto err; | |
2425 | } | |
2426 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2427 | btrfs_release_path(root, path); |
39279cc3 | 2428 | |
e02119d5 CM |
2429 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2430 | inode, dir->i_ino); | |
49eb7e46 | 2431 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2432 | |
2433 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2434 | dir, index); | |
2435 | BUG_ON(ret); | |
39279cc3 CM |
2436 | err: |
2437 | btrfs_free_path(path); | |
e02119d5 CM |
2438 | if (ret) |
2439 | goto out; | |
2440 | ||
2441 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2442 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2443 | btrfs_update_inode(trans, root, dir); | |
2444 | btrfs_drop_nlink(inode); | |
2445 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2446 | out: |
39279cc3 CM |
2447 | return ret; |
2448 | } | |
2449 | ||
2450 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2451 | { | |
2452 | struct btrfs_root *root; | |
2453 | struct btrfs_trans_handle *trans; | |
7b128766 | 2454 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2455 | int ret; |
1832a6d5 | 2456 | unsigned long nr = 0; |
39279cc3 CM |
2457 | |
2458 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2459 | |
39279cc3 | 2460 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2461 | |
39279cc3 | 2462 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2463 | |
2464 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2465 | ||
e02119d5 CM |
2466 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2467 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2468 | |
2469 | if (inode->i_nlink == 0) | |
2470 | ret = btrfs_orphan_add(trans, inode); | |
2471 | ||
d3c2fdcf | 2472 | nr = trans->blocks_used; |
5f39d397 | 2473 | |
89ce8a63 | 2474 | btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2475 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2476 | return ret; |
2477 | } | |
2478 | ||
4df27c4d YZ |
2479 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2480 | struct btrfs_root *root, | |
2481 | struct inode *dir, u64 objectid, | |
2482 | const char *name, int name_len) | |
2483 | { | |
2484 | struct btrfs_path *path; | |
2485 | struct extent_buffer *leaf; | |
2486 | struct btrfs_dir_item *di; | |
2487 | struct btrfs_key key; | |
2488 | u64 index; | |
2489 | int ret; | |
2490 | ||
2491 | path = btrfs_alloc_path(); | |
2492 | if (!path) | |
2493 | return -ENOMEM; | |
2494 | ||
2495 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2496 | name, name_len, -1); | |
2497 | BUG_ON(!di || IS_ERR(di)); | |
2498 | ||
2499 | leaf = path->nodes[0]; | |
2500 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2501 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2502 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2503 | BUG_ON(ret); | |
2504 | btrfs_release_path(root, path); | |
2505 | ||
2506 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2507 | objectid, root->root_key.objectid, | |
2508 | dir->i_ino, &index, name, name_len); | |
2509 | if (ret < 0) { | |
2510 | BUG_ON(ret != -ENOENT); | |
2511 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2512 | name, name_len); | |
2513 | BUG_ON(!di || IS_ERR(di)); | |
2514 | ||
2515 | leaf = path->nodes[0]; | |
2516 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2517 | btrfs_release_path(root, path); | |
2518 | index = key.offset; | |
2519 | } | |
2520 | ||
2521 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2522 | index, name, name_len, -1); | |
2523 | BUG_ON(!di || IS_ERR(di)); | |
2524 | ||
2525 | leaf = path->nodes[0]; | |
2526 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2527 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2528 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2529 | BUG_ON(ret); | |
2530 | btrfs_release_path(root, path); | |
2531 | ||
2532 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2533 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2534 | ret = btrfs_update_inode(trans, root, dir); | |
2535 | BUG_ON(ret); | |
2536 | dir->i_sb->s_dirt = 1; | |
2537 | ||
2538 | btrfs_free_path(path); | |
2539 | return 0; | |
2540 | } | |
2541 | ||
39279cc3 CM |
2542 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2543 | { | |
2544 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2545 | int err = 0; |
39279cc3 CM |
2546 | int ret; |
2547 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2548 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2549 | unsigned long nr = 0; |
39279cc3 | 2550 | |
3394e160 | 2551 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2552 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2553 | return -ENOTEMPTY; |
2554 | ||
39279cc3 CM |
2555 | trans = btrfs_start_transaction(root, 1); |
2556 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2557 | |
4df27c4d YZ |
2558 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2559 | err = btrfs_unlink_subvol(trans, root, dir, | |
2560 | BTRFS_I(inode)->location.objectid, | |
2561 | dentry->d_name.name, | |
2562 | dentry->d_name.len); | |
2563 | goto out; | |
2564 | } | |
2565 | ||
7b128766 JB |
2566 | err = btrfs_orphan_add(trans, inode); |
2567 | if (err) | |
4df27c4d | 2568 | goto out; |
7b128766 | 2569 | |
39279cc3 | 2570 | /* now the directory is empty */ |
e02119d5 CM |
2571 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2572 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2573 | if (!err) |
dbe674a9 | 2574 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2575 | out: |
d3c2fdcf | 2576 | nr = trans->blocks_used; |
89ce8a63 | 2577 | ret = btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2578 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2579 | |
39279cc3 CM |
2580 | if (ret && !err) |
2581 | err = ret; | |
2582 | return err; | |
2583 | } | |
2584 | ||
d20f7043 | 2585 | #if 0 |
323ac95b CM |
2586 | /* |
2587 | * when truncating bytes in a file, it is possible to avoid reading | |
2588 | * the leaves that contain only checksum items. This can be the | |
2589 | * majority of the IO required to delete a large file, but it must | |
2590 | * be done carefully. | |
2591 | * | |
2592 | * The keys in the level just above the leaves are checked to make sure | |
2593 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2594 | * after the new size. | |
2595 | * | |
2596 | * Then the key for the next leaf is checked to make sure it also has | |
2597 | * a checksum item for the same file. If it does, we know our target leaf | |
2598 | * contains only checksum items, and it can be safely freed without reading | |
2599 | * it. | |
2600 | * | |
2601 | * This is just an optimization targeted at large files. It may do | |
2602 | * nothing. It will return 0 unless things went badly. | |
2603 | */ | |
2604 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2605 | struct btrfs_root *root, | |
2606 | struct btrfs_path *path, | |
2607 | struct inode *inode, u64 new_size) | |
2608 | { | |
2609 | struct btrfs_key key; | |
2610 | int ret; | |
2611 | int nritems; | |
2612 | struct btrfs_key found_key; | |
2613 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2614 | struct btrfs_leaf_ref *ref; |
2615 | u64 leaf_gen; | |
2616 | u64 leaf_start; | |
323ac95b CM |
2617 | |
2618 | path->lowest_level = 1; | |
2619 | key.objectid = inode->i_ino; | |
2620 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2621 | key.offset = new_size; | |
2622 | again: | |
2623 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2624 | if (ret < 0) | |
2625 | goto out; | |
2626 | ||
2627 | if (path->nodes[1] == NULL) { | |
2628 | ret = 0; | |
2629 | goto out; | |
2630 | } | |
2631 | ret = 0; | |
2632 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2633 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2634 | ||
2635 | if (!nritems) | |
2636 | goto out; | |
2637 | ||
2638 | if (path->slots[1] >= nritems) | |
2639 | goto next_node; | |
2640 | ||
2641 | /* did we find a key greater than anything we want to delete? */ | |
2642 | if (found_key.objectid > inode->i_ino || | |
2643 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2644 | goto out; | |
2645 | ||
2646 | /* we check the next key in the node to make sure the leave contains | |
2647 | * only checksum items. This comparison doesn't work if our | |
2648 | * leaf is the last one in the node | |
2649 | */ | |
2650 | if (path->slots[1] + 1 >= nritems) { | |
2651 | next_node: | |
2652 | /* search forward from the last key in the node, this | |
2653 | * will bring us into the next node in the tree | |
2654 | */ | |
2655 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2656 | ||
2657 | /* unlikely, but we inc below, so check to be safe */ | |
2658 | if (found_key.offset == (u64)-1) | |
2659 | goto out; | |
2660 | ||
2661 | /* search_forward needs a path with locks held, do the | |
2662 | * search again for the original key. It is possible | |
2663 | * this will race with a balance and return a path that | |
2664 | * we could modify, but this drop is just an optimization | |
2665 | * and is allowed to miss some leaves. | |
2666 | */ | |
2667 | btrfs_release_path(root, path); | |
2668 | found_key.offset++; | |
2669 | ||
2670 | /* setup a max key for search_forward */ | |
2671 | other_key.offset = (u64)-1; | |
2672 | other_key.type = key.type; | |
2673 | other_key.objectid = key.objectid; | |
2674 | ||
2675 | path->keep_locks = 1; | |
2676 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2677 | path, 0, 0); | |
2678 | path->keep_locks = 0; | |
2679 | if (ret || found_key.objectid != key.objectid || | |
2680 | found_key.type != key.type) { | |
2681 | ret = 0; | |
2682 | goto out; | |
2683 | } | |
2684 | ||
2685 | key.offset = found_key.offset; | |
2686 | btrfs_release_path(root, path); | |
2687 | cond_resched(); | |
2688 | goto again; | |
2689 | } | |
2690 | ||
2691 | /* we know there's one more slot after us in the tree, | |
2692 | * read that key so we can verify it is also a checksum item | |
2693 | */ | |
2694 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2695 | ||
2696 | if (found_key.objectid < inode->i_ino) | |
2697 | goto next_key; | |
2698 | ||
2699 | if (found_key.type != key.type || found_key.offset < new_size) | |
2700 | goto next_key; | |
2701 | ||
2702 | /* | |
2703 | * if the key for the next leaf isn't a csum key from this objectid, | |
2704 | * we can't be sure there aren't good items inside this leaf. | |
2705 | * Bail out | |
2706 | */ | |
2707 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2708 | goto out; | |
2709 | ||
5b84e8d6 YZ |
2710 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2711 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2712 | /* |
2713 | * it is safe to delete this leaf, it contains only | |
2714 | * csum items from this inode at an offset >= new_size | |
2715 | */ | |
5b84e8d6 | 2716 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2717 | BUG_ON(ret); |
2718 | ||
5b84e8d6 YZ |
2719 | if (root->ref_cows && leaf_gen < trans->transid) { |
2720 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2721 | if (ref) { | |
2722 | ref->root_gen = root->root_key.offset; | |
2723 | ref->bytenr = leaf_start; | |
2724 | ref->owner = 0; | |
2725 | ref->generation = leaf_gen; | |
2726 | ref->nritems = 0; | |
2727 | ||
bd56b302 CM |
2728 | btrfs_sort_leaf_ref(ref); |
2729 | ||
5b84e8d6 YZ |
2730 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2731 | WARN_ON(ret); | |
2732 | btrfs_free_leaf_ref(root, ref); | |
2733 | } else { | |
2734 | WARN_ON(1); | |
2735 | } | |
2736 | } | |
323ac95b CM |
2737 | next_key: |
2738 | btrfs_release_path(root, path); | |
2739 | ||
2740 | if (other_key.objectid == inode->i_ino && | |
2741 | other_key.type == key.type && other_key.offset > key.offset) { | |
2742 | key.offset = other_key.offset; | |
2743 | cond_resched(); | |
2744 | goto again; | |
2745 | } | |
2746 | ret = 0; | |
2747 | out: | |
2748 | /* fixup any changes we've made to the path */ | |
2749 | path->lowest_level = 0; | |
2750 | path->keep_locks = 0; | |
2751 | btrfs_release_path(root, path); | |
2752 | return ret; | |
2753 | } | |
2754 | ||
d20f7043 CM |
2755 | #endif |
2756 | ||
39279cc3 CM |
2757 | /* |
2758 | * this can truncate away extent items, csum items and directory items. | |
2759 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2760 | * any higher than new_size |
39279cc3 CM |
2761 | * |
2762 | * csum items that cross the new i_size are truncated to the new size | |
2763 | * as well. | |
7b128766 JB |
2764 | * |
2765 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2766 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2767 | */ |
e02119d5 CM |
2768 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2769 | struct btrfs_root *root, | |
2770 | struct inode *inode, | |
2771 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2772 | { |
2773 | int ret; | |
2774 | struct btrfs_path *path; | |
2775 | struct btrfs_key key; | |
5f39d397 | 2776 | struct btrfs_key found_key; |
06d9a8d7 | 2777 | u32 found_type = (u8)-1; |
5f39d397 | 2778 | struct extent_buffer *leaf; |
39279cc3 CM |
2779 | struct btrfs_file_extent_item *fi; |
2780 | u64 extent_start = 0; | |
db94535d | 2781 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2782 | u64 extent_offset = 0; |
39279cc3 CM |
2783 | u64 item_end = 0; |
2784 | int found_extent; | |
2785 | int del_item; | |
85e21bac CM |
2786 | int pending_del_nr = 0; |
2787 | int pending_del_slot = 0; | |
179e29e4 | 2788 | int extent_type = -1; |
771ed689 | 2789 | int encoding; |
3b951516 | 2790 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2791 | |
e02119d5 | 2792 | if (root->ref_cows) |
5b21f2ed | 2793 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2794 | path = btrfs_alloc_path(); |
2795 | BUG_ON(!path); | |
33c17ad5 | 2796 | path->reada = -1; |
5f39d397 | 2797 | |
39279cc3 CM |
2798 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2799 | key.objectid = inode->i_ino; | |
2800 | key.offset = (u64)-1; | |
5f39d397 CM |
2801 | key.type = (u8)-1; |
2802 | ||
85e21bac | 2803 | search_again: |
b9473439 | 2804 | path->leave_spinning = 1; |
85e21bac | 2805 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2806 | if (ret < 0) |
85e21bac | 2807 | goto error; |
d397712b | 2808 | |
85e21bac | 2809 | if (ret > 0) { |
e02119d5 CM |
2810 | /* there are no items in the tree for us to truncate, we're |
2811 | * done | |
2812 | */ | |
2813 | if (path->slots[0] == 0) { | |
2814 | ret = 0; | |
2815 | goto error; | |
2816 | } | |
85e21bac CM |
2817 | path->slots[0]--; |
2818 | } | |
2819 | ||
d397712b | 2820 | while (1) { |
39279cc3 | 2821 | fi = NULL; |
5f39d397 CM |
2822 | leaf = path->nodes[0]; |
2823 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2824 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2825 | encoding = 0; |
39279cc3 | 2826 | |
5f39d397 | 2827 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2828 | break; |
5f39d397 | 2829 | |
85e21bac | 2830 | if (found_type < min_type) |
39279cc3 CM |
2831 | break; |
2832 | ||
5f39d397 | 2833 | item_end = found_key.offset; |
39279cc3 | 2834 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2835 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2836 | struct btrfs_file_extent_item); |
179e29e4 | 2837 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2838 | encoding = btrfs_file_extent_compression(leaf, fi); |
2839 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2840 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2841 | ||
179e29e4 | 2842 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2843 | item_end += |
db94535d | 2844 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2845 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2846 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2847 | fi); |
39279cc3 | 2848 | } |
008630c1 | 2849 | item_end--; |
39279cc3 | 2850 | } |
e02119d5 | 2851 | if (item_end < new_size) { |
d397712b | 2852 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2853 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2854 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2855 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2856 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2857 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2858 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2859 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2860 | else if (found_type) |
b888db2b | 2861 | found_type--; |
d397712b | 2862 | else |
b888db2b | 2863 | break; |
a61721d5 | 2864 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2865 | goto next; |
39279cc3 | 2866 | } |
e02119d5 | 2867 | if (found_key.offset >= new_size) |
39279cc3 CM |
2868 | del_item = 1; |
2869 | else | |
2870 | del_item = 0; | |
2871 | found_extent = 0; | |
2872 | ||
2873 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2874 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2875 | goto delete; | |
2876 | ||
2877 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2878 | u64 num_dec; |
db94535d | 2879 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2880 | if (!del_item && !encoding) { |
db94535d CM |
2881 | u64 orig_num_bytes = |
2882 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2883 | extent_num_bytes = new_size - |
5f39d397 | 2884 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2885 | extent_num_bytes = extent_num_bytes & |
2886 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2887 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2888 | extent_num_bytes); | |
2889 | num_dec = (orig_num_bytes - | |
9069218d | 2890 | extent_num_bytes); |
e02119d5 | 2891 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2892 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2893 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2894 | } else { |
db94535d CM |
2895 | extent_num_bytes = |
2896 | btrfs_file_extent_disk_num_bytes(leaf, | |
2897 | fi); | |
5d4f98a2 YZ |
2898 | extent_offset = found_key.offset - |
2899 | btrfs_file_extent_offset(leaf, fi); | |
2900 | ||
39279cc3 | 2901 | /* FIXME blocksize != 4096 */ |
9069218d | 2902 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2903 | if (extent_start != 0) { |
2904 | found_extent = 1; | |
e02119d5 | 2905 | if (root->ref_cows) |
a76a3cd4 | 2906 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2907 | } |
39279cc3 | 2908 | } |
9069218d | 2909 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2910 | /* |
2911 | * we can't truncate inline items that have had | |
2912 | * special encodings | |
2913 | */ | |
2914 | if (!del_item && | |
2915 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2916 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2917 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2918 | u32 size = new_size - found_key.offset; |
2919 | ||
2920 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2921 | inode_sub_bytes(inode, item_end + 1 - |
2922 | new_size); | |
e02119d5 CM |
2923 | } |
2924 | size = | |
2925 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2926 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2927 | size, 1); |
9069218d | 2928 | BUG_ON(ret); |
e02119d5 | 2929 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2930 | inode_sub_bytes(inode, item_end + 1 - |
2931 | found_key.offset); | |
9069218d | 2932 | } |
39279cc3 | 2933 | } |
179e29e4 | 2934 | delete: |
39279cc3 | 2935 | if (del_item) { |
85e21bac CM |
2936 | if (!pending_del_nr) { |
2937 | /* no pending yet, add ourselves */ | |
2938 | pending_del_slot = path->slots[0]; | |
2939 | pending_del_nr = 1; | |
2940 | } else if (pending_del_nr && | |
2941 | path->slots[0] + 1 == pending_del_slot) { | |
2942 | /* hop on the pending chunk */ | |
2943 | pending_del_nr++; | |
2944 | pending_del_slot = path->slots[0]; | |
2945 | } else { | |
d397712b | 2946 | BUG(); |
85e21bac | 2947 | } |
39279cc3 CM |
2948 | } else { |
2949 | break; | |
2950 | } | |
5d4f98a2 | 2951 | if (found_extent && root->ref_cows) { |
b9473439 | 2952 | btrfs_set_path_blocking(path); |
39279cc3 | 2953 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
2954 | extent_num_bytes, 0, |
2955 | btrfs_header_owner(leaf), | |
2956 | inode->i_ino, extent_offset); | |
39279cc3 CM |
2957 | BUG_ON(ret); |
2958 | } | |
85e21bac CM |
2959 | next: |
2960 | if (path->slots[0] == 0) { | |
2961 | if (pending_del_nr) | |
2962 | goto del_pending; | |
2963 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2964 | if (found_type == BTRFS_INODE_ITEM_KEY) |
2965 | break; | |
85e21bac CM |
2966 | goto search_again; |
2967 | } | |
2968 | ||
2969 | path->slots[0]--; | |
2970 | if (pending_del_nr && | |
2971 | path->slots[0] + 1 != pending_del_slot) { | |
2972 | struct btrfs_key debug; | |
2973 | del_pending: | |
2974 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2975 | pending_del_slot); | |
2976 | ret = btrfs_del_items(trans, root, path, | |
2977 | pending_del_slot, | |
2978 | pending_del_nr); | |
2979 | BUG_ON(ret); | |
2980 | pending_del_nr = 0; | |
2981 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2982 | if (found_type == BTRFS_INODE_ITEM_KEY) |
2983 | break; | |
85e21bac CM |
2984 | goto search_again; |
2985 | } | |
39279cc3 CM |
2986 | } |
2987 | ret = 0; | |
2988 | error: | |
85e21bac CM |
2989 | if (pending_del_nr) { |
2990 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
2991 | pending_del_nr); | |
2992 | } | |
39279cc3 | 2993 | btrfs_free_path(path); |
39279cc3 CM |
2994 | return ret; |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * taken from block_truncate_page, but does cow as it zeros out | |
2999 | * any bytes left in the last page in the file. | |
3000 | */ | |
3001 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3002 | { | |
3003 | struct inode *inode = mapping->host; | |
db94535d | 3004 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3005 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3006 | struct btrfs_ordered_extent *ordered; | |
3007 | char *kaddr; | |
db94535d | 3008 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3009 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3010 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3011 | struct page *page; | |
39279cc3 | 3012 | int ret = 0; |
a52d9a80 | 3013 | u64 page_start; |
e6dcd2dc | 3014 | u64 page_end; |
39279cc3 CM |
3015 | |
3016 | if ((offset & (blocksize - 1)) == 0) | |
3017 | goto out; | |
3018 | ||
3019 | ret = -ENOMEM; | |
211c17f5 | 3020 | again: |
39279cc3 CM |
3021 | page = grab_cache_page(mapping, index); |
3022 | if (!page) | |
3023 | goto out; | |
e6dcd2dc CM |
3024 | |
3025 | page_start = page_offset(page); | |
3026 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3027 | ||
39279cc3 | 3028 | if (!PageUptodate(page)) { |
9ebefb18 | 3029 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3030 | lock_page(page); |
211c17f5 CM |
3031 | if (page->mapping != mapping) { |
3032 | unlock_page(page); | |
3033 | page_cache_release(page); | |
3034 | goto again; | |
3035 | } | |
39279cc3 CM |
3036 | if (!PageUptodate(page)) { |
3037 | ret = -EIO; | |
89642229 | 3038 | goto out_unlock; |
39279cc3 CM |
3039 | } |
3040 | } | |
211c17f5 | 3041 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3042 | |
3043 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3044 | set_page_extent_mapped(page); | |
3045 | ||
3046 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3047 | if (ordered) { | |
3048 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3049 | unlock_page(page); | |
3050 | page_cache_release(page); | |
eb84ae03 | 3051 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3052 | btrfs_put_ordered_extent(ordered); |
3053 | goto again; | |
3054 | } | |
3055 | ||
9ed74f2d JB |
3056 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3057 | if (ret) { | |
3058 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3059 | goto out_unlock; | |
3060 | } | |
3061 | ||
e6dcd2dc CM |
3062 | ret = 0; |
3063 | if (offset != PAGE_CACHE_SIZE) { | |
3064 | kaddr = kmap(page); | |
3065 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3066 | flush_dcache_page(page); | |
3067 | kunmap(page); | |
3068 | } | |
247e743c | 3069 | ClearPageChecked(page); |
e6dcd2dc CM |
3070 | set_page_dirty(page); |
3071 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3072 | |
89642229 | 3073 | out_unlock: |
39279cc3 CM |
3074 | unlock_page(page); |
3075 | page_cache_release(page); | |
3076 | out: | |
3077 | return ret; | |
3078 | } | |
3079 | ||
9036c102 | 3080 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3081 | { |
9036c102 YZ |
3082 | struct btrfs_trans_handle *trans; |
3083 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3084 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3085 | struct extent_map *em; | |
3086 | u64 mask = root->sectorsize - 1; | |
3087 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3088 | u64 block_end = (size + mask) & ~mask; | |
3089 | u64 last_byte; | |
3090 | u64 cur_offset; | |
3091 | u64 hole_size; | |
9ed74f2d | 3092 | int err = 0; |
39279cc3 | 3093 | |
9036c102 YZ |
3094 | if (size <= hole_start) |
3095 | return 0; | |
3096 | ||
9036c102 | 3097 | btrfs_truncate_page(inode->i_mapping, inode->i_size); |
2bf5a725 | 3098 | |
9036c102 YZ |
3099 | while (1) { |
3100 | struct btrfs_ordered_extent *ordered; | |
3101 | btrfs_wait_ordered_range(inode, hole_start, | |
3102 | block_end - hole_start); | |
3103 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3104 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3105 | if (!ordered) | |
3106 | break; | |
3107 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3108 | btrfs_put_ordered_extent(ordered); | |
3109 | } | |
39279cc3 | 3110 | |
9036c102 YZ |
3111 | trans = btrfs_start_transaction(root, 1); |
3112 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3113 | |
9036c102 YZ |
3114 | cur_offset = hole_start; |
3115 | while (1) { | |
3116 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3117 | block_end - cur_offset, 0); | |
3118 | BUG_ON(IS_ERR(em) || !em); | |
3119 | last_byte = min(extent_map_end(em), block_end); | |
3120 | last_byte = (last_byte + mask) & ~mask; | |
3121 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3122 | u64 hint_byte = 0; |
9036c102 | 3123 | hole_size = last_byte - cur_offset; |
771ed689 CM |
3124 | err = btrfs_drop_extents(trans, root, inode, |
3125 | cur_offset, | |
3126 | cur_offset + hole_size, | |
e980b50c | 3127 | block_end, |
a1ed835e | 3128 | cur_offset, &hint_byte, 1); |
771ed689 CM |
3129 | if (err) |
3130 | break; | |
9ed74f2d JB |
3131 | |
3132 | err = btrfs_reserve_metadata_space(root, 1); | |
3133 | if (err) | |
3134 | break; | |
3135 | ||
9036c102 YZ |
3136 | err = btrfs_insert_file_extent(trans, root, |
3137 | inode->i_ino, cur_offset, 0, | |
3138 | 0, hole_size, 0, hole_size, | |
3139 | 0, 0, 0); | |
3140 | btrfs_drop_extent_cache(inode, hole_start, | |
3141 | last_byte - 1, 0); | |
9ed74f2d | 3142 | btrfs_unreserve_metadata_space(root, 1); |
9036c102 YZ |
3143 | } |
3144 | free_extent_map(em); | |
3145 | cur_offset = last_byte; | |
3146 | if (err || cur_offset >= block_end) | |
3147 | break; | |
3148 | } | |
1832a6d5 | 3149 | |
9036c102 YZ |
3150 | btrfs_end_transaction(trans, root); |
3151 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3152 | return err; | |
3153 | } | |
39279cc3 | 3154 | |
9036c102 YZ |
3155 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3156 | { | |
3157 | struct inode *inode = dentry->d_inode; | |
3158 | int err; | |
39279cc3 | 3159 | |
9036c102 YZ |
3160 | err = inode_change_ok(inode, attr); |
3161 | if (err) | |
3162 | return err; | |
2bf5a725 | 3163 | |
5a3f23d5 CM |
3164 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3165 | if (attr->ia_size > inode->i_size) { | |
3166 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3167 | if (err) | |
3168 | return err; | |
3169 | } else if (inode->i_size > 0 && | |
3170 | attr->ia_size == 0) { | |
3171 | ||
3172 | /* we're truncating a file that used to have good | |
3173 | * data down to zero. Make sure it gets into | |
3174 | * the ordered flush list so that any new writes | |
3175 | * get down to disk quickly. | |
3176 | */ | |
3177 | BTRFS_I(inode)->ordered_data_close = 1; | |
3178 | } | |
39279cc3 | 3179 | } |
9036c102 | 3180 | |
39279cc3 | 3181 | err = inode_setattr(inode, attr); |
33268eaf JB |
3182 | |
3183 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3184 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3185 | return err; |
3186 | } | |
61295eb8 | 3187 | |
39279cc3 CM |
3188 | void btrfs_delete_inode(struct inode *inode) |
3189 | { | |
3190 | struct btrfs_trans_handle *trans; | |
3191 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3192 | unsigned long nr; |
39279cc3 CM |
3193 | int ret; |
3194 | ||
3195 | truncate_inode_pages(&inode->i_data, 0); | |
3196 | if (is_bad_inode(inode)) { | |
7b128766 | 3197 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3198 | goto no_delete; |
3199 | } | |
4a096752 | 3200 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3201 | |
76dda93c YZ |
3202 | if (inode->i_nlink > 0) { |
3203 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3204 | goto no_delete; | |
3205 | } | |
3206 | ||
dbe674a9 | 3207 | btrfs_i_size_write(inode, 0); |
180591bc | 3208 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3209 | |
39279cc3 | 3210 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3211 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3212 | if (ret) { |
3213 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3214 | goto no_delete_lock; |
7b128766 JB |
3215 | } |
3216 | ||
3217 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3218 | |
d3c2fdcf | 3219 | nr = trans->blocks_used; |
85e21bac | 3220 | clear_inode(inode); |
5f39d397 | 3221 | |
39279cc3 | 3222 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3223 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3224 | return; |
54aa1f4d CM |
3225 | |
3226 | no_delete_lock: | |
d3c2fdcf | 3227 | nr = trans->blocks_used; |
54aa1f4d | 3228 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3229 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3230 | no_delete: |
3231 | clear_inode(inode); | |
3232 | } | |
3233 | ||
3234 | /* | |
3235 | * this returns the key found in the dir entry in the location pointer. | |
3236 | * If no dir entries were found, location->objectid is 0. | |
3237 | */ | |
3238 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3239 | struct btrfs_key *location) | |
3240 | { | |
3241 | const char *name = dentry->d_name.name; | |
3242 | int namelen = dentry->d_name.len; | |
3243 | struct btrfs_dir_item *di; | |
3244 | struct btrfs_path *path; | |
3245 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3246 | int ret = 0; |
39279cc3 CM |
3247 | |
3248 | path = btrfs_alloc_path(); | |
3249 | BUG_ON(!path); | |
3954401f | 3250 | |
39279cc3 CM |
3251 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3252 | namelen, 0); | |
0d9f7f3e Y |
3253 | if (IS_ERR(di)) |
3254 | ret = PTR_ERR(di); | |
d397712b CM |
3255 | |
3256 | if (!di || IS_ERR(di)) | |
3954401f | 3257 | goto out_err; |
d397712b | 3258 | |
5f39d397 | 3259 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3260 | out: |
39279cc3 CM |
3261 | btrfs_free_path(path); |
3262 | return ret; | |
3954401f CM |
3263 | out_err: |
3264 | location->objectid = 0; | |
3265 | goto out; | |
39279cc3 CM |
3266 | } |
3267 | ||
3268 | /* | |
3269 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3270 | * needs to be changed to reflect the root directory of the tree root. This | |
3271 | * is kind of like crossing a mount point. | |
3272 | */ | |
3273 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3274 | struct inode *dir, |
3275 | struct dentry *dentry, | |
3276 | struct btrfs_key *location, | |
3277 | struct btrfs_root **sub_root) | |
39279cc3 | 3278 | { |
4df27c4d YZ |
3279 | struct btrfs_path *path; |
3280 | struct btrfs_root *new_root; | |
3281 | struct btrfs_root_ref *ref; | |
3282 | struct extent_buffer *leaf; | |
3283 | int ret; | |
3284 | int err = 0; | |
39279cc3 | 3285 | |
4df27c4d YZ |
3286 | path = btrfs_alloc_path(); |
3287 | if (!path) { | |
3288 | err = -ENOMEM; | |
3289 | goto out; | |
3290 | } | |
39279cc3 | 3291 | |
4df27c4d YZ |
3292 | err = -ENOENT; |
3293 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3294 | BTRFS_I(dir)->root->root_key.objectid, | |
3295 | location->objectid); | |
3296 | if (ret) { | |
3297 | if (ret < 0) | |
3298 | err = ret; | |
3299 | goto out; | |
3300 | } | |
39279cc3 | 3301 | |
4df27c4d YZ |
3302 | leaf = path->nodes[0]; |
3303 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3304 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3305 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3306 | goto out; | |
39279cc3 | 3307 | |
4df27c4d YZ |
3308 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3309 | (unsigned long)(ref + 1), | |
3310 | dentry->d_name.len); | |
3311 | if (ret) | |
3312 | goto out; | |
3313 | ||
3314 | btrfs_release_path(root->fs_info->tree_root, path); | |
3315 | ||
3316 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3317 | if (IS_ERR(new_root)) { | |
3318 | err = PTR_ERR(new_root); | |
3319 | goto out; | |
3320 | } | |
3321 | ||
3322 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3323 | err = -ENOENT; | |
3324 | goto out; | |
3325 | } | |
3326 | ||
3327 | *sub_root = new_root; | |
3328 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3329 | location->type = BTRFS_INODE_ITEM_KEY; | |
3330 | location->offset = 0; | |
3331 | err = 0; | |
3332 | out: | |
3333 | btrfs_free_path(path); | |
3334 | return err; | |
39279cc3 CM |
3335 | } |
3336 | ||
5d4f98a2 YZ |
3337 | static void inode_tree_add(struct inode *inode) |
3338 | { | |
3339 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3340 | struct btrfs_inode *entry; | |
03e860bd FNP |
3341 | struct rb_node **p; |
3342 | struct rb_node *parent; | |
03e860bd FNP |
3343 | again: |
3344 | p = &root->inode_tree.rb_node; | |
3345 | parent = NULL; | |
5d4f98a2 | 3346 | |
76dda93c YZ |
3347 | if (hlist_unhashed(&inode->i_hash)) |
3348 | return; | |
3349 | ||
5d4f98a2 YZ |
3350 | spin_lock(&root->inode_lock); |
3351 | while (*p) { | |
3352 | parent = *p; | |
3353 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3354 | ||
3355 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3356 | p = &parent->rb_left; |
5d4f98a2 | 3357 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3358 | p = &parent->rb_right; |
5d4f98a2 YZ |
3359 | else { |
3360 | WARN_ON(!(entry->vfs_inode.i_state & | |
3361 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd FNP |
3362 | rb_erase(parent, &root->inode_tree); |
3363 | RB_CLEAR_NODE(parent); | |
3364 | spin_unlock(&root->inode_lock); | |
3365 | goto again; | |
5d4f98a2 YZ |
3366 | } |
3367 | } | |
3368 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3369 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3370 | spin_unlock(&root->inode_lock); | |
3371 | } | |
3372 | ||
3373 | static void inode_tree_del(struct inode *inode) | |
3374 | { | |
3375 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3376 | int empty = 0; |
5d4f98a2 | 3377 | |
03e860bd | 3378 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3379 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3380 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3381 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3382 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3383 | } |
03e860bd | 3384 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3385 | |
3386 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3387 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3388 | spin_lock(&root->inode_lock); | |
3389 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3390 | spin_unlock(&root->inode_lock); | |
3391 | if (empty) | |
3392 | btrfs_add_dead_root(root); | |
3393 | } | |
3394 | } | |
3395 | ||
3396 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3397 | { | |
3398 | struct rb_node *node; | |
3399 | struct rb_node *prev; | |
3400 | struct btrfs_inode *entry; | |
3401 | struct inode *inode; | |
3402 | u64 objectid = 0; | |
3403 | ||
3404 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3405 | ||
3406 | spin_lock(&root->inode_lock); | |
3407 | again: | |
3408 | node = root->inode_tree.rb_node; | |
3409 | prev = NULL; | |
3410 | while (node) { | |
3411 | prev = node; | |
3412 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3413 | ||
3414 | if (objectid < entry->vfs_inode.i_ino) | |
3415 | node = node->rb_left; | |
3416 | else if (objectid > entry->vfs_inode.i_ino) | |
3417 | node = node->rb_right; | |
3418 | else | |
3419 | break; | |
3420 | } | |
3421 | if (!node) { | |
3422 | while (prev) { | |
3423 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3424 | if (objectid <= entry->vfs_inode.i_ino) { | |
3425 | node = prev; | |
3426 | break; | |
3427 | } | |
3428 | prev = rb_next(prev); | |
3429 | } | |
3430 | } | |
3431 | while (node) { | |
3432 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3433 | objectid = entry->vfs_inode.i_ino + 1; | |
3434 | inode = igrab(&entry->vfs_inode); | |
3435 | if (inode) { | |
3436 | spin_unlock(&root->inode_lock); | |
3437 | if (atomic_read(&inode->i_count) > 1) | |
3438 | d_prune_aliases(inode); | |
3439 | /* | |
3440 | * btrfs_drop_inode will remove it from | |
3441 | * the inode cache when its usage count | |
3442 | * hits zero. | |
3443 | */ | |
3444 | iput(inode); | |
3445 | cond_resched(); | |
3446 | spin_lock(&root->inode_lock); | |
3447 | goto again; | |
3448 | } | |
3449 | ||
3450 | if (cond_resched_lock(&root->inode_lock)) | |
3451 | goto again; | |
3452 | ||
3453 | node = rb_next(node); | |
3454 | } | |
3455 | spin_unlock(&root->inode_lock); | |
3456 | return 0; | |
5d4f98a2 YZ |
3457 | } |
3458 | ||
e02119d5 | 3459 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3460 | { |
e02119d5 CM |
3461 | struct btrfs_inode *bi = BTRFS_I(inode); |
3462 | ||
e02119d5 | 3463 | bi->generation = 0; |
c3027eb5 | 3464 | bi->sequence = 0; |
e02119d5 CM |
3465 | bi->last_trans = 0; |
3466 | bi->logged_trans = 0; | |
3467 | bi->delalloc_bytes = 0; | |
6a63209f | 3468 | bi->reserved_bytes = 0; |
e02119d5 CM |
3469 | bi->disk_i_size = 0; |
3470 | bi->flags = 0; | |
3471 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3472 | bi->last_unlink_trans = 0; |
2757495c | 3473 | bi->ordered_data_close = 0; |
d1310b2e CM |
3474 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3475 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3476 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3477 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3478 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3479 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3480 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3481 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3482 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3483 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3484 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3485 | } | |
3486 | ||
3487 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3488 | { | |
3489 | struct btrfs_iget_args *args = p; | |
3490 | inode->i_ino = args->ino; | |
3491 | init_btrfs_i(inode); | |
3492 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3493 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3494 | return 0; |
3495 | } | |
3496 | ||
3497 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3498 | { | |
3499 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3500 | return args->ino == inode->i_ino && |
3501 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3502 | } |
3503 | ||
5d4f98a2 YZ |
3504 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3505 | u64 objectid, | |
3506 | struct btrfs_root *root) | |
39279cc3 CM |
3507 | { |
3508 | struct inode *inode; | |
3509 | struct btrfs_iget_args args; | |
3510 | args.ino = objectid; | |
3511 | args.root = root; | |
3512 | ||
3513 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3514 | btrfs_init_locked_inode, | |
3515 | (void *)&args); | |
3516 | return inode; | |
3517 | } | |
3518 | ||
1a54ef8c BR |
3519 | /* Get an inode object given its location and corresponding root. |
3520 | * Returns in *is_new if the inode was read from disk | |
3521 | */ | |
3522 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3523 | struct btrfs_root *root) |
1a54ef8c BR |
3524 | { |
3525 | struct inode *inode; | |
3526 | ||
3527 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3528 | if (!inode) | |
5d4f98a2 | 3529 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3530 | |
3531 | if (inode->i_state & I_NEW) { | |
3532 | BTRFS_I(inode)->root = root; | |
3533 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3534 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3535 | |
3536 | inode_tree_add(inode); | |
1a54ef8c | 3537 | unlock_new_inode(inode); |
1a54ef8c BR |
3538 | } |
3539 | ||
3540 | return inode; | |
3541 | } | |
3542 | ||
4df27c4d YZ |
3543 | static struct inode *new_simple_dir(struct super_block *s, |
3544 | struct btrfs_key *key, | |
3545 | struct btrfs_root *root) | |
3546 | { | |
3547 | struct inode *inode = new_inode(s); | |
3548 | ||
3549 | if (!inode) | |
3550 | return ERR_PTR(-ENOMEM); | |
3551 | ||
3552 | init_btrfs_i(inode); | |
3553 | ||
3554 | BTRFS_I(inode)->root = root; | |
3555 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3556 | BTRFS_I(inode)->dummy_inode = 1; | |
3557 | ||
3558 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3559 | inode->i_op = &simple_dir_inode_operations; | |
3560 | inode->i_fop = &simple_dir_operations; | |
3561 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3562 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3563 | ||
3564 | return inode; | |
3565 | } | |
3566 | ||
3de4586c | 3567 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3568 | { |
d397712b | 3569 | struct inode *inode; |
4df27c4d | 3570 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3571 | struct btrfs_root *sub_root = root; |
3572 | struct btrfs_key location; | |
76dda93c | 3573 | int index; |
5d4f98a2 | 3574 | int ret; |
39279cc3 | 3575 | |
76dda93c YZ |
3576 | dentry->d_op = &btrfs_dentry_operations; |
3577 | ||
39279cc3 CM |
3578 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3579 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3580 | |
39279cc3 | 3581 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3582 | |
39279cc3 CM |
3583 | if (ret < 0) |
3584 | return ERR_PTR(ret); | |
5f39d397 | 3585 | |
4df27c4d YZ |
3586 | if (location.objectid == 0) |
3587 | return NULL; | |
3588 | ||
3589 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3590 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3591 | return inode; | |
3592 | } | |
3593 | ||
3594 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3595 | ||
76dda93c | 3596 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3597 | ret = fixup_tree_root_location(root, dir, dentry, |
3598 | &location, &sub_root); | |
3599 | if (ret < 0) { | |
3600 | if (ret != -ENOENT) | |
3601 | inode = ERR_PTR(ret); | |
3602 | else | |
3603 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3604 | } else { | |
5d4f98a2 | 3605 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3606 | } |
76dda93c YZ |
3607 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3608 | ||
3de4586c CM |
3609 | return inode; |
3610 | } | |
3611 | ||
76dda93c YZ |
3612 | static int btrfs_dentry_delete(struct dentry *dentry) |
3613 | { | |
3614 | struct btrfs_root *root; | |
3615 | ||
3616 | if (!dentry->d_inode) | |
3617 | return 0; | |
3618 | ||
3619 | root = BTRFS_I(dentry->d_inode)->root; | |
3620 | if (btrfs_root_refs(&root->root_item) == 0) | |
3621 | return 1; | |
3622 | return 0; | |
3623 | } | |
3624 | ||
3de4586c CM |
3625 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3626 | struct nameidata *nd) | |
3627 | { | |
3628 | struct inode *inode; | |
3629 | ||
3de4586c CM |
3630 | inode = btrfs_lookup_dentry(dir, dentry); |
3631 | if (IS_ERR(inode)) | |
3632 | return ERR_CAST(inode); | |
7b128766 | 3633 | |
39279cc3 CM |
3634 | return d_splice_alias(inode, dentry); |
3635 | } | |
3636 | ||
39279cc3 CM |
3637 | static unsigned char btrfs_filetype_table[] = { |
3638 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3639 | }; | |
3640 | ||
cbdf5a24 DW |
3641 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3642 | filldir_t filldir) | |
39279cc3 | 3643 | { |
6da6abae | 3644 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3645 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3646 | struct btrfs_item *item; | |
3647 | struct btrfs_dir_item *di; | |
3648 | struct btrfs_key key; | |
5f39d397 | 3649 | struct btrfs_key found_key; |
39279cc3 CM |
3650 | struct btrfs_path *path; |
3651 | int ret; | |
3652 | u32 nritems; | |
5f39d397 | 3653 | struct extent_buffer *leaf; |
39279cc3 CM |
3654 | int slot; |
3655 | int advance; | |
3656 | unsigned char d_type; | |
3657 | int over = 0; | |
3658 | u32 di_cur; | |
3659 | u32 di_total; | |
3660 | u32 di_len; | |
3661 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3662 | char tmp_name[32]; |
3663 | char *name_ptr; | |
3664 | int name_len; | |
39279cc3 CM |
3665 | |
3666 | /* FIXME, use a real flag for deciding about the key type */ | |
3667 | if (root->fs_info->tree_root == root) | |
3668 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3669 | |
3954401f CM |
3670 | /* special case for "." */ |
3671 | if (filp->f_pos == 0) { | |
3672 | over = filldir(dirent, ".", 1, | |
3673 | 1, inode->i_ino, | |
3674 | DT_DIR); | |
3675 | if (over) | |
3676 | return 0; | |
3677 | filp->f_pos = 1; | |
3678 | } | |
3954401f CM |
3679 | /* special case for .., just use the back ref */ |
3680 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3681 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3682 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3683 | 2, pino, DT_DIR); |
3954401f | 3684 | if (over) |
49593bfa | 3685 | return 0; |
3954401f CM |
3686 | filp->f_pos = 2; |
3687 | } | |
49593bfa DW |
3688 | path = btrfs_alloc_path(); |
3689 | path->reada = 2; | |
3690 | ||
39279cc3 CM |
3691 | btrfs_set_key_type(&key, key_type); |
3692 | key.offset = filp->f_pos; | |
49593bfa | 3693 | key.objectid = inode->i_ino; |
5f39d397 | 3694 | |
39279cc3 CM |
3695 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3696 | if (ret < 0) | |
3697 | goto err; | |
3698 | advance = 0; | |
49593bfa DW |
3699 | |
3700 | while (1) { | |
5f39d397 CM |
3701 | leaf = path->nodes[0]; |
3702 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3703 | slot = path->slots[0]; |
3704 | if (advance || slot >= nritems) { | |
49593bfa | 3705 | if (slot >= nritems - 1) { |
39279cc3 CM |
3706 | ret = btrfs_next_leaf(root, path); |
3707 | if (ret) | |
3708 | break; | |
5f39d397 CM |
3709 | leaf = path->nodes[0]; |
3710 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3711 | slot = path->slots[0]; |
3712 | } else { | |
3713 | slot++; | |
3714 | path->slots[0]++; | |
3715 | } | |
3716 | } | |
3de4586c | 3717 | |
39279cc3 | 3718 | advance = 1; |
5f39d397 CM |
3719 | item = btrfs_item_nr(leaf, slot); |
3720 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3721 | ||
3722 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3723 | break; |
5f39d397 | 3724 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3725 | break; |
5f39d397 | 3726 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3727 | continue; |
5f39d397 CM |
3728 | |
3729 | filp->f_pos = found_key.offset; | |
49593bfa | 3730 | |
39279cc3 CM |
3731 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3732 | di_cur = 0; | |
5f39d397 | 3733 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3734 | |
3735 | while (di_cur < di_total) { | |
5f39d397 CM |
3736 | struct btrfs_key location; |
3737 | ||
3738 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3739 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3740 | name_ptr = tmp_name; |
3741 | } else { | |
3742 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3743 | if (!name_ptr) { |
3744 | ret = -ENOMEM; | |
3745 | goto err; | |
3746 | } | |
5f39d397 CM |
3747 | } |
3748 | read_extent_buffer(leaf, name_ptr, | |
3749 | (unsigned long)(di + 1), name_len); | |
3750 | ||
3751 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3752 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3753 | |
3754 | /* is this a reference to our own snapshot? If so | |
3755 | * skip it | |
3756 | */ | |
3757 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3758 | location.objectid == root->root_key.objectid) { | |
3759 | over = 0; | |
3760 | goto skip; | |
3761 | } | |
5f39d397 | 3762 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3763 | found_key.offset, location.objectid, |
39279cc3 | 3764 | d_type); |
5f39d397 | 3765 | |
3de4586c | 3766 | skip: |
5f39d397 CM |
3767 | if (name_ptr != tmp_name) |
3768 | kfree(name_ptr); | |
3769 | ||
39279cc3 CM |
3770 | if (over) |
3771 | goto nopos; | |
5103e947 | 3772 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3773 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3774 | di_cur += di_len; |
3775 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3776 | } | |
3777 | } | |
49593bfa DW |
3778 | |
3779 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3780 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3781 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3782 | else |
3783 | filp->f_pos++; | |
39279cc3 CM |
3784 | nopos: |
3785 | ret = 0; | |
3786 | err: | |
39279cc3 | 3787 | btrfs_free_path(path); |
39279cc3 CM |
3788 | return ret; |
3789 | } | |
3790 | ||
3791 | int btrfs_write_inode(struct inode *inode, int wait) | |
3792 | { | |
3793 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3794 | struct btrfs_trans_handle *trans; | |
3795 | int ret = 0; | |
3796 | ||
c146afad | 3797 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3798 | return 0; |
3799 | ||
39279cc3 | 3800 | if (wait) { |
f9295749 | 3801 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3802 | btrfs_set_trans_block_group(trans, inode); |
3803 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3804 | } |
3805 | return ret; | |
3806 | } | |
3807 | ||
3808 | /* | |
54aa1f4d | 3809 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3810 | * inode changes. But, it is most likely to find the inode in cache. |
3811 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3812 | * to keep or drop this code. | |
3813 | */ | |
3814 | void btrfs_dirty_inode(struct inode *inode) | |
3815 | { | |
3816 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3817 | struct btrfs_trans_handle *trans; | |
3818 | ||
f9295749 | 3819 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3820 | btrfs_set_trans_block_group(trans, inode); |
3821 | btrfs_update_inode(trans, root, inode); | |
3822 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3823 | } |
3824 | ||
d352ac68 CM |
3825 | /* |
3826 | * find the highest existing sequence number in a directory | |
3827 | * and then set the in-memory index_cnt variable to reflect | |
3828 | * free sequence numbers | |
3829 | */ | |
aec7477b JB |
3830 | static int btrfs_set_inode_index_count(struct inode *inode) |
3831 | { | |
3832 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3833 | struct btrfs_key key, found_key; | |
3834 | struct btrfs_path *path; | |
3835 | struct extent_buffer *leaf; | |
3836 | int ret; | |
3837 | ||
3838 | key.objectid = inode->i_ino; | |
3839 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3840 | key.offset = (u64)-1; | |
3841 | ||
3842 | path = btrfs_alloc_path(); | |
3843 | if (!path) | |
3844 | return -ENOMEM; | |
3845 | ||
3846 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3847 | if (ret < 0) | |
3848 | goto out; | |
3849 | /* FIXME: we should be able to handle this */ | |
3850 | if (ret == 0) | |
3851 | goto out; | |
3852 | ret = 0; | |
3853 | ||
3854 | /* | |
3855 | * MAGIC NUMBER EXPLANATION: | |
3856 | * since we search a directory based on f_pos we have to start at 2 | |
3857 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3858 | * else has to start at 2 | |
3859 | */ | |
3860 | if (path->slots[0] == 0) { | |
3861 | BTRFS_I(inode)->index_cnt = 2; | |
3862 | goto out; | |
3863 | } | |
3864 | ||
3865 | path->slots[0]--; | |
3866 | ||
3867 | leaf = path->nodes[0]; | |
3868 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3869 | ||
3870 | if (found_key.objectid != inode->i_ino || | |
3871 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3872 | BTRFS_I(inode)->index_cnt = 2; | |
3873 | goto out; | |
3874 | } | |
3875 | ||
3876 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3877 | out: | |
3878 | btrfs_free_path(path); | |
3879 | return ret; | |
3880 | } | |
3881 | ||
d352ac68 CM |
3882 | /* |
3883 | * helper to find a free sequence number in a given directory. This current | |
3884 | * code is very simple, later versions will do smarter things in the btree | |
3885 | */ | |
3de4586c | 3886 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3887 | { |
3888 | int ret = 0; | |
3889 | ||
3890 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3891 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3892 | if (ret) |
aec7477b JB |
3893 | return ret; |
3894 | } | |
3895 | ||
00e4e6b3 | 3896 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3897 | BTRFS_I(dir)->index_cnt++; |
3898 | ||
3899 | return ret; | |
3900 | } | |
3901 | ||
39279cc3 CM |
3902 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3903 | struct btrfs_root *root, | |
aec7477b | 3904 | struct inode *dir, |
9c58309d | 3905 | const char *name, int name_len, |
d2fb3437 YZ |
3906 | u64 ref_objectid, u64 objectid, |
3907 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3908 | { |
3909 | struct inode *inode; | |
5f39d397 | 3910 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3911 | struct btrfs_key *location; |
5f39d397 | 3912 | struct btrfs_path *path; |
9c58309d CM |
3913 | struct btrfs_inode_ref *ref; |
3914 | struct btrfs_key key[2]; | |
3915 | u32 sizes[2]; | |
3916 | unsigned long ptr; | |
39279cc3 CM |
3917 | int ret; |
3918 | int owner; | |
3919 | ||
5f39d397 CM |
3920 | path = btrfs_alloc_path(); |
3921 | BUG_ON(!path); | |
3922 | ||
39279cc3 CM |
3923 | inode = new_inode(root->fs_info->sb); |
3924 | if (!inode) | |
3925 | return ERR_PTR(-ENOMEM); | |
3926 | ||
aec7477b | 3927 | if (dir) { |
3de4586c | 3928 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
3929 | if (ret) { |
3930 | iput(inode); | |
aec7477b | 3931 | return ERR_PTR(ret); |
09771430 | 3932 | } |
aec7477b JB |
3933 | } |
3934 | /* | |
3935 | * index_cnt is ignored for everything but a dir, | |
3936 | * btrfs_get_inode_index_count has an explanation for the magic | |
3937 | * number | |
3938 | */ | |
e02119d5 | 3939 | init_btrfs_i(inode); |
aec7477b | 3940 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3941 | BTRFS_I(inode)->root = root; |
e02119d5 | 3942 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 3943 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 3944 | |
39279cc3 CM |
3945 | if (mode & S_IFDIR) |
3946 | owner = 0; | |
3947 | else | |
3948 | owner = 1; | |
d2fb3437 YZ |
3949 | BTRFS_I(inode)->block_group = |
3950 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
3951 | |
3952 | key[0].objectid = objectid; | |
3953 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3954 | key[0].offset = 0; | |
3955 | ||
3956 | key[1].objectid = objectid; | |
3957 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3958 | key[1].offset = ref_objectid; | |
3959 | ||
3960 | sizes[0] = sizeof(struct btrfs_inode_item); | |
3961 | sizes[1] = name_len + sizeof(*ref); | |
3962 | ||
b9473439 | 3963 | path->leave_spinning = 1; |
9c58309d CM |
3964 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
3965 | if (ret != 0) | |
5f39d397 CM |
3966 | goto fail; |
3967 | ||
79683f2d | 3968 | inode->i_uid = current_fsuid(); |
8c087b51 | 3969 | |
42f15d77 | 3970 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
3971 | inode->i_gid = dir->i_gid; |
3972 | if (S_ISDIR(mode)) | |
3973 | mode |= S_ISGID; | |
3974 | } else | |
3975 | inode->i_gid = current_fsgid(); | |
3976 | ||
39279cc3 CM |
3977 | inode->i_mode = mode; |
3978 | inode->i_ino = objectid; | |
a76a3cd4 | 3979 | inode_set_bytes(inode, 0); |
39279cc3 | 3980 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
3981 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3982 | struct btrfs_inode_item); | |
e02119d5 | 3983 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
3984 | |
3985 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3986 | struct btrfs_inode_ref); | |
3987 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 3988 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
3989 | ptr = (unsigned long)(ref + 1); |
3990 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
3991 | ||
5f39d397 CM |
3992 | btrfs_mark_buffer_dirty(path->nodes[0]); |
3993 | btrfs_free_path(path); | |
3994 | ||
39279cc3 CM |
3995 | location = &BTRFS_I(inode)->location; |
3996 | location->objectid = objectid; | |
39279cc3 CM |
3997 | location->offset = 0; |
3998 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
3999 | ||
6cbff00f CH |
4000 | btrfs_inherit_iflags(inode, dir); |
4001 | ||
94272164 CM |
4002 | if ((mode & S_IFREG)) { |
4003 | if (btrfs_test_opt(root, NODATASUM)) | |
4004 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4005 | if (btrfs_test_opt(root, NODATACOW)) | |
4006 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4007 | } | |
4008 | ||
39279cc3 | 4009 | insert_inode_hash(inode); |
5d4f98a2 | 4010 | inode_tree_add(inode); |
39279cc3 | 4011 | return inode; |
5f39d397 | 4012 | fail: |
aec7477b JB |
4013 | if (dir) |
4014 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4015 | btrfs_free_path(path); |
09771430 | 4016 | iput(inode); |
5f39d397 | 4017 | return ERR_PTR(ret); |
39279cc3 CM |
4018 | } |
4019 | ||
4020 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4021 | { | |
4022 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4023 | } | |
4024 | ||
d352ac68 CM |
4025 | /* |
4026 | * utility function to add 'inode' into 'parent_inode' with | |
4027 | * a give name and a given sequence number. | |
4028 | * if 'add_backref' is true, also insert a backref from the | |
4029 | * inode to the parent directory. | |
4030 | */ | |
e02119d5 CM |
4031 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4032 | struct inode *parent_inode, struct inode *inode, | |
4033 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4034 | { |
4df27c4d | 4035 | int ret = 0; |
39279cc3 | 4036 | struct btrfs_key key; |
e02119d5 | 4037 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4038 | |
4df27c4d YZ |
4039 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4040 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4041 | } else { | |
4042 | key.objectid = inode->i_ino; | |
4043 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4044 | key.offset = 0; | |
4045 | } | |
4046 | ||
4047 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4048 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4049 | key.objectid, root->root_key.objectid, | |
4050 | parent_inode->i_ino, | |
4051 | index, name, name_len); | |
4052 | } else if (add_backref) { | |
4053 | ret = btrfs_insert_inode_ref(trans, root, | |
4054 | name, name_len, inode->i_ino, | |
4055 | parent_inode->i_ino, index); | |
4056 | } | |
39279cc3 | 4057 | |
39279cc3 | 4058 | if (ret == 0) { |
4df27c4d YZ |
4059 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4060 | parent_inode->i_ino, &key, | |
4061 | btrfs_inode_type(inode), index); | |
4062 | BUG_ON(ret); | |
4063 | ||
dbe674a9 | 4064 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4065 | name_len * 2); |
79c44584 | 4066 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4067 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4068 | } |
4069 | return ret; | |
4070 | } | |
4071 | ||
4072 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4073 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4074 | int backref, u64 index) |
39279cc3 | 4075 | { |
e02119d5 CM |
4076 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4077 | inode, dentry->d_name.name, | |
4078 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4079 | if (!err) { |
4080 | d_instantiate(dentry, inode); | |
4081 | return 0; | |
4082 | } | |
4083 | if (err > 0) | |
4084 | err = -EEXIST; | |
4085 | return err; | |
4086 | } | |
4087 | ||
618e21d5 JB |
4088 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4089 | int mode, dev_t rdev) | |
4090 | { | |
4091 | struct btrfs_trans_handle *trans; | |
4092 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4093 | struct inode *inode = NULL; |
618e21d5 JB |
4094 | int err; |
4095 | int drop_inode = 0; | |
4096 | u64 objectid; | |
1832a6d5 | 4097 | unsigned long nr = 0; |
00e4e6b3 | 4098 | u64 index = 0; |
618e21d5 JB |
4099 | |
4100 | if (!new_valid_dev(rdev)) | |
4101 | return -EINVAL; | |
4102 | ||
9ed74f2d JB |
4103 | /* |
4104 | * 2 for inode item and ref | |
4105 | * 2 for dir items | |
4106 | * 1 for xattr if selinux is on | |
4107 | */ | |
4108 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4109 | if (err) |
9ed74f2d | 4110 | return err; |
1832a6d5 | 4111 | |
618e21d5 | 4112 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4113 | if (!trans) |
4114 | goto fail; | |
618e21d5 JB |
4115 | btrfs_set_trans_block_group(trans, dir); |
4116 | ||
4117 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4118 | if (err) { | |
4119 | err = -ENOSPC; | |
4120 | goto out_unlock; | |
4121 | } | |
4122 | ||
aec7477b | 4123 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4124 | dentry->d_name.len, |
4125 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4126 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4127 | err = PTR_ERR(inode); |
4128 | if (IS_ERR(inode)) | |
4129 | goto out_unlock; | |
4130 | ||
0279b4cd | 4131 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4132 | if (err) { |
4133 | drop_inode = 1; | |
4134 | goto out_unlock; | |
4135 | } | |
4136 | ||
618e21d5 | 4137 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4138 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4139 | if (err) |
4140 | drop_inode = 1; | |
4141 | else { | |
4142 | inode->i_op = &btrfs_special_inode_operations; | |
4143 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4144 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4145 | } |
618e21d5 JB |
4146 | btrfs_update_inode_block_group(trans, inode); |
4147 | btrfs_update_inode_block_group(trans, dir); | |
4148 | out_unlock: | |
d3c2fdcf | 4149 | nr = trans->blocks_used; |
89ce8a63 | 4150 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4151 | fail: |
9ed74f2d | 4152 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4153 | if (drop_inode) { |
4154 | inode_dec_link_count(inode); | |
4155 | iput(inode); | |
4156 | } | |
d3c2fdcf | 4157 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4158 | return err; |
4159 | } | |
4160 | ||
39279cc3 CM |
4161 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4162 | int mode, struct nameidata *nd) | |
4163 | { | |
4164 | struct btrfs_trans_handle *trans; | |
4165 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4166 | struct inode *inode = NULL; |
39279cc3 CM |
4167 | int err; |
4168 | int drop_inode = 0; | |
1832a6d5 | 4169 | unsigned long nr = 0; |
39279cc3 | 4170 | u64 objectid; |
00e4e6b3 | 4171 | u64 index = 0; |
39279cc3 | 4172 | |
9ed74f2d JB |
4173 | /* |
4174 | * 2 for inode item and ref | |
4175 | * 2 for dir items | |
4176 | * 1 for xattr if selinux is on | |
4177 | */ | |
4178 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4179 | if (err) |
9ed74f2d JB |
4180 | return err; |
4181 | ||
39279cc3 | 4182 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4183 | if (!trans) |
4184 | goto fail; | |
39279cc3 CM |
4185 | btrfs_set_trans_block_group(trans, dir); |
4186 | ||
4187 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4188 | if (err) { | |
4189 | err = -ENOSPC; | |
4190 | goto out_unlock; | |
4191 | } | |
4192 | ||
aec7477b | 4193 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4194 | dentry->d_name.len, |
4195 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4196 | objectid, BTRFS_I(dir)->block_group, mode, |
4197 | &index); | |
39279cc3 CM |
4198 | err = PTR_ERR(inode); |
4199 | if (IS_ERR(inode)) | |
4200 | goto out_unlock; | |
4201 | ||
0279b4cd | 4202 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4203 | if (err) { |
4204 | drop_inode = 1; | |
4205 | goto out_unlock; | |
4206 | } | |
4207 | ||
39279cc3 | 4208 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4209 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4210 | if (err) |
4211 | drop_inode = 1; | |
4212 | else { | |
4213 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4214 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4215 | inode->i_fop = &btrfs_file_operations; |
4216 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4217 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4218 | } |
39279cc3 CM |
4219 | btrfs_update_inode_block_group(trans, inode); |
4220 | btrfs_update_inode_block_group(trans, dir); | |
4221 | out_unlock: | |
d3c2fdcf | 4222 | nr = trans->blocks_used; |
ab78c84d | 4223 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4224 | fail: |
9ed74f2d | 4225 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4226 | if (drop_inode) { |
4227 | inode_dec_link_count(inode); | |
4228 | iput(inode); | |
4229 | } | |
d3c2fdcf | 4230 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4231 | return err; |
4232 | } | |
4233 | ||
4234 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4235 | struct dentry *dentry) | |
4236 | { | |
4237 | struct btrfs_trans_handle *trans; | |
4238 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4239 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4240 | u64 index; |
1832a6d5 | 4241 | unsigned long nr = 0; |
39279cc3 CM |
4242 | int err; |
4243 | int drop_inode = 0; | |
4244 | ||
4245 | if (inode->i_nlink == 0) | |
4246 | return -ENOENT; | |
4247 | ||
9ed74f2d JB |
4248 | /* |
4249 | * 1 item for inode ref | |
4250 | * 2 items for dir items | |
4251 | */ | |
4252 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4253 | if (err) |
9ed74f2d JB |
4254 | return err; |
4255 | ||
4256 | btrfs_inc_nlink(inode); | |
4257 | ||
3de4586c | 4258 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4259 | if (err) |
4260 | goto fail; | |
4261 | ||
39279cc3 | 4262 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4263 | |
39279cc3 CM |
4264 | btrfs_set_trans_block_group(trans, dir); |
4265 | atomic_inc(&inode->i_count); | |
aec7477b | 4266 | |
00e4e6b3 | 4267 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4268 | |
a5719521 | 4269 | if (err) { |
54aa1f4d | 4270 | drop_inode = 1; |
a5719521 YZ |
4271 | } else { |
4272 | btrfs_update_inode_block_group(trans, dir); | |
4273 | err = btrfs_update_inode(trans, root, inode); | |
4274 | BUG_ON(err); | |
4275 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4276 | } | |
39279cc3 | 4277 | |
d3c2fdcf | 4278 | nr = trans->blocks_used; |
ab78c84d | 4279 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4280 | fail: |
9ed74f2d | 4281 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4282 | if (drop_inode) { |
4283 | inode_dec_link_count(inode); | |
4284 | iput(inode); | |
4285 | } | |
d3c2fdcf | 4286 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4287 | return err; |
4288 | } | |
4289 | ||
39279cc3 CM |
4290 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4291 | { | |
b9d86667 | 4292 | struct inode *inode = NULL; |
39279cc3 CM |
4293 | struct btrfs_trans_handle *trans; |
4294 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4295 | int err = 0; | |
4296 | int drop_on_err = 0; | |
b9d86667 | 4297 | u64 objectid = 0; |
00e4e6b3 | 4298 | u64 index = 0; |
d3c2fdcf | 4299 | unsigned long nr = 1; |
39279cc3 | 4300 | |
9ed74f2d JB |
4301 | /* |
4302 | * 2 items for inode and ref | |
4303 | * 2 items for dir items | |
4304 | * 1 for xattr if selinux is on | |
4305 | */ | |
4306 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4307 | if (err) |
9ed74f2d | 4308 | return err; |
1832a6d5 | 4309 | |
39279cc3 | 4310 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4311 | if (!trans) { |
4312 | err = -ENOMEM; | |
39279cc3 CM |
4313 | goto out_unlock; |
4314 | } | |
9ed74f2d | 4315 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4316 | |
4317 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4318 | if (err) { | |
4319 | err = -ENOSPC; | |
4320 | goto out_unlock; | |
4321 | } | |
4322 | ||
aec7477b | 4323 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4324 | dentry->d_name.len, |
4325 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4326 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4327 | &index); | |
39279cc3 CM |
4328 | if (IS_ERR(inode)) { |
4329 | err = PTR_ERR(inode); | |
4330 | goto out_fail; | |
4331 | } | |
5f39d397 | 4332 | |
39279cc3 | 4333 | drop_on_err = 1; |
33268eaf | 4334 | |
0279b4cd | 4335 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4336 | if (err) |
4337 | goto out_fail; | |
4338 | ||
39279cc3 CM |
4339 | inode->i_op = &btrfs_dir_inode_operations; |
4340 | inode->i_fop = &btrfs_dir_file_operations; | |
4341 | btrfs_set_trans_block_group(trans, inode); | |
4342 | ||
dbe674a9 | 4343 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4344 | err = btrfs_update_inode(trans, root, inode); |
4345 | if (err) | |
4346 | goto out_fail; | |
5f39d397 | 4347 | |
e02119d5 CM |
4348 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4349 | inode, dentry->d_name.name, | |
4350 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4351 | if (err) |
4352 | goto out_fail; | |
5f39d397 | 4353 | |
39279cc3 CM |
4354 | d_instantiate(dentry, inode); |
4355 | drop_on_err = 0; | |
39279cc3 CM |
4356 | btrfs_update_inode_block_group(trans, inode); |
4357 | btrfs_update_inode_block_group(trans, dir); | |
4358 | ||
4359 | out_fail: | |
d3c2fdcf | 4360 | nr = trans->blocks_used; |
ab78c84d | 4361 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4362 | |
39279cc3 | 4363 | out_unlock: |
9ed74f2d | 4364 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4365 | if (drop_on_err) |
4366 | iput(inode); | |
d3c2fdcf | 4367 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4368 | return err; |
4369 | } | |
4370 | ||
d352ac68 CM |
4371 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4372 | * and an extent that you want to insert, deal with overlap and insert | |
4373 | * the new extent into the tree. | |
4374 | */ | |
3b951516 CM |
4375 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4376 | struct extent_map *existing, | |
e6dcd2dc CM |
4377 | struct extent_map *em, |
4378 | u64 map_start, u64 map_len) | |
3b951516 CM |
4379 | { |
4380 | u64 start_diff; | |
3b951516 | 4381 | |
e6dcd2dc CM |
4382 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4383 | start_diff = map_start - em->start; | |
4384 | em->start = map_start; | |
4385 | em->len = map_len; | |
c8b97818 CM |
4386 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4387 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4388 | em->block_start += start_diff; |
c8b97818 CM |
4389 | em->block_len -= start_diff; |
4390 | } | |
e6dcd2dc | 4391 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4392 | } |
4393 | ||
c8b97818 CM |
4394 | static noinline int uncompress_inline(struct btrfs_path *path, |
4395 | struct inode *inode, struct page *page, | |
4396 | size_t pg_offset, u64 extent_offset, | |
4397 | struct btrfs_file_extent_item *item) | |
4398 | { | |
4399 | int ret; | |
4400 | struct extent_buffer *leaf = path->nodes[0]; | |
4401 | char *tmp; | |
4402 | size_t max_size; | |
4403 | unsigned long inline_size; | |
4404 | unsigned long ptr; | |
4405 | ||
4406 | WARN_ON(pg_offset != 0); | |
4407 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4408 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4409 | btrfs_item_nr(leaf, path->slots[0])); | |
4410 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4411 | ptr = btrfs_file_extent_inline_start(item); | |
4412 | ||
4413 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4414 | ||
5b050f04 | 4415 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4416 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4417 | inline_size, max_size); | |
4418 | if (ret) { | |
4419 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4420 | unsigned long copy_size = min_t(u64, | |
4421 | PAGE_CACHE_SIZE - pg_offset, | |
4422 | max_size - extent_offset); | |
4423 | memset(kaddr + pg_offset, 0, copy_size); | |
4424 | kunmap_atomic(kaddr, KM_USER0); | |
4425 | } | |
4426 | kfree(tmp); | |
4427 | return 0; | |
4428 | } | |
4429 | ||
d352ac68 CM |
4430 | /* |
4431 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4432 | * the ugly parts come from merging extents from the disk with the in-ram |
4433 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4434 | * where the in-ram extents might be locked pending data=ordered completion. |
4435 | * | |
4436 | * This also copies inline extents directly into the page. | |
4437 | */ | |
d397712b | 4438 | |
a52d9a80 | 4439 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4440 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4441 | int create) |
4442 | { | |
4443 | int ret; | |
4444 | int err = 0; | |
db94535d | 4445 | u64 bytenr; |
a52d9a80 CM |
4446 | u64 extent_start = 0; |
4447 | u64 extent_end = 0; | |
4448 | u64 objectid = inode->i_ino; | |
4449 | u32 found_type; | |
f421950f | 4450 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4451 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4452 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4453 | struct extent_buffer *leaf; |
4454 | struct btrfs_key found_key; | |
a52d9a80 CM |
4455 | struct extent_map *em = NULL; |
4456 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4457 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4458 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4459 | int compressed; |
a52d9a80 | 4460 | |
a52d9a80 | 4461 | again: |
890871be | 4462 | read_lock(&em_tree->lock); |
d1310b2e | 4463 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4464 | if (em) |
4465 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4466 | read_unlock(&em_tree->lock); |
d1310b2e | 4467 | |
a52d9a80 | 4468 | if (em) { |
e1c4b745 CM |
4469 | if (em->start > start || em->start + em->len <= start) |
4470 | free_extent_map(em); | |
4471 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4472 | free_extent_map(em); |
4473 | else | |
4474 | goto out; | |
a52d9a80 | 4475 | } |
d1310b2e | 4476 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4477 | if (!em) { |
d1310b2e CM |
4478 | err = -ENOMEM; |
4479 | goto out; | |
a52d9a80 | 4480 | } |
e6dcd2dc | 4481 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4482 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4483 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4484 | em->len = (u64)-1; |
c8b97818 | 4485 | em->block_len = (u64)-1; |
f421950f CM |
4486 | |
4487 | if (!path) { | |
4488 | path = btrfs_alloc_path(); | |
4489 | BUG_ON(!path); | |
4490 | } | |
4491 | ||
179e29e4 CM |
4492 | ret = btrfs_lookup_file_extent(trans, root, path, |
4493 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4494 | if (ret < 0) { |
4495 | err = ret; | |
4496 | goto out; | |
4497 | } | |
4498 | ||
4499 | if (ret != 0) { | |
4500 | if (path->slots[0] == 0) | |
4501 | goto not_found; | |
4502 | path->slots[0]--; | |
4503 | } | |
4504 | ||
5f39d397 CM |
4505 | leaf = path->nodes[0]; |
4506 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4507 | struct btrfs_file_extent_item); |
a52d9a80 | 4508 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4509 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4510 | found_type = btrfs_key_type(&found_key); | |
4511 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4512 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4513 | goto not_found; | |
4514 | } | |
4515 | ||
5f39d397 CM |
4516 | found_type = btrfs_file_extent_type(leaf, item); |
4517 | extent_start = found_key.offset; | |
c8b97818 | 4518 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4519 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4520 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4521 | extent_end = extent_start + |
db94535d | 4522 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4523 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4524 | size_t size; | |
4525 | size = btrfs_file_extent_inline_len(leaf, item); | |
4526 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4527 | ~((u64)root->sectorsize - 1); | |
4528 | } | |
4529 | ||
4530 | if (start >= extent_end) { | |
4531 | path->slots[0]++; | |
4532 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4533 | ret = btrfs_next_leaf(root, path); | |
4534 | if (ret < 0) { | |
4535 | err = ret; | |
4536 | goto out; | |
a52d9a80 | 4537 | } |
9036c102 YZ |
4538 | if (ret > 0) |
4539 | goto not_found; | |
4540 | leaf = path->nodes[0]; | |
a52d9a80 | 4541 | } |
9036c102 YZ |
4542 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4543 | if (found_key.objectid != objectid || | |
4544 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4545 | goto not_found; | |
4546 | if (start + len <= found_key.offset) | |
4547 | goto not_found; | |
4548 | em->start = start; | |
4549 | em->len = found_key.offset - start; | |
4550 | goto not_found_em; | |
4551 | } | |
4552 | ||
d899e052 YZ |
4553 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4554 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4555 | em->start = extent_start; |
4556 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4557 | em->orig_start = extent_start - |
4558 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4559 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4560 | if (bytenr == 0) { | |
5f39d397 | 4561 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4562 | goto insert; |
4563 | } | |
c8b97818 CM |
4564 | if (compressed) { |
4565 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4566 | em->block_start = bytenr; | |
4567 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4568 | item); | |
4569 | } else { | |
4570 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4571 | em->block_start = bytenr; | |
4572 | em->block_len = em->len; | |
d899e052 YZ |
4573 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4574 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4575 | } |
a52d9a80 CM |
4576 | goto insert; |
4577 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4578 | unsigned long ptr; |
a52d9a80 | 4579 | char *map; |
3326d1b0 CM |
4580 | size_t size; |
4581 | size_t extent_offset; | |
4582 | size_t copy_size; | |
a52d9a80 | 4583 | |
689f9346 | 4584 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4585 | if (!page || create) { |
689f9346 | 4586 | em->start = extent_start; |
9036c102 | 4587 | em->len = extent_end - extent_start; |
689f9346 Y |
4588 | goto out; |
4589 | } | |
5f39d397 | 4590 | |
9036c102 YZ |
4591 | size = btrfs_file_extent_inline_len(leaf, item); |
4592 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4593 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4594 | size - extent_offset); |
3326d1b0 | 4595 | em->start = extent_start + extent_offset; |
70dec807 CM |
4596 | em->len = (copy_size + root->sectorsize - 1) & |
4597 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4598 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4599 | if (compressed) |
4600 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4601 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4602 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4603 | if (btrfs_file_extent_compression(leaf, item) == |
4604 | BTRFS_COMPRESS_ZLIB) { | |
4605 | ret = uncompress_inline(path, inode, page, | |
4606 | pg_offset, | |
4607 | extent_offset, item); | |
4608 | BUG_ON(ret); | |
4609 | } else { | |
4610 | map = kmap(page); | |
4611 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4612 | copy_size); | |
93c82d57 CM |
4613 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4614 | memset(map + pg_offset + copy_size, 0, | |
4615 | PAGE_CACHE_SIZE - pg_offset - | |
4616 | copy_size); | |
4617 | } | |
c8b97818 CM |
4618 | kunmap(page); |
4619 | } | |
179e29e4 CM |
4620 | flush_dcache_page(page); |
4621 | } else if (create && PageUptodate(page)) { | |
4622 | if (!trans) { | |
4623 | kunmap(page); | |
4624 | free_extent_map(em); | |
4625 | em = NULL; | |
4626 | btrfs_release_path(root, path); | |
f9295749 | 4627 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4628 | goto again; |
4629 | } | |
c8b97818 | 4630 | map = kmap(page); |
70dec807 | 4631 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4632 | copy_size); |
c8b97818 | 4633 | kunmap(page); |
179e29e4 | 4634 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4635 | } |
d1310b2e CM |
4636 | set_extent_uptodate(io_tree, em->start, |
4637 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4638 | goto insert; |
4639 | } else { | |
d397712b | 4640 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4641 | WARN_ON(1); |
4642 | } | |
4643 | not_found: | |
4644 | em->start = start; | |
d1310b2e | 4645 | em->len = len; |
a52d9a80 | 4646 | not_found_em: |
5f39d397 | 4647 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4648 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4649 | insert: |
4650 | btrfs_release_path(root, path); | |
d1310b2e | 4651 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4652 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4653 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4654 | (unsigned long long)em->len, | |
4655 | (unsigned long long)start, | |
4656 | (unsigned long long)len); | |
a52d9a80 CM |
4657 | err = -EIO; |
4658 | goto out; | |
4659 | } | |
d1310b2e CM |
4660 | |
4661 | err = 0; | |
890871be | 4662 | write_lock(&em_tree->lock); |
a52d9a80 | 4663 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4664 | /* it is possible that someone inserted the extent into the tree |
4665 | * while we had the lock dropped. It is also possible that | |
4666 | * an overlapping map exists in the tree | |
4667 | */ | |
a52d9a80 | 4668 | if (ret == -EEXIST) { |
3b951516 | 4669 | struct extent_map *existing; |
e6dcd2dc CM |
4670 | |
4671 | ret = 0; | |
4672 | ||
3b951516 | 4673 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4674 | if (existing && (existing->start > start || |
4675 | existing->start + existing->len <= start)) { | |
4676 | free_extent_map(existing); | |
4677 | existing = NULL; | |
4678 | } | |
3b951516 CM |
4679 | if (!existing) { |
4680 | existing = lookup_extent_mapping(em_tree, em->start, | |
4681 | em->len); | |
4682 | if (existing) { | |
4683 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4684 | em, start, |
4685 | root->sectorsize); | |
3b951516 CM |
4686 | free_extent_map(existing); |
4687 | if (err) { | |
4688 | free_extent_map(em); | |
4689 | em = NULL; | |
4690 | } | |
4691 | } else { | |
4692 | err = -EIO; | |
3b951516 CM |
4693 | free_extent_map(em); |
4694 | em = NULL; | |
4695 | } | |
4696 | } else { | |
4697 | free_extent_map(em); | |
4698 | em = existing; | |
e6dcd2dc | 4699 | err = 0; |
a52d9a80 | 4700 | } |
a52d9a80 | 4701 | } |
890871be | 4702 | write_unlock(&em_tree->lock); |
a52d9a80 | 4703 | out: |
f421950f CM |
4704 | if (path) |
4705 | btrfs_free_path(path); | |
a52d9a80 CM |
4706 | if (trans) { |
4707 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4708 | if (!err) |
a52d9a80 CM |
4709 | err = ret; |
4710 | } | |
a52d9a80 CM |
4711 | if (err) { |
4712 | free_extent_map(em); | |
a52d9a80 CM |
4713 | return ERR_PTR(err); |
4714 | } | |
4715 | return em; | |
4716 | } | |
4717 | ||
16432985 CM |
4718 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4719 | const struct iovec *iov, loff_t offset, | |
4720 | unsigned long nr_segs) | |
4721 | { | |
e1c4b745 | 4722 | return -EINVAL; |
16432985 CM |
4723 | } |
4724 | ||
1506fcc8 YS |
4725 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4726 | __u64 start, __u64 len) | |
4727 | { | |
4728 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4729 | } | |
4730 | ||
a52d9a80 | 4731 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4732 | { |
d1310b2e CM |
4733 | struct extent_io_tree *tree; |
4734 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4735 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4736 | } |
1832a6d5 | 4737 | |
a52d9a80 | 4738 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4739 | { |
d1310b2e | 4740 | struct extent_io_tree *tree; |
b888db2b CM |
4741 | |
4742 | ||
4743 | if (current->flags & PF_MEMALLOC) { | |
4744 | redirty_page_for_writepage(wbc, page); | |
4745 | unlock_page(page); | |
4746 | return 0; | |
4747 | } | |
d1310b2e | 4748 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4749 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4750 | } |
4751 | ||
f421950f CM |
4752 | int btrfs_writepages(struct address_space *mapping, |
4753 | struct writeback_control *wbc) | |
b293f02e | 4754 | { |
d1310b2e | 4755 | struct extent_io_tree *tree; |
771ed689 | 4756 | |
d1310b2e | 4757 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4758 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4759 | } | |
4760 | ||
3ab2fb5a CM |
4761 | static int |
4762 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4763 | struct list_head *pages, unsigned nr_pages) | |
4764 | { | |
d1310b2e CM |
4765 | struct extent_io_tree *tree; |
4766 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4767 | return extent_readpages(tree, mapping, pages, nr_pages, |
4768 | btrfs_get_extent); | |
4769 | } | |
e6dcd2dc | 4770 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4771 | { |
d1310b2e CM |
4772 | struct extent_io_tree *tree; |
4773 | struct extent_map_tree *map; | |
a52d9a80 | 4774 | int ret; |
8c2383c3 | 4775 | |
d1310b2e CM |
4776 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4777 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4778 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4779 | if (ret == 1) { |
4780 | ClearPagePrivate(page); | |
4781 | set_page_private(page, 0); | |
4782 | page_cache_release(page); | |
39279cc3 | 4783 | } |
a52d9a80 | 4784 | return ret; |
39279cc3 CM |
4785 | } |
4786 | ||
e6dcd2dc CM |
4787 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4788 | { | |
98509cfc CM |
4789 | if (PageWriteback(page) || PageDirty(page)) |
4790 | return 0; | |
b335b003 | 4791 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4792 | } |
4793 | ||
a52d9a80 | 4794 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4795 | { |
d1310b2e | 4796 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4797 | struct btrfs_ordered_extent *ordered; |
4798 | u64 page_start = page_offset(page); | |
4799 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4800 | |
8b62b72b CM |
4801 | |
4802 | /* | |
4803 | * we have the page locked, so new writeback can't start, | |
4804 | * and the dirty bit won't be cleared while we are here. | |
4805 | * | |
4806 | * Wait for IO on this page so that we can safely clear | |
4807 | * the PagePrivate2 bit and do ordered accounting | |
4808 | */ | |
e6dcd2dc | 4809 | wait_on_page_writeback(page); |
8b62b72b | 4810 | |
d1310b2e | 4811 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4812 | if (offset) { |
4813 | btrfs_releasepage(page, GFP_NOFS); | |
4814 | return; | |
4815 | } | |
e6dcd2dc CM |
4816 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4817 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4818 | page_offset(page)); | |
4819 | if (ordered) { | |
eb84ae03 CM |
4820 | /* |
4821 | * IO on this page will never be started, so we need | |
4822 | * to account for any ordered extents now | |
4823 | */ | |
e6dcd2dc CM |
4824 | clear_extent_bit(tree, page_start, page_end, |
4825 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
2c64c53d | 4826 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
8b62b72b CM |
4827 | /* |
4828 | * whoever cleared the private bit is responsible | |
4829 | * for the finish_ordered_io | |
4830 | */ | |
4831 | if (TestClearPagePrivate2(page)) { | |
4832 | btrfs_finish_ordered_io(page->mapping->host, | |
4833 | page_start, page_end); | |
4834 | } | |
e6dcd2dc CM |
4835 | btrfs_put_ordered_extent(ordered); |
4836 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4837 | } | |
4838 | clear_extent_bit(tree, page_start, page_end, | |
8b62b72b | 4839 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, |
2c64c53d | 4840 | 1, 1, NULL, GFP_NOFS); |
e6dcd2dc CM |
4841 | __btrfs_releasepage(page, GFP_NOFS); |
4842 | ||
4a096752 | 4843 | ClearPageChecked(page); |
9ad6b7bc | 4844 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4845 | ClearPagePrivate(page); |
4846 | set_page_private(page, 0); | |
4847 | page_cache_release(page); | |
4848 | } | |
39279cc3 CM |
4849 | } |
4850 | ||
9ebefb18 CM |
4851 | /* |
4852 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4853 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4854 | * be careful to check for EOF conditions here. We set the page up correctly | |
4855 | * for a written page which means we get ENOSPC checking when writing into | |
4856 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4857 | * support these features. | |
4858 | * | |
4859 | * We are not allowed to take the i_mutex here so we have to play games to | |
4860 | * protect against truncate races as the page could now be beyond EOF. Because | |
4861 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4862 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4863 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4864 | * unlock the page. | |
4865 | */ | |
c2ec175c | 4866 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4867 | { |
c2ec175c | 4868 | struct page *page = vmf->page; |
6da6abae | 4869 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4870 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4871 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4872 | struct btrfs_ordered_extent *ordered; | |
4873 | char *kaddr; | |
4874 | unsigned long zero_start; | |
9ebefb18 | 4875 | loff_t size; |
1832a6d5 | 4876 | int ret; |
a52d9a80 | 4877 | u64 page_start; |
e6dcd2dc | 4878 | u64 page_end; |
9ebefb18 | 4879 | |
6a63209f | 4880 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4881 | if (ret) { |
4882 | if (ret == -ENOMEM) | |
4883 | ret = VM_FAULT_OOM; | |
4884 | else /* -ENOSPC, -EIO, etc */ | |
4885 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4886 | goto out; |
56a76f82 | 4887 | } |
1832a6d5 | 4888 | |
9ed74f2d JB |
4889 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
4890 | if (ret) { | |
4891 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
4892 | ret = VM_FAULT_SIGBUS; | |
4893 | goto out; | |
4894 | } | |
4895 | ||
56a76f82 | 4896 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4897 | again: |
9ebefb18 | 4898 | lock_page(page); |
9ebefb18 | 4899 | size = i_size_read(inode); |
e6dcd2dc CM |
4900 | page_start = page_offset(page); |
4901 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4902 | |
9ebefb18 | 4903 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4904 | (page_start >= size)) { |
6a63209f | 4905 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
4906 | /* page got truncated out from underneath us */ |
4907 | goto out_unlock; | |
4908 | } | |
e6dcd2dc CM |
4909 | wait_on_page_writeback(page); |
4910 | ||
4911 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4912 | set_page_extent_mapped(page); | |
4913 | ||
eb84ae03 CM |
4914 | /* |
4915 | * we can't set the delalloc bits if there are pending ordered | |
4916 | * extents. Drop our locks and wait for them to finish | |
4917 | */ | |
e6dcd2dc CM |
4918 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4919 | if (ordered) { | |
4920 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4921 | unlock_page(page); | |
eb84ae03 | 4922 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4923 | btrfs_put_ordered_extent(ordered); |
4924 | goto again; | |
4925 | } | |
4926 | ||
fbf19087 JB |
4927 | /* |
4928 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
4929 | * if it was already dirty, so for space accounting reasons we need to | |
4930 | * clear any delalloc bits for the range we are fixing to save. There | |
4931 | * is probably a better way to do this, but for now keep consistent with | |
4932 | * prepare_pages in the normal write path. | |
4933 | */ | |
4934 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
4935 | EXTENT_DIRTY | EXTENT_DELALLOC, GFP_NOFS); | |
4936 | ||
9ed74f2d JB |
4937 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
4938 | if (ret) { | |
4939 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4940 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 4941 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
4942 | goto out_unlock; |
4943 | } | |
e6dcd2dc | 4944 | ret = 0; |
9ebefb18 CM |
4945 | |
4946 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4947 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4948 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4949 | else |
e6dcd2dc | 4950 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4951 | |
e6dcd2dc CM |
4952 | if (zero_start != PAGE_CACHE_SIZE) { |
4953 | kaddr = kmap(page); | |
4954 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4955 | flush_dcache_page(page); | |
4956 | kunmap(page); | |
4957 | } | |
247e743c | 4958 | ClearPageChecked(page); |
e6dcd2dc | 4959 | set_page_dirty(page); |
50a9b214 | 4960 | SetPageUptodate(page); |
5a3f23d5 CM |
4961 | |
4962 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
e6dcd2dc | 4963 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
4964 | |
4965 | out_unlock: | |
9ed74f2d | 4966 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
4967 | if (!ret) |
4968 | return VM_FAULT_LOCKED; | |
9ebefb18 | 4969 | unlock_page(page); |
1832a6d5 | 4970 | out: |
9ebefb18 CM |
4971 | return ret; |
4972 | } | |
4973 | ||
39279cc3 CM |
4974 | static void btrfs_truncate(struct inode *inode) |
4975 | { | |
4976 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4977 | int ret; | |
4978 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 4979 | unsigned long nr; |
dbe674a9 | 4980 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
4981 | |
4982 | if (!S_ISREG(inode->i_mode)) | |
4983 | return; | |
4984 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4985 | return; | |
4986 | ||
4987 | btrfs_truncate_page(inode->i_mapping, inode->i_size); | |
4a096752 | 4988 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 4989 | |
39279cc3 | 4990 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
4991 | |
4992 | /* | |
4993 | * setattr is responsible for setting the ordered_data_close flag, | |
4994 | * but that is only tested during the last file release. That | |
4995 | * could happen well after the next commit, leaving a great big | |
4996 | * window where new writes may get lost if someone chooses to write | |
4997 | * to this file after truncating to zero | |
4998 | * | |
4999 | * The inode doesn't have any dirty data here, and so if we commit | |
5000 | * this is a noop. If someone immediately starts writing to the inode | |
5001 | * it is very likely we'll catch some of their writes in this | |
5002 | * transaction, and the commit will find this file on the ordered | |
5003 | * data list with good things to send down. | |
5004 | * | |
5005 | * This is a best effort solution, there is still a window where | |
5006 | * using truncate to replace the contents of the file will | |
5007 | * end up with a zero length file after a crash. | |
5008 | */ | |
5009 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5010 | btrfs_add_ordered_operation(trans, root, inode); | |
5011 | ||
39279cc3 | 5012 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 5013 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 5014 | |
7b128766 JB |
5015 | ret = btrfs_orphan_add(trans, inode); |
5016 | if (ret) | |
5017 | goto out; | |
39279cc3 | 5018 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 5019 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 5020 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 5021 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 5022 | |
7b128766 JB |
5023 | ret = btrfs_orphan_del(trans, inode); |
5024 | BUG_ON(ret); | |
5025 | ||
5026 | out: | |
5027 | nr = trans->blocks_used; | |
89ce8a63 | 5028 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5029 | BUG_ON(ret); |
d3c2fdcf | 5030 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5031 | } |
5032 | ||
d352ac68 CM |
5033 | /* |
5034 | * create a new subvolume directory/inode (helper for the ioctl). | |
5035 | */ | |
d2fb3437 | 5036 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5037 | struct btrfs_root *new_root, |
d2fb3437 | 5038 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5039 | { |
39279cc3 | 5040 | struct inode *inode; |
76dda93c | 5041 | int err; |
00e4e6b3 | 5042 | u64 index = 0; |
39279cc3 | 5043 | |
aec7477b | 5044 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5045 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5046 | if (IS_ERR(inode)) |
f46b5a66 | 5047 | return PTR_ERR(inode); |
39279cc3 CM |
5048 | inode->i_op = &btrfs_dir_inode_operations; |
5049 | inode->i_fop = &btrfs_dir_file_operations; | |
5050 | ||
39279cc3 | 5051 | inode->i_nlink = 1; |
dbe674a9 | 5052 | btrfs_i_size_write(inode, 0); |
3b96362c | 5053 | |
76dda93c YZ |
5054 | err = btrfs_update_inode(trans, new_root, inode); |
5055 | BUG_ON(err); | |
cb8e7090 | 5056 | |
76dda93c | 5057 | iput(inode); |
cb8e7090 | 5058 | return 0; |
39279cc3 CM |
5059 | } |
5060 | ||
d352ac68 CM |
5061 | /* helper function for file defrag and space balancing. This |
5062 | * forces readahead on a given range of bytes in an inode | |
5063 | */ | |
edbd8d4e | 5064 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5065 | struct file_ra_state *ra, struct file *file, |
5066 | pgoff_t offset, pgoff_t last_index) | |
5067 | { | |
8e7bf94f | 5068 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5069 | |
86479a04 CM |
5070 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5071 | return offset + req_size; | |
86479a04 CM |
5072 | } |
5073 | ||
39279cc3 CM |
5074 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5075 | { | |
5076 | struct btrfs_inode *ei; | |
5077 | ||
5078 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5079 | if (!ei) | |
5080 | return NULL; | |
15ee9bc7 | 5081 | ei->last_trans = 0; |
e02119d5 | 5082 | ei->logged_trans = 0; |
9ed74f2d JB |
5083 | ei->delalloc_extents = 0; |
5084 | ei->delalloc_reserved_extents = 0; | |
e6dcd2dc | 5085 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5086 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5087 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5088 | return &ei->vfs_inode; |
5089 | } | |
5090 | ||
5091 | void btrfs_destroy_inode(struct inode *inode) | |
5092 | { | |
e6dcd2dc | 5093 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5094 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5095 | ||
39279cc3 CM |
5096 | WARN_ON(!list_empty(&inode->i_dentry)); |
5097 | WARN_ON(inode->i_data.nrpages); | |
5098 | ||
5a3f23d5 CM |
5099 | /* |
5100 | * Make sure we're properly removed from the ordered operation | |
5101 | * lists. | |
5102 | */ | |
5103 | smp_mb(); | |
5104 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5105 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5106 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5107 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5108 | } | |
5109 | ||
5110 | spin_lock(&root->list_lock); | |
7b128766 JB |
5111 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
5112 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
5113 | " list\n", inode->i_ino); | |
5114 | dump_stack(); | |
5115 | } | |
5a3f23d5 | 5116 | spin_unlock(&root->list_lock); |
7b128766 | 5117 | |
d397712b | 5118 | while (1) { |
e6dcd2dc CM |
5119 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5120 | if (!ordered) | |
5121 | break; | |
5122 | else { | |
d397712b CM |
5123 | printk(KERN_ERR "btrfs found ordered " |
5124 | "extent %llu %llu on inode cleanup\n", | |
5125 | (unsigned long long)ordered->file_offset, | |
5126 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5127 | btrfs_remove_ordered_extent(inode, ordered); |
5128 | btrfs_put_ordered_extent(ordered); | |
5129 | btrfs_put_ordered_extent(ordered); | |
5130 | } | |
5131 | } | |
5d4f98a2 | 5132 | inode_tree_del(inode); |
5b21f2ed | 5133 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
5134 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5135 | } | |
5136 | ||
76dda93c YZ |
5137 | void btrfs_drop_inode(struct inode *inode) |
5138 | { | |
5139 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5140 | ||
5141 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5142 | generic_delete_inode(inode); | |
5143 | else | |
5144 | generic_drop_inode(inode); | |
5145 | } | |
5146 | ||
0ee0fda0 | 5147 | static void init_once(void *foo) |
39279cc3 CM |
5148 | { |
5149 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5150 | ||
5151 | inode_init_once(&ei->vfs_inode); | |
5152 | } | |
5153 | ||
5154 | void btrfs_destroy_cachep(void) | |
5155 | { | |
5156 | if (btrfs_inode_cachep) | |
5157 | kmem_cache_destroy(btrfs_inode_cachep); | |
5158 | if (btrfs_trans_handle_cachep) | |
5159 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5160 | if (btrfs_transaction_cachep) | |
5161 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5162 | if (btrfs_path_cachep) |
5163 | kmem_cache_destroy(btrfs_path_cachep); | |
5164 | } | |
5165 | ||
5166 | int btrfs_init_cachep(void) | |
5167 | { | |
9601e3f6 CH |
5168 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5169 | sizeof(struct btrfs_inode), 0, | |
5170 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5171 | if (!btrfs_inode_cachep) |
5172 | goto fail; | |
9601e3f6 CH |
5173 | |
5174 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5175 | sizeof(struct btrfs_trans_handle), 0, | |
5176 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5177 | if (!btrfs_trans_handle_cachep) |
5178 | goto fail; | |
9601e3f6 CH |
5179 | |
5180 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5181 | sizeof(struct btrfs_transaction), 0, | |
5182 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5183 | if (!btrfs_transaction_cachep) |
5184 | goto fail; | |
9601e3f6 CH |
5185 | |
5186 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5187 | sizeof(struct btrfs_path), 0, | |
5188 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5189 | if (!btrfs_path_cachep) |
5190 | goto fail; | |
9601e3f6 | 5191 | |
39279cc3 CM |
5192 | return 0; |
5193 | fail: | |
5194 | btrfs_destroy_cachep(); | |
5195 | return -ENOMEM; | |
5196 | } | |
5197 | ||
5198 | static int btrfs_getattr(struct vfsmount *mnt, | |
5199 | struct dentry *dentry, struct kstat *stat) | |
5200 | { | |
5201 | struct inode *inode = dentry->d_inode; | |
5202 | generic_fillattr(inode, stat); | |
3394e160 | 5203 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5204 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5205 | stat->blocks = (inode_get_bytes(inode) + |
5206 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5207 | return 0; |
5208 | } | |
5209 | ||
d397712b CM |
5210 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5211 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5212 | { |
5213 | struct btrfs_trans_handle *trans; | |
5214 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5215 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5216 | struct inode *new_inode = new_dentry->d_inode; |
5217 | struct inode *old_inode = old_dentry->d_inode; | |
5218 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5219 | u64 index = 0; |
4df27c4d | 5220 | u64 root_objectid; |
39279cc3 CM |
5221 | int ret; |
5222 | ||
f679a840 YZ |
5223 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5224 | return -EPERM; | |
5225 | ||
4df27c4d YZ |
5226 | /* we only allow rename subvolume link between subvolumes */ |
5227 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5228 | return -EXDEV; |
5229 | ||
4df27c4d YZ |
5230 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5231 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5232 | return -ENOTEMPTY; |
5f39d397 | 5233 | |
4df27c4d YZ |
5234 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5235 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5236 | return -ENOTEMPTY; | |
0660b5af | 5237 | |
9ed74f2d JB |
5238 | /* |
5239 | * 2 items for dir items | |
5240 | * 1 item for orphan entry | |
5241 | * 1 item for ref | |
5242 | */ | |
5243 | ret = btrfs_reserve_metadata_space(root, 4); | |
1832a6d5 | 5244 | if (ret) |
4df27c4d | 5245 | return ret; |
1832a6d5 | 5246 | |
5a3f23d5 CM |
5247 | /* |
5248 | * we're using rename to replace one file with another. | |
5249 | * and the replacement file is large. Start IO on it now so | |
5250 | * we don't add too much work to the end of the transaction | |
5251 | */ | |
4baf8c92 | 5252 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5253 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5254 | filemap_flush(old_inode->i_mapping); | |
5255 | ||
76dda93c YZ |
5256 | /* close the racy window with snapshot create/destroy ioctl */ |
5257 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5258 | down_read(&root->fs_info->subvol_sem); | |
5259 | ||
39279cc3 | 5260 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5261 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5262 | |
4df27c4d YZ |
5263 | if (dest != root) |
5264 | btrfs_record_root_in_trans(trans, dest); | |
5265 | ||
a5719521 YZ |
5266 | ret = btrfs_set_inode_index(new_dir, &index); |
5267 | if (ret) | |
5268 | goto out_fail; | |
5a3f23d5 | 5269 | |
a5719521 | 5270 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5271 | /* force full log commit if subvolume involved. */ |
5272 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5273 | } else { | |
a5719521 YZ |
5274 | ret = btrfs_insert_inode_ref(trans, dest, |
5275 | new_dentry->d_name.name, | |
5276 | new_dentry->d_name.len, | |
5277 | old_inode->i_ino, | |
5278 | new_dir->i_ino, index); | |
5279 | if (ret) | |
5280 | goto out_fail; | |
4df27c4d YZ |
5281 | /* |
5282 | * this is an ugly little race, but the rename is required | |
5283 | * to make sure that if we crash, the inode is either at the | |
5284 | * old name or the new one. pinning the log transaction lets | |
5285 | * us make sure we don't allow a log commit to come in after | |
5286 | * we unlink the name but before we add the new name back in. | |
5287 | */ | |
5288 | btrfs_pin_log_trans(root); | |
5289 | } | |
a5719521 YZ |
5290 | /* |
5291 | * make sure the inode gets flushed if it is replacing | |
5292 | * something. | |
5293 | */ | |
5294 | if (new_inode && new_inode->i_size && | |
5295 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5296 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5297 | } | |
39279cc3 | 5298 | |
39279cc3 CM |
5299 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5300 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5301 | old_inode->i_ctime = ctime; | |
5f39d397 | 5302 | |
12fcfd22 CM |
5303 | if (old_dentry->d_parent != new_dentry->d_parent) |
5304 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5305 | ||
4df27c4d YZ |
5306 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5307 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5308 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5309 | old_dentry->d_name.name, | |
5310 | old_dentry->d_name.len); | |
5311 | } else { | |
5312 | btrfs_inc_nlink(old_dentry->d_inode); | |
5313 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5314 | old_dentry->d_inode, | |
5315 | old_dentry->d_name.name, | |
5316 | old_dentry->d_name.len); | |
5317 | } | |
5318 | BUG_ON(ret); | |
39279cc3 CM |
5319 | |
5320 | if (new_inode) { | |
5321 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5322 | if (unlikely(new_inode->i_ino == |
5323 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5324 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5325 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5326 | root_objectid, | |
5327 | new_dentry->d_name.name, | |
5328 | new_dentry->d_name.len); | |
5329 | BUG_ON(new_inode->i_nlink == 0); | |
5330 | } else { | |
5331 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5332 | new_dentry->d_inode, | |
5333 | new_dentry->d_name.name, | |
5334 | new_dentry->d_name.len); | |
5335 | } | |
5336 | BUG_ON(ret); | |
7b128766 | 5337 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5338 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5339 | BUG_ON(ret); |
7b128766 | 5340 | } |
39279cc3 | 5341 | } |
aec7477b | 5342 | |
4df27c4d YZ |
5343 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5344 | new_dentry->d_name.name, | |
a5719521 | 5345 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5346 | BUG_ON(ret); |
39279cc3 | 5347 | |
4df27c4d YZ |
5348 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5349 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5350 | new_dentry->d_parent); | |
5351 | btrfs_end_log_trans(root); | |
5352 | } | |
a5719521 | 5353 | out_fail: |
ab78c84d | 5354 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5355 | |
76dda93c YZ |
5356 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5357 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d JB |
5358 | |
5359 | btrfs_unreserve_metadata_space(root, 4); | |
39279cc3 CM |
5360 | return ret; |
5361 | } | |
5362 | ||
d352ac68 CM |
5363 | /* |
5364 | * some fairly slow code that needs optimization. This walks the list | |
5365 | * of all the inodes with pending delalloc and forces them to disk. | |
5366 | */ | |
ea8c2819 CM |
5367 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5368 | { | |
5369 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5370 | struct btrfs_inode *binode; | |
5b21f2ed | 5371 | struct inode *inode; |
ea8c2819 | 5372 | |
c146afad YZ |
5373 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5374 | return -EROFS; | |
5375 | ||
75eff68e | 5376 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5377 | while (!list_empty(head)) { |
ea8c2819 CM |
5378 | binode = list_entry(head->next, struct btrfs_inode, |
5379 | delalloc_inodes); | |
5b21f2ed ZY |
5380 | inode = igrab(&binode->vfs_inode); |
5381 | if (!inode) | |
5382 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5383 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5384 | if (inode) { |
8c8bee1d | 5385 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5386 | iput(inode); |
5387 | } | |
5388 | cond_resched(); | |
75eff68e | 5389 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5390 | } |
75eff68e | 5391 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5392 | |
5393 | /* the filemap_flush will queue IO into the worker threads, but | |
5394 | * we have to make sure the IO is actually started and that | |
5395 | * ordered extents get created before we return | |
5396 | */ | |
5397 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5398 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5399 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5400 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5401 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5402 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5403 | } |
5404 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5405 | return 0; |
5406 | } | |
5407 | ||
39279cc3 CM |
5408 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5409 | const char *symname) | |
5410 | { | |
5411 | struct btrfs_trans_handle *trans; | |
5412 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5413 | struct btrfs_path *path; | |
5414 | struct btrfs_key key; | |
1832a6d5 | 5415 | struct inode *inode = NULL; |
39279cc3 CM |
5416 | int err; |
5417 | int drop_inode = 0; | |
5418 | u64 objectid; | |
00e4e6b3 | 5419 | u64 index = 0 ; |
39279cc3 CM |
5420 | int name_len; |
5421 | int datasize; | |
5f39d397 | 5422 | unsigned long ptr; |
39279cc3 | 5423 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5424 | struct extent_buffer *leaf; |
1832a6d5 | 5425 | unsigned long nr = 0; |
39279cc3 CM |
5426 | |
5427 | name_len = strlen(symname) + 1; | |
5428 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5429 | return -ENAMETOOLONG; | |
1832a6d5 | 5430 | |
9ed74f2d JB |
5431 | /* |
5432 | * 2 items for inode item and ref | |
5433 | * 2 items for dir items | |
5434 | * 1 item for xattr if selinux is on | |
5435 | */ | |
5436 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5437 | if (err) |
9ed74f2d | 5438 | return err; |
1832a6d5 | 5439 | |
39279cc3 | 5440 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5441 | if (!trans) |
5442 | goto out_fail; | |
39279cc3 CM |
5443 | btrfs_set_trans_block_group(trans, dir); |
5444 | ||
5445 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5446 | if (err) { | |
5447 | err = -ENOSPC; | |
5448 | goto out_unlock; | |
5449 | } | |
5450 | ||
aec7477b | 5451 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5452 | dentry->d_name.len, |
5453 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5454 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5455 | &index); | |
39279cc3 CM |
5456 | err = PTR_ERR(inode); |
5457 | if (IS_ERR(inode)) | |
5458 | goto out_unlock; | |
5459 | ||
0279b4cd | 5460 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5461 | if (err) { |
5462 | drop_inode = 1; | |
5463 | goto out_unlock; | |
5464 | } | |
5465 | ||
39279cc3 | 5466 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5467 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5468 | if (err) |
5469 | drop_inode = 1; | |
5470 | else { | |
5471 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5472 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5473 | inode->i_fop = &btrfs_file_operations; |
5474 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5475 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5476 | } |
39279cc3 CM |
5477 | btrfs_update_inode_block_group(trans, inode); |
5478 | btrfs_update_inode_block_group(trans, dir); | |
5479 | if (drop_inode) | |
5480 | goto out_unlock; | |
5481 | ||
5482 | path = btrfs_alloc_path(); | |
5483 | BUG_ON(!path); | |
5484 | key.objectid = inode->i_ino; | |
5485 | key.offset = 0; | |
39279cc3 CM |
5486 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5487 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5488 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5489 | datasize); | |
54aa1f4d CM |
5490 | if (err) { |
5491 | drop_inode = 1; | |
5492 | goto out_unlock; | |
5493 | } | |
5f39d397 CM |
5494 | leaf = path->nodes[0]; |
5495 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5496 | struct btrfs_file_extent_item); | |
5497 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5498 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5499 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5500 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5501 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5502 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5503 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5504 | ||
39279cc3 | 5505 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5506 | write_extent_buffer(leaf, symname, ptr, name_len); |
5507 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5508 | btrfs_free_path(path); |
5f39d397 | 5509 | |
39279cc3 CM |
5510 | inode->i_op = &btrfs_symlink_inode_operations; |
5511 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5512 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5513 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5514 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5515 | err = btrfs_update_inode(trans, root, inode); |
5516 | if (err) | |
5517 | drop_inode = 1; | |
39279cc3 CM |
5518 | |
5519 | out_unlock: | |
d3c2fdcf | 5520 | nr = trans->blocks_used; |
ab78c84d | 5521 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5522 | out_fail: |
9ed74f2d | 5523 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5524 | if (drop_inode) { |
5525 | inode_dec_link_count(inode); | |
5526 | iput(inode); | |
5527 | } | |
d3c2fdcf | 5528 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5529 | return err; |
5530 | } | |
16432985 | 5531 | |
546888da CM |
5532 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5533 | struct inode *inode, u64 start, u64 end, | |
e980b50c | 5534 | u64 locked_end, u64 alloc_hint, int mode) |
d899e052 | 5535 | { |
d899e052 YZ |
5536 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5537 | struct btrfs_key ins; | |
5538 | u64 alloc_size; | |
5539 | u64 cur_offset = start; | |
5540 | u64 num_bytes = end - start; | |
5541 | int ret = 0; | |
5542 | ||
d899e052 YZ |
5543 | while (num_bytes > 0) { |
5544 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d JB |
5545 | |
5546 | ret = btrfs_reserve_metadata_space(root, 1); | |
5547 | if (ret) | |
5548 | goto out; | |
5549 | ||
d899e052 YZ |
5550 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5551 | root->sectorsize, 0, alloc_hint, | |
5552 | (u64)-1, &ins, 1); | |
5553 | if (ret) { | |
5554 | WARN_ON(1); | |
5555 | goto out; | |
5556 | } | |
5557 | ret = insert_reserved_file_extent(trans, inode, | |
5558 | cur_offset, ins.objectid, | |
5559 | ins.offset, ins.offset, | |
e980b50c CM |
5560 | ins.offset, locked_end, |
5561 | 0, 0, 0, | |
d899e052 YZ |
5562 | BTRFS_FILE_EXTENT_PREALLOC); |
5563 | BUG_ON(ret); | |
a1ed835e CM |
5564 | btrfs_drop_extent_cache(inode, cur_offset, |
5565 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5566 | num_bytes -= ins.offset; |
5567 | cur_offset += ins.offset; | |
5568 | alloc_hint = ins.objectid + ins.offset; | |
9ed74f2d | 5569 | btrfs_unreserve_metadata_space(root, 1); |
d899e052 YZ |
5570 | } |
5571 | out: | |
5572 | if (cur_offset > start) { | |
5573 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5574 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5575 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5576 | cur_offset > i_size_read(inode)) | |
5577 | btrfs_i_size_write(inode, cur_offset); | |
5578 | ret = btrfs_update_inode(trans, root, inode); | |
5579 | BUG_ON(ret); | |
5580 | } | |
5581 | ||
d899e052 YZ |
5582 | return ret; |
5583 | } | |
5584 | ||
5585 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5586 | loff_t offset, loff_t len) | |
5587 | { | |
5588 | u64 cur_offset; | |
5589 | u64 last_byte; | |
5590 | u64 alloc_start; | |
5591 | u64 alloc_end; | |
5592 | u64 alloc_hint = 0; | |
e980b50c | 5593 | u64 locked_end; |
d899e052 YZ |
5594 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5595 | struct extent_map *em; | |
546888da | 5596 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5597 | struct btrfs_root *root; |
d899e052 YZ |
5598 | int ret; |
5599 | ||
5600 | alloc_start = offset & ~mask; | |
5601 | alloc_end = (offset + len + mask) & ~mask; | |
5602 | ||
546888da CM |
5603 | /* |
5604 | * wait for ordered IO before we have any locks. We'll loop again | |
5605 | * below with the locks held. | |
5606 | */ | |
5607 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5608 | ||
d899e052 YZ |
5609 | mutex_lock(&inode->i_mutex); |
5610 | if (alloc_start > inode->i_size) { | |
5611 | ret = btrfs_cont_expand(inode, alloc_start); | |
5612 | if (ret) | |
5613 | goto out; | |
5614 | } | |
5615 | ||
a970b0a1 JB |
5616 | root = BTRFS_I(inode)->root; |
5617 | ||
5618 | ret = btrfs_check_data_free_space(root, inode, | |
5619 | alloc_end - alloc_start); | |
5620 | if (ret) | |
5621 | goto out; | |
5622 | ||
e980b50c | 5623 | locked_end = alloc_end - 1; |
d899e052 YZ |
5624 | while (1) { |
5625 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5626 | |
5627 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5628 | if (!trans) { | |
5629 | ret = -EIO; | |
a970b0a1 | 5630 | goto out_free; |
546888da CM |
5631 | } |
5632 | ||
5633 | /* the extent lock is ordered inside the running | |
5634 | * transaction | |
5635 | */ | |
e980b50c CM |
5636 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5637 | GFP_NOFS); | |
d899e052 YZ |
5638 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5639 | alloc_end - 1); | |
5640 | if (ordered && | |
5641 | ordered->file_offset + ordered->len > alloc_start && | |
5642 | ordered->file_offset < alloc_end) { | |
5643 | btrfs_put_ordered_extent(ordered); | |
5644 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5645 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5646 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5647 | ||
5648 | /* | |
5649 | * we can't wait on the range with the transaction | |
5650 | * running or with the extent lock held | |
5651 | */ | |
d899e052 YZ |
5652 | btrfs_wait_ordered_range(inode, alloc_start, |
5653 | alloc_end - alloc_start); | |
5654 | } else { | |
5655 | if (ordered) | |
5656 | btrfs_put_ordered_extent(ordered); | |
5657 | break; | |
5658 | } | |
5659 | } | |
5660 | ||
5661 | cur_offset = alloc_start; | |
5662 | while (1) { | |
5663 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5664 | alloc_end - cur_offset, 0); | |
5665 | BUG_ON(IS_ERR(em) || !em); | |
5666 | last_byte = min(extent_map_end(em), alloc_end); | |
5667 | last_byte = (last_byte + mask) & ~mask; | |
5668 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5669 | ret = prealloc_file_range(trans, inode, cur_offset, |
e980b50c CM |
5670 | last_byte, locked_end + 1, |
5671 | alloc_hint, mode); | |
d899e052 YZ |
5672 | if (ret < 0) { |
5673 | free_extent_map(em); | |
5674 | break; | |
5675 | } | |
5676 | } | |
5677 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5678 | alloc_hint = em->block_start; | |
5679 | free_extent_map(em); | |
5680 | ||
5681 | cur_offset = last_byte; | |
5682 | if (cur_offset >= alloc_end) { | |
5683 | ret = 0; | |
5684 | break; | |
5685 | } | |
5686 | } | |
e980b50c | 5687 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5688 | GFP_NOFS); |
546888da CM |
5689 | |
5690 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5691 | out_free: |
5692 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5693 | out: |
5694 | mutex_unlock(&inode->i_mutex); | |
5695 | return ret; | |
5696 | } | |
5697 | ||
e6dcd2dc CM |
5698 | static int btrfs_set_page_dirty(struct page *page) |
5699 | { | |
e6dcd2dc CM |
5700 | return __set_page_dirty_nobuffers(page); |
5701 | } | |
5702 | ||
0ee0fda0 | 5703 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5704 | { |
6cbff00f | 5705 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5706 | return -EACCES; |
33268eaf | 5707 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5708 | } |
39279cc3 CM |
5709 | |
5710 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 5711 | .getattr = btrfs_getattr, |
39279cc3 CM |
5712 | .lookup = btrfs_lookup, |
5713 | .create = btrfs_create, | |
5714 | .unlink = btrfs_unlink, | |
5715 | .link = btrfs_link, | |
5716 | .mkdir = btrfs_mkdir, | |
5717 | .rmdir = btrfs_rmdir, | |
5718 | .rename = btrfs_rename, | |
5719 | .symlink = btrfs_symlink, | |
5720 | .setattr = btrfs_setattr, | |
618e21d5 | 5721 | .mknod = btrfs_mknod, |
95819c05 CH |
5722 | .setxattr = btrfs_setxattr, |
5723 | .getxattr = btrfs_getxattr, | |
5103e947 | 5724 | .listxattr = btrfs_listxattr, |
95819c05 | 5725 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5726 | .permission = btrfs_permission, |
39279cc3 | 5727 | }; |
39279cc3 CM |
5728 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
5729 | .lookup = btrfs_lookup, | |
fdebe2bd | 5730 | .permission = btrfs_permission, |
39279cc3 | 5731 | }; |
76dda93c | 5732 | |
39279cc3 CM |
5733 | static struct file_operations btrfs_dir_file_operations = { |
5734 | .llseek = generic_file_llseek, | |
5735 | .read = generic_read_dir, | |
cbdf5a24 | 5736 | .readdir = btrfs_real_readdir, |
34287aa3 | 5737 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5738 | #ifdef CONFIG_COMPAT |
34287aa3 | 5739 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5740 | #endif |
6bf13c0c | 5741 | .release = btrfs_release_file, |
e02119d5 | 5742 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5743 | }; |
5744 | ||
d1310b2e | 5745 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5746 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5747 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5748 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5749 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5750 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5751 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5752 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5753 | .set_bit_hook = btrfs_set_bit_hook, |
5754 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
5755 | .merge_extent_hook = btrfs_merge_extent_hook, |
5756 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
5757 | }; |
5758 | ||
35054394 CM |
5759 | /* |
5760 | * btrfs doesn't support the bmap operation because swapfiles | |
5761 | * use bmap to make a mapping of extents in the file. They assume | |
5762 | * these extents won't change over the life of the file and they | |
5763 | * use the bmap result to do IO directly to the drive. | |
5764 | * | |
5765 | * the btrfs bmap call would return logical addresses that aren't | |
5766 | * suitable for IO and they also will change frequently as COW | |
5767 | * operations happen. So, swapfile + btrfs == corruption. | |
5768 | * | |
5769 | * For now we're avoiding this by dropping bmap. | |
5770 | */ | |
39279cc3 CM |
5771 | static struct address_space_operations btrfs_aops = { |
5772 | .readpage = btrfs_readpage, | |
5773 | .writepage = btrfs_writepage, | |
b293f02e | 5774 | .writepages = btrfs_writepages, |
3ab2fb5a | 5775 | .readpages = btrfs_readpages, |
39279cc3 | 5776 | .sync_page = block_sync_page, |
16432985 | 5777 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5778 | .invalidatepage = btrfs_invalidatepage, |
5779 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5780 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5781 | }; |
5782 | ||
5783 | static struct address_space_operations btrfs_symlink_aops = { | |
5784 | .readpage = btrfs_readpage, | |
5785 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5786 | .invalidatepage = btrfs_invalidatepage, |
5787 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5788 | }; |
5789 | ||
5790 | static struct inode_operations btrfs_file_inode_operations = { | |
5791 | .truncate = btrfs_truncate, | |
5792 | .getattr = btrfs_getattr, | |
5793 | .setattr = btrfs_setattr, | |
95819c05 CH |
5794 | .setxattr = btrfs_setxattr, |
5795 | .getxattr = btrfs_getxattr, | |
5103e947 | 5796 | .listxattr = btrfs_listxattr, |
95819c05 | 5797 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5798 | .permission = btrfs_permission, |
d899e052 | 5799 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5800 | .fiemap = btrfs_fiemap, |
39279cc3 | 5801 | }; |
618e21d5 JB |
5802 | static struct inode_operations btrfs_special_inode_operations = { |
5803 | .getattr = btrfs_getattr, | |
5804 | .setattr = btrfs_setattr, | |
fdebe2bd | 5805 | .permission = btrfs_permission, |
95819c05 CH |
5806 | .setxattr = btrfs_setxattr, |
5807 | .getxattr = btrfs_getxattr, | |
33268eaf | 5808 | .listxattr = btrfs_listxattr, |
95819c05 | 5809 | .removexattr = btrfs_removexattr, |
618e21d5 | 5810 | }; |
39279cc3 CM |
5811 | static struct inode_operations btrfs_symlink_inode_operations = { |
5812 | .readlink = generic_readlink, | |
5813 | .follow_link = page_follow_link_light, | |
5814 | .put_link = page_put_link, | |
fdebe2bd | 5815 | .permission = btrfs_permission, |
0279b4cd JO |
5816 | .setxattr = btrfs_setxattr, |
5817 | .getxattr = btrfs_getxattr, | |
5818 | .listxattr = btrfs_listxattr, | |
5819 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5820 | }; |
76dda93c YZ |
5821 | |
5822 | struct dentry_operations btrfs_dentry_operations = { | |
5823 | .d_delete = btrfs_dentry_delete, | |
5824 | }; |