]>
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 | ||
6e1d5dcc AD |
58 | static const struct inode_operations btrfs_dir_inode_operations; |
59 | static const struct inode_operations btrfs_symlink_inode_operations; | |
60 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
61 | static const struct inode_operations btrfs_special_inode_operations; | |
62 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
63 | static const struct address_space_operations btrfs_aops; |
64 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 65 | static const 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 | ||
920bbbfb | 233 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 234 | &hint_byte, 1); |
c8b97818 CM |
235 | BUG_ON(ret); |
236 | ||
237 | if (isize > actual_end) | |
238 | inline_len = min_t(u64, isize, actual_end); | |
239 | ret = insert_inline_extent(trans, root, inode, start, | |
240 | inline_len, compressed_size, | |
241 | compressed_pages); | |
242 | BUG_ON(ret); | |
a1ed835e | 243 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
244 | return 0; |
245 | } | |
246 | ||
771ed689 CM |
247 | struct async_extent { |
248 | u64 start; | |
249 | u64 ram_size; | |
250 | u64 compressed_size; | |
251 | struct page **pages; | |
252 | unsigned long nr_pages; | |
253 | struct list_head list; | |
254 | }; | |
255 | ||
256 | struct async_cow { | |
257 | struct inode *inode; | |
258 | struct btrfs_root *root; | |
259 | struct page *locked_page; | |
260 | u64 start; | |
261 | u64 end; | |
262 | struct list_head extents; | |
263 | struct btrfs_work work; | |
264 | }; | |
265 | ||
266 | static noinline int add_async_extent(struct async_cow *cow, | |
267 | u64 start, u64 ram_size, | |
268 | u64 compressed_size, | |
269 | struct page **pages, | |
270 | unsigned long nr_pages) | |
271 | { | |
272 | struct async_extent *async_extent; | |
273 | ||
274 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
275 | async_extent->start = start; | |
276 | async_extent->ram_size = ram_size; | |
277 | async_extent->compressed_size = compressed_size; | |
278 | async_extent->pages = pages; | |
279 | async_extent->nr_pages = nr_pages; | |
280 | list_add_tail(&async_extent->list, &cow->extents); | |
281 | return 0; | |
282 | } | |
283 | ||
d352ac68 | 284 | /* |
771ed689 CM |
285 | * we create compressed extents in two phases. The first |
286 | * phase compresses a range of pages that have already been | |
287 | * locked (both pages and state bits are locked). | |
c8b97818 | 288 | * |
771ed689 CM |
289 | * This is done inside an ordered work queue, and the compression |
290 | * is spread across many cpus. The actual IO submission is step | |
291 | * two, and the ordered work queue takes care of making sure that | |
292 | * happens in the same order things were put onto the queue by | |
293 | * writepages and friends. | |
c8b97818 | 294 | * |
771ed689 CM |
295 | * If this code finds it can't get good compression, it puts an |
296 | * entry onto the work queue to write the uncompressed bytes. This | |
297 | * makes sure that both compressed inodes and uncompressed inodes | |
298 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 299 | */ |
771ed689 CM |
300 | static noinline int compress_file_range(struct inode *inode, |
301 | struct page *locked_page, | |
302 | u64 start, u64 end, | |
303 | struct async_cow *async_cow, | |
304 | int *num_added) | |
b888db2b CM |
305 | { |
306 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
307 | struct btrfs_trans_handle *trans; | |
db94535d | 308 | u64 num_bytes; |
c8b97818 CM |
309 | u64 orig_start; |
310 | u64 disk_num_bytes; | |
db94535d | 311 | u64 blocksize = root->sectorsize; |
c8b97818 | 312 | u64 actual_end; |
42dc7bab | 313 | u64 isize = i_size_read(inode); |
e6dcd2dc | 314 | int ret = 0; |
c8b97818 CM |
315 | struct page **pages = NULL; |
316 | unsigned long nr_pages; | |
317 | unsigned long nr_pages_ret = 0; | |
318 | unsigned long total_compressed = 0; | |
319 | unsigned long total_in = 0; | |
320 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 321 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
322 | int i; |
323 | int will_compress; | |
b888db2b | 324 | |
c8b97818 CM |
325 | orig_start = start; |
326 | ||
42dc7bab | 327 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
328 | again: |
329 | will_compress = 0; | |
330 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
331 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 332 | |
f03d9301 CM |
333 | /* |
334 | * we don't want to send crud past the end of i_size through | |
335 | * compression, that's just a waste of CPU time. So, if the | |
336 | * end of the file is before the start of our current | |
337 | * requested range of bytes, we bail out to the uncompressed | |
338 | * cleanup code that can deal with all of this. | |
339 | * | |
340 | * It isn't really the fastest way to fix things, but this is a | |
341 | * very uncommon corner. | |
342 | */ | |
343 | if (actual_end <= start) | |
344 | goto cleanup_and_bail_uncompressed; | |
345 | ||
c8b97818 CM |
346 | total_compressed = actual_end - start; |
347 | ||
348 | /* we want to make sure that amount of ram required to uncompress | |
349 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
350 | * of a compressed extent to 128k. This is a crucial number |
351 | * because it also controls how easily we can spread reads across | |
352 | * cpus for decompression. | |
353 | * | |
354 | * We also want to make sure the amount of IO required to do | |
355 | * a random read is reasonably small, so we limit the size of | |
356 | * a compressed extent to 128k. | |
c8b97818 CM |
357 | */ |
358 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 359 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 360 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
361 | disk_num_bytes = num_bytes; |
362 | total_in = 0; | |
363 | ret = 0; | |
db94535d | 364 | |
771ed689 CM |
365 | /* |
366 | * we do compression for mount -o compress and when the | |
367 | * inode has not been flagged as nocompress. This flag can | |
368 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 369 | */ |
6cbff00f | 370 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
371 | btrfs_test_opt(root, COMPRESS)) { |
372 | WARN_ON(pages); | |
cfbc246e | 373 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 374 | |
c8b97818 CM |
375 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
376 | total_compressed, pages, | |
377 | nr_pages, &nr_pages_ret, | |
378 | &total_in, | |
379 | &total_compressed, | |
380 | max_compressed); | |
381 | ||
382 | if (!ret) { | |
383 | unsigned long offset = total_compressed & | |
384 | (PAGE_CACHE_SIZE - 1); | |
385 | struct page *page = pages[nr_pages_ret - 1]; | |
386 | char *kaddr; | |
387 | ||
388 | /* zero the tail end of the last page, we might be | |
389 | * sending it down to disk | |
390 | */ | |
391 | if (offset) { | |
392 | kaddr = kmap_atomic(page, KM_USER0); | |
393 | memset(kaddr + offset, 0, | |
394 | PAGE_CACHE_SIZE - offset); | |
395 | kunmap_atomic(kaddr, KM_USER0); | |
396 | } | |
397 | will_compress = 1; | |
398 | } | |
399 | } | |
400 | if (start == 0) { | |
771ed689 CM |
401 | trans = btrfs_join_transaction(root, 1); |
402 | BUG_ON(!trans); | |
403 | btrfs_set_trans_block_group(trans, inode); | |
404 | ||
c8b97818 | 405 | /* lets try to make an inline extent */ |
771ed689 | 406 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 407 | /* we didn't compress the entire range, try |
771ed689 | 408 | * to make an uncompressed inline extent. |
c8b97818 CM |
409 | */ |
410 | ret = cow_file_range_inline(trans, root, inode, | |
411 | start, end, 0, NULL); | |
412 | } else { | |
771ed689 | 413 | /* try making a compressed inline extent */ |
c8b97818 CM |
414 | ret = cow_file_range_inline(trans, root, inode, |
415 | start, end, | |
416 | total_compressed, pages); | |
417 | } | |
771ed689 | 418 | btrfs_end_transaction(trans, root); |
c8b97818 | 419 | if (ret == 0) { |
771ed689 CM |
420 | /* |
421 | * inline extent creation worked, we don't need | |
422 | * to create any more async work items. Unlock | |
423 | * and free up our temp pages. | |
424 | */ | |
c8b97818 | 425 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
426 | &BTRFS_I(inode)->io_tree, |
427 | start, end, NULL, | |
428 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 429 | EXTENT_CLEAR_DELALLOC | |
32c00aff | 430 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e | 431 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c8b97818 CM |
432 | ret = 0; |
433 | goto free_pages_out; | |
434 | } | |
435 | } | |
436 | ||
437 | if (will_compress) { | |
438 | /* | |
439 | * we aren't doing an inline extent round the compressed size | |
440 | * up to a block size boundary so the allocator does sane | |
441 | * things | |
442 | */ | |
443 | total_compressed = (total_compressed + blocksize - 1) & | |
444 | ~(blocksize - 1); | |
445 | ||
446 | /* | |
447 | * one last check to make sure the compression is really a | |
448 | * win, compare the page count read with the blocks on disk | |
449 | */ | |
450 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
451 | ~(PAGE_CACHE_SIZE - 1); | |
452 | if (total_compressed >= total_in) { | |
453 | will_compress = 0; | |
454 | } else { | |
455 | disk_num_bytes = total_compressed; | |
456 | num_bytes = total_in; | |
457 | } | |
458 | } | |
459 | if (!will_compress && pages) { | |
460 | /* | |
461 | * the compression code ran but failed to make things smaller, | |
462 | * free any pages it allocated and our page pointer array | |
463 | */ | |
464 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 465 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
466 | page_cache_release(pages[i]); |
467 | } | |
468 | kfree(pages); | |
469 | pages = NULL; | |
470 | total_compressed = 0; | |
471 | nr_pages_ret = 0; | |
472 | ||
473 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 474 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 475 | } |
771ed689 CM |
476 | if (will_compress) { |
477 | *num_added += 1; | |
c8b97818 | 478 | |
771ed689 CM |
479 | /* the async work queues will take care of doing actual |
480 | * allocation on disk for these compressed pages, | |
481 | * and will submit them to the elevator. | |
482 | */ | |
483 | add_async_extent(async_cow, start, num_bytes, | |
484 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 485 | |
42dc7bab | 486 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
487 | start += num_bytes; |
488 | pages = NULL; | |
489 | cond_resched(); | |
490 | goto again; | |
491 | } | |
492 | } else { | |
f03d9301 | 493 | cleanup_and_bail_uncompressed: |
771ed689 CM |
494 | /* |
495 | * No compression, but we still need to write the pages in | |
496 | * the file we've been given so far. redirty the locked | |
497 | * page if it corresponds to our extent and set things up | |
498 | * for the async work queue to run cow_file_range to do | |
499 | * the normal delalloc dance | |
500 | */ | |
501 | if (page_offset(locked_page) >= start && | |
502 | page_offset(locked_page) <= end) { | |
503 | __set_page_dirty_nobuffers(locked_page); | |
504 | /* unlocked later on in the async handlers */ | |
505 | } | |
506 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
507 | *num_added += 1; | |
508 | } | |
3b951516 | 509 | |
771ed689 CM |
510 | out: |
511 | return 0; | |
512 | ||
513 | free_pages_out: | |
514 | for (i = 0; i < nr_pages_ret; i++) { | |
515 | WARN_ON(pages[i]->mapping); | |
516 | page_cache_release(pages[i]); | |
517 | } | |
d397712b | 518 | kfree(pages); |
771ed689 CM |
519 | |
520 | goto out; | |
521 | } | |
522 | ||
523 | /* | |
524 | * phase two of compressed writeback. This is the ordered portion | |
525 | * of the code, which only gets called in the order the work was | |
526 | * queued. We walk all the async extents created by compress_file_range | |
527 | * and send them down to the disk. | |
528 | */ | |
529 | static noinline int submit_compressed_extents(struct inode *inode, | |
530 | struct async_cow *async_cow) | |
531 | { | |
532 | struct async_extent *async_extent; | |
533 | u64 alloc_hint = 0; | |
534 | struct btrfs_trans_handle *trans; | |
535 | struct btrfs_key ins; | |
536 | struct extent_map *em; | |
537 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
538 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
539 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 540 | int ret = 0; |
771ed689 CM |
541 | |
542 | if (list_empty(&async_cow->extents)) | |
543 | return 0; | |
544 | ||
545 | trans = btrfs_join_transaction(root, 1); | |
546 | ||
d397712b | 547 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
548 | async_extent = list_entry(async_cow->extents.next, |
549 | struct async_extent, list); | |
550 | list_del(&async_extent->list); | |
c8b97818 | 551 | |
771ed689 CM |
552 | io_tree = &BTRFS_I(inode)->io_tree; |
553 | ||
f5a84ee3 | 554 | retry: |
771ed689 CM |
555 | /* did the compression code fall back to uncompressed IO? */ |
556 | if (!async_extent->pages) { | |
557 | int page_started = 0; | |
558 | unsigned long nr_written = 0; | |
559 | ||
560 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
561 | async_extent->start + |
562 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
563 | |
564 | /* allocate blocks */ | |
f5a84ee3 JB |
565 | ret = cow_file_range(inode, async_cow->locked_page, |
566 | async_extent->start, | |
567 | async_extent->start + | |
568 | async_extent->ram_size - 1, | |
569 | &page_started, &nr_written, 0); | |
771ed689 CM |
570 | |
571 | /* | |
572 | * if page_started, cow_file_range inserted an | |
573 | * inline extent and took care of all the unlocking | |
574 | * and IO for us. Otherwise, we need to submit | |
575 | * all those pages down to the drive. | |
576 | */ | |
f5a84ee3 | 577 | if (!page_started && !ret) |
771ed689 CM |
578 | extent_write_locked_range(io_tree, |
579 | inode, async_extent->start, | |
d397712b | 580 | async_extent->start + |
771ed689 CM |
581 | async_extent->ram_size - 1, |
582 | btrfs_get_extent, | |
583 | WB_SYNC_ALL); | |
584 | kfree(async_extent); | |
585 | cond_resched(); | |
586 | continue; | |
587 | } | |
588 | ||
589 | lock_extent(io_tree, async_extent->start, | |
590 | async_extent->start + async_extent->ram_size - 1, | |
591 | GFP_NOFS); | |
c8b97818 | 592 | /* |
771ed689 CM |
593 | * here we're doing allocation and writeback of the |
594 | * compressed pages | |
c8b97818 | 595 | */ |
771ed689 CM |
596 | btrfs_drop_extent_cache(inode, async_extent->start, |
597 | async_extent->start + | |
598 | async_extent->ram_size - 1, 0); | |
599 | ||
600 | ret = btrfs_reserve_extent(trans, root, | |
601 | async_extent->compressed_size, | |
602 | async_extent->compressed_size, | |
603 | 0, alloc_hint, | |
604 | (u64)-1, &ins, 1); | |
f5a84ee3 JB |
605 | if (ret) { |
606 | int i; | |
607 | for (i = 0; i < async_extent->nr_pages; i++) { | |
608 | WARN_ON(async_extent->pages[i]->mapping); | |
609 | page_cache_release(async_extent->pages[i]); | |
610 | } | |
611 | kfree(async_extent->pages); | |
612 | async_extent->nr_pages = 0; | |
613 | async_extent->pages = NULL; | |
614 | unlock_extent(io_tree, async_extent->start, | |
615 | async_extent->start + | |
616 | async_extent->ram_size - 1, GFP_NOFS); | |
617 | goto retry; | |
618 | } | |
619 | ||
771ed689 CM |
620 | em = alloc_extent_map(GFP_NOFS); |
621 | em->start = async_extent->start; | |
622 | em->len = async_extent->ram_size; | |
445a6944 | 623 | em->orig_start = em->start; |
c8b97818 | 624 | |
771ed689 CM |
625 | em->block_start = ins.objectid; |
626 | em->block_len = ins.offset; | |
627 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
628 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
629 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
630 | ||
d397712b | 631 | while (1) { |
890871be | 632 | write_lock(&em_tree->lock); |
771ed689 | 633 | ret = add_extent_mapping(em_tree, em); |
890871be | 634 | write_unlock(&em_tree->lock); |
771ed689 CM |
635 | if (ret != -EEXIST) { |
636 | free_extent_map(em); | |
637 | break; | |
638 | } | |
639 | btrfs_drop_extent_cache(inode, async_extent->start, | |
640 | async_extent->start + | |
641 | async_extent->ram_size - 1, 0); | |
642 | } | |
643 | ||
644 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
645 | ins.objectid, | |
646 | async_extent->ram_size, | |
647 | ins.offset, | |
648 | BTRFS_ORDERED_COMPRESSED); | |
649 | BUG_ON(ret); | |
650 | ||
651 | btrfs_end_transaction(trans, root); | |
652 | ||
653 | /* | |
654 | * clear dirty, set writeback and unlock the pages. | |
655 | */ | |
656 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
657 | &BTRFS_I(inode)->io_tree, |
658 | async_extent->start, | |
659 | async_extent->start + | |
660 | async_extent->ram_size - 1, | |
661 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
662 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 663 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 664 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
665 | |
666 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
667 | async_extent->start, |
668 | async_extent->ram_size, | |
669 | ins.objectid, | |
670 | ins.offset, async_extent->pages, | |
671 | async_extent->nr_pages); | |
771ed689 CM |
672 | |
673 | BUG_ON(ret); | |
674 | trans = btrfs_join_transaction(root, 1); | |
675 | alloc_hint = ins.objectid + ins.offset; | |
676 | kfree(async_extent); | |
677 | cond_resched(); | |
678 | } | |
679 | ||
680 | btrfs_end_transaction(trans, root); | |
681 | return 0; | |
682 | } | |
683 | ||
684 | /* | |
685 | * when extent_io.c finds a delayed allocation range in the file, | |
686 | * the call backs end up in this code. The basic idea is to | |
687 | * allocate extents on disk for the range, and create ordered data structs | |
688 | * in ram to track those extents. | |
689 | * | |
690 | * locked_page is the page that writepage had locked already. We use | |
691 | * it to make sure we don't do extra locks or unlocks. | |
692 | * | |
693 | * *page_started is set to one if we unlock locked_page and do everything | |
694 | * required to start IO on it. It may be clean and already done with | |
695 | * IO when we return. | |
696 | */ | |
697 | static noinline int cow_file_range(struct inode *inode, | |
698 | struct page *locked_page, | |
699 | u64 start, u64 end, int *page_started, | |
700 | unsigned long *nr_written, | |
701 | int unlock) | |
702 | { | |
703 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
704 | struct btrfs_trans_handle *trans; | |
705 | u64 alloc_hint = 0; | |
706 | u64 num_bytes; | |
707 | unsigned long ram_size; | |
708 | u64 disk_num_bytes; | |
709 | u64 cur_alloc_size; | |
710 | u64 blocksize = root->sectorsize; | |
711 | u64 actual_end; | |
42dc7bab | 712 | u64 isize = i_size_read(inode); |
771ed689 CM |
713 | struct btrfs_key ins; |
714 | struct extent_map *em; | |
715 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
716 | int ret = 0; | |
717 | ||
718 | trans = btrfs_join_transaction(root, 1); | |
719 | BUG_ON(!trans); | |
720 | btrfs_set_trans_block_group(trans, inode); | |
721 | ||
42dc7bab | 722 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
723 | |
724 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
725 | num_bytes = max(blocksize, num_bytes); | |
726 | disk_num_bytes = num_bytes; | |
727 | ret = 0; | |
728 | ||
729 | if (start == 0) { | |
730 | /* lets try to make an inline extent */ | |
731 | ret = cow_file_range_inline(trans, root, inode, | |
732 | start, end, 0, NULL); | |
733 | if (ret == 0) { | |
734 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
735 | &BTRFS_I(inode)->io_tree, |
736 | start, end, NULL, | |
737 | EXTENT_CLEAR_UNLOCK_PAGE | | |
738 | EXTENT_CLEAR_UNLOCK | | |
739 | EXTENT_CLEAR_DELALLOC | | |
32c00aff | 740 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e CM |
741 | EXTENT_CLEAR_DIRTY | |
742 | EXTENT_SET_WRITEBACK | | |
743 | EXTENT_END_WRITEBACK); | |
771ed689 CM |
744 | *nr_written = *nr_written + |
745 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
746 | *page_started = 1; | |
747 | ret = 0; | |
748 | goto out; | |
749 | } | |
750 | } | |
751 | ||
752 | BUG_ON(disk_num_bytes > | |
753 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
754 | ||
b917b7c3 CM |
755 | |
756 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
757 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
758 | start, num_bytes); | |
759 | if (em) { | |
6346c939 JB |
760 | /* |
761 | * if block start isn't an actual block number then find the | |
762 | * first block in this inode and use that as a hint. If that | |
763 | * block is also bogus then just don't worry about it. | |
764 | */ | |
765 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
766 | free_extent_map(em); | |
767 | em = search_extent_mapping(em_tree, 0, 0); | |
768 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
769 | alloc_hint = em->block_start; | |
770 | if (em) | |
771 | free_extent_map(em); | |
772 | } else { | |
773 | alloc_hint = em->block_start; | |
774 | free_extent_map(em); | |
775 | } | |
b917b7c3 CM |
776 | } |
777 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
778 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
779 | ||
d397712b | 780 | while (disk_num_bytes > 0) { |
a791e35e CM |
781 | unsigned long op; |
782 | ||
c8b97818 | 783 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 784 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 785 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 786 | (u64)-1, &ins, 1); |
d397712b CM |
787 | BUG_ON(ret); |
788 | ||
e6dcd2dc CM |
789 | em = alloc_extent_map(GFP_NOFS); |
790 | em->start = start; | |
445a6944 | 791 | em->orig_start = em->start; |
771ed689 CM |
792 | ram_size = ins.offset; |
793 | em->len = ins.offset; | |
c8b97818 | 794 | |
e6dcd2dc | 795 | em->block_start = ins.objectid; |
c8b97818 | 796 | em->block_len = ins.offset; |
e6dcd2dc | 797 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 798 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 799 | |
d397712b | 800 | while (1) { |
890871be | 801 | write_lock(&em_tree->lock); |
e6dcd2dc | 802 | ret = add_extent_mapping(em_tree, em); |
890871be | 803 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
804 | if (ret != -EEXIST) { |
805 | free_extent_map(em); | |
806 | break; | |
807 | } | |
808 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 809 | start + ram_size - 1, 0); |
e6dcd2dc CM |
810 | } |
811 | ||
98d20f67 | 812 | cur_alloc_size = ins.offset; |
e6dcd2dc | 813 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 814 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 815 | BUG_ON(ret); |
c8b97818 | 816 | |
17d217fe YZ |
817 | if (root->root_key.objectid == |
818 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
819 | ret = btrfs_reloc_clone_csums(inode, start, | |
820 | cur_alloc_size); | |
821 | BUG_ON(ret); | |
822 | } | |
823 | ||
d397712b | 824 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 825 | break; |
d397712b | 826 | |
c8b97818 CM |
827 | /* we're not doing compressed IO, don't unlock the first |
828 | * page (which the caller expects to stay locked), don't | |
829 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
830 | * |
831 | * Do set the Private2 bit so we know this page was properly | |
832 | * setup for writepage | |
c8b97818 | 833 | */ |
a791e35e CM |
834 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
835 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
836 | EXTENT_SET_PRIVATE2; | |
837 | ||
c8b97818 CM |
838 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
839 | start, start + ram_size - 1, | |
a791e35e | 840 | locked_page, op); |
c8b97818 | 841 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
842 | num_bytes -= cur_alloc_size; |
843 | alloc_hint = ins.objectid + ins.offset; | |
844 | start += cur_alloc_size; | |
b888db2b | 845 | } |
b888db2b | 846 | out: |
771ed689 | 847 | ret = 0; |
b888db2b | 848 | btrfs_end_transaction(trans, root); |
c8b97818 | 849 | |
be20aa9d | 850 | return ret; |
771ed689 | 851 | } |
c8b97818 | 852 | |
771ed689 CM |
853 | /* |
854 | * work queue call back to started compression on a file and pages | |
855 | */ | |
856 | static noinline void async_cow_start(struct btrfs_work *work) | |
857 | { | |
858 | struct async_cow *async_cow; | |
859 | int num_added = 0; | |
860 | async_cow = container_of(work, struct async_cow, work); | |
861 | ||
862 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
863 | async_cow->start, async_cow->end, async_cow, | |
864 | &num_added); | |
865 | if (num_added == 0) | |
866 | async_cow->inode = NULL; | |
867 | } | |
868 | ||
869 | /* | |
870 | * work queue call back to submit previously compressed pages | |
871 | */ | |
872 | static noinline void async_cow_submit(struct btrfs_work *work) | |
873 | { | |
874 | struct async_cow *async_cow; | |
875 | struct btrfs_root *root; | |
876 | unsigned long nr_pages; | |
877 | ||
878 | async_cow = container_of(work, struct async_cow, work); | |
879 | ||
880 | root = async_cow->root; | |
881 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
882 | PAGE_CACHE_SHIFT; | |
883 | ||
884 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
885 | ||
886 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
887 | 5 * 1042 * 1024 && | |
888 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
889 | wake_up(&root->fs_info->async_submit_wait); | |
890 | ||
d397712b | 891 | if (async_cow->inode) |
771ed689 | 892 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 893 | } |
c8b97818 | 894 | |
771ed689 CM |
895 | static noinline void async_cow_free(struct btrfs_work *work) |
896 | { | |
897 | struct async_cow *async_cow; | |
898 | async_cow = container_of(work, struct async_cow, work); | |
899 | kfree(async_cow); | |
900 | } | |
901 | ||
902 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
903 | u64 start, u64 end, int *page_started, | |
904 | unsigned long *nr_written) | |
905 | { | |
906 | struct async_cow *async_cow; | |
907 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
908 | unsigned long nr_pages; | |
909 | u64 cur_end; | |
910 | int limit = 10 * 1024 * 1042; | |
911 | ||
a3429ab7 CM |
912 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
913 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 914 | while (start < end) { |
771ed689 CM |
915 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
916 | async_cow->inode = inode; | |
917 | async_cow->root = root; | |
918 | async_cow->locked_page = locked_page; | |
919 | async_cow->start = start; | |
920 | ||
6cbff00f | 921 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
922 | cur_end = end; |
923 | else | |
924 | cur_end = min(end, start + 512 * 1024 - 1); | |
925 | ||
926 | async_cow->end = cur_end; | |
927 | INIT_LIST_HEAD(&async_cow->extents); | |
928 | ||
929 | async_cow->work.func = async_cow_start; | |
930 | async_cow->work.ordered_func = async_cow_submit; | |
931 | async_cow->work.ordered_free = async_cow_free; | |
932 | async_cow->work.flags = 0; | |
933 | ||
771ed689 CM |
934 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
935 | PAGE_CACHE_SHIFT; | |
936 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
937 | ||
938 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
939 | &async_cow->work); | |
940 | ||
941 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
942 | wait_event(root->fs_info->async_submit_wait, | |
943 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
944 | limit)); | |
945 | } | |
946 | ||
d397712b | 947 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
948 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
949 | wait_event(root->fs_info->async_submit_wait, | |
950 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
951 | 0)); | |
952 | } | |
953 | ||
954 | *nr_written += nr_pages; | |
955 | start = cur_end + 1; | |
956 | } | |
957 | *page_started = 1; | |
958 | return 0; | |
be20aa9d CM |
959 | } |
960 | ||
d397712b | 961 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
962 | u64 bytenr, u64 num_bytes) |
963 | { | |
964 | int ret; | |
965 | struct btrfs_ordered_sum *sums; | |
966 | LIST_HEAD(list); | |
967 | ||
07d400a6 YZ |
968 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
969 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
970 | if (ret == 0 && list_empty(&list)) |
971 | return 0; | |
972 | ||
973 | while (!list_empty(&list)) { | |
974 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
975 | list_del(&sums->list); | |
976 | kfree(sums); | |
977 | } | |
978 | return 1; | |
979 | } | |
980 | ||
d352ac68 CM |
981 | /* |
982 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
983 | * of the extents that exist in the file, and COWs the file as required. | |
984 | * | |
985 | * If no cow copies or snapshots exist, we write directly to the existing | |
986 | * blocks on disk | |
987 | */ | |
7f366cfe CM |
988 | static noinline int run_delalloc_nocow(struct inode *inode, |
989 | struct page *locked_page, | |
771ed689 CM |
990 | u64 start, u64 end, int *page_started, int force, |
991 | unsigned long *nr_written) | |
be20aa9d | 992 | { |
be20aa9d | 993 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 994 | struct btrfs_trans_handle *trans; |
be20aa9d | 995 | struct extent_buffer *leaf; |
be20aa9d | 996 | struct btrfs_path *path; |
80ff3856 | 997 | struct btrfs_file_extent_item *fi; |
be20aa9d | 998 | struct btrfs_key found_key; |
80ff3856 YZ |
999 | u64 cow_start; |
1000 | u64 cur_offset; | |
1001 | u64 extent_end; | |
5d4f98a2 | 1002 | u64 extent_offset; |
80ff3856 YZ |
1003 | u64 disk_bytenr; |
1004 | u64 num_bytes; | |
1005 | int extent_type; | |
1006 | int ret; | |
d899e052 | 1007 | int type; |
80ff3856 YZ |
1008 | int nocow; |
1009 | int check_prev = 1; | |
be20aa9d CM |
1010 | |
1011 | path = btrfs_alloc_path(); | |
1012 | BUG_ON(!path); | |
7ea394f1 YZ |
1013 | trans = btrfs_join_transaction(root, 1); |
1014 | BUG_ON(!trans); | |
be20aa9d | 1015 | |
80ff3856 YZ |
1016 | cow_start = (u64)-1; |
1017 | cur_offset = start; | |
1018 | while (1) { | |
1019 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1020 | cur_offset, 0); | |
1021 | BUG_ON(ret < 0); | |
1022 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1023 | leaf = path->nodes[0]; | |
1024 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1025 | path->slots[0] - 1); | |
1026 | if (found_key.objectid == inode->i_ino && | |
1027 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1028 | path->slots[0]--; | |
1029 | } | |
1030 | check_prev = 0; | |
1031 | next_slot: | |
1032 | leaf = path->nodes[0]; | |
1033 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1034 | ret = btrfs_next_leaf(root, path); | |
1035 | if (ret < 0) | |
1036 | BUG_ON(1); | |
1037 | if (ret > 0) | |
1038 | break; | |
1039 | leaf = path->nodes[0]; | |
1040 | } | |
be20aa9d | 1041 | |
80ff3856 YZ |
1042 | nocow = 0; |
1043 | disk_bytenr = 0; | |
17d217fe | 1044 | num_bytes = 0; |
80ff3856 YZ |
1045 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1046 | ||
1047 | if (found_key.objectid > inode->i_ino || | |
1048 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1049 | found_key.offset > end) | |
1050 | break; | |
1051 | ||
1052 | if (found_key.offset > cur_offset) { | |
1053 | extent_end = found_key.offset; | |
e9061e21 | 1054 | extent_type = 0; |
80ff3856 YZ |
1055 | goto out_check; |
1056 | } | |
1057 | ||
1058 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1059 | struct btrfs_file_extent_item); | |
1060 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1061 | ||
d899e052 YZ |
1062 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1063 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1064 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1065 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1066 | extent_end = found_key.offset + |
1067 | btrfs_file_extent_num_bytes(leaf, fi); | |
1068 | if (extent_end <= start) { | |
1069 | path->slots[0]++; | |
1070 | goto next_slot; | |
1071 | } | |
17d217fe YZ |
1072 | if (disk_bytenr == 0) |
1073 | goto out_check; | |
80ff3856 YZ |
1074 | if (btrfs_file_extent_compression(leaf, fi) || |
1075 | btrfs_file_extent_encryption(leaf, fi) || | |
1076 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1077 | goto out_check; | |
d899e052 YZ |
1078 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1079 | goto out_check; | |
d2fb3437 | 1080 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1081 | goto out_check; |
17d217fe | 1082 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1083 | found_key.offset - |
1084 | extent_offset, disk_bytenr)) | |
17d217fe | 1085 | goto out_check; |
5d4f98a2 | 1086 | disk_bytenr += extent_offset; |
17d217fe YZ |
1087 | disk_bytenr += cur_offset - found_key.offset; |
1088 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1089 | /* | |
1090 | * force cow if csum exists in the range. | |
1091 | * this ensure that csum for a given extent are | |
1092 | * either valid or do not exist. | |
1093 | */ | |
1094 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1095 | goto out_check; | |
80ff3856 YZ |
1096 | nocow = 1; |
1097 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1098 | extent_end = found_key.offset + | |
1099 | btrfs_file_extent_inline_len(leaf, fi); | |
1100 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1101 | } else { | |
1102 | BUG_ON(1); | |
1103 | } | |
1104 | out_check: | |
1105 | if (extent_end <= start) { | |
1106 | path->slots[0]++; | |
1107 | goto next_slot; | |
1108 | } | |
1109 | if (!nocow) { | |
1110 | if (cow_start == (u64)-1) | |
1111 | cow_start = cur_offset; | |
1112 | cur_offset = extent_end; | |
1113 | if (cur_offset > end) | |
1114 | break; | |
1115 | path->slots[0]++; | |
1116 | goto next_slot; | |
7ea394f1 YZ |
1117 | } |
1118 | ||
1119 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1120 | if (cow_start != (u64)-1) { |
1121 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1122 | found_key.offset - 1, page_started, |
1123 | nr_written, 1); | |
80ff3856 YZ |
1124 | BUG_ON(ret); |
1125 | cow_start = (u64)-1; | |
7ea394f1 | 1126 | } |
80ff3856 | 1127 | |
d899e052 YZ |
1128 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1129 | struct extent_map *em; | |
1130 | struct extent_map_tree *em_tree; | |
1131 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1132 | em = alloc_extent_map(GFP_NOFS); | |
1133 | em->start = cur_offset; | |
445a6944 | 1134 | em->orig_start = em->start; |
d899e052 YZ |
1135 | em->len = num_bytes; |
1136 | em->block_len = num_bytes; | |
1137 | em->block_start = disk_bytenr; | |
1138 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1139 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1140 | while (1) { | |
890871be | 1141 | write_lock(&em_tree->lock); |
d899e052 | 1142 | ret = add_extent_mapping(em_tree, em); |
890871be | 1143 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1144 | if (ret != -EEXIST) { |
1145 | free_extent_map(em); | |
1146 | break; | |
1147 | } | |
1148 | btrfs_drop_extent_cache(inode, em->start, | |
1149 | em->start + em->len - 1, 0); | |
1150 | } | |
1151 | type = BTRFS_ORDERED_PREALLOC; | |
1152 | } else { | |
1153 | type = BTRFS_ORDERED_NOCOW; | |
1154 | } | |
80ff3856 YZ |
1155 | |
1156 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1157 | num_bytes, num_bytes, type); |
1158 | BUG_ON(ret); | |
771ed689 | 1159 | |
d899e052 | 1160 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1161 | cur_offset, cur_offset + num_bytes - 1, |
1162 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1163 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1164 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1165 | cur_offset = extent_end; |
1166 | if (cur_offset > end) | |
1167 | break; | |
be20aa9d | 1168 | } |
80ff3856 YZ |
1169 | btrfs_release_path(root, path); |
1170 | ||
1171 | if (cur_offset <= end && cow_start == (u64)-1) | |
1172 | cow_start = cur_offset; | |
1173 | if (cow_start != (u64)-1) { | |
1174 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1175 | page_started, nr_written, 1); |
80ff3856 YZ |
1176 | BUG_ON(ret); |
1177 | } | |
1178 | ||
1179 | ret = btrfs_end_transaction(trans, root); | |
1180 | BUG_ON(ret); | |
7ea394f1 | 1181 | btrfs_free_path(path); |
80ff3856 | 1182 | return 0; |
be20aa9d CM |
1183 | } |
1184 | ||
d352ac68 CM |
1185 | /* |
1186 | * extent_io.c call back to do delayed allocation processing | |
1187 | */ | |
c8b97818 | 1188 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1189 | u64 start, u64 end, int *page_started, |
1190 | unsigned long *nr_written) | |
be20aa9d | 1191 | { |
be20aa9d | 1192 | int ret; |
7f366cfe | 1193 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1194 | |
6cbff00f | 1195 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1196 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1197 | page_started, 1, nr_written); |
6cbff00f | 1198 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1199 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1200 | page_started, 0, nr_written); |
7f366cfe CM |
1201 | else if (!btrfs_test_opt(root, COMPRESS)) |
1202 | ret = cow_file_range(inode, locked_page, start, end, | |
1203 | page_started, nr_written, 1); | |
be20aa9d | 1204 | else |
771ed689 | 1205 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1206 | page_started, nr_written); |
b888db2b CM |
1207 | return ret; |
1208 | } | |
1209 | ||
9ed74f2d JB |
1210 | static int btrfs_split_extent_hook(struct inode *inode, |
1211 | struct extent_state *orig, u64 split) | |
1212 | { | |
1213 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1214 | u64 size; | |
1215 | ||
1216 | if (!(orig->state & EXTENT_DELALLOC)) | |
1217 | return 0; | |
1218 | ||
1219 | size = orig->end - orig->start + 1; | |
1220 | if (size > root->fs_info->max_extent) { | |
1221 | u64 num_extents; | |
1222 | u64 new_size; | |
1223 | ||
1224 | new_size = orig->end - split + 1; | |
1225 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1226 | root->fs_info->max_extent); | |
1227 | ||
1228 | /* | |
32c00aff JB |
1229 | * if we break a large extent up then leave oustanding_extents |
1230 | * be, since we've already accounted for the large extent. | |
9ed74f2d JB |
1231 | */ |
1232 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1233 | root->fs_info->max_extent) < num_extents) | |
1234 | return 0; | |
1235 | } | |
1236 | ||
32c00aff JB |
1237 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1238 | BTRFS_I(inode)->outstanding_extents++; | |
1239 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1240 | |
1241 | return 0; | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1246 | * extents so we can keep track of new extents that are just merged onto old | |
1247 | * extents, such as when we are doing sequential writes, so we can properly | |
1248 | * account for the metadata space we'll need. | |
1249 | */ | |
1250 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1251 | struct extent_state *new, | |
1252 | struct extent_state *other) | |
1253 | { | |
1254 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1255 | u64 new_size, old_size; | |
1256 | u64 num_extents; | |
1257 | ||
1258 | /* not delalloc, ignore it */ | |
1259 | if (!(other->state & EXTENT_DELALLOC)) | |
1260 | return 0; | |
1261 | ||
1262 | old_size = other->end - other->start + 1; | |
1263 | if (new->start < other->start) | |
1264 | new_size = other->end - new->start + 1; | |
1265 | else | |
1266 | new_size = new->end - other->start + 1; | |
1267 | ||
1268 | /* we're not bigger than the max, unreserve the space and go */ | |
1269 | if (new_size <= root->fs_info->max_extent) { | |
32c00aff JB |
1270 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1271 | BTRFS_I(inode)->outstanding_extents--; | |
1272 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1273 | return 0; |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * If we grew by another max_extent, just return, we want to keep that | |
1278 | * reserved amount. | |
1279 | */ | |
1280 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1281 | root->fs_info->max_extent); | |
1282 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1283 | root->fs_info->max_extent) > num_extents) | |
1284 | return 0; | |
1285 | ||
32c00aff JB |
1286 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1287 | BTRFS_I(inode)->outstanding_extents--; | |
1288 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1289 | |
1290 | return 0; | |
1291 | } | |
1292 | ||
d352ac68 CM |
1293 | /* |
1294 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1295 | * bytes in this file, and to maintain the list of inodes that | |
1296 | * have pending delalloc work to be done. | |
1297 | */ | |
b2950863 | 1298 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1299 | unsigned long old, unsigned long bits) |
291d673e | 1300 | { |
9ed74f2d | 1301 | |
75eff68e CM |
1302 | /* |
1303 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1304 | * but in this case, we are only testeing for the DELALLOC | |
1305 | * bit, which is only set or cleared with irqs on | |
1306 | */ | |
b0c68f8b | 1307 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1308 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d | 1309 | |
32c00aff JB |
1310 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1311 | BTRFS_I(inode)->outstanding_extents++; | |
1312 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
6a63209f | 1313 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1314 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1315 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1316 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1317 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1318 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1319 | &root->fs_info->delalloc_inodes); | |
1320 | } | |
75eff68e | 1321 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1322 | } |
1323 | return 0; | |
1324 | } | |
1325 | ||
d352ac68 CM |
1326 | /* |
1327 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1328 | */ | |
9ed74f2d JB |
1329 | static int btrfs_clear_bit_hook(struct inode *inode, |
1330 | struct extent_state *state, unsigned long bits) | |
291d673e | 1331 | { |
75eff68e CM |
1332 | /* |
1333 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1334 | * but in this case, we are only testeing for the DELALLOC | |
1335 | * bit, which is only set or cleared with irqs on | |
1336 | */ | |
9ed74f2d | 1337 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1338 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1339 | |
32c00aff JB |
1340 | if (bits & EXTENT_DO_ACCOUNTING) { |
1341 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
1342 | BTRFS_I(inode)->outstanding_extents--; | |
1343 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
1344 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1345 | } | |
9ed74f2d | 1346 | |
75eff68e | 1347 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1348 | if (state->end - state->start + 1 > |
1349 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1350 | printk(KERN_INFO "btrfs warning: delalloc account " |
1351 | "%llu %llu\n", | |
9ed74f2d JB |
1352 | (unsigned long long) |
1353 | state->end - state->start + 1, | |
d397712b CM |
1354 | (unsigned long long) |
1355 | root->fs_info->delalloc_bytes); | |
6a63209f | 1356 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1357 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1358 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1359 | } else { |
6a63209f | 1360 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1361 | state->end - |
1362 | state->start + 1); | |
1363 | root->fs_info->delalloc_bytes -= state->end - | |
1364 | state->start + 1; | |
1365 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1366 | state->start + 1; | |
b0c68f8b | 1367 | } |
ea8c2819 CM |
1368 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1369 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1370 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1371 | } | |
75eff68e | 1372 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1373 | } |
1374 | return 0; | |
1375 | } | |
1376 | ||
d352ac68 CM |
1377 | /* |
1378 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1379 | * we don't create bios that span stripes or chunks | |
1380 | */ | |
239b14b3 | 1381 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1382 | size_t size, struct bio *bio, |
1383 | unsigned long bio_flags) | |
239b14b3 CM |
1384 | { |
1385 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1386 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1387 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1388 | u64 length = 0; |
1389 | u64 map_length; | |
239b14b3 CM |
1390 | int ret; |
1391 | ||
771ed689 CM |
1392 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1393 | return 0; | |
1394 | ||
f2d8d74d | 1395 | length = bio->bi_size; |
239b14b3 CM |
1396 | map_tree = &root->fs_info->mapping_tree; |
1397 | map_length = length; | |
cea9e445 | 1398 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1399 | &map_length, NULL, 0); |
cea9e445 | 1400 | |
d397712b | 1401 | if (map_length < length + size) |
239b14b3 | 1402 | return 1; |
239b14b3 CM |
1403 | return 0; |
1404 | } | |
1405 | ||
d352ac68 CM |
1406 | /* |
1407 | * in order to insert checksums into the metadata in large chunks, | |
1408 | * we wait until bio submission time. All the pages in the bio are | |
1409 | * checksummed and sums are attached onto the ordered extent record. | |
1410 | * | |
1411 | * At IO completion time the cums attached on the ordered extent record | |
1412 | * are inserted into the btree | |
1413 | */ | |
d397712b CM |
1414 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1415 | struct bio *bio, int mirror_num, | |
1416 | unsigned long bio_flags) | |
065631f6 | 1417 | { |
065631f6 | 1418 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1419 | int ret = 0; |
e015640f | 1420 | |
d20f7043 | 1421 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1422 | BUG_ON(ret); |
4a69a410 CM |
1423 | return 0; |
1424 | } | |
e015640f | 1425 | |
4a69a410 CM |
1426 | /* |
1427 | * in order to insert checksums into the metadata in large chunks, | |
1428 | * we wait until bio submission time. All the pages in the bio are | |
1429 | * checksummed and sums are attached onto the ordered extent record. | |
1430 | * | |
1431 | * At IO completion time the cums attached on the ordered extent record | |
1432 | * are inserted into the btree | |
1433 | */ | |
b2950863 | 1434 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1435 | int mirror_num, unsigned long bio_flags) |
1436 | { | |
1437 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1438 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1439 | } |
1440 | ||
d352ac68 | 1441 | /* |
cad321ad CM |
1442 | * extent_io.c submission hook. This does the right thing for csum calculation |
1443 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1444 | */ |
b2950863 | 1445 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1446 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1447 | { |
1448 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1449 | int ret = 0; | |
19b9bdb0 | 1450 | int skip_sum; |
44b8bd7e | 1451 | |
6cbff00f | 1452 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1453 | |
e6dcd2dc CM |
1454 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1455 | BUG_ON(ret); | |
065631f6 | 1456 | |
4d1b5fb4 | 1457 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1458 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1459 | return btrfs_submit_compressed_read(inode, bio, |
1460 | mirror_num, bio_flags); | |
d20f7043 CM |
1461 | } else if (!skip_sum) |
1462 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1463 | goto mapit; |
19b9bdb0 | 1464 | } else if (!skip_sum) { |
17d217fe YZ |
1465 | /* csum items have already been cloned */ |
1466 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1467 | goto mapit; | |
19b9bdb0 CM |
1468 | /* we're doing a write, do the async checksumming */ |
1469 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1470 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1471 | bio_flags, __btrfs_submit_bio_start, |
1472 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1473 | } |
1474 | ||
0b86a832 | 1475 | mapit: |
8b712842 | 1476 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1477 | } |
6885f308 | 1478 | |
d352ac68 CM |
1479 | /* |
1480 | * given a list of ordered sums record them in the inode. This happens | |
1481 | * at IO completion time based on sums calculated at bio submission time. | |
1482 | */ | |
ba1da2f4 | 1483 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1484 | struct inode *inode, u64 file_offset, |
1485 | struct list_head *list) | |
1486 | { | |
e6dcd2dc CM |
1487 | struct btrfs_ordered_sum *sum; |
1488 | ||
1489 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1490 | |
1491 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1492 | btrfs_csum_file_blocks(trans, |
1493 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1494 | } |
1495 | return 0; | |
1496 | } | |
1497 | ||
ea8c2819 CM |
1498 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1499 | { | |
d397712b | 1500 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1501 | WARN_ON(1); |
ea8c2819 CM |
1502 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1503 | GFP_NOFS); | |
1504 | } | |
1505 | ||
d352ac68 | 1506 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1507 | struct btrfs_writepage_fixup { |
1508 | struct page *page; | |
1509 | struct btrfs_work work; | |
1510 | }; | |
1511 | ||
b2950863 | 1512 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1513 | { |
1514 | struct btrfs_writepage_fixup *fixup; | |
1515 | struct btrfs_ordered_extent *ordered; | |
1516 | struct page *page; | |
1517 | struct inode *inode; | |
1518 | u64 page_start; | |
1519 | u64 page_end; | |
1520 | ||
1521 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1522 | page = fixup->page; | |
4a096752 | 1523 | again: |
247e743c CM |
1524 | lock_page(page); |
1525 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1526 | ClearPageChecked(page); | |
1527 | goto out_page; | |
1528 | } | |
1529 | ||
1530 | inode = page->mapping->host; | |
1531 | page_start = page_offset(page); | |
1532 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1533 | ||
1534 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1535 | |
1536 | /* already ordered? We're done */ | |
8b62b72b | 1537 | if (PagePrivate2(page)) |
247e743c | 1538 | goto out; |
4a096752 CM |
1539 | |
1540 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1541 | if (ordered) { | |
1542 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1543 | page_end, GFP_NOFS); | |
1544 | unlock_page(page); | |
1545 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1546 | goto again; | |
1547 | } | |
247e743c | 1548 | |
ea8c2819 | 1549 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1550 | ClearPageChecked(page); |
1551 | out: | |
1552 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1553 | out_page: | |
1554 | unlock_page(page); | |
1555 | page_cache_release(page); | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * There are a few paths in the higher layers of the kernel that directly | |
1560 | * set the page dirty bit without asking the filesystem if it is a | |
1561 | * good idea. This causes problems because we want to make sure COW | |
1562 | * properly happens and the data=ordered rules are followed. | |
1563 | * | |
c8b97818 | 1564 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1565 | * hasn't been properly setup for IO. We kick off an async process |
1566 | * to fix it up. The async helper will wait for ordered extents, set | |
1567 | * the delalloc bit and make it safe to write the page. | |
1568 | */ | |
b2950863 | 1569 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1570 | { |
1571 | struct inode *inode = page->mapping->host; | |
1572 | struct btrfs_writepage_fixup *fixup; | |
1573 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1574 | |
8b62b72b CM |
1575 | /* this page is properly in the ordered list */ |
1576 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1577 | return 0; |
1578 | ||
1579 | if (PageChecked(page)) | |
1580 | return -EAGAIN; | |
1581 | ||
1582 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1583 | if (!fixup) | |
1584 | return -EAGAIN; | |
f421950f | 1585 | |
247e743c CM |
1586 | SetPageChecked(page); |
1587 | page_cache_get(page); | |
1588 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1589 | fixup->page = page; | |
1590 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1591 | return -EAGAIN; | |
1592 | } | |
1593 | ||
d899e052 YZ |
1594 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1595 | struct inode *inode, u64 file_pos, | |
1596 | u64 disk_bytenr, u64 disk_num_bytes, | |
1597 | u64 num_bytes, u64 ram_bytes, | |
1598 | u8 compression, u8 encryption, | |
1599 | u16 other_encoding, int extent_type) | |
1600 | { | |
1601 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1602 | struct btrfs_file_extent_item *fi; | |
1603 | struct btrfs_path *path; | |
1604 | struct extent_buffer *leaf; | |
1605 | struct btrfs_key ins; | |
1606 | u64 hint; | |
1607 | int ret; | |
1608 | ||
1609 | path = btrfs_alloc_path(); | |
1610 | BUG_ON(!path); | |
1611 | ||
b9473439 | 1612 | path->leave_spinning = 1; |
a1ed835e CM |
1613 | |
1614 | /* | |
1615 | * we may be replacing one extent in the tree with another. | |
1616 | * The new extent is pinned in the extent map, and we don't want | |
1617 | * to drop it from the cache until it is completely in the btree. | |
1618 | * | |
1619 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1620 | * the caller is expected to unpin it and allow it to be merged | |
1621 | * with the others. | |
1622 | */ | |
920bbbfb YZ |
1623 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1624 | &hint, 0); | |
d899e052 YZ |
1625 | BUG_ON(ret); |
1626 | ||
1627 | ins.objectid = inode->i_ino; | |
1628 | ins.offset = file_pos; | |
1629 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1630 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1631 | BUG_ON(ret); | |
1632 | leaf = path->nodes[0]; | |
1633 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1634 | struct btrfs_file_extent_item); | |
1635 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1636 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1637 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1638 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1639 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1640 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1641 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1642 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1643 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1644 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1645 | |
1646 | btrfs_unlock_up_safe(path, 1); | |
1647 | btrfs_set_lock_blocking(leaf); | |
1648 | ||
d899e052 YZ |
1649 | btrfs_mark_buffer_dirty(leaf); |
1650 | ||
1651 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1652 | |
1653 | ins.objectid = disk_bytenr; | |
1654 | ins.offset = disk_num_bytes; | |
1655 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1656 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1657 | root->root_key.objectid, | |
1658 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1659 | BUG_ON(ret); |
d899e052 | 1660 | btrfs_free_path(path); |
b9473439 | 1661 | |
d899e052 YZ |
1662 | return 0; |
1663 | } | |
1664 | ||
5d13a98f CM |
1665 | /* |
1666 | * helper function for btrfs_finish_ordered_io, this | |
1667 | * just reads in some of the csum leaves to prime them into ram | |
1668 | * before we start the transaction. It limits the amount of btree | |
1669 | * reads required while inside the transaction. | |
1670 | */ | |
1671 | static noinline void reada_csum(struct btrfs_root *root, | |
1672 | struct btrfs_path *path, | |
1673 | struct btrfs_ordered_extent *ordered_extent) | |
1674 | { | |
1675 | struct btrfs_ordered_sum *sum; | |
1676 | u64 bytenr; | |
1677 | ||
1678 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1679 | list); | |
1680 | bytenr = sum->sums[0].bytenr; | |
1681 | ||
1682 | /* | |
1683 | * we don't care about the results, the point of this search is | |
1684 | * just to get the btree leaves into ram | |
1685 | */ | |
1686 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1687 | } | |
1688 | ||
d352ac68 CM |
1689 | /* as ordered data IO finishes, this gets called so we can finish |
1690 | * an ordered extent if the range of bytes in the file it covers are | |
1691 | * fully written. | |
1692 | */ | |
211f90e6 | 1693 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1694 | { |
e6dcd2dc CM |
1695 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1696 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1697 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1698 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1699 | struct btrfs_path *path; |
d899e052 | 1700 | int compressed = 0; |
e6dcd2dc CM |
1701 | int ret; |
1702 | ||
1703 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1704 | if (!ret) |
e6dcd2dc | 1705 | return 0; |
e6dcd2dc | 1706 | |
b7ec40d7 CM |
1707 | /* |
1708 | * before we join the transaction, try to do some of our IO. | |
1709 | * This will limit the amount of IO that we have to do with | |
1710 | * the transaction running. We're unlikely to need to do any | |
1711 | * IO if the file extents are new, the disk_i_size checks | |
1712 | * covers the most common case. | |
1713 | */ | |
1714 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1715 | path = btrfs_alloc_path(); | |
1716 | if (path) { | |
1717 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1718 | inode->i_ino, | |
1719 | start, 0); | |
5d13a98f CM |
1720 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1721 | start); | |
1722 | if (!list_empty(&ordered_extent->list)) { | |
1723 | btrfs_release_path(root, path); | |
1724 | reada_csum(root, path, ordered_extent); | |
1725 | } | |
b7ec40d7 CM |
1726 | btrfs_free_path(path); |
1727 | } | |
1728 | } | |
1729 | ||
f9295749 | 1730 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc | 1731 | |
5d13a98f CM |
1732 | if (!ordered_extent) |
1733 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1734 | BUG_ON(!ordered_extent); |
7ea394f1 YZ |
1735 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1736 | goto nocow; | |
e6dcd2dc CM |
1737 | |
1738 | lock_extent(io_tree, ordered_extent->file_offset, | |
1739 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1740 | GFP_NOFS); | |
1741 | ||
c8b97818 | 1742 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1743 | compressed = 1; |
1744 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1745 | BUG_ON(compressed); | |
920bbbfb | 1746 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1747 | ordered_extent->file_offset, |
1748 | ordered_extent->file_offset + | |
1749 | ordered_extent->len); | |
1750 | BUG_ON(ret); | |
1751 | } else { | |
1752 | ret = insert_reserved_file_extent(trans, inode, | |
1753 | ordered_extent->file_offset, | |
1754 | ordered_extent->start, | |
1755 | ordered_extent->disk_len, | |
1756 | ordered_extent->len, | |
1757 | ordered_extent->len, | |
1758 | compressed, 0, 0, | |
1759 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1760 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1761 | ordered_extent->file_offset, | |
1762 | ordered_extent->len); | |
d899e052 YZ |
1763 | BUG_ON(ret); |
1764 | } | |
e6dcd2dc CM |
1765 | unlock_extent(io_tree, ordered_extent->file_offset, |
1766 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1767 | GFP_NOFS); | |
7ea394f1 | 1768 | nocow: |
e6dcd2dc CM |
1769 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1770 | &ordered_extent->list); | |
1771 | ||
34353029 | 1772 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1773 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1774 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1775 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1776 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1777 | |
e6dcd2dc CM |
1778 | /* once for us */ |
1779 | btrfs_put_ordered_extent(ordered_extent); | |
1780 | /* once for the tree */ | |
1781 | btrfs_put_ordered_extent(ordered_extent); | |
1782 | ||
e6dcd2dc CM |
1783 | btrfs_end_transaction(trans, root); |
1784 | return 0; | |
1785 | } | |
1786 | ||
b2950863 | 1787 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1788 | struct extent_state *state, int uptodate) |
1789 | { | |
8b62b72b | 1790 | ClearPagePrivate2(page); |
211f90e6 CM |
1791 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1792 | } | |
1793 | ||
d352ac68 CM |
1794 | /* |
1795 | * When IO fails, either with EIO or csum verification fails, we | |
1796 | * try other mirrors that might have a good copy of the data. This | |
1797 | * io_failure_record is used to record state as we go through all the | |
1798 | * mirrors. If another mirror has good data, the page is set up to date | |
1799 | * and things continue. If a good mirror can't be found, the original | |
1800 | * bio end_io callback is called to indicate things have failed. | |
1801 | */ | |
7e38326f CM |
1802 | struct io_failure_record { |
1803 | struct page *page; | |
1804 | u64 start; | |
1805 | u64 len; | |
1806 | u64 logical; | |
d20f7043 | 1807 | unsigned long bio_flags; |
7e38326f CM |
1808 | int last_mirror; |
1809 | }; | |
1810 | ||
b2950863 | 1811 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1812 | struct page *page, u64 start, u64 end, |
1813 | struct extent_state *state) | |
7e38326f CM |
1814 | { |
1815 | struct io_failure_record *failrec = NULL; | |
1816 | u64 private; | |
1817 | struct extent_map *em; | |
1818 | struct inode *inode = page->mapping->host; | |
1819 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1820 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1821 | struct bio *bio; |
1822 | int num_copies; | |
1823 | int ret; | |
1259ab75 | 1824 | int rw; |
7e38326f CM |
1825 | u64 logical; |
1826 | ||
1827 | ret = get_state_private(failure_tree, start, &private); | |
1828 | if (ret) { | |
7e38326f CM |
1829 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1830 | if (!failrec) | |
1831 | return -ENOMEM; | |
1832 | failrec->start = start; | |
1833 | failrec->len = end - start + 1; | |
1834 | failrec->last_mirror = 0; | |
d20f7043 | 1835 | failrec->bio_flags = 0; |
7e38326f | 1836 | |
890871be | 1837 | read_lock(&em_tree->lock); |
3b951516 CM |
1838 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1839 | if (em->start > start || em->start + em->len < start) { | |
1840 | free_extent_map(em); | |
1841 | em = NULL; | |
1842 | } | |
890871be | 1843 | read_unlock(&em_tree->lock); |
7e38326f CM |
1844 | |
1845 | if (!em || IS_ERR(em)) { | |
1846 | kfree(failrec); | |
1847 | return -EIO; | |
1848 | } | |
1849 | logical = start - em->start; | |
1850 | logical = em->block_start + logical; | |
d20f7043 CM |
1851 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1852 | logical = em->block_start; | |
1853 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1854 | } | |
7e38326f CM |
1855 | failrec->logical = logical; |
1856 | free_extent_map(em); | |
1857 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1858 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1859 | set_state_private(failure_tree, start, |
1860 | (u64)(unsigned long)failrec); | |
7e38326f | 1861 | } else { |
587f7704 | 1862 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1863 | } |
1864 | num_copies = btrfs_num_copies( | |
1865 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1866 | failrec->logical, failrec->len); | |
1867 | failrec->last_mirror++; | |
1868 | if (!state) { | |
cad321ad | 1869 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1870 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1871 | failrec->start, | |
1872 | EXTENT_LOCKED); | |
1873 | if (state && state->start != failrec->start) | |
1874 | state = NULL; | |
cad321ad | 1875 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1876 | } |
1877 | if (!state || failrec->last_mirror > num_copies) { | |
1878 | set_state_private(failure_tree, failrec->start, 0); | |
1879 | clear_extent_bits(failure_tree, failrec->start, | |
1880 | failrec->start + failrec->len - 1, | |
1881 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1882 | kfree(failrec); | |
1883 | return -EIO; | |
1884 | } | |
1885 | bio = bio_alloc(GFP_NOFS, 1); | |
1886 | bio->bi_private = state; | |
1887 | bio->bi_end_io = failed_bio->bi_end_io; | |
1888 | bio->bi_sector = failrec->logical >> 9; | |
1889 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1890 | bio->bi_size = 0; |
d20f7043 | 1891 | |
7e38326f | 1892 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1893 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1894 | rw = WRITE; | |
1895 | else | |
1896 | rw = READ; | |
1897 | ||
1898 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1899 | failrec->last_mirror, |
d20f7043 | 1900 | failrec->bio_flags); |
1259ab75 CM |
1901 | return 0; |
1902 | } | |
1903 | ||
d352ac68 CM |
1904 | /* |
1905 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1906 | * to see if we need to process or clean up an io_failure_record | |
1907 | */ | |
b2950863 | 1908 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1909 | { |
1910 | u64 private; | |
1911 | u64 private_failure; | |
1912 | struct io_failure_record *failure; | |
1913 | int ret; | |
1914 | ||
1915 | private = 0; | |
1916 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1917 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1918 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1919 | start, &private_failure); | |
1920 | if (ret == 0) { | |
1921 | failure = (struct io_failure_record *)(unsigned long) | |
1922 | private_failure; | |
1923 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1924 | failure->start, 0); | |
1925 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1926 | failure->start, | |
1927 | failure->start + failure->len - 1, | |
1928 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1929 | GFP_NOFS); | |
1930 | kfree(failure); | |
1931 | } | |
1932 | } | |
7e38326f CM |
1933 | return 0; |
1934 | } | |
1935 | ||
d352ac68 CM |
1936 | /* |
1937 | * when reads are done, we need to check csums to verify the data is correct | |
1938 | * if there's a match, we allow the bio to finish. If not, we go through | |
1939 | * the io_failure_record routines to find good copies | |
1940 | */ | |
b2950863 | 1941 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1942 | struct extent_state *state) |
07157aac | 1943 | { |
35ebb934 | 1944 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1945 | struct inode *inode = page->mapping->host; |
d1310b2e | 1946 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1947 | char *kaddr; |
aadfeb6e | 1948 | u64 private = ~(u32)0; |
07157aac | 1949 | int ret; |
ff79f819 CM |
1950 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1951 | u32 csum = ~(u32)0; | |
d1310b2e | 1952 | |
d20f7043 CM |
1953 | if (PageChecked(page)) { |
1954 | ClearPageChecked(page); | |
1955 | goto good; | |
1956 | } | |
6cbff00f CH |
1957 | |
1958 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1959 | return 0; |
1960 | ||
1961 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1962 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1963 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1964 | GFP_NOFS); | |
b6cda9bc | 1965 | return 0; |
17d217fe | 1966 | } |
d20f7043 | 1967 | |
c2e639f0 | 1968 | if (state && state->start == start) { |
70dec807 CM |
1969 | private = state->private; |
1970 | ret = 0; | |
1971 | } else { | |
1972 | ret = get_state_private(io_tree, start, &private); | |
1973 | } | |
9ab86c8e | 1974 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1975 | if (ret) |
07157aac | 1976 | goto zeroit; |
d397712b | 1977 | |
ff79f819 CM |
1978 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1979 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1980 | if (csum != private) |
07157aac | 1981 | goto zeroit; |
d397712b | 1982 | |
9ab86c8e | 1983 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1984 | good: |
7e38326f CM |
1985 | /* if the io failure tree for this inode is non-empty, |
1986 | * check to see if we've recovered from a failed IO | |
1987 | */ | |
1259ab75 | 1988 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1989 | return 0; |
1990 | ||
1991 | zeroit: | |
193f284d CM |
1992 | if (printk_ratelimit()) { |
1993 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1994 | "private %llu\n", page->mapping->host->i_ino, | |
1995 | (unsigned long long)start, csum, | |
1996 | (unsigned long long)private); | |
1997 | } | |
db94535d CM |
1998 | memset(kaddr + offset, 1, end - start + 1); |
1999 | flush_dcache_page(page); | |
9ab86c8e | 2000 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2001 | if (private == 0) |
2002 | return 0; | |
7e38326f | 2003 | return -EIO; |
07157aac | 2004 | } |
b888db2b | 2005 | |
7b128766 JB |
2006 | /* |
2007 | * This creates an orphan entry for the given inode in case something goes | |
2008 | * wrong in the middle of an unlink/truncate. | |
2009 | */ | |
2010 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2011 | { | |
2012 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2013 | int ret = 0; | |
2014 | ||
bcc63abb | 2015 | spin_lock(&root->list_lock); |
7b128766 JB |
2016 | |
2017 | /* already on the orphan list, we're good */ | |
2018 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2019 | spin_unlock(&root->list_lock); |
7b128766 JB |
2020 | return 0; |
2021 | } | |
2022 | ||
2023 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2024 | ||
bcc63abb | 2025 | spin_unlock(&root->list_lock); |
7b128766 JB |
2026 | |
2027 | /* | |
2028 | * insert an orphan item to track this unlinked/truncated file | |
2029 | */ | |
2030 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2031 | ||
2032 | return ret; | |
2033 | } | |
2034 | ||
2035 | /* | |
2036 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2037 | * item for this particular inode. | |
2038 | */ | |
2039 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2040 | { | |
2041 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2042 | int ret = 0; | |
2043 | ||
bcc63abb | 2044 | spin_lock(&root->list_lock); |
7b128766 JB |
2045 | |
2046 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2047 | spin_unlock(&root->list_lock); |
7b128766 JB |
2048 | return 0; |
2049 | } | |
2050 | ||
2051 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2052 | if (!trans) { | |
bcc63abb | 2053 | spin_unlock(&root->list_lock); |
7b128766 JB |
2054 | return 0; |
2055 | } | |
2056 | ||
bcc63abb | 2057 | spin_unlock(&root->list_lock); |
7b128766 JB |
2058 | |
2059 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2060 | ||
2061 | return ret; | |
2062 | } | |
2063 | ||
2064 | /* | |
2065 | * this cleans up any orphans that may be left on the list from the last use | |
2066 | * of this root. | |
2067 | */ | |
2068 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2069 | { | |
2070 | struct btrfs_path *path; | |
2071 | struct extent_buffer *leaf; | |
2072 | struct btrfs_item *item; | |
2073 | struct btrfs_key key, found_key; | |
2074 | struct btrfs_trans_handle *trans; | |
2075 | struct inode *inode; | |
2076 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2077 | ||
7b128766 JB |
2078 | path = btrfs_alloc_path(); |
2079 | if (!path) | |
2080 | return; | |
2081 | path->reada = -1; | |
2082 | ||
2083 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2084 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2085 | key.offset = (u64)-1; | |
2086 | ||
7b128766 JB |
2087 | |
2088 | while (1) { | |
2089 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2090 | if (ret < 0) { | |
2091 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2092 | "\n", ret); | |
2093 | break; | |
2094 | } | |
2095 | ||
2096 | /* | |
2097 | * if ret == 0 means we found what we were searching for, which | |
2098 | * is weird, but possible, so only screw with path if we didnt | |
2099 | * find the key and see if we have stuff that matches | |
2100 | */ | |
2101 | if (ret > 0) { | |
2102 | if (path->slots[0] == 0) | |
2103 | break; | |
2104 | path->slots[0]--; | |
2105 | } | |
2106 | ||
2107 | /* pull out the item */ | |
2108 | leaf = path->nodes[0]; | |
2109 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2110 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2111 | ||
2112 | /* make sure the item matches what we want */ | |
2113 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2114 | break; | |
2115 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2116 | break; | |
2117 | ||
2118 | /* release the path since we're done with it */ | |
2119 | btrfs_release_path(root, path); | |
2120 | ||
2121 | /* | |
2122 | * this is where we are basically btrfs_lookup, without the | |
2123 | * crossing root thing. we store the inode number in the | |
2124 | * offset of the orphan item. | |
2125 | */ | |
5d4f98a2 YZ |
2126 | found_key.objectid = found_key.offset; |
2127 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2128 | found_key.offset = 0; | |
2129 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2130 | if (IS_ERR(inode)) | |
7b128766 JB |
2131 | break; |
2132 | ||
7b128766 JB |
2133 | /* |
2134 | * add this inode to the orphan list so btrfs_orphan_del does | |
2135 | * the proper thing when we hit it | |
2136 | */ | |
bcc63abb | 2137 | spin_lock(&root->list_lock); |
7b128766 | 2138 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2139 | spin_unlock(&root->list_lock); |
7b128766 JB |
2140 | |
2141 | /* | |
2142 | * if this is a bad inode, means we actually succeeded in | |
2143 | * removing the inode, but not the orphan record, which means | |
2144 | * we need to manually delete the orphan since iput will just | |
2145 | * do a destroy_inode | |
2146 | */ | |
2147 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2148 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2149 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2150 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2151 | iput(inode); |
2152 | continue; | |
2153 | } | |
2154 | ||
2155 | /* if we have links, this was a truncate, lets do that */ | |
2156 | if (inode->i_nlink) { | |
2157 | nr_truncate++; | |
2158 | btrfs_truncate(inode); | |
2159 | } else { | |
2160 | nr_unlink++; | |
2161 | } | |
2162 | ||
2163 | /* this will do delete_inode and everything for us */ | |
2164 | iput(inode); | |
2165 | } | |
2166 | ||
2167 | if (nr_unlink) | |
2168 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2169 | if (nr_truncate) | |
2170 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2171 | ||
2172 | btrfs_free_path(path); | |
7b128766 JB |
2173 | } |
2174 | ||
46a53cca CM |
2175 | /* |
2176 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2177 | * don't find any xattrs, we know there can't be any acls. | |
2178 | * | |
2179 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2180 | */ | |
2181 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2182 | int slot, u64 objectid) | |
2183 | { | |
2184 | u32 nritems = btrfs_header_nritems(leaf); | |
2185 | struct btrfs_key found_key; | |
2186 | int scanned = 0; | |
2187 | ||
2188 | slot++; | |
2189 | while (slot < nritems) { | |
2190 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2191 | ||
2192 | /* we found a different objectid, there must not be acls */ | |
2193 | if (found_key.objectid != objectid) | |
2194 | return 0; | |
2195 | ||
2196 | /* we found an xattr, assume we've got an acl */ | |
2197 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2198 | return 1; | |
2199 | ||
2200 | /* | |
2201 | * we found a key greater than an xattr key, there can't | |
2202 | * be any acls later on | |
2203 | */ | |
2204 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2205 | return 0; | |
2206 | ||
2207 | slot++; | |
2208 | scanned++; | |
2209 | ||
2210 | /* | |
2211 | * it goes inode, inode backrefs, xattrs, extents, | |
2212 | * so if there are a ton of hard links to an inode there can | |
2213 | * be a lot of backrefs. Don't waste time searching too hard, | |
2214 | * this is just an optimization | |
2215 | */ | |
2216 | if (scanned >= 8) | |
2217 | break; | |
2218 | } | |
2219 | /* we hit the end of the leaf before we found an xattr or | |
2220 | * something larger than an xattr. We have to assume the inode | |
2221 | * has acls | |
2222 | */ | |
2223 | return 1; | |
2224 | } | |
2225 | ||
d352ac68 CM |
2226 | /* |
2227 | * read an inode from the btree into the in-memory inode | |
2228 | */ | |
5d4f98a2 | 2229 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2230 | { |
2231 | struct btrfs_path *path; | |
5f39d397 | 2232 | struct extent_buffer *leaf; |
39279cc3 | 2233 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2234 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2235 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2236 | struct btrfs_key location; | |
46a53cca | 2237 | int maybe_acls; |
39279cc3 | 2238 | u64 alloc_group_block; |
618e21d5 | 2239 | u32 rdev; |
39279cc3 CM |
2240 | int ret; |
2241 | ||
2242 | path = btrfs_alloc_path(); | |
2243 | BUG_ON(!path); | |
39279cc3 | 2244 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2245 | |
39279cc3 | 2246 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2247 | if (ret) |
39279cc3 | 2248 | goto make_bad; |
39279cc3 | 2249 | |
5f39d397 CM |
2250 | leaf = path->nodes[0]; |
2251 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2252 | struct btrfs_inode_item); | |
2253 | ||
2254 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2255 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2256 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2257 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2258 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2259 | |
2260 | tspec = btrfs_inode_atime(inode_item); | |
2261 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2262 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2263 | ||
2264 | tspec = btrfs_inode_mtime(inode_item); | |
2265 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2266 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2267 | ||
2268 | tspec = btrfs_inode_ctime(inode_item); | |
2269 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2270 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2271 | ||
a76a3cd4 | 2272 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2273 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2274 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2275 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2276 | inode->i_rdev = 0; |
5f39d397 CM |
2277 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2278 | ||
aec7477b | 2279 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2280 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2281 | |
5f39d397 | 2282 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2283 | |
46a53cca CM |
2284 | /* |
2285 | * try to precache a NULL acl entry for files that don't have | |
2286 | * any xattrs or acls | |
2287 | */ | |
2288 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2289 | if (!maybe_acls) |
2290 | cache_no_acl(inode); | |
46a53cca | 2291 | |
d2fb3437 YZ |
2292 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2293 | alloc_group_block, 0); | |
39279cc3 CM |
2294 | btrfs_free_path(path); |
2295 | inode_item = NULL; | |
2296 | ||
39279cc3 | 2297 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2298 | case S_IFREG: |
2299 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2300 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2301 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2302 | inode->i_fop = &btrfs_file_operations; |
2303 | inode->i_op = &btrfs_file_inode_operations; | |
2304 | break; | |
2305 | case S_IFDIR: | |
2306 | inode->i_fop = &btrfs_dir_file_operations; | |
2307 | if (root == root->fs_info->tree_root) | |
2308 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2309 | else | |
2310 | inode->i_op = &btrfs_dir_inode_operations; | |
2311 | break; | |
2312 | case S_IFLNK: | |
2313 | inode->i_op = &btrfs_symlink_inode_operations; | |
2314 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2315 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2316 | break; |
618e21d5 | 2317 | default: |
0279b4cd | 2318 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2319 | init_special_inode(inode, inode->i_mode, rdev); |
2320 | break; | |
39279cc3 | 2321 | } |
6cbff00f CH |
2322 | |
2323 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2324 | return; |
2325 | ||
2326 | make_bad: | |
39279cc3 | 2327 | btrfs_free_path(path); |
39279cc3 CM |
2328 | make_bad_inode(inode); |
2329 | } | |
2330 | ||
d352ac68 CM |
2331 | /* |
2332 | * given a leaf and an inode, copy the inode fields into the leaf | |
2333 | */ | |
e02119d5 CM |
2334 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2335 | struct extent_buffer *leaf, | |
5f39d397 | 2336 | struct btrfs_inode_item *item, |
39279cc3 CM |
2337 | struct inode *inode) |
2338 | { | |
5f39d397 CM |
2339 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2340 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2341 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2342 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2343 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2344 | ||
2345 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2346 | inode->i_atime.tv_sec); | |
2347 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2348 | inode->i_atime.tv_nsec); | |
2349 | ||
2350 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2351 | inode->i_mtime.tv_sec); | |
2352 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2353 | inode->i_mtime.tv_nsec); | |
2354 | ||
2355 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2356 | inode->i_ctime.tv_sec); | |
2357 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2358 | inode->i_ctime.tv_nsec); | |
2359 | ||
a76a3cd4 | 2360 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2361 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2362 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2363 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2364 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2365 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2366 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2367 | } |
2368 | ||
d352ac68 CM |
2369 | /* |
2370 | * copy everything in the in-memory inode into the btree. | |
2371 | */ | |
d397712b CM |
2372 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2373 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2374 | { |
2375 | struct btrfs_inode_item *inode_item; | |
2376 | struct btrfs_path *path; | |
5f39d397 | 2377 | struct extent_buffer *leaf; |
39279cc3 CM |
2378 | int ret; |
2379 | ||
2380 | path = btrfs_alloc_path(); | |
2381 | BUG_ON(!path); | |
b9473439 | 2382 | path->leave_spinning = 1; |
39279cc3 CM |
2383 | ret = btrfs_lookup_inode(trans, root, path, |
2384 | &BTRFS_I(inode)->location, 1); | |
2385 | if (ret) { | |
2386 | if (ret > 0) | |
2387 | ret = -ENOENT; | |
2388 | goto failed; | |
2389 | } | |
2390 | ||
b4ce94de | 2391 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2392 | leaf = path->nodes[0]; |
2393 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2394 | struct btrfs_inode_item); |
2395 | ||
e02119d5 | 2396 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2397 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2398 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2399 | ret = 0; |
2400 | failed: | |
39279cc3 CM |
2401 | btrfs_free_path(path); |
2402 | return ret; | |
2403 | } | |
2404 | ||
2405 | ||
d352ac68 CM |
2406 | /* |
2407 | * unlink helper that gets used here in inode.c and in the tree logging | |
2408 | * recovery code. It remove a link in a directory with a given name, and | |
2409 | * also drops the back refs in the inode to the directory | |
2410 | */ | |
e02119d5 CM |
2411 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2412 | struct btrfs_root *root, | |
2413 | struct inode *dir, struct inode *inode, | |
2414 | const char *name, int name_len) | |
39279cc3 CM |
2415 | { |
2416 | struct btrfs_path *path; | |
39279cc3 | 2417 | int ret = 0; |
5f39d397 | 2418 | struct extent_buffer *leaf; |
39279cc3 | 2419 | struct btrfs_dir_item *di; |
5f39d397 | 2420 | struct btrfs_key key; |
aec7477b | 2421 | u64 index; |
39279cc3 CM |
2422 | |
2423 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2424 | if (!path) { |
2425 | ret = -ENOMEM; | |
2426 | goto err; | |
2427 | } | |
2428 | ||
b9473439 | 2429 | path->leave_spinning = 1; |
39279cc3 CM |
2430 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2431 | name, name_len, -1); | |
2432 | if (IS_ERR(di)) { | |
2433 | ret = PTR_ERR(di); | |
2434 | goto err; | |
2435 | } | |
2436 | if (!di) { | |
2437 | ret = -ENOENT; | |
2438 | goto err; | |
2439 | } | |
5f39d397 CM |
2440 | leaf = path->nodes[0]; |
2441 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2442 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2443 | if (ret) |
2444 | goto err; | |
39279cc3 CM |
2445 | btrfs_release_path(root, path); |
2446 | ||
aec7477b | 2447 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2448 | inode->i_ino, |
2449 | dir->i_ino, &index); | |
aec7477b | 2450 | if (ret) { |
d397712b | 2451 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2452 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2453 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2454 | goto err; |
2455 | } | |
2456 | ||
39279cc3 | 2457 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2458 | index, name, name_len, -1); |
39279cc3 CM |
2459 | if (IS_ERR(di)) { |
2460 | ret = PTR_ERR(di); | |
2461 | goto err; | |
2462 | } | |
2463 | if (!di) { | |
2464 | ret = -ENOENT; | |
2465 | goto err; | |
2466 | } | |
2467 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2468 | btrfs_release_path(root, path); |
39279cc3 | 2469 | |
e02119d5 CM |
2470 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2471 | inode, dir->i_ino); | |
49eb7e46 | 2472 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2473 | |
2474 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2475 | dir, index); | |
2476 | BUG_ON(ret); | |
39279cc3 CM |
2477 | err: |
2478 | btrfs_free_path(path); | |
e02119d5 CM |
2479 | if (ret) |
2480 | goto out; | |
2481 | ||
2482 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2483 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2484 | btrfs_update_inode(trans, root, dir); | |
2485 | btrfs_drop_nlink(inode); | |
2486 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2487 | out: |
39279cc3 CM |
2488 | return ret; |
2489 | } | |
2490 | ||
2491 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2492 | { | |
2493 | struct btrfs_root *root; | |
2494 | struct btrfs_trans_handle *trans; | |
7b128766 | 2495 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2496 | int ret; |
1832a6d5 | 2497 | unsigned long nr = 0; |
39279cc3 CM |
2498 | |
2499 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2500 | |
5df6a9f6 JB |
2501 | /* |
2502 | * 5 items for unlink inode | |
2503 | * 1 for orphan | |
2504 | */ | |
2505 | ret = btrfs_reserve_metadata_space(root, 6); | |
2506 | if (ret) | |
2507 | return ret; | |
2508 | ||
39279cc3 | 2509 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2510 | if (IS_ERR(trans)) { |
2511 | btrfs_unreserve_metadata_space(root, 6); | |
2512 | return PTR_ERR(trans); | |
2513 | } | |
5f39d397 | 2514 | |
39279cc3 | 2515 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2516 | |
2517 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2518 | ||
e02119d5 CM |
2519 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2520 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2521 | |
2522 | if (inode->i_nlink == 0) | |
2523 | ret = btrfs_orphan_add(trans, inode); | |
2524 | ||
d3c2fdcf | 2525 | nr = trans->blocks_used; |
5f39d397 | 2526 | |
89ce8a63 | 2527 | btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2528 | btrfs_unreserve_metadata_space(root, 6); |
d3c2fdcf | 2529 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2530 | return ret; |
2531 | } | |
2532 | ||
4df27c4d YZ |
2533 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2534 | struct btrfs_root *root, | |
2535 | struct inode *dir, u64 objectid, | |
2536 | const char *name, int name_len) | |
2537 | { | |
2538 | struct btrfs_path *path; | |
2539 | struct extent_buffer *leaf; | |
2540 | struct btrfs_dir_item *di; | |
2541 | struct btrfs_key key; | |
2542 | u64 index; | |
2543 | int ret; | |
2544 | ||
2545 | path = btrfs_alloc_path(); | |
2546 | if (!path) | |
2547 | return -ENOMEM; | |
2548 | ||
2549 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2550 | name, name_len, -1); | |
2551 | BUG_ON(!di || IS_ERR(di)); | |
2552 | ||
2553 | leaf = path->nodes[0]; | |
2554 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2555 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2556 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2557 | BUG_ON(ret); | |
2558 | btrfs_release_path(root, path); | |
2559 | ||
2560 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2561 | objectid, root->root_key.objectid, | |
2562 | dir->i_ino, &index, name, name_len); | |
2563 | if (ret < 0) { | |
2564 | BUG_ON(ret != -ENOENT); | |
2565 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2566 | name, name_len); | |
2567 | BUG_ON(!di || IS_ERR(di)); | |
2568 | ||
2569 | leaf = path->nodes[0]; | |
2570 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2571 | btrfs_release_path(root, path); | |
2572 | index = key.offset; | |
2573 | } | |
2574 | ||
2575 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2576 | index, name, name_len, -1); | |
2577 | BUG_ON(!di || IS_ERR(di)); | |
2578 | ||
2579 | leaf = path->nodes[0]; | |
2580 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2581 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2582 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2583 | BUG_ON(ret); | |
2584 | btrfs_release_path(root, path); | |
2585 | ||
2586 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2587 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2588 | ret = btrfs_update_inode(trans, root, dir); | |
2589 | BUG_ON(ret); | |
2590 | dir->i_sb->s_dirt = 1; | |
2591 | ||
2592 | btrfs_free_path(path); | |
2593 | return 0; | |
2594 | } | |
2595 | ||
39279cc3 CM |
2596 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2597 | { | |
2598 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2599 | int err = 0; |
39279cc3 CM |
2600 | int ret; |
2601 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2602 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2603 | unsigned long nr = 0; |
39279cc3 | 2604 | |
3394e160 | 2605 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2606 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2607 | return -ENOTEMPTY; |
2608 | ||
5df6a9f6 JB |
2609 | ret = btrfs_reserve_metadata_space(root, 5); |
2610 | if (ret) | |
2611 | return ret; | |
2612 | ||
39279cc3 | 2613 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2614 | if (IS_ERR(trans)) { |
2615 | btrfs_unreserve_metadata_space(root, 5); | |
2616 | return PTR_ERR(trans); | |
2617 | } | |
2618 | ||
39279cc3 | 2619 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 2620 | |
4df27c4d YZ |
2621 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2622 | err = btrfs_unlink_subvol(trans, root, dir, | |
2623 | BTRFS_I(inode)->location.objectid, | |
2624 | dentry->d_name.name, | |
2625 | dentry->d_name.len); | |
2626 | goto out; | |
2627 | } | |
2628 | ||
7b128766 JB |
2629 | err = btrfs_orphan_add(trans, inode); |
2630 | if (err) | |
4df27c4d | 2631 | goto out; |
7b128766 | 2632 | |
39279cc3 | 2633 | /* now the directory is empty */ |
e02119d5 CM |
2634 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2635 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2636 | if (!err) |
dbe674a9 | 2637 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2638 | out: |
d3c2fdcf | 2639 | nr = trans->blocks_used; |
89ce8a63 | 2640 | ret = btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2641 | btrfs_unreserve_metadata_space(root, 5); |
d3c2fdcf | 2642 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2643 | |
39279cc3 CM |
2644 | if (ret && !err) |
2645 | err = ret; | |
2646 | return err; | |
2647 | } | |
2648 | ||
d20f7043 | 2649 | #if 0 |
323ac95b CM |
2650 | /* |
2651 | * when truncating bytes in a file, it is possible to avoid reading | |
2652 | * the leaves that contain only checksum items. This can be the | |
2653 | * majority of the IO required to delete a large file, but it must | |
2654 | * be done carefully. | |
2655 | * | |
2656 | * The keys in the level just above the leaves are checked to make sure | |
2657 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2658 | * after the new size. | |
2659 | * | |
2660 | * Then the key for the next leaf is checked to make sure it also has | |
2661 | * a checksum item for the same file. If it does, we know our target leaf | |
2662 | * contains only checksum items, and it can be safely freed without reading | |
2663 | * it. | |
2664 | * | |
2665 | * This is just an optimization targeted at large files. It may do | |
2666 | * nothing. It will return 0 unless things went badly. | |
2667 | */ | |
2668 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2669 | struct btrfs_root *root, | |
2670 | struct btrfs_path *path, | |
2671 | struct inode *inode, u64 new_size) | |
2672 | { | |
2673 | struct btrfs_key key; | |
2674 | int ret; | |
2675 | int nritems; | |
2676 | struct btrfs_key found_key; | |
2677 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2678 | struct btrfs_leaf_ref *ref; |
2679 | u64 leaf_gen; | |
2680 | u64 leaf_start; | |
323ac95b CM |
2681 | |
2682 | path->lowest_level = 1; | |
2683 | key.objectid = inode->i_ino; | |
2684 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2685 | key.offset = new_size; | |
2686 | again: | |
2687 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2688 | if (ret < 0) | |
2689 | goto out; | |
2690 | ||
2691 | if (path->nodes[1] == NULL) { | |
2692 | ret = 0; | |
2693 | goto out; | |
2694 | } | |
2695 | ret = 0; | |
2696 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2697 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2698 | ||
2699 | if (!nritems) | |
2700 | goto out; | |
2701 | ||
2702 | if (path->slots[1] >= nritems) | |
2703 | goto next_node; | |
2704 | ||
2705 | /* did we find a key greater than anything we want to delete? */ | |
2706 | if (found_key.objectid > inode->i_ino || | |
2707 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2708 | goto out; | |
2709 | ||
2710 | /* we check the next key in the node to make sure the leave contains | |
2711 | * only checksum items. This comparison doesn't work if our | |
2712 | * leaf is the last one in the node | |
2713 | */ | |
2714 | if (path->slots[1] + 1 >= nritems) { | |
2715 | next_node: | |
2716 | /* search forward from the last key in the node, this | |
2717 | * will bring us into the next node in the tree | |
2718 | */ | |
2719 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2720 | ||
2721 | /* unlikely, but we inc below, so check to be safe */ | |
2722 | if (found_key.offset == (u64)-1) | |
2723 | goto out; | |
2724 | ||
2725 | /* search_forward needs a path with locks held, do the | |
2726 | * search again for the original key. It is possible | |
2727 | * this will race with a balance and return a path that | |
2728 | * we could modify, but this drop is just an optimization | |
2729 | * and is allowed to miss some leaves. | |
2730 | */ | |
2731 | btrfs_release_path(root, path); | |
2732 | found_key.offset++; | |
2733 | ||
2734 | /* setup a max key for search_forward */ | |
2735 | other_key.offset = (u64)-1; | |
2736 | other_key.type = key.type; | |
2737 | other_key.objectid = key.objectid; | |
2738 | ||
2739 | path->keep_locks = 1; | |
2740 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2741 | path, 0, 0); | |
2742 | path->keep_locks = 0; | |
2743 | if (ret || found_key.objectid != key.objectid || | |
2744 | found_key.type != key.type) { | |
2745 | ret = 0; | |
2746 | goto out; | |
2747 | } | |
2748 | ||
2749 | key.offset = found_key.offset; | |
2750 | btrfs_release_path(root, path); | |
2751 | cond_resched(); | |
2752 | goto again; | |
2753 | } | |
2754 | ||
2755 | /* we know there's one more slot after us in the tree, | |
2756 | * read that key so we can verify it is also a checksum item | |
2757 | */ | |
2758 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2759 | ||
2760 | if (found_key.objectid < inode->i_ino) | |
2761 | goto next_key; | |
2762 | ||
2763 | if (found_key.type != key.type || found_key.offset < new_size) | |
2764 | goto next_key; | |
2765 | ||
2766 | /* | |
2767 | * if the key for the next leaf isn't a csum key from this objectid, | |
2768 | * we can't be sure there aren't good items inside this leaf. | |
2769 | * Bail out | |
2770 | */ | |
2771 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2772 | goto out; | |
2773 | ||
5b84e8d6 YZ |
2774 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2775 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2776 | /* |
2777 | * it is safe to delete this leaf, it contains only | |
2778 | * csum items from this inode at an offset >= new_size | |
2779 | */ | |
5b84e8d6 | 2780 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2781 | BUG_ON(ret); |
2782 | ||
5b84e8d6 YZ |
2783 | if (root->ref_cows && leaf_gen < trans->transid) { |
2784 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2785 | if (ref) { | |
2786 | ref->root_gen = root->root_key.offset; | |
2787 | ref->bytenr = leaf_start; | |
2788 | ref->owner = 0; | |
2789 | ref->generation = leaf_gen; | |
2790 | ref->nritems = 0; | |
2791 | ||
bd56b302 CM |
2792 | btrfs_sort_leaf_ref(ref); |
2793 | ||
5b84e8d6 YZ |
2794 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2795 | WARN_ON(ret); | |
2796 | btrfs_free_leaf_ref(root, ref); | |
2797 | } else { | |
2798 | WARN_ON(1); | |
2799 | } | |
2800 | } | |
323ac95b CM |
2801 | next_key: |
2802 | btrfs_release_path(root, path); | |
2803 | ||
2804 | if (other_key.objectid == inode->i_ino && | |
2805 | other_key.type == key.type && other_key.offset > key.offset) { | |
2806 | key.offset = other_key.offset; | |
2807 | cond_resched(); | |
2808 | goto again; | |
2809 | } | |
2810 | ret = 0; | |
2811 | out: | |
2812 | /* fixup any changes we've made to the path */ | |
2813 | path->lowest_level = 0; | |
2814 | path->keep_locks = 0; | |
2815 | btrfs_release_path(root, path); | |
2816 | return ret; | |
2817 | } | |
2818 | ||
d20f7043 CM |
2819 | #endif |
2820 | ||
39279cc3 CM |
2821 | /* |
2822 | * this can truncate away extent items, csum items and directory items. | |
2823 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2824 | * any higher than new_size |
39279cc3 CM |
2825 | * |
2826 | * csum items that cross the new i_size are truncated to the new size | |
2827 | * as well. | |
7b128766 JB |
2828 | * |
2829 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2830 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2831 | */ |
e02119d5 CM |
2832 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2833 | struct btrfs_root *root, | |
2834 | struct inode *inode, | |
2835 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2836 | { |
2837 | int ret; | |
2838 | struct btrfs_path *path; | |
2839 | struct btrfs_key key; | |
5f39d397 | 2840 | struct btrfs_key found_key; |
06d9a8d7 | 2841 | u32 found_type = (u8)-1; |
5f39d397 | 2842 | struct extent_buffer *leaf; |
39279cc3 CM |
2843 | struct btrfs_file_extent_item *fi; |
2844 | u64 extent_start = 0; | |
db94535d | 2845 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2846 | u64 extent_offset = 0; |
39279cc3 CM |
2847 | u64 item_end = 0; |
2848 | int found_extent; | |
2849 | int del_item; | |
85e21bac CM |
2850 | int pending_del_nr = 0; |
2851 | int pending_del_slot = 0; | |
179e29e4 | 2852 | int extent_type = -1; |
771ed689 | 2853 | int encoding; |
3b951516 | 2854 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2855 | |
e02119d5 | 2856 | if (root->ref_cows) |
5b21f2ed | 2857 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2858 | path = btrfs_alloc_path(); |
2859 | BUG_ON(!path); | |
33c17ad5 | 2860 | path->reada = -1; |
5f39d397 | 2861 | |
39279cc3 CM |
2862 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2863 | key.objectid = inode->i_ino; | |
2864 | key.offset = (u64)-1; | |
5f39d397 CM |
2865 | key.type = (u8)-1; |
2866 | ||
85e21bac | 2867 | search_again: |
b9473439 | 2868 | path->leave_spinning = 1; |
85e21bac | 2869 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2870 | if (ret < 0) |
85e21bac | 2871 | goto error; |
d397712b | 2872 | |
85e21bac | 2873 | if (ret > 0) { |
e02119d5 CM |
2874 | /* there are no items in the tree for us to truncate, we're |
2875 | * done | |
2876 | */ | |
2877 | if (path->slots[0] == 0) { | |
2878 | ret = 0; | |
2879 | goto error; | |
2880 | } | |
85e21bac CM |
2881 | path->slots[0]--; |
2882 | } | |
2883 | ||
d397712b | 2884 | while (1) { |
39279cc3 | 2885 | fi = NULL; |
5f39d397 CM |
2886 | leaf = path->nodes[0]; |
2887 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2888 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2889 | encoding = 0; |
39279cc3 | 2890 | |
5f39d397 | 2891 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2892 | break; |
5f39d397 | 2893 | |
85e21bac | 2894 | if (found_type < min_type) |
39279cc3 CM |
2895 | break; |
2896 | ||
5f39d397 | 2897 | item_end = found_key.offset; |
39279cc3 | 2898 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2899 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2900 | struct btrfs_file_extent_item); |
179e29e4 | 2901 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2902 | encoding = btrfs_file_extent_compression(leaf, fi); |
2903 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2904 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2905 | ||
179e29e4 | 2906 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2907 | item_end += |
db94535d | 2908 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2909 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2910 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2911 | fi); |
39279cc3 | 2912 | } |
008630c1 | 2913 | item_end--; |
39279cc3 | 2914 | } |
e02119d5 | 2915 | if (item_end < new_size) { |
d397712b | 2916 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2917 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2918 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2919 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2920 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2921 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2922 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2923 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2924 | else if (found_type) |
b888db2b | 2925 | found_type--; |
d397712b | 2926 | else |
b888db2b | 2927 | break; |
a61721d5 | 2928 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2929 | goto next; |
39279cc3 | 2930 | } |
e02119d5 | 2931 | if (found_key.offset >= new_size) |
39279cc3 CM |
2932 | del_item = 1; |
2933 | else | |
2934 | del_item = 0; | |
2935 | found_extent = 0; | |
2936 | ||
2937 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2938 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2939 | goto delete; | |
2940 | ||
2941 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2942 | u64 num_dec; |
db94535d | 2943 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2944 | if (!del_item && !encoding) { |
db94535d CM |
2945 | u64 orig_num_bytes = |
2946 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2947 | extent_num_bytes = new_size - |
5f39d397 | 2948 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2949 | extent_num_bytes = extent_num_bytes & |
2950 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2951 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2952 | extent_num_bytes); | |
2953 | num_dec = (orig_num_bytes - | |
9069218d | 2954 | extent_num_bytes); |
e02119d5 | 2955 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2956 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2957 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2958 | } else { |
db94535d CM |
2959 | extent_num_bytes = |
2960 | btrfs_file_extent_disk_num_bytes(leaf, | |
2961 | fi); | |
5d4f98a2 YZ |
2962 | extent_offset = found_key.offset - |
2963 | btrfs_file_extent_offset(leaf, fi); | |
2964 | ||
39279cc3 | 2965 | /* FIXME blocksize != 4096 */ |
9069218d | 2966 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2967 | if (extent_start != 0) { |
2968 | found_extent = 1; | |
e02119d5 | 2969 | if (root->ref_cows) |
a76a3cd4 | 2970 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2971 | } |
39279cc3 | 2972 | } |
9069218d | 2973 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2974 | /* |
2975 | * we can't truncate inline items that have had | |
2976 | * special encodings | |
2977 | */ | |
2978 | if (!del_item && | |
2979 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2980 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2981 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2982 | u32 size = new_size - found_key.offset; |
2983 | ||
2984 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2985 | inode_sub_bytes(inode, item_end + 1 - |
2986 | new_size); | |
e02119d5 CM |
2987 | } |
2988 | size = | |
2989 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2990 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2991 | size, 1); |
9069218d | 2992 | BUG_ON(ret); |
e02119d5 | 2993 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2994 | inode_sub_bytes(inode, item_end + 1 - |
2995 | found_key.offset); | |
9069218d | 2996 | } |
39279cc3 | 2997 | } |
179e29e4 | 2998 | delete: |
39279cc3 | 2999 | if (del_item) { |
85e21bac CM |
3000 | if (!pending_del_nr) { |
3001 | /* no pending yet, add ourselves */ | |
3002 | pending_del_slot = path->slots[0]; | |
3003 | pending_del_nr = 1; | |
3004 | } else if (pending_del_nr && | |
3005 | path->slots[0] + 1 == pending_del_slot) { | |
3006 | /* hop on the pending chunk */ | |
3007 | pending_del_nr++; | |
3008 | pending_del_slot = path->slots[0]; | |
3009 | } else { | |
d397712b | 3010 | BUG(); |
85e21bac | 3011 | } |
39279cc3 CM |
3012 | } else { |
3013 | break; | |
3014 | } | |
5d4f98a2 | 3015 | if (found_extent && root->ref_cows) { |
b9473439 | 3016 | btrfs_set_path_blocking(path); |
39279cc3 | 3017 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3018 | extent_num_bytes, 0, |
3019 | btrfs_header_owner(leaf), | |
3020 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3021 | BUG_ON(ret); |
3022 | } | |
85e21bac CM |
3023 | next: |
3024 | if (path->slots[0] == 0) { | |
3025 | if (pending_del_nr) | |
3026 | goto del_pending; | |
3027 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3028 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3029 | break; | |
85e21bac CM |
3030 | goto search_again; |
3031 | } | |
3032 | ||
3033 | path->slots[0]--; | |
3034 | if (pending_del_nr && | |
3035 | path->slots[0] + 1 != pending_del_slot) { | |
3036 | struct btrfs_key debug; | |
3037 | del_pending: | |
3038 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
3039 | pending_del_slot); | |
3040 | ret = btrfs_del_items(trans, root, path, | |
3041 | pending_del_slot, | |
3042 | pending_del_nr); | |
3043 | BUG_ON(ret); | |
3044 | pending_del_nr = 0; | |
3045 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3046 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3047 | break; | |
85e21bac CM |
3048 | goto search_again; |
3049 | } | |
39279cc3 CM |
3050 | } |
3051 | ret = 0; | |
3052 | error: | |
85e21bac CM |
3053 | if (pending_del_nr) { |
3054 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3055 | pending_del_nr); | |
3056 | } | |
39279cc3 | 3057 | btrfs_free_path(path); |
39279cc3 CM |
3058 | return ret; |
3059 | } | |
3060 | ||
3061 | /* | |
3062 | * taken from block_truncate_page, but does cow as it zeros out | |
3063 | * any bytes left in the last page in the file. | |
3064 | */ | |
3065 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3066 | { | |
3067 | struct inode *inode = mapping->host; | |
db94535d | 3068 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3069 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3070 | struct btrfs_ordered_extent *ordered; | |
3071 | char *kaddr; | |
db94535d | 3072 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3073 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3074 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3075 | struct page *page; | |
39279cc3 | 3076 | int ret = 0; |
a52d9a80 | 3077 | u64 page_start; |
e6dcd2dc | 3078 | u64 page_end; |
39279cc3 CM |
3079 | |
3080 | if ((offset & (blocksize - 1)) == 0) | |
3081 | goto out; | |
5d5e103a JB |
3082 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
3083 | if (ret) | |
3084 | goto out; | |
3085 | ||
3086 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); | |
3087 | if (ret) | |
3088 | goto out; | |
39279cc3 CM |
3089 | |
3090 | ret = -ENOMEM; | |
211c17f5 | 3091 | again: |
39279cc3 | 3092 | page = grab_cache_page(mapping, index); |
5d5e103a JB |
3093 | if (!page) { |
3094 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3095 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 | 3096 | goto out; |
5d5e103a | 3097 | } |
e6dcd2dc CM |
3098 | |
3099 | page_start = page_offset(page); | |
3100 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3101 | ||
39279cc3 | 3102 | if (!PageUptodate(page)) { |
9ebefb18 | 3103 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3104 | lock_page(page); |
211c17f5 CM |
3105 | if (page->mapping != mapping) { |
3106 | unlock_page(page); | |
3107 | page_cache_release(page); | |
3108 | goto again; | |
3109 | } | |
39279cc3 CM |
3110 | if (!PageUptodate(page)) { |
3111 | ret = -EIO; | |
89642229 | 3112 | goto out_unlock; |
39279cc3 CM |
3113 | } |
3114 | } | |
211c17f5 | 3115 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3116 | |
3117 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3118 | set_page_extent_mapped(page); | |
3119 | ||
3120 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3121 | if (ordered) { | |
3122 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3123 | unlock_page(page); | |
3124 | page_cache_release(page); | |
eb84ae03 | 3125 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3126 | btrfs_put_ordered_extent(ordered); |
3127 | goto again; | |
3128 | } | |
3129 | ||
5d5e103a JB |
3130 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
3131 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, | |
3132 | GFP_NOFS); | |
3133 | ||
9ed74f2d JB |
3134 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3135 | if (ret) { | |
3136 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3137 | goto out_unlock; | |
3138 | } | |
3139 | ||
e6dcd2dc CM |
3140 | ret = 0; |
3141 | if (offset != PAGE_CACHE_SIZE) { | |
3142 | kaddr = kmap(page); | |
3143 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3144 | flush_dcache_page(page); | |
3145 | kunmap(page); | |
3146 | } | |
247e743c | 3147 | ClearPageChecked(page); |
e6dcd2dc CM |
3148 | set_page_dirty(page); |
3149 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3150 | |
89642229 | 3151 | out_unlock: |
5d5e103a JB |
3152 | if (ret) |
3153 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3154 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 CM |
3155 | unlock_page(page); |
3156 | page_cache_release(page); | |
3157 | out: | |
3158 | return ret; | |
3159 | } | |
3160 | ||
9036c102 | 3161 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3162 | { |
9036c102 YZ |
3163 | struct btrfs_trans_handle *trans; |
3164 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3165 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3166 | struct extent_map *em; | |
3167 | u64 mask = root->sectorsize - 1; | |
3168 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3169 | u64 block_end = (size + mask) & ~mask; | |
3170 | u64 last_byte; | |
3171 | u64 cur_offset; | |
3172 | u64 hole_size; | |
9ed74f2d | 3173 | int err = 0; |
39279cc3 | 3174 | |
9036c102 YZ |
3175 | if (size <= hole_start) |
3176 | return 0; | |
3177 | ||
5d5e103a JB |
3178 | err = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
3179 | if (err) | |
3180 | return err; | |
2bf5a725 | 3181 | |
9036c102 YZ |
3182 | while (1) { |
3183 | struct btrfs_ordered_extent *ordered; | |
3184 | btrfs_wait_ordered_range(inode, hole_start, | |
3185 | block_end - hole_start); | |
3186 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3187 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3188 | if (!ordered) | |
3189 | break; | |
3190 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3191 | btrfs_put_ordered_extent(ordered); | |
3192 | } | |
39279cc3 | 3193 | |
9036c102 YZ |
3194 | trans = btrfs_start_transaction(root, 1); |
3195 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3196 | |
9036c102 YZ |
3197 | cur_offset = hole_start; |
3198 | while (1) { | |
3199 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3200 | block_end - cur_offset, 0); | |
3201 | BUG_ON(IS_ERR(em) || !em); | |
3202 | last_byte = min(extent_map_end(em), block_end); | |
3203 | last_byte = (last_byte + mask) & ~mask; | |
3204 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3205 | u64 hint_byte = 0; |
9036c102 | 3206 | hole_size = last_byte - cur_offset; |
920bbbfb | 3207 | err = btrfs_drop_extents(trans, inode, cur_offset, |
771ed689 | 3208 | cur_offset + hole_size, |
920bbbfb | 3209 | &hint_byte, 1); |
771ed689 CM |
3210 | if (err) |
3211 | break; | |
9ed74f2d JB |
3212 | |
3213 | err = btrfs_reserve_metadata_space(root, 1); | |
3214 | if (err) | |
3215 | break; | |
3216 | ||
9036c102 YZ |
3217 | err = btrfs_insert_file_extent(trans, root, |
3218 | inode->i_ino, cur_offset, 0, | |
3219 | 0, hole_size, 0, hole_size, | |
3220 | 0, 0, 0); | |
3221 | btrfs_drop_extent_cache(inode, hole_start, | |
3222 | last_byte - 1, 0); | |
9ed74f2d | 3223 | btrfs_unreserve_metadata_space(root, 1); |
9036c102 YZ |
3224 | } |
3225 | free_extent_map(em); | |
3226 | cur_offset = last_byte; | |
3227 | if (err || cur_offset >= block_end) | |
3228 | break; | |
3229 | } | |
1832a6d5 | 3230 | |
9036c102 YZ |
3231 | btrfs_end_transaction(trans, root); |
3232 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3233 | return err; | |
3234 | } | |
39279cc3 | 3235 | |
9036c102 YZ |
3236 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3237 | { | |
3238 | struct inode *inode = dentry->d_inode; | |
3239 | int err; | |
39279cc3 | 3240 | |
9036c102 YZ |
3241 | err = inode_change_ok(inode, attr); |
3242 | if (err) | |
3243 | return err; | |
2bf5a725 | 3244 | |
5a3f23d5 CM |
3245 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3246 | if (attr->ia_size > inode->i_size) { | |
3247 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3248 | if (err) | |
3249 | return err; | |
3250 | } else if (inode->i_size > 0 && | |
3251 | attr->ia_size == 0) { | |
3252 | ||
3253 | /* we're truncating a file that used to have good | |
3254 | * data down to zero. Make sure it gets into | |
3255 | * the ordered flush list so that any new writes | |
3256 | * get down to disk quickly. | |
3257 | */ | |
3258 | BTRFS_I(inode)->ordered_data_close = 1; | |
3259 | } | |
39279cc3 | 3260 | } |
9036c102 | 3261 | |
39279cc3 | 3262 | err = inode_setattr(inode, attr); |
33268eaf JB |
3263 | |
3264 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3265 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3266 | return err; |
3267 | } | |
61295eb8 | 3268 | |
39279cc3 CM |
3269 | void btrfs_delete_inode(struct inode *inode) |
3270 | { | |
3271 | struct btrfs_trans_handle *trans; | |
3272 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3273 | unsigned long nr; |
39279cc3 CM |
3274 | int ret; |
3275 | ||
3276 | truncate_inode_pages(&inode->i_data, 0); | |
3277 | if (is_bad_inode(inode)) { | |
7b128766 | 3278 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3279 | goto no_delete; |
3280 | } | |
4a096752 | 3281 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3282 | |
76dda93c YZ |
3283 | if (inode->i_nlink > 0) { |
3284 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3285 | goto no_delete; | |
3286 | } | |
3287 | ||
dbe674a9 | 3288 | btrfs_i_size_write(inode, 0); |
180591bc | 3289 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3290 | |
39279cc3 | 3291 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3292 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3293 | if (ret) { |
3294 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3295 | goto no_delete_lock; |
7b128766 JB |
3296 | } |
3297 | ||
3298 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3299 | |
d3c2fdcf | 3300 | nr = trans->blocks_used; |
85e21bac | 3301 | clear_inode(inode); |
5f39d397 | 3302 | |
39279cc3 | 3303 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3304 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3305 | return; |
54aa1f4d CM |
3306 | |
3307 | no_delete_lock: | |
d3c2fdcf | 3308 | nr = trans->blocks_used; |
54aa1f4d | 3309 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3310 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3311 | no_delete: |
3312 | clear_inode(inode); | |
3313 | } | |
3314 | ||
3315 | /* | |
3316 | * this returns the key found in the dir entry in the location pointer. | |
3317 | * If no dir entries were found, location->objectid is 0. | |
3318 | */ | |
3319 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3320 | struct btrfs_key *location) | |
3321 | { | |
3322 | const char *name = dentry->d_name.name; | |
3323 | int namelen = dentry->d_name.len; | |
3324 | struct btrfs_dir_item *di; | |
3325 | struct btrfs_path *path; | |
3326 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3327 | int ret = 0; |
39279cc3 CM |
3328 | |
3329 | path = btrfs_alloc_path(); | |
3330 | BUG_ON(!path); | |
3954401f | 3331 | |
39279cc3 CM |
3332 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3333 | namelen, 0); | |
0d9f7f3e Y |
3334 | if (IS_ERR(di)) |
3335 | ret = PTR_ERR(di); | |
d397712b CM |
3336 | |
3337 | if (!di || IS_ERR(di)) | |
3954401f | 3338 | goto out_err; |
d397712b | 3339 | |
5f39d397 | 3340 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3341 | out: |
39279cc3 CM |
3342 | btrfs_free_path(path); |
3343 | return ret; | |
3954401f CM |
3344 | out_err: |
3345 | location->objectid = 0; | |
3346 | goto out; | |
39279cc3 CM |
3347 | } |
3348 | ||
3349 | /* | |
3350 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3351 | * needs to be changed to reflect the root directory of the tree root. This | |
3352 | * is kind of like crossing a mount point. | |
3353 | */ | |
3354 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3355 | struct inode *dir, |
3356 | struct dentry *dentry, | |
3357 | struct btrfs_key *location, | |
3358 | struct btrfs_root **sub_root) | |
39279cc3 | 3359 | { |
4df27c4d YZ |
3360 | struct btrfs_path *path; |
3361 | struct btrfs_root *new_root; | |
3362 | struct btrfs_root_ref *ref; | |
3363 | struct extent_buffer *leaf; | |
3364 | int ret; | |
3365 | int err = 0; | |
39279cc3 | 3366 | |
4df27c4d YZ |
3367 | path = btrfs_alloc_path(); |
3368 | if (!path) { | |
3369 | err = -ENOMEM; | |
3370 | goto out; | |
3371 | } | |
39279cc3 | 3372 | |
4df27c4d YZ |
3373 | err = -ENOENT; |
3374 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3375 | BTRFS_I(dir)->root->root_key.objectid, | |
3376 | location->objectid); | |
3377 | if (ret) { | |
3378 | if (ret < 0) | |
3379 | err = ret; | |
3380 | goto out; | |
3381 | } | |
39279cc3 | 3382 | |
4df27c4d YZ |
3383 | leaf = path->nodes[0]; |
3384 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3385 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3386 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3387 | goto out; | |
39279cc3 | 3388 | |
4df27c4d YZ |
3389 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3390 | (unsigned long)(ref + 1), | |
3391 | dentry->d_name.len); | |
3392 | if (ret) | |
3393 | goto out; | |
3394 | ||
3395 | btrfs_release_path(root->fs_info->tree_root, path); | |
3396 | ||
3397 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3398 | if (IS_ERR(new_root)) { | |
3399 | err = PTR_ERR(new_root); | |
3400 | goto out; | |
3401 | } | |
3402 | ||
3403 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3404 | err = -ENOENT; | |
3405 | goto out; | |
3406 | } | |
3407 | ||
3408 | *sub_root = new_root; | |
3409 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3410 | location->type = BTRFS_INODE_ITEM_KEY; | |
3411 | location->offset = 0; | |
3412 | err = 0; | |
3413 | out: | |
3414 | btrfs_free_path(path); | |
3415 | return err; | |
39279cc3 CM |
3416 | } |
3417 | ||
5d4f98a2 YZ |
3418 | static void inode_tree_add(struct inode *inode) |
3419 | { | |
3420 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3421 | struct btrfs_inode *entry; | |
03e860bd NP |
3422 | struct rb_node **p; |
3423 | struct rb_node *parent; | |
03e860bd NP |
3424 | again: |
3425 | p = &root->inode_tree.rb_node; | |
3426 | parent = NULL; | |
5d4f98a2 | 3427 | |
76dda93c YZ |
3428 | if (hlist_unhashed(&inode->i_hash)) |
3429 | return; | |
3430 | ||
5d4f98a2 YZ |
3431 | spin_lock(&root->inode_lock); |
3432 | while (*p) { | |
3433 | parent = *p; | |
3434 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3435 | ||
3436 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3437 | p = &parent->rb_left; |
5d4f98a2 | 3438 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3439 | p = &parent->rb_right; |
5d4f98a2 YZ |
3440 | else { |
3441 | WARN_ON(!(entry->vfs_inode.i_state & | |
3442 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd NP |
3443 | rb_erase(parent, &root->inode_tree); |
3444 | RB_CLEAR_NODE(parent); | |
3445 | spin_unlock(&root->inode_lock); | |
3446 | goto again; | |
5d4f98a2 YZ |
3447 | } |
3448 | } | |
3449 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3450 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3451 | spin_unlock(&root->inode_lock); | |
3452 | } | |
3453 | ||
3454 | static void inode_tree_del(struct inode *inode) | |
3455 | { | |
3456 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3457 | int empty = 0; |
5d4f98a2 | 3458 | |
03e860bd | 3459 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3460 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3461 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3462 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3463 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3464 | } |
03e860bd | 3465 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3466 | |
3467 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3468 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3469 | spin_lock(&root->inode_lock); | |
3470 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3471 | spin_unlock(&root->inode_lock); | |
3472 | if (empty) | |
3473 | btrfs_add_dead_root(root); | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3478 | { | |
3479 | struct rb_node *node; | |
3480 | struct rb_node *prev; | |
3481 | struct btrfs_inode *entry; | |
3482 | struct inode *inode; | |
3483 | u64 objectid = 0; | |
3484 | ||
3485 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3486 | ||
3487 | spin_lock(&root->inode_lock); | |
3488 | again: | |
3489 | node = root->inode_tree.rb_node; | |
3490 | prev = NULL; | |
3491 | while (node) { | |
3492 | prev = node; | |
3493 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3494 | ||
3495 | if (objectid < entry->vfs_inode.i_ino) | |
3496 | node = node->rb_left; | |
3497 | else if (objectid > entry->vfs_inode.i_ino) | |
3498 | node = node->rb_right; | |
3499 | else | |
3500 | break; | |
3501 | } | |
3502 | if (!node) { | |
3503 | while (prev) { | |
3504 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3505 | if (objectid <= entry->vfs_inode.i_ino) { | |
3506 | node = prev; | |
3507 | break; | |
3508 | } | |
3509 | prev = rb_next(prev); | |
3510 | } | |
3511 | } | |
3512 | while (node) { | |
3513 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3514 | objectid = entry->vfs_inode.i_ino + 1; | |
3515 | inode = igrab(&entry->vfs_inode); | |
3516 | if (inode) { | |
3517 | spin_unlock(&root->inode_lock); | |
3518 | if (atomic_read(&inode->i_count) > 1) | |
3519 | d_prune_aliases(inode); | |
3520 | /* | |
3521 | * btrfs_drop_inode will remove it from | |
3522 | * the inode cache when its usage count | |
3523 | * hits zero. | |
3524 | */ | |
3525 | iput(inode); | |
3526 | cond_resched(); | |
3527 | spin_lock(&root->inode_lock); | |
3528 | goto again; | |
3529 | } | |
3530 | ||
3531 | if (cond_resched_lock(&root->inode_lock)) | |
3532 | goto again; | |
3533 | ||
3534 | node = rb_next(node); | |
3535 | } | |
3536 | spin_unlock(&root->inode_lock); | |
3537 | return 0; | |
5d4f98a2 YZ |
3538 | } |
3539 | ||
e02119d5 | 3540 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3541 | { |
e02119d5 CM |
3542 | struct btrfs_inode *bi = BTRFS_I(inode); |
3543 | ||
e02119d5 | 3544 | bi->generation = 0; |
c3027eb5 | 3545 | bi->sequence = 0; |
e02119d5 | 3546 | bi->last_trans = 0; |
257c62e1 | 3547 | bi->last_sub_trans = 0; |
e02119d5 CM |
3548 | bi->logged_trans = 0; |
3549 | bi->delalloc_bytes = 0; | |
6a63209f | 3550 | bi->reserved_bytes = 0; |
e02119d5 CM |
3551 | bi->disk_i_size = 0; |
3552 | bi->flags = 0; | |
3553 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3554 | bi->last_unlink_trans = 0; |
2757495c | 3555 | bi->ordered_data_close = 0; |
d1310b2e CM |
3556 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3557 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3558 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3559 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3560 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3561 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3562 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3563 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3564 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3565 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3566 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3567 | } | |
3568 | ||
3569 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3570 | { | |
3571 | struct btrfs_iget_args *args = p; | |
3572 | inode->i_ino = args->ino; | |
3573 | init_btrfs_i(inode); | |
3574 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3575 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3576 | return 0; |
3577 | } | |
3578 | ||
3579 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3580 | { | |
3581 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3582 | return args->ino == inode->i_ino && |
3583 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3584 | } |
3585 | ||
5d4f98a2 YZ |
3586 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3587 | u64 objectid, | |
3588 | struct btrfs_root *root) | |
39279cc3 CM |
3589 | { |
3590 | struct inode *inode; | |
3591 | struct btrfs_iget_args args; | |
3592 | args.ino = objectid; | |
3593 | args.root = root; | |
3594 | ||
3595 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3596 | btrfs_init_locked_inode, | |
3597 | (void *)&args); | |
3598 | return inode; | |
3599 | } | |
3600 | ||
1a54ef8c BR |
3601 | /* Get an inode object given its location and corresponding root. |
3602 | * Returns in *is_new if the inode was read from disk | |
3603 | */ | |
3604 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3605 | struct btrfs_root *root) |
1a54ef8c BR |
3606 | { |
3607 | struct inode *inode; | |
3608 | ||
3609 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3610 | if (!inode) | |
5d4f98a2 | 3611 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3612 | |
3613 | if (inode->i_state & I_NEW) { | |
3614 | BTRFS_I(inode)->root = root; | |
3615 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3616 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3617 | |
3618 | inode_tree_add(inode); | |
1a54ef8c | 3619 | unlock_new_inode(inode); |
1a54ef8c BR |
3620 | } |
3621 | ||
3622 | return inode; | |
3623 | } | |
3624 | ||
4df27c4d YZ |
3625 | static struct inode *new_simple_dir(struct super_block *s, |
3626 | struct btrfs_key *key, | |
3627 | struct btrfs_root *root) | |
3628 | { | |
3629 | struct inode *inode = new_inode(s); | |
3630 | ||
3631 | if (!inode) | |
3632 | return ERR_PTR(-ENOMEM); | |
3633 | ||
3634 | init_btrfs_i(inode); | |
3635 | ||
3636 | BTRFS_I(inode)->root = root; | |
3637 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3638 | BTRFS_I(inode)->dummy_inode = 1; | |
3639 | ||
3640 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3641 | inode->i_op = &simple_dir_inode_operations; | |
3642 | inode->i_fop = &simple_dir_operations; | |
3643 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3644 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3645 | ||
3646 | return inode; | |
3647 | } | |
3648 | ||
3de4586c | 3649 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3650 | { |
d397712b | 3651 | struct inode *inode; |
4df27c4d | 3652 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3653 | struct btrfs_root *sub_root = root; |
3654 | struct btrfs_key location; | |
76dda93c | 3655 | int index; |
5d4f98a2 | 3656 | int ret; |
39279cc3 | 3657 | |
76dda93c YZ |
3658 | dentry->d_op = &btrfs_dentry_operations; |
3659 | ||
39279cc3 CM |
3660 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3661 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3662 | |
39279cc3 | 3663 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3664 | |
39279cc3 CM |
3665 | if (ret < 0) |
3666 | return ERR_PTR(ret); | |
5f39d397 | 3667 | |
4df27c4d YZ |
3668 | if (location.objectid == 0) |
3669 | return NULL; | |
3670 | ||
3671 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3672 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3673 | return inode; | |
3674 | } | |
3675 | ||
3676 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3677 | ||
76dda93c | 3678 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3679 | ret = fixup_tree_root_location(root, dir, dentry, |
3680 | &location, &sub_root); | |
3681 | if (ret < 0) { | |
3682 | if (ret != -ENOENT) | |
3683 | inode = ERR_PTR(ret); | |
3684 | else | |
3685 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3686 | } else { | |
5d4f98a2 | 3687 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3688 | } |
76dda93c YZ |
3689 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3690 | ||
3de4586c CM |
3691 | return inode; |
3692 | } | |
3693 | ||
76dda93c YZ |
3694 | static int btrfs_dentry_delete(struct dentry *dentry) |
3695 | { | |
3696 | struct btrfs_root *root; | |
3697 | ||
efefb143 YZ |
3698 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3699 | dentry = dentry->d_parent; | |
76dda93c | 3700 | |
efefb143 YZ |
3701 | if (dentry->d_inode) { |
3702 | root = BTRFS_I(dentry->d_inode)->root; | |
3703 | if (btrfs_root_refs(&root->root_item) == 0) | |
3704 | return 1; | |
3705 | } | |
76dda93c YZ |
3706 | return 0; |
3707 | } | |
3708 | ||
3de4586c CM |
3709 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3710 | struct nameidata *nd) | |
3711 | { | |
3712 | struct inode *inode; | |
3713 | ||
3de4586c CM |
3714 | inode = btrfs_lookup_dentry(dir, dentry); |
3715 | if (IS_ERR(inode)) | |
3716 | return ERR_CAST(inode); | |
7b128766 | 3717 | |
39279cc3 CM |
3718 | return d_splice_alias(inode, dentry); |
3719 | } | |
3720 | ||
39279cc3 CM |
3721 | static unsigned char btrfs_filetype_table[] = { |
3722 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3723 | }; | |
3724 | ||
cbdf5a24 DW |
3725 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3726 | filldir_t filldir) | |
39279cc3 | 3727 | { |
6da6abae | 3728 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3729 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3730 | struct btrfs_item *item; | |
3731 | struct btrfs_dir_item *di; | |
3732 | struct btrfs_key key; | |
5f39d397 | 3733 | struct btrfs_key found_key; |
39279cc3 CM |
3734 | struct btrfs_path *path; |
3735 | int ret; | |
3736 | u32 nritems; | |
5f39d397 | 3737 | struct extent_buffer *leaf; |
39279cc3 CM |
3738 | int slot; |
3739 | int advance; | |
3740 | unsigned char d_type; | |
3741 | int over = 0; | |
3742 | u32 di_cur; | |
3743 | u32 di_total; | |
3744 | u32 di_len; | |
3745 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3746 | char tmp_name[32]; |
3747 | char *name_ptr; | |
3748 | int name_len; | |
39279cc3 CM |
3749 | |
3750 | /* FIXME, use a real flag for deciding about the key type */ | |
3751 | if (root->fs_info->tree_root == root) | |
3752 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3753 | |
3954401f CM |
3754 | /* special case for "." */ |
3755 | if (filp->f_pos == 0) { | |
3756 | over = filldir(dirent, ".", 1, | |
3757 | 1, inode->i_ino, | |
3758 | DT_DIR); | |
3759 | if (over) | |
3760 | return 0; | |
3761 | filp->f_pos = 1; | |
3762 | } | |
3954401f CM |
3763 | /* special case for .., just use the back ref */ |
3764 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3765 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3766 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3767 | 2, pino, DT_DIR); |
3954401f | 3768 | if (over) |
49593bfa | 3769 | return 0; |
3954401f CM |
3770 | filp->f_pos = 2; |
3771 | } | |
49593bfa DW |
3772 | path = btrfs_alloc_path(); |
3773 | path->reada = 2; | |
3774 | ||
39279cc3 CM |
3775 | btrfs_set_key_type(&key, key_type); |
3776 | key.offset = filp->f_pos; | |
49593bfa | 3777 | key.objectid = inode->i_ino; |
5f39d397 | 3778 | |
39279cc3 CM |
3779 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3780 | if (ret < 0) | |
3781 | goto err; | |
3782 | advance = 0; | |
49593bfa DW |
3783 | |
3784 | while (1) { | |
5f39d397 CM |
3785 | leaf = path->nodes[0]; |
3786 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3787 | slot = path->slots[0]; |
3788 | if (advance || slot >= nritems) { | |
49593bfa | 3789 | if (slot >= nritems - 1) { |
39279cc3 CM |
3790 | ret = btrfs_next_leaf(root, path); |
3791 | if (ret) | |
3792 | break; | |
5f39d397 CM |
3793 | leaf = path->nodes[0]; |
3794 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3795 | slot = path->slots[0]; |
3796 | } else { | |
3797 | slot++; | |
3798 | path->slots[0]++; | |
3799 | } | |
3800 | } | |
3de4586c | 3801 | |
39279cc3 | 3802 | advance = 1; |
5f39d397 CM |
3803 | item = btrfs_item_nr(leaf, slot); |
3804 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3805 | ||
3806 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3807 | break; |
5f39d397 | 3808 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3809 | break; |
5f39d397 | 3810 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3811 | continue; |
5f39d397 CM |
3812 | |
3813 | filp->f_pos = found_key.offset; | |
49593bfa | 3814 | |
39279cc3 CM |
3815 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3816 | di_cur = 0; | |
5f39d397 | 3817 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3818 | |
3819 | while (di_cur < di_total) { | |
5f39d397 CM |
3820 | struct btrfs_key location; |
3821 | ||
3822 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3823 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3824 | name_ptr = tmp_name; |
3825 | } else { | |
3826 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3827 | if (!name_ptr) { |
3828 | ret = -ENOMEM; | |
3829 | goto err; | |
3830 | } | |
5f39d397 CM |
3831 | } |
3832 | read_extent_buffer(leaf, name_ptr, | |
3833 | (unsigned long)(di + 1), name_len); | |
3834 | ||
3835 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3836 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3837 | |
3838 | /* is this a reference to our own snapshot? If so | |
3839 | * skip it | |
3840 | */ | |
3841 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3842 | location.objectid == root->root_key.objectid) { | |
3843 | over = 0; | |
3844 | goto skip; | |
3845 | } | |
5f39d397 | 3846 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3847 | found_key.offset, location.objectid, |
39279cc3 | 3848 | d_type); |
5f39d397 | 3849 | |
3de4586c | 3850 | skip: |
5f39d397 CM |
3851 | if (name_ptr != tmp_name) |
3852 | kfree(name_ptr); | |
3853 | ||
39279cc3 CM |
3854 | if (over) |
3855 | goto nopos; | |
5103e947 | 3856 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3857 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3858 | di_cur += di_len; |
3859 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3860 | } | |
3861 | } | |
49593bfa DW |
3862 | |
3863 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3864 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3865 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3866 | else |
3867 | filp->f_pos++; | |
39279cc3 CM |
3868 | nopos: |
3869 | ret = 0; | |
3870 | err: | |
39279cc3 | 3871 | btrfs_free_path(path); |
39279cc3 CM |
3872 | return ret; |
3873 | } | |
3874 | ||
3875 | int btrfs_write_inode(struct inode *inode, int wait) | |
3876 | { | |
3877 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3878 | struct btrfs_trans_handle *trans; | |
3879 | int ret = 0; | |
3880 | ||
c146afad | 3881 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3882 | return 0; |
3883 | ||
39279cc3 | 3884 | if (wait) { |
f9295749 | 3885 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3886 | btrfs_set_trans_block_group(trans, inode); |
3887 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3888 | } |
3889 | return ret; | |
3890 | } | |
3891 | ||
3892 | /* | |
54aa1f4d | 3893 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3894 | * inode changes. But, it is most likely to find the inode in cache. |
3895 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3896 | * to keep or drop this code. | |
3897 | */ | |
3898 | void btrfs_dirty_inode(struct inode *inode) | |
3899 | { | |
3900 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3901 | struct btrfs_trans_handle *trans; | |
3902 | ||
f9295749 | 3903 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3904 | btrfs_set_trans_block_group(trans, inode); |
3905 | btrfs_update_inode(trans, root, inode); | |
3906 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3907 | } |
3908 | ||
d352ac68 CM |
3909 | /* |
3910 | * find the highest existing sequence number in a directory | |
3911 | * and then set the in-memory index_cnt variable to reflect | |
3912 | * free sequence numbers | |
3913 | */ | |
aec7477b JB |
3914 | static int btrfs_set_inode_index_count(struct inode *inode) |
3915 | { | |
3916 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3917 | struct btrfs_key key, found_key; | |
3918 | struct btrfs_path *path; | |
3919 | struct extent_buffer *leaf; | |
3920 | int ret; | |
3921 | ||
3922 | key.objectid = inode->i_ino; | |
3923 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3924 | key.offset = (u64)-1; | |
3925 | ||
3926 | path = btrfs_alloc_path(); | |
3927 | if (!path) | |
3928 | return -ENOMEM; | |
3929 | ||
3930 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3931 | if (ret < 0) | |
3932 | goto out; | |
3933 | /* FIXME: we should be able to handle this */ | |
3934 | if (ret == 0) | |
3935 | goto out; | |
3936 | ret = 0; | |
3937 | ||
3938 | /* | |
3939 | * MAGIC NUMBER EXPLANATION: | |
3940 | * since we search a directory based on f_pos we have to start at 2 | |
3941 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3942 | * else has to start at 2 | |
3943 | */ | |
3944 | if (path->slots[0] == 0) { | |
3945 | BTRFS_I(inode)->index_cnt = 2; | |
3946 | goto out; | |
3947 | } | |
3948 | ||
3949 | path->slots[0]--; | |
3950 | ||
3951 | leaf = path->nodes[0]; | |
3952 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3953 | ||
3954 | if (found_key.objectid != inode->i_ino || | |
3955 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3956 | BTRFS_I(inode)->index_cnt = 2; | |
3957 | goto out; | |
3958 | } | |
3959 | ||
3960 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3961 | out: | |
3962 | btrfs_free_path(path); | |
3963 | return ret; | |
3964 | } | |
3965 | ||
d352ac68 CM |
3966 | /* |
3967 | * helper to find a free sequence number in a given directory. This current | |
3968 | * code is very simple, later versions will do smarter things in the btree | |
3969 | */ | |
3de4586c | 3970 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3971 | { |
3972 | int ret = 0; | |
3973 | ||
3974 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3975 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3976 | if (ret) |
aec7477b JB |
3977 | return ret; |
3978 | } | |
3979 | ||
00e4e6b3 | 3980 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3981 | BTRFS_I(dir)->index_cnt++; |
3982 | ||
3983 | return ret; | |
3984 | } | |
3985 | ||
39279cc3 CM |
3986 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3987 | struct btrfs_root *root, | |
aec7477b | 3988 | struct inode *dir, |
9c58309d | 3989 | const char *name, int name_len, |
d2fb3437 YZ |
3990 | u64 ref_objectid, u64 objectid, |
3991 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3992 | { |
3993 | struct inode *inode; | |
5f39d397 | 3994 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3995 | struct btrfs_key *location; |
5f39d397 | 3996 | struct btrfs_path *path; |
9c58309d CM |
3997 | struct btrfs_inode_ref *ref; |
3998 | struct btrfs_key key[2]; | |
3999 | u32 sizes[2]; | |
4000 | unsigned long ptr; | |
39279cc3 CM |
4001 | int ret; |
4002 | int owner; | |
4003 | ||
5f39d397 CM |
4004 | path = btrfs_alloc_path(); |
4005 | BUG_ON(!path); | |
4006 | ||
39279cc3 CM |
4007 | inode = new_inode(root->fs_info->sb); |
4008 | if (!inode) | |
4009 | return ERR_PTR(-ENOMEM); | |
4010 | ||
aec7477b | 4011 | if (dir) { |
3de4586c | 4012 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
4013 | if (ret) { |
4014 | iput(inode); | |
aec7477b | 4015 | return ERR_PTR(ret); |
09771430 | 4016 | } |
aec7477b JB |
4017 | } |
4018 | /* | |
4019 | * index_cnt is ignored for everything but a dir, | |
4020 | * btrfs_get_inode_index_count has an explanation for the magic | |
4021 | * number | |
4022 | */ | |
e02119d5 | 4023 | init_btrfs_i(inode); |
aec7477b | 4024 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 4025 | BTRFS_I(inode)->root = root; |
e02119d5 | 4026 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 4027 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4028 | |
39279cc3 CM |
4029 | if (mode & S_IFDIR) |
4030 | owner = 0; | |
4031 | else | |
4032 | owner = 1; | |
d2fb3437 YZ |
4033 | BTRFS_I(inode)->block_group = |
4034 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4035 | |
4036 | key[0].objectid = objectid; | |
4037 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4038 | key[0].offset = 0; | |
4039 | ||
4040 | key[1].objectid = objectid; | |
4041 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4042 | key[1].offset = ref_objectid; | |
4043 | ||
4044 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4045 | sizes[1] = name_len + sizeof(*ref); | |
4046 | ||
b9473439 | 4047 | path->leave_spinning = 1; |
9c58309d CM |
4048 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4049 | if (ret != 0) | |
5f39d397 CM |
4050 | goto fail; |
4051 | ||
79683f2d | 4052 | inode->i_uid = current_fsuid(); |
8c087b51 | 4053 | |
42f15d77 | 4054 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4055 | inode->i_gid = dir->i_gid; |
4056 | if (S_ISDIR(mode)) | |
4057 | mode |= S_ISGID; | |
4058 | } else | |
4059 | inode->i_gid = current_fsgid(); | |
4060 | ||
39279cc3 CM |
4061 | inode->i_mode = mode; |
4062 | inode->i_ino = objectid; | |
a76a3cd4 | 4063 | inode_set_bytes(inode, 0); |
39279cc3 | 4064 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4065 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4066 | struct btrfs_inode_item); | |
e02119d5 | 4067 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4068 | |
4069 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4070 | struct btrfs_inode_ref); | |
4071 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4072 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4073 | ptr = (unsigned long)(ref + 1); |
4074 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4075 | ||
5f39d397 CM |
4076 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4077 | btrfs_free_path(path); | |
4078 | ||
39279cc3 CM |
4079 | location = &BTRFS_I(inode)->location; |
4080 | location->objectid = objectid; | |
39279cc3 CM |
4081 | location->offset = 0; |
4082 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4083 | ||
6cbff00f CH |
4084 | btrfs_inherit_iflags(inode, dir); |
4085 | ||
94272164 CM |
4086 | if ((mode & S_IFREG)) { |
4087 | if (btrfs_test_opt(root, NODATASUM)) | |
4088 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4089 | if (btrfs_test_opt(root, NODATACOW)) | |
4090 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4091 | } | |
4092 | ||
39279cc3 | 4093 | insert_inode_hash(inode); |
5d4f98a2 | 4094 | inode_tree_add(inode); |
39279cc3 | 4095 | return inode; |
5f39d397 | 4096 | fail: |
aec7477b JB |
4097 | if (dir) |
4098 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4099 | btrfs_free_path(path); |
09771430 | 4100 | iput(inode); |
5f39d397 | 4101 | return ERR_PTR(ret); |
39279cc3 CM |
4102 | } |
4103 | ||
4104 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4105 | { | |
4106 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4107 | } | |
4108 | ||
d352ac68 CM |
4109 | /* |
4110 | * utility function to add 'inode' into 'parent_inode' with | |
4111 | * a give name and a given sequence number. | |
4112 | * if 'add_backref' is true, also insert a backref from the | |
4113 | * inode to the parent directory. | |
4114 | */ | |
e02119d5 CM |
4115 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4116 | struct inode *parent_inode, struct inode *inode, | |
4117 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4118 | { |
4df27c4d | 4119 | int ret = 0; |
39279cc3 | 4120 | struct btrfs_key key; |
e02119d5 | 4121 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4122 | |
4df27c4d YZ |
4123 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4124 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4125 | } else { | |
4126 | key.objectid = inode->i_ino; | |
4127 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4128 | key.offset = 0; | |
4129 | } | |
4130 | ||
4131 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4132 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4133 | key.objectid, root->root_key.objectid, | |
4134 | parent_inode->i_ino, | |
4135 | index, name, name_len); | |
4136 | } else if (add_backref) { | |
4137 | ret = btrfs_insert_inode_ref(trans, root, | |
4138 | name, name_len, inode->i_ino, | |
4139 | parent_inode->i_ino, index); | |
4140 | } | |
39279cc3 | 4141 | |
39279cc3 | 4142 | if (ret == 0) { |
4df27c4d YZ |
4143 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4144 | parent_inode->i_ino, &key, | |
4145 | btrfs_inode_type(inode), index); | |
4146 | BUG_ON(ret); | |
4147 | ||
dbe674a9 | 4148 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4149 | name_len * 2); |
79c44584 | 4150 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4151 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4152 | } |
4153 | return ret; | |
4154 | } | |
4155 | ||
4156 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4157 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4158 | int backref, u64 index) |
39279cc3 | 4159 | { |
e02119d5 CM |
4160 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4161 | inode, dentry->d_name.name, | |
4162 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4163 | if (!err) { |
4164 | d_instantiate(dentry, inode); | |
4165 | return 0; | |
4166 | } | |
4167 | if (err > 0) | |
4168 | err = -EEXIST; | |
4169 | return err; | |
4170 | } | |
4171 | ||
618e21d5 JB |
4172 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4173 | int mode, dev_t rdev) | |
4174 | { | |
4175 | struct btrfs_trans_handle *trans; | |
4176 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4177 | struct inode *inode = NULL; |
618e21d5 JB |
4178 | int err; |
4179 | int drop_inode = 0; | |
4180 | u64 objectid; | |
1832a6d5 | 4181 | unsigned long nr = 0; |
00e4e6b3 | 4182 | u64 index = 0; |
618e21d5 JB |
4183 | |
4184 | if (!new_valid_dev(rdev)) | |
4185 | return -EINVAL; | |
4186 | ||
9ed74f2d JB |
4187 | /* |
4188 | * 2 for inode item and ref | |
4189 | * 2 for dir items | |
4190 | * 1 for xattr if selinux is on | |
4191 | */ | |
4192 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4193 | if (err) |
9ed74f2d | 4194 | return err; |
1832a6d5 | 4195 | |
618e21d5 | 4196 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4197 | if (!trans) |
4198 | goto fail; | |
618e21d5 JB |
4199 | btrfs_set_trans_block_group(trans, dir); |
4200 | ||
4201 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4202 | if (err) { | |
4203 | err = -ENOSPC; | |
4204 | goto out_unlock; | |
4205 | } | |
4206 | ||
aec7477b | 4207 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4208 | dentry->d_name.len, |
4209 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4210 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4211 | err = PTR_ERR(inode); |
4212 | if (IS_ERR(inode)) | |
4213 | goto out_unlock; | |
4214 | ||
0279b4cd | 4215 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4216 | if (err) { |
4217 | drop_inode = 1; | |
4218 | goto out_unlock; | |
4219 | } | |
4220 | ||
618e21d5 | 4221 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4222 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4223 | if (err) |
4224 | drop_inode = 1; | |
4225 | else { | |
4226 | inode->i_op = &btrfs_special_inode_operations; | |
4227 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4228 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4229 | } |
618e21d5 JB |
4230 | btrfs_update_inode_block_group(trans, inode); |
4231 | btrfs_update_inode_block_group(trans, dir); | |
4232 | out_unlock: | |
d3c2fdcf | 4233 | nr = trans->blocks_used; |
89ce8a63 | 4234 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4235 | fail: |
9ed74f2d | 4236 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4237 | if (drop_inode) { |
4238 | inode_dec_link_count(inode); | |
4239 | iput(inode); | |
4240 | } | |
d3c2fdcf | 4241 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4242 | return err; |
4243 | } | |
4244 | ||
39279cc3 CM |
4245 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4246 | int mode, struct nameidata *nd) | |
4247 | { | |
4248 | struct btrfs_trans_handle *trans; | |
4249 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4250 | struct inode *inode = NULL; |
39279cc3 CM |
4251 | int err; |
4252 | int drop_inode = 0; | |
1832a6d5 | 4253 | unsigned long nr = 0; |
39279cc3 | 4254 | u64 objectid; |
00e4e6b3 | 4255 | u64 index = 0; |
39279cc3 | 4256 | |
9ed74f2d JB |
4257 | /* |
4258 | * 2 for inode item and ref | |
4259 | * 2 for dir items | |
4260 | * 1 for xattr if selinux is on | |
4261 | */ | |
4262 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4263 | if (err) |
9ed74f2d JB |
4264 | return err; |
4265 | ||
39279cc3 | 4266 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4267 | if (!trans) |
4268 | goto fail; | |
39279cc3 CM |
4269 | btrfs_set_trans_block_group(trans, dir); |
4270 | ||
4271 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4272 | if (err) { | |
4273 | err = -ENOSPC; | |
4274 | goto out_unlock; | |
4275 | } | |
4276 | ||
aec7477b | 4277 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4278 | dentry->d_name.len, |
4279 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4280 | objectid, BTRFS_I(dir)->block_group, mode, |
4281 | &index); | |
39279cc3 CM |
4282 | err = PTR_ERR(inode); |
4283 | if (IS_ERR(inode)) | |
4284 | goto out_unlock; | |
4285 | ||
0279b4cd | 4286 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4287 | if (err) { |
4288 | drop_inode = 1; | |
4289 | goto out_unlock; | |
4290 | } | |
4291 | ||
39279cc3 | 4292 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4293 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4294 | if (err) |
4295 | drop_inode = 1; | |
4296 | else { | |
4297 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4298 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4299 | inode->i_fop = &btrfs_file_operations; |
4300 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4301 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4302 | } |
39279cc3 CM |
4303 | btrfs_update_inode_block_group(trans, inode); |
4304 | btrfs_update_inode_block_group(trans, dir); | |
4305 | out_unlock: | |
d3c2fdcf | 4306 | nr = trans->blocks_used; |
ab78c84d | 4307 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4308 | fail: |
9ed74f2d | 4309 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4310 | if (drop_inode) { |
4311 | inode_dec_link_count(inode); | |
4312 | iput(inode); | |
4313 | } | |
d3c2fdcf | 4314 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4315 | return err; |
4316 | } | |
4317 | ||
4318 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4319 | struct dentry *dentry) | |
4320 | { | |
4321 | struct btrfs_trans_handle *trans; | |
4322 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4323 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4324 | u64 index; |
1832a6d5 | 4325 | unsigned long nr = 0; |
39279cc3 CM |
4326 | int err; |
4327 | int drop_inode = 0; | |
4328 | ||
4329 | if (inode->i_nlink == 0) | |
4330 | return -ENOENT; | |
4331 | ||
9ed74f2d JB |
4332 | /* |
4333 | * 1 item for inode ref | |
4334 | * 2 items for dir items | |
4335 | */ | |
4336 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4337 | if (err) |
9ed74f2d JB |
4338 | return err; |
4339 | ||
4340 | btrfs_inc_nlink(inode); | |
4341 | ||
3de4586c | 4342 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4343 | if (err) |
4344 | goto fail; | |
4345 | ||
39279cc3 | 4346 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4347 | |
39279cc3 CM |
4348 | btrfs_set_trans_block_group(trans, dir); |
4349 | atomic_inc(&inode->i_count); | |
aec7477b | 4350 | |
00e4e6b3 | 4351 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4352 | |
a5719521 | 4353 | if (err) { |
54aa1f4d | 4354 | drop_inode = 1; |
a5719521 YZ |
4355 | } else { |
4356 | btrfs_update_inode_block_group(trans, dir); | |
4357 | err = btrfs_update_inode(trans, root, inode); | |
4358 | BUG_ON(err); | |
4359 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4360 | } | |
39279cc3 | 4361 | |
d3c2fdcf | 4362 | nr = trans->blocks_used; |
ab78c84d | 4363 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4364 | fail: |
9ed74f2d | 4365 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4366 | if (drop_inode) { |
4367 | inode_dec_link_count(inode); | |
4368 | iput(inode); | |
4369 | } | |
d3c2fdcf | 4370 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4371 | return err; |
4372 | } | |
4373 | ||
39279cc3 CM |
4374 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4375 | { | |
b9d86667 | 4376 | struct inode *inode = NULL; |
39279cc3 CM |
4377 | struct btrfs_trans_handle *trans; |
4378 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4379 | int err = 0; | |
4380 | int drop_on_err = 0; | |
b9d86667 | 4381 | u64 objectid = 0; |
00e4e6b3 | 4382 | u64 index = 0; |
d3c2fdcf | 4383 | unsigned long nr = 1; |
39279cc3 | 4384 | |
9ed74f2d JB |
4385 | /* |
4386 | * 2 items for inode and ref | |
4387 | * 2 items for dir items | |
4388 | * 1 for xattr if selinux is on | |
4389 | */ | |
4390 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4391 | if (err) |
9ed74f2d | 4392 | return err; |
1832a6d5 | 4393 | |
39279cc3 | 4394 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4395 | if (!trans) { |
4396 | err = -ENOMEM; | |
39279cc3 CM |
4397 | goto out_unlock; |
4398 | } | |
9ed74f2d | 4399 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4400 | |
4401 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4402 | if (err) { | |
4403 | err = -ENOSPC; | |
4404 | goto out_unlock; | |
4405 | } | |
4406 | ||
aec7477b | 4407 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4408 | dentry->d_name.len, |
4409 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4410 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4411 | &index); | |
39279cc3 CM |
4412 | if (IS_ERR(inode)) { |
4413 | err = PTR_ERR(inode); | |
4414 | goto out_fail; | |
4415 | } | |
5f39d397 | 4416 | |
39279cc3 | 4417 | drop_on_err = 1; |
33268eaf | 4418 | |
0279b4cd | 4419 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4420 | if (err) |
4421 | goto out_fail; | |
4422 | ||
39279cc3 CM |
4423 | inode->i_op = &btrfs_dir_inode_operations; |
4424 | inode->i_fop = &btrfs_dir_file_operations; | |
4425 | btrfs_set_trans_block_group(trans, inode); | |
4426 | ||
dbe674a9 | 4427 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4428 | err = btrfs_update_inode(trans, root, inode); |
4429 | if (err) | |
4430 | goto out_fail; | |
5f39d397 | 4431 | |
e02119d5 CM |
4432 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4433 | inode, dentry->d_name.name, | |
4434 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4435 | if (err) |
4436 | goto out_fail; | |
5f39d397 | 4437 | |
39279cc3 CM |
4438 | d_instantiate(dentry, inode); |
4439 | drop_on_err = 0; | |
39279cc3 CM |
4440 | btrfs_update_inode_block_group(trans, inode); |
4441 | btrfs_update_inode_block_group(trans, dir); | |
4442 | ||
4443 | out_fail: | |
d3c2fdcf | 4444 | nr = trans->blocks_used; |
ab78c84d | 4445 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4446 | |
39279cc3 | 4447 | out_unlock: |
9ed74f2d | 4448 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4449 | if (drop_on_err) |
4450 | iput(inode); | |
d3c2fdcf | 4451 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4452 | return err; |
4453 | } | |
4454 | ||
d352ac68 CM |
4455 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4456 | * and an extent that you want to insert, deal with overlap and insert | |
4457 | * the new extent into the tree. | |
4458 | */ | |
3b951516 CM |
4459 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4460 | struct extent_map *existing, | |
e6dcd2dc CM |
4461 | struct extent_map *em, |
4462 | u64 map_start, u64 map_len) | |
3b951516 CM |
4463 | { |
4464 | u64 start_diff; | |
3b951516 | 4465 | |
e6dcd2dc CM |
4466 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4467 | start_diff = map_start - em->start; | |
4468 | em->start = map_start; | |
4469 | em->len = map_len; | |
c8b97818 CM |
4470 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4471 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4472 | em->block_start += start_diff; |
c8b97818 CM |
4473 | em->block_len -= start_diff; |
4474 | } | |
e6dcd2dc | 4475 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4476 | } |
4477 | ||
c8b97818 CM |
4478 | static noinline int uncompress_inline(struct btrfs_path *path, |
4479 | struct inode *inode, struct page *page, | |
4480 | size_t pg_offset, u64 extent_offset, | |
4481 | struct btrfs_file_extent_item *item) | |
4482 | { | |
4483 | int ret; | |
4484 | struct extent_buffer *leaf = path->nodes[0]; | |
4485 | char *tmp; | |
4486 | size_t max_size; | |
4487 | unsigned long inline_size; | |
4488 | unsigned long ptr; | |
4489 | ||
4490 | WARN_ON(pg_offset != 0); | |
4491 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4492 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4493 | btrfs_item_nr(leaf, path->slots[0])); | |
4494 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4495 | ptr = btrfs_file_extent_inline_start(item); | |
4496 | ||
4497 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4498 | ||
5b050f04 | 4499 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4500 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4501 | inline_size, max_size); | |
4502 | if (ret) { | |
4503 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4504 | unsigned long copy_size = min_t(u64, | |
4505 | PAGE_CACHE_SIZE - pg_offset, | |
4506 | max_size - extent_offset); | |
4507 | memset(kaddr + pg_offset, 0, copy_size); | |
4508 | kunmap_atomic(kaddr, KM_USER0); | |
4509 | } | |
4510 | kfree(tmp); | |
4511 | return 0; | |
4512 | } | |
4513 | ||
d352ac68 CM |
4514 | /* |
4515 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4516 | * the ugly parts come from merging extents from the disk with the in-ram |
4517 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4518 | * where the in-ram extents might be locked pending data=ordered completion. |
4519 | * | |
4520 | * This also copies inline extents directly into the page. | |
4521 | */ | |
d397712b | 4522 | |
a52d9a80 | 4523 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4524 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4525 | int create) |
4526 | { | |
4527 | int ret; | |
4528 | int err = 0; | |
db94535d | 4529 | u64 bytenr; |
a52d9a80 CM |
4530 | u64 extent_start = 0; |
4531 | u64 extent_end = 0; | |
4532 | u64 objectid = inode->i_ino; | |
4533 | u32 found_type; | |
f421950f | 4534 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4535 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4536 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4537 | struct extent_buffer *leaf; |
4538 | struct btrfs_key found_key; | |
a52d9a80 CM |
4539 | struct extent_map *em = NULL; |
4540 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4541 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4542 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4543 | int compressed; |
a52d9a80 | 4544 | |
a52d9a80 | 4545 | again: |
890871be | 4546 | read_lock(&em_tree->lock); |
d1310b2e | 4547 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4548 | if (em) |
4549 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4550 | read_unlock(&em_tree->lock); |
d1310b2e | 4551 | |
a52d9a80 | 4552 | if (em) { |
e1c4b745 CM |
4553 | if (em->start > start || em->start + em->len <= start) |
4554 | free_extent_map(em); | |
4555 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4556 | free_extent_map(em); |
4557 | else | |
4558 | goto out; | |
a52d9a80 | 4559 | } |
d1310b2e | 4560 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4561 | if (!em) { |
d1310b2e CM |
4562 | err = -ENOMEM; |
4563 | goto out; | |
a52d9a80 | 4564 | } |
e6dcd2dc | 4565 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4566 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4567 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4568 | em->len = (u64)-1; |
c8b97818 | 4569 | em->block_len = (u64)-1; |
f421950f CM |
4570 | |
4571 | if (!path) { | |
4572 | path = btrfs_alloc_path(); | |
4573 | BUG_ON(!path); | |
4574 | } | |
4575 | ||
179e29e4 CM |
4576 | ret = btrfs_lookup_file_extent(trans, root, path, |
4577 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4578 | if (ret < 0) { |
4579 | err = ret; | |
4580 | goto out; | |
4581 | } | |
4582 | ||
4583 | if (ret != 0) { | |
4584 | if (path->slots[0] == 0) | |
4585 | goto not_found; | |
4586 | path->slots[0]--; | |
4587 | } | |
4588 | ||
5f39d397 CM |
4589 | leaf = path->nodes[0]; |
4590 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4591 | struct btrfs_file_extent_item); |
a52d9a80 | 4592 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4593 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4594 | found_type = btrfs_key_type(&found_key); | |
4595 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4596 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4597 | goto not_found; | |
4598 | } | |
4599 | ||
5f39d397 CM |
4600 | found_type = btrfs_file_extent_type(leaf, item); |
4601 | extent_start = found_key.offset; | |
c8b97818 | 4602 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4603 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4604 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4605 | extent_end = extent_start + |
db94535d | 4606 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4607 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4608 | size_t size; | |
4609 | size = btrfs_file_extent_inline_len(leaf, item); | |
4610 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4611 | ~((u64)root->sectorsize - 1); | |
4612 | } | |
4613 | ||
4614 | if (start >= extent_end) { | |
4615 | path->slots[0]++; | |
4616 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4617 | ret = btrfs_next_leaf(root, path); | |
4618 | if (ret < 0) { | |
4619 | err = ret; | |
4620 | goto out; | |
a52d9a80 | 4621 | } |
9036c102 YZ |
4622 | if (ret > 0) |
4623 | goto not_found; | |
4624 | leaf = path->nodes[0]; | |
a52d9a80 | 4625 | } |
9036c102 YZ |
4626 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4627 | if (found_key.objectid != objectid || | |
4628 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4629 | goto not_found; | |
4630 | if (start + len <= found_key.offset) | |
4631 | goto not_found; | |
4632 | em->start = start; | |
4633 | em->len = found_key.offset - start; | |
4634 | goto not_found_em; | |
4635 | } | |
4636 | ||
d899e052 YZ |
4637 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4638 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4639 | em->start = extent_start; |
4640 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4641 | em->orig_start = extent_start - |
4642 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4643 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4644 | if (bytenr == 0) { | |
5f39d397 | 4645 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4646 | goto insert; |
4647 | } | |
c8b97818 CM |
4648 | if (compressed) { |
4649 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4650 | em->block_start = bytenr; | |
4651 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4652 | item); | |
4653 | } else { | |
4654 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4655 | em->block_start = bytenr; | |
4656 | em->block_len = em->len; | |
d899e052 YZ |
4657 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4658 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4659 | } |
a52d9a80 CM |
4660 | goto insert; |
4661 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4662 | unsigned long ptr; |
a52d9a80 | 4663 | char *map; |
3326d1b0 CM |
4664 | size_t size; |
4665 | size_t extent_offset; | |
4666 | size_t copy_size; | |
a52d9a80 | 4667 | |
689f9346 | 4668 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4669 | if (!page || create) { |
689f9346 | 4670 | em->start = extent_start; |
9036c102 | 4671 | em->len = extent_end - extent_start; |
689f9346 Y |
4672 | goto out; |
4673 | } | |
5f39d397 | 4674 | |
9036c102 YZ |
4675 | size = btrfs_file_extent_inline_len(leaf, item); |
4676 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4677 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4678 | size - extent_offset); |
3326d1b0 | 4679 | em->start = extent_start + extent_offset; |
70dec807 CM |
4680 | em->len = (copy_size + root->sectorsize - 1) & |
4681 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4682 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4683 | if (compressed) |
4684 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4685 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4686 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4687 | if (btrfs_file_extent_compression(leaf, item) == |
4688 | BTRFS_COMPRESS_ZLIB) { | |
4689 | ret = uncompress_inline(path, inode, page, | |
4690 | pg_offset, | |
4691 | extent_offset, item); | |
4692 | BUG_ON(ret); | |
4693 | } else { | |
4694 | map = kmap(page); | |
4695 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4696 | copy_size); | |
93c82d57 CM |
4697 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4698 | memset(map + pg_offset + copy_size, 0, | |
4699 | PAGE_CACHE_SIZE - pg_offset - | |
4700 | copy_size); | |
4701 | } | |
c8b97818 CM |
4702 | kunmap(page); |
4703 | } | |
179e29e4 CM |
4704 | flush_dcache_page(page); |
4705 | } else if (create && PageUptodate(page)) { | |
4706 | if (!trans) { | |
4707 | kunmap(page); | |
4708 | free_extent_map(em); | |
4709 | em = NULL; | |
4710 | btrfs_release_path(root, path); | |
f9295749 | 4711 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4712 | goto again; |
4713 | } | |
c8b97818 | 4714 | map = kmap(page); |
70dec807 | 4715 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4716 | copy_size); |
c8b97818 | 4717 | kunmap(page); |
179e29e4 | 4718 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4719 | } |
d1310b2e CM |
4720 | set_extent_uptodate(io_tree, em->start, |
4721 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4722 | goto insert; |
4723 | } else { | |
d397712b | 4724 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4725 | WARN_ON(1); |
4726 | } | |
4727 | not_found: | |
4728 | em->start = start; | |
d1310b2e | 4729 | em->len = len; |
a52d9a80 | 4730 | not_found_em: |
5f39d397 | 4731 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4732 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4733 | insert: |
4734 | btrfs_release_path(root, path); | |
d1310b2e | 4735 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4736 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4737 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4738 | (unsigned long long)em->len, | |
4739 | (unsigned long long)start, | |
4740 | (unsigned long long)len); | |
a52d9a80 CM |
4741 | err = -EIO; |
4742 | goto out; | |
4743 | } | |
d1310b2e CM |
4744 | |
4745 | err = 0; | |
890871be | 4746 | write_lock(&em_tree->lock); |
a52d9a80 | 4747 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4748 | /* it is possible that someone inserted the extent into the tree |
4749 | * while we had the lock dropped. It is also possible that | |
4750 | * an overlapping map exists in the tree | |
4751 | */ | |
a52d9a80 | 4752 | if (ret == -EEXIST) { |
3b951516 | 4753 | struct extent_map *existing; |
e6dcd2dc CM |
4754 | |
4755 | ret = 0; | |
4756 | ||
3b951516 | 4757 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4758 | if (existing && (existing->start > start || |
4759 | existing->start + existing->len <= start)) { | |
4760 | free_extent_map(existing); | |
4761 | existing = NULL; | |
4762 | } | |
3b951516 CM |
4763 | if (!existing) { |
4764 | existing = lookup_extent_mapping(em_tree, em->start, | |
4765 | em->len); | |
4766 | if (existing) { | |
4767 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4768 | em, start, |
4769 | root->sectorsize); | |
3b951516 CM |
4770 | free_extent_map(existing); |
4771 | if (err) { | |
4772 | free_extent_map(em); | |
4773 | em = NULL; | |
4774 | } | |
4775 | } else { | |
4776 | err = -EIO; | |
3b951516 CM |
4777 | free_extent_map(em); |
4778 | em = NULL; | |
4779 | } | |
4780 | } else { | |
4781 | free_extent_map(em); | |
4782 | em = existing; | |
e6dcd2dc | 4783 | err = 0; |
a52d9a80 | 4784 | } |
a52d9a80 | 4785 | } |
890871be | 4786 | write_unlock(&em_tree->lock); |
a52d9a80 | 4787 | out: |
f421950f CM |
4788 | if (path) |
4789 | btrfs_free_path(path); | |
a52d9a80 CM |
4790 | if (trans) { |
4791 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4792 | if (!err) |
a52d9a80 CM |
4793 | err = ret; |
4794 | } | |
a52d9a80 CM |
4795 | if (err) { |
4796 | free_extent_map(em); | |
a52d9a80 CM |
4797 | return ERR_PTR(err); |
4798 | } | |
4799 | return em; | |
4800 | } | |
4801 | ||
16432985 CM |
4802 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4803 | const struct iovec *iov, loff_t offset, | |
4804 | unsigned long nr_segs) | |
4805 | { | |
e1c4b745 | 4806 | return -EINVAL; |
16432985 CM |
4807 | } |
4808 | ||
1506fcc8 YS |
4809 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4810 | __u64 start, __u64 len) | |
4811 | { | |
4812 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4813 | } | |
4814 | ||
a52d9a80 | 4815 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4816 | { |
d1310b2e CM |
4817 | struct extent_io_tree *tree; |
4818 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4819 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4820 | } |
1832a6d5 | 4821 | |
a52d9a80 | 4822 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4823 | { |
d1310b2e | 4824 | struct extent_io_tree *tree; |
b888db2b CM |
4825 | |
4826 | ||
4827 | if (current->flags & PF_MEMALLOC) { | |
4828 | redirty_page_for_writepage(wbc, page); | |
4829 | unlock_page(page); | |
4830 | return 0; | |
4831 | } | |
d1310b2e | 4832 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4833 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4834 | } |
4835 | ||
f421950f CM |
4836 | int btrfs_writepages(struct address_space *mapping, |
4837 | struct writeback_control *wbc) | |
b293f02e | 4838 | { |
d1310b2e | 4839 | struct extent_io_tree *tree; |
771ed689 | 4840 | |
d1310b2e | 4841 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4842 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4843 | } | |
4844 | ||
3ab2fb5a CM |
4845 | static int |
4846 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4847 | struct list_head *pages, unsigned nr_pages) | |
4848 | { | |
d1310b2e CM |
4849 | struct extent_io_tree *tree; |
4850 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4851 | return extent_readpages(tree, mapping, pages, nr_pages, |
4852 | btrfs_get_extent); | |
4853 | } | |
e6dcd2dc | 4854 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4855 | { |
d1310b2e CM |
4856 | struct extent_io_tree *tree; |
4857 | struct extent_map_tree *map; | |
a52d9a80 | 4858 | int ret; |
8c2383c3 | 4859 | |
d1310b2e CM |
4860 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4861 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4862 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4863 | if (ret == 1) { |
4864 | ClearPagePrivate(page); | |
4865 | set_page_private(page, 0); | |
4866 | page_cache_release(page); | |
39279cc3 | 4867 | } |
a52d9a80 | 4868 | return ret; |
39279cc3 CM |
4869 | } |
4870 | ||
e6dcd2dc CM |
4871 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4872 | { | |
98509cfc CM |
4873 | if (PageWriteback(page) || PageDirty(page)) |
4874 | return 0; | |
b335b003 | 4875 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4876 | } |
4877 | ||
a52d9a80 | 4878 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4879 | { |
d1310b2e | 4880 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4881 | struct btrfs_ordered_extent *ordered; |
4882 | u64 page_start = page_offset(page); | |
4883 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4884 | |
8b62b72b CM |
4885 | |
4886 | /* | |
4887 | * we have the page locked, so new writeback can't start, | |
4888 | * and the dirty bit won't be cleared while we are here. | |
4889 | * | |
4890 | * Wait for IO on this page so that we can safely clear | |
4891 | * the PagePrivate2 bit and do ordered accounting | |
4892 | */ | |
e6dcd2dc | 4893 | wait_on_page_writeback(page); |
8b62b72b | 4894 | |
d1310b2e | 4895 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4896 | if (offset) { |
4897 | btrfs_releasepage(page, GFP_NOFS); | |
4898 | return; | |
4899 | } | |
e6dcd2dc CM |
4900 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4901 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4902 | page_offset(page)); | |
4903 | if (ordered) { | |
eb84ae03 CM |
4904 | /* |
4905 | * IO on this page will never be started, so we need | |
4906 | * to account for any ordered extents now | |
4907 | */ | |
e6dcd2dc CM |
4908 | clear_extent_bit(tree, page_start, page_end, |
4909 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff JB |
4910 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
4911 | NULL, GFP_NOFS); | |
8b62b72b CM |
4912 | /* |
4913 | * whoever cleared the private bit is responsible | |
4914 | * for the finish_ordered_io | |
4915 | */ | |
4916 | if (TestClearPagePrivate2(page)) { | |
4917 | btrfs_finish_ordered_io(page->mapping->host, | |
4918 | page_start, page_end); | |
4919 | } | |
e6dcd2dc CM |
4920 | btrfs_put_ordered_extent(ordered); |
4921 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4922 | } | |
4923 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff JB |
4924 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4925 | EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); | |
e6dcd2dc CM |
4926 | __btrfs_releasepage(page, GFP_NOFS); |
4927 | ||
4a096752 | 4928 | ClearPageChecked(page); |
9ad6b7bc | 4929 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4930 | ClearPagePrivate(page); |
4931 | set_page_private(page, 0); | |
4932 | page_cache_release(page); | |
4933 | } | |
39279cc3 CM |
4934 | } |
4935 | ||
9ebefb18 CM |
4936 | /* |
4937 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4938 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4939 | * be careful to check for EOF conditions here. We set the page up correctly | |
4940 | * for a written page which means we get ENOSPC checking when writing into | |
4941 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4942 | * support these features. | |
4943 | * | |
4944 | * We are not allowed to take the i_mutex here so we have to play games to | |
4945 | * protect against truncate races as the page could now be beyond EOF. Because | |
4946 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4947 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4948 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4949 | * unlock the page. | |
4950 | */ | |
c2ec175c | 4951 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4952 | { |
c2ec175c | 4953 | struct page *page = vmf->page; |
6da6abae | 4954 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4955 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4956 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4957 | struct btrfs_ordered_extent *ordered; | |
4958 | char *kaddr; | |
4959 | unsigned long zero_start; | |
9ebefb18 | 4960 | loff_t size; |
1832a6d5 | 4961 | int ret; |
a52d9a80 | 4962 | u64 page_start; |
e6dcd2dc | 4963 | u64 page_end; |
9ebefb18 | 4964 | |
6a63209f | 4965 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4966 | if (ret) { |
4967 | if (ret == -ENOMEM) | |
4968 | ret = VM_FAULT_OOM; | |
4969 | else /* -ENOSPC, -EIO, etc */ | |
4970 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4971 | goto out; |
56a76f82 | 4972 | } |
1832a6d5 | 4973 | |
9ed74f2d JB |
4974 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
4975 | if (ret) { | |
4976 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
4977 | ret = VM_FAULT_SIGBUS; | |
4978 | goto out; | |
4979 | } | |
4980 | ||
56a76f82 | 4981 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4982 | again: |
9ebefb18 | 4983 | lock_page(page); |
9ebefb18 | 4984 | size = i_size_read(inode); |
e6dcd2dc CM |
4985 | page_start = page_offset(page); |
4986 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4987 | |
9ebefb18 | 4988 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4989 | (page_start >= size)) { |
6a63209f | 4990 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
4991 | /* page got truncated out from underneath us */ |
4992 | goto out_unlock; | |
4993 | } | |
e6dcd2dc CM |
4994 | wait_on_page_writeback(page); |
4995 | ||
4996 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4997 | set_page_extent_mapped(page); | |
4998 | ||
eb84ae03 CM |
4999 | /* |
5000 | * we can't set the delalloc bits if there are pending ordered | |
5001 | * extents. Drop our locks and wait for them to finish | |
5002 | */ | |
e6dcd2dc CM |
5003 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5004 | if (ordered) { | |
5005 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5006 | unlock_page(page); | |
eb84ae03 | 5007 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5008 | btrfs_put_ordered_extent(ordered); |
5009 | goto again; | |
5010 | } | |
5011 | ||
fbf19087 JB |
5012 | /* |
5013 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
5014 | * if it was already dirty, so for space accounting reasons we need to | |
5015 | * clear any delalloc bits for the range we are fixing to save. There | |
5016 | * is probably a better way to do this, but for now keep consistent with | |
5017 | * prepare_pages in the normal write path. | |
5018 | */ | |
5019 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
32c00aff JB |
5020 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
5021 | GFP_NOFS); | |
fbf19087 | 5022 | |
9ed74f2d JB |
5023 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
5024 | if (ret) { | |
5025 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5026 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 5027 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
5028 | goto out_unlock; |
5029 | } | |
e6dcd2dc | 5030 | ret = 0; |
9ebefb18 CM |
5031 | |
5032 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 5033 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 5034 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 5035 | else |
e6dcd2dc | 5036 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 5037 | |
e6dcd2dc CM |
5038 | if (zero_start != PAGE_CACHE_SIZE) { |
5039 | kaddr = kmap(page); | |
5040 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
5041 | flush_dcache_page(page); | |
5042 | kunmap(page); | |
5043 | } | |
247e743c | 5044 | ClearPageChecked(page); |
e6dcd2dc | 5045 | set_page_dirty(page); |
50a9b214 | 5046 | SetPageUptodate(page); |
5a3f23d5 | 5047 | |
257c62e1 CM |
5048 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5049 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5050 | ||
e6dcd2dc | 5051 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
5052 | |
5053 | out_unlock: | |
9ed74f2d | 5054 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
5055 | if (!ret) |
5056 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5057 | unlock_page(page); |
1832a6d5 | 5058 | out: |
9ebefb18 CM |
5059 | return ret; |
5060 | } | |
5061 | ||
39279cc3 CM |
5062 | static void btrfs_truncate(struct inode *inode) |
5063 | { | |
5064 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5065 | int ret; | |
5066 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5067 | unsigned long nr; |
dbe674a9 | 5068 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
5069 | |
5070 | if (!S_ISREG(inode->i_mode)) | |
5071 | return; | |
5072 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
5073 | return; | |
5074 | ||
5d5e103a JB |
5075 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5076 | if (ret) | |
5077 | return; | |
4a096752 | 5078 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 5079 | |
39279cc3 | 5080 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
5081 | |
5082 | /* | |
5083 | * setattr is responsible for setting the ordered_data_close flag, | |
5084 | * but that is only tested during the last file release. That | |
5085 | * could happen well after the next commit, leaving a great big | |
5086 | * window where new writes may get lost if someone chooses to write | |
5087 | * to this file after truncating to zero | |
5088 | * | |
5089 | * The inode doesn't have any dirty data here, and so if we commit | |
5090 | * this is a noop. If someone immediately starts writing to the inode | |
5091 | * it is very likely we'll catch some of their writes in this | |
5092 | * transaction, and the commit will find this file on the ordered | |
5093 | * data list with good things to send down. | |
5094 | * | |
5095 | * This is a best effort solution, there is still a window where | |
5096 | * using truncate to replace the contents of the file will | |
5097 | * end up with a zero length file after a crash. | |
5098 | */ | |
5099 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5100 | btrfs_add_ordered_operation(trans, root, inode); | |
5101 | ||
39279cc3 | 5102 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 5103 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 5104 | |
7b128766 JB |
5105 | ret = btrfs_orphan_add(trans, inode); |
5106 | if (ret) | |
5107 | goto out; | |
39279cc3 | 5108 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 5109 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 5110 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 5111 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 5112 | |
7b128766 JB |
5113 | ret = btrfs_orphan_del(trans, inode); |
5114 | BUG_ON(ret); | |
5115 | ||
5116 | out: | |
5117 | nr = trans->blocks_used; | |
89ce8a63 | 5118 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5119 | BUG_ON(ret); |
d3c2fdcf | 5120 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5121 | } |
5122 | ||
d352ac68 CM |
5123 | /* |
5124 | * create a new subvolume directory/inode (helper for the ioctl). | |
5125 | */ | |
d2fb3437 | 5126 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5127 | struct btrfs_root *new_root, |
d2fb3437 | 5128 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5129 | { |
39279cc3 | 5130 | struct inode *inode; |
76dda93c | 5131 | int err; |
00e4e6b3 | 5132 | u64 index = 0; |
39279cc3 | 5133 | |
aec7477b | 5134 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5135 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5136 | if (IS_ERR(inode)) |
f46b5a66 | 5137 | return PTR_ERR(inode); |
39279cc3 CM |
5138 | inode->i_op = &btrfs_dir_inode_operations; |
5139 | inode->i_fop = &btrfs_dir_file_operations; | |
5140 | ||
39279cc3 | 5141 | inode->i_nlink = 1; |
dbe674a9 | 5142 | btrfs_i_size_write(inode, 0); |
3b96362c | 5143 | |
76dda93c YZ |
5144 | err = btrfs_update_inode(trans, new_root, inode); |
5145 | BUG_ON(err); | |
cb8e7090 | 5146 | |
76dda93c | 5147 | iput(inode); |
cb8e7090 | 5148 | return 0; |
39279cc3 CM |
5149 | } |
5150 | ||
d352ac68 CM |
5151 | /* helper function for file defrag and space balancing. This |
5152 | * forces readahead on a given range of bytes in an inode | |
5153 | */ | |
edbd8d4e | 5154 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5155 | struct file_ra_state *ra, struct file *file, |
5156 | pgoff_t offset, pgoff_t last_index) | |
5157 | { | |
8e7bf94f | 5158 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5159 | |
86479a04 CM |
5160 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5161 | return offset + req_size; | |
86479a04 CM |
5162 | } |
5163 | ||
39279cc3 CM |
5164 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5165 | { | |
5166 | struct btrfs_inode *ei; | |
5167 | ||
5168 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5169 | if (!ei) | |
5170 | return NULL; | |
15ee9bc7 | 5171 | ei->last_trans = 0; |
257c62e1 | 5172 | ei->last_sub_trans = 0; |
e02119d5 | 5173 | ei->logged_trans = 0; |
32c00aff JB |
5174 | ei->outstanding_extents = 0; |
5175 | ei->reserved_extents = 0; | |
a6dbd429 | 5176 | ei->root = NULL; |
32c00aff | 5177 | spin_lock_init(&ei->accounting_lock); |
e6dcd2dc | 5178 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5179 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5180 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5181 | return &ei->vfs_inode; |
5182 | } | |
5183 | ||
5184 | void btrfs_destroy_inode(struct inode *inode) | |
5185 | { | |
e6dcd2dc | 5186 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5187 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5188 | ||
39279cc3 CM |
5189 | WARN_ON(!list_empty(&inode->i_dentry)); |
5190 | WARN_ON(inode->i_data.nrpages); | |
5191 | ||
a6dbd429 JB |
5192 | /* |
5193 | * This can happen where we create an inode, but somebody else also | |
5194 | * created the same inode and we need to destroy the one we already | |
5195 | * created. | |
5196 | */ | |
5197 | if (!root) | |
5198 | goto free; | |
5199 | ||
5a3f23d5 CM |
5200 | /* |
5201 | * Make sure we're properly removed from the ordered operation | |
5202 | * lists. | |
5203 | */ | |
5204 | smp_mb(); | |
5205 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5206 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5207 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5208 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5209 | } | |
5210 | ||
5211 | spin_lock(&root->list_lock); | |
7b128766 JB |
5212 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
5213 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
5214 | " list\n", inode->i_ino); | |
5215 | dump_stack(); | |
5216 | } | |
5a3f23d5 | 5217 | spin_unlock(&root->list_lock); |
7b128766 | 5218 | |
d397712b | 5219 | while (1) { |
e6dcd2dc CM |
5220 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5221 | if (!ordered) | |
5222 | break; | |
5223 | else { | |
d397712b CM |
5224 | printk(KERN_ERR "btrfs found ordered " |
5225 | "extent %llu %llu on inode cleanup\n", | |
5226 | (unsigned long long)ordered->file_offset, | |
5227 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5228 | btrfs_remove_ordered_extent(inode, ordered); |
5229 | btrfs_put_ordered_extent(ordered); | |
5230 | btrfs_put_ordered_extent(ordered); | |
5231 | } | |
5232 | } | |
5d4f98a2 | 5233 | inode_tree_del(inode); |
5b21f2ed | 5234 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 5235 | free: |
39279cc3 CM |
5236 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5237 | } | |
5238 | ||
76dda93c YZ |
5239 | void btrfs_drop_inode(struct inode *inode) |
5240 | { | |
5241 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5242 | ||
5243 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5244 | generic_delete_inode(inode); | |
5245 | else | |
5246 | generic_drop_inode(inode); | |
5247 | } | |
5248 | ||
0ee0fda0 | 5249 | static void init_once(void *foo) |
39279cc3 CM |
5250 | { |
5251 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5252 | ||
5253 | inode_init_once(&ei->vfs_inode); | |
5254 | } | |
5255 | ||
5256 | void btrfs_destroy_cachep(void) | |
5257 | { | |
5258 | if (btrfs_inode_cachep) | |
5259 | kmem_cache_destroy(btrfs_inode_cachep); | |
5260 | if (btrfs_trans_handle_cachep) | |
5261 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5262 | if (btrfs_transaction_cachep) | |
5263 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5264 | if (btrfs_path_cachep) |
5265 | kmem_cache_destroy(btrfs_path_cachep); | |
5266 | } | |
5267 | ||
5268 | int btrfs_init_cachep(void) | |
5269 | { | |
9601e3f6 CH |
5270 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5271 | sizeof(struct btrfs_inode), 0, | |
5272 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5273 | if (!btrfs_inode_cachep) |
5274 | goto fail; | |
9601e3f6 CH |
5275 | |
5276 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5277 | sizeof(struct btrfs_trans_handle), 0, | |
5278 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5279 | if (!btrfs_trans_handle_cachep) |
5280 | goto fail; | |
9601e3f6 CH |
5281 | |
5282 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5283 | sizeof(struct btrfs_transaction), 0, | |
5284 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5285 | if (!btrfs_transaction_cachep) |
5286 | goto fail; | |
9601e3f6 CH |
5287 | |
5288 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5289 | sizeof(struct btrfs_path), 0, | |
5290 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5291 | if (!btrfs_path_cachep) |
5292 | goto fail; | |
9601e3f6 | 5293 | |
39279cc3 CM |
5294 | return 0; |
5295 | fail: | |
5296 | btrfs_destroy_cachep(); | |
5297 | return -ENOMEM; | |
5298 | } | |
5299 | ||
5300 | static int btrfs_getattr(struct vfsmount *mnt, | |
5301 | struct dentry *dentry, struct kstat *stat) | |
5302 | { | |
5303 | struct inode *inode = dentry->d_inode; | |
5304 | generic_fillattr(inode, stat); | |
3394e160 | 5305 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5306 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5307 | stat->blocks = (inode_get_bytes(inode) + |
5308 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5309 | return 0; |
5310 | } | |
5311 | ||
d397712b CM |
5312 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5313 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5314 | { |
5315 | struct btrfs_trans_handle *trans; | |
5316 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5317 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5318 | struct inode *new_inode = new_dentry->d_inode; |
5319 | struct inode *old_inode = old_dentry->d_inode; | |
5320 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5321 | u64 index = 0; |
4df27c4d | 5322 | u64 root_objectid; |
39279cc3 CM |
5323 | int ret; |
5324 | ||
f679a840 YZ |
5325 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5326 | return -EPERM; | |
5327 | ||
4df27c4d YZ |
5328 | /* we only allow rename subvolume link between subvolumes */ |
5329 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5330 | return -EXDEV; |
5331 | ||
4df27c4d YZ |
5332 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5333 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5334 | return -ENOTEMPTY; |
5f39d397 | 5335 | |
4df27c4d YZ |
5336 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5337 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5338 | return -ENOTEMPTY; | |
0660b5af | 5339 | |
9ed74f2d | 5340 | /* |
5df6a9f6 JB |
5341 | * We want to reserve the absolute worst case amount of items. So if |
5342 | * both inodes are subvols and we need to unlink them then that would | |
5343 | * require 4 item modifications, but if they are both normal inodes it | |
5344 | * would require 5 item modifications, so we'll assume their normal | |
5345 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
5346 | * should cover the worst case number of items we'll modify. | |
9ed74f2d | 5347 | */ |
5df6a9f6 | 5348 | ret = btrfs_reserve_metadata_space(root, 11); |
1832a6d5 | 5349 | if (ret) |
4df27c4d | 5350 | return ret; |
1832a6d5 | 5351 | |
5a3f23d5 CM |
5352 | /* |
5353 | * we're using rename to replace one file with another. | |
5354 | * and the replacement file is large. Start IO on it now so | |
5355 | * we don't add too much work to the end of the transaction | |
5356 | */ | |
4baf8c92 | 5357 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5358 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5359 | filemap_flush(old_inode->i_mapping); | |
5360 | ||
76dda93c YZ |
5361 | /* close the racy window with snapshot create/destroy ioctl */ |
5362 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5363 | down_read(&root->fs_info->subvol_sem); | |
5364 | ||
39279cc3 | 5365 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5366 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5367 | |
4df27c4d YZ |
5368 | if (dest != root) |
5369 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 5370 | |
a5719521 YZ |
5371 | ret = btrfs_set_inode_index(new_dir, &index); |
5372 | if (ret) | |
5373 | goto out_fail; | |
5a3f23d5 | 5374 | |
a5719521 | 5375 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5376 | /* force full log commit if subvolume involved. */ |
5377 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5378 | } else { | |
a5719521 YZ |
5379 | ret = btrfs_insert_inode_ref(trans, dest, |
5380 | new_dentry->d_name.name, | |
5381 | new_dentry->d_name.len, | |
5382 | old_inode->i_ino, | |
5383 | new_dir->i_ino, index); | |
5384 | if (ret) | |
5385 | goto out_fail; | |
4df27c4d YZ |
5386 | /* |
5387 | * this is an ugly little race, but the rename is required | |
5388 | * to make sure that if we crash, the inode is either at the | |
5389 | * old name or the new one. pinning the log transaction lets | |
5390 | * us make sure we don't allow a log commit to come in after | |
5391 | * we unlink the name but before we add the new name back in. | |
5392 | */ | |
5393 | btrfs_pin_log_trans(root); | |
5394 | } | |
5a3f23d5 CM |
5395 | /* |
5396 | * make sure the inode gets flushed if it is replacing | |
5397 | * something. | |
5398 | */ | |
5399 | if (new_inode && new_inode->i_size && | |
5400 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5401 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5402 | } | |
5403 | ||
39279cc3 CM |
5404 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5405 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5406 | old_inode->i_ctime = ctime; | |
5f39d397 | 5407 | |
12fcfd22 CM |
5408 | if (old_dentry->d_parent != new_dentry->d_parent) |
5409 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5410 | ||
4df27c4d YZ |
5411 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5412 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5413 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5414 | old_dentry->d_name.name, | |
5415 | old_dentry->d_name.len); | |
5416 | } else { | |
5417 | btrfs_inc_nlink(old_dentry->d_inode); | |
5418 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5419 | old_dentry->d_inode, | |
5420 | old_dentry->d_name.name, | |
5421 | old_dentry->d_name.len); | |
5422 | } | |
5423 | BUG_ON(ret); | |
39279cc3 CM |
5424 | |
5425 | if (new_inode) { | |
5426 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5427 | if (unlikely(new_inode->i_ino == |
5428 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5429 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5430 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5431 | root_objectid, | |
5432 | new_dentry->d_name.name, | |
5433 | new_dentry->d_name.len); | |
5434 | BUG_ON(new_inode->i_nlink == 0); | |
5435 | } else { | |
5436 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5437 | new_dentry->d_inode, | |
5438 | new_dentry->d_name.name, | |
5439 | new_dentry->d_name.len); | |
5440 | } | |
5441 | BUG_ON(ret); | |
7b128766 | 5442 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5443 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5444 | BUG_ON(ret); |
7b128766 | 5445 | } |
39279cc3 | 5446 | } |
aec7477b | 5447 | |
4df27c4d YZ |
5448 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5449 | new_dentry->d_name.name, | |
a5719521 | 5450 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5451 | BUG_ON(ret); |
39279cc3 | 5452 | |
4df27c4d YZ |
5453 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5454 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5455 | new_dentry->d_parent); | |
5456 | btrfs_end_log_trans(root); | |
5457 | } | |
39279cc3 | 5458 | out_fail: |
ab78c84d | 5459 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5460 | |
76dda93c YZ |
5461 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5462 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 5463 | |
5df6a9f6 | 5464 | btrfs_unreserve_metadata_space(root, 11); |
39279cc3 CM |
5465 | return ret; |
5466 | } | |
5467 | ||
d352ac68 CM |
5468 | /* |
5469 | * some fairly slow code that needs optimization. This walks the list | |
5470 | * of all the inodes with pending delalloc and forces them to disk. | |
5471 | */ | |
ea8c2819 CM |
5472 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5473 | { | |
5474 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5475 | struct btrfs_inode *binode; | |
5b21f2ed | 5476 | struct inode *inode; |
ea8c2819 | 5477 | |
c146afad YZ |
5478 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5479 | return -EROFS; | |
5480 | ||
75eff68e | 5481 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5482 | while (!list_empty(head)) { |
ea8c2819 CM |
5483 | binode = list_entry(head->next, struct btrfs_inode, |
5484 | delalloc_inodes); | |
5b21f2ed ZY |
5485 | inode = igrab(&binode->vfs_inode); |
5486 | if (!inode) | |
5487 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5488 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5489 | if (inode) { |
8c8bee1d | 5490 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5491 | iput(inode); |
5492 | } | |
5493 | cond_resched(); | |
75eff68e | 5494 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5495 | } |
75eff68e | 5496 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5497 | |
5498 | /* the filemap_flush will queue IO into the worker threads, but | |
5499 | * we have to make sure the IO is actually started and that | |
5500 | * ordered extents get created before we return | |
5501 | */ | |
5502 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5503 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5504 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5505 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5506 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5507 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5508 | } |
5509 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5510 | return 0; |
5511 | } | |
5512 | ||
39279cc3 CM |
5513 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5514 | const char *symname) | |
5515 | { | |
5516 | struct btrfs_trans_handle *trans; | |
5517 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5518 | struct btrfs_path *path; | |
5519 | struct btrfs_key key; | |
1832a6d5 | 5520 | struct inode *inode = NULL; |
39279cc3 CM |
5521 | int err; |
5522 | int drop_inode = 0; | |
5523 | u64 objectid; | |
00e4e6b3 | 5524 | u64 index = 0 ; |
39279cc3 CM |
5525 | int name_len; |
5526 | int datasize; | |
5f39d397 | 5527 | unsigned long ptr; |
39279cc3 | 5528 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5529 | struct extent_buffer *leaf; |
1832a6d5 | 5530 | unsigned long nr = 0; |
39279cc3 CM |
5531 | |
5532 | name_len = strlen(symname) + 1; | |
5533 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5534 | return -ENAMETOOLONG; | |
1832a6d5 | 5535 | |
9ed74f2d JB |
5536 | /* |
5537 | * 2 items for inode item and ref | |
5538 | * 2 items for dir items | |
5539 | * 1 item for xattr if selinux is on | |
5540 | */ | |
5541 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5542 | if (err) |
9ed74f2d | 5543 | return err; |
1832a6d5 | 5544 | |
39279cc3 | 5545 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5546 | if (!trans) |
5547 | goto out_fail; | |
39279cc3 CM |
5548 | btrfs_set_trans_block_group(trans, dir); |
5549 | ||
5550 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5551 | if (err) { | |
5552 | err = -ENOSPC; | |
5553 | goto out_unlock; | |
5554 | } | |
5555 | ||
aec7477b | 5556 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5557 | dentry->d_name.len, |
5558 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5559 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5560 | &index); | |
39279cc3 CM |
5561 | err = PTR_ERR(inode); |
5562 | if (IS_ERR(inode)) | |
5563 | goto out_unlock; | |
5564 | ||
0279b4cd | 5565 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5566 | if (err) { |
5567 | drop_inode = 1; | |
5568 | goto out_unlock; | |
5569 | } | |
5570 | ||
39279cc3 | 5571 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5572 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5573 | if (err) |
5574 | drop_inode = 1; | |
5575 | else { | |
5576 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5577 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5578 | inode->i_fop = &btrfs_file_operations; |
5579 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5580 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5581 | } |
39279cc3 CM |
5582 | btrfs_update_inode_block_group(trans, inode); |
5583 | btrfs_update_inode_block_group(trans, dir); | |
5584 | if (drop_inode) | |
5585 | goto out_unlock; | |
5586 | ||
5587 | path = btrfs_alloc_path(); | |
5588 | BUG_ON(!path); | |
5589 | key.objectid = inode->i_ino; | |
5590 | key.offset = 0; | |
39279cc3 CM |
5591 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5592 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5593 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5594 | datasize); | |
54aa1f4d CM |
5595 | if (err) { |
5596 | drop_inode = 1; | |
5597 | goto out_unlock; | |
5598 | } | |
5f39d397 CM |
5599 | leaf = path->nodes[0]; |
5600 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5601 | struct btrfs_file_extent_item); | |
5602 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5603 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5604 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5605 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5606 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5607 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5608 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5609 | ||
39279cc3 | 5610 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5611 | write_extent_buffer(leaf, symname, ptr, name_len); |
5612 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5613 | btrfs_free_path(path); |
5f39d397 | 5614 | |
39279cc3 CM |
5615 | inode->i_op = &btrfs_symlink_inode_operations; |
5616 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5617 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5618 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5619 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5620 | err = btrfs_update_inode(trans, root, inode); |
5621 | if (err) | |
5622 | drop_inode = 1; | |
39279cc3 CM |
5623 | |
5624 | out_unlock: | |
d3c2fdcf | 5625 | nr = trans->blocks_used; |
ab78c84d | 5626 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5627 | out_fail: |
9ed74f2d | 5628 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5629 | if (drop_inode) { |
5630 | inode_dec_link_count(inode); | |
5631 | iput(inode); | |
5632 | } | |
d3c2fdcf | 5633 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5634 | return err; |
5635 | } | |
16432985 | 5636 | |
546888da CM |
5637 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5638 | struct inode *inode, u64 start, u64 end, | |
920bbbfb | 5639 | u64 alloc_hint, int mode) |
d899e052 | 5640 | { |
d899e052 YZ |
5641 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5642 | struct btrfs_key ins; | |
5643 | u64 alloc_size; | |
5644 | u64 cur_offset = start; | |
5645 | u64 num_bytes = end - start; | |
5646 | int ret = 0; | |
5647 | ||
d899e052 YZ |
5648 | while (num_bytes > 0) { |
5649 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d JB |
5650 | |
5651 | ret = btrfs_reserve_metadata_space(root, 1); | |
5652 | if (ret) | |
5653 | goto out; | |
5654 | ||
d899e052 YZ |
5655 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5656 | root->sectorsize, 0, alloc_hint, | |
5657 | (u64)-1, &ins, 1); | |
5658 | if (ret) { | |
5659 | WARN_ON(1); | |
5660 | goto out; | |
5661 | } | |
5662 | ret = insert_reserved_file_extent(trans, inode, | |
5663 | cur_offset, ins.objectid, | |
5664 | ins.offset, ins.offset, | |
920bbbfb | 5665 | ins.offset, 0, 0, 0, |
d899e052 YZ |
5666 | BTRFS_FILE_EXTENT_PREALLOC); |
5667 | BUG_ON(ret); | |
a1ed835e CM |
5668 | btrfs_drop_extent_cache(inode, cur_offset, |
5669 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5670 | num_bytes -= ins.offset; |
5671 | cur_offset += ins.offset; | |
5672 | alloc_hint = ins.objectid + ins.offset; | |
9ed74f2d | 5673 | btrfs_unreserve_metadata_space(root, 1); |
d899e052 YZ |
5674 | } |
5675 | out: | |
5676 | if (cur_offset > start) { | |
5677 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5678 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5679 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5680 | cur_offset > i_size_read(inode)) | |
5681 | btrfs_i_size_write(inode, cur_offset); | |
5682 | ret = btrfs_update_inode(trans, root, inode); | |
5683 | BUG_ON(ret); | |
5684 | } | |
5685 | ||
d899e052 YZ |
5686 | return ret; |
5687 | } | |
5688 | ||
5689 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5690 | loff_t offset, loff_t len) | |
5691 | { | |
5692 | u64 cur_offset; | |
5693 | u64 last_byte; | |
5694 | u64 alloc_start; | |
5695 | u64 alloc_end; | |
5696 | u64 alloc_hint = 0; | |
e980b50c | 5697 | u64 locked_end; |
d899e052 YZ |
5698 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5699 | struct extent_map *em; | |
546888da | 5700 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5701 | struct btrfs_root *root; |
d899e052 YZ |
5702 | int ret; |
5703 | ||
5704 | alloc_start = offset & ~mask; | |
5705 | alloc_end = (offset + len + mask) & ~mask; | |
5706 | ||
546888da CM |
5707 | /* |
5708 | * wait for ordered IO before we have any locks. We'll loop again | |
5709 | * below with the locks held. | |
5710 | */ | |
5711 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5712 | ||
d899e052 YZ |
5713 | mutex_lock(&inode->i_mutex); |
5714 | if (alloc_start > inode->i_size) { | |
5715 | ret = btrfs_cont_expand(inode, alloc_start); | |
5716 | if (ret) | |
5717 | goto out; | |
5718 | } | |
5719 | ||
a970b0a1 JB |
5720 | root = BTRFS_I(inode)->root; |
5721 | ||
5722 | ret = btrfs_check_data_free_space(root, inode, | |
5723 | alloc_end - alloc_start); | |
5724 | if (ret) | |
5725 | goto out; | |
5726 | ||
e980b50c | 5727 | locked_end = alloc_end - 1; |
d899e052 YZ |
5728 | while (1) { |
5729 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5730 | |
5731 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5732 | if (!trans) { | |
5733 | ret = -EIO; | |
a970b0a1 | 5734 | goto out_free; |
546888da CM |
5735 | } |
5736 | ||
5737 | /* the extent lock is ordered inside the running | |
5738 | * transaction | |
5739 | */ | |
e980b50c CM |
5740 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5741 | GFP_NOFS); | |
d899e052 YZ |
5742 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5743 | alloc_end - 1); | |
5744 | if (ordered && | |
5745 | ordered->file_offset + ordered->len > alloc_start && | |
5746 | ordered->file_offset < alloc_end) { | |
5747 | btrfs_put_ordered_extent(ordered); | |
5748 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5749 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5750 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5751 | ||
5752 | /* | |
5753 | * we can't wait on the range with the transaction | |
5754 | * running or with the extent lock held | |
5755 | */ | |
d899e052 YZ |
5756 | btrfs_wait_ordered_range(inode, alloc_start, |
5757 | alloc_end - alloc_start); | |
5758 | } else { | |
5759 | if (ordered) | |
5760 | btrfs_put_ordered_extent(ordered); | |
5761 | break; | |
5762 | } | |
5763 | } | |
5764 | ||
5765 | cur_offset = alloc_start; | |
5766 | while (1) { | |
5767 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5768 | alloc_end - cur_offset, 0); | |
5769 | BUG_ON(IS_ERR(em) || !em); | |
5770 | last_byte = min(extent_map_end(em), alloc_end); | |
5771 | last_byte = (last_byte + mask) & ~mask; | |
5772 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5773 | ret = prealloc_file_range(trans, inode, cur_offset, |
920bbbfb | 5774 | last_byte, alloc_hint, mode); |
d899e052 YZ |
5775 | if (ret < 0) { |
5776 | free_extent_map(em); | |
5777 | break; | |
5778 | } | |
5779 | } | |
5780 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5781 | alloc_hint = em->block_start; | |
5782 | free_extent_map(em); | |
5783 | ||
5784 | cur_offset = last_byte; | |
5785 | if (cur_offset >= alloc_end) { | |
5786 | ret = 0; | |
5787 | break; | |
5788 | } | |
5789 | } | |
e980b50c | 5790 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5791 | GFP_NOFS); |
546888da CM |
5792 | |
5793 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5794 | out_free: |
5795 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5796 | out: |
5797 | mutex_unlock(&inode->i_mutex); | |
5798 | return ret; | |
5799 | } | |
5800 | ||
e6dcd2dc CM |
5801 | static int btrfs_set_page_dirty(struct page *page) |
5802 | { | |
e6dcd2dc CM |
5803 | return __set_page_dirty_nobuffers(page); |
5804 | } | |
5805 | ||
0ee0fda0 | 5806 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5807 | { |
6cbff00f | 5808 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5809 | return -EACCES; |
33268eaf | 5810 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5811 | } |
39279cc3 | 5812 | |
6e1d5dcc | 5813 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 5814 | .getattr = btrfs_getattr, |
39279cc3 CM |
5815 | .lookup = btrfs_lookup, |
5816 | .create = btrfs_create, | |
5817 | .unlink = btrfs_unlink, | |
5818 | .link = btrfs_link, | |
5819 | .mkdir = btrfs_mkdir, | |
5820 | .rmdir = btrfs_rmdir, | |
5821 | .rename = btrfs_rename, | |
5822 | .symlink = btrfs_symlink, | |
5823 | .setattr = btrfs_setattr, | |
618e21d5 | 5824 | .mknod = btrfs_mknod, |
95819c05 CH |
5825 | .setxattr = btrfs_setxattr, |
5826 | .getxattr = btrfs_getxattr, | |
5103e947 | 5827 | .listxattr = btrfs_listxattr, |
95819c05 | 5828 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5829 | .permission = btrfs_permission, |
39279cc3 | 5830 | }; |
6e1d5dcc | 5831 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 5832 | .lookup = btrfs_lookup, |
fdebe2bd | 5833 | .permission = btrfs_permission, |
39279cc3 | 5834 | }; |
76dda93c | 5835 | |
828c0950 | 5836 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
5837 | .llseek = generic_file_llseek, |
5838 | .read = generic_read_dir, | |
cbdf5a24 | 5839 | .readdir = btrfs_real_readdir, |
34287aa3 | 5840 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5841 | #ifdef CONFIG_COMPAT |
34287aa3 | 5842 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5843 | #endif |
6bf13c0c | 5844 | .release = btrfs_release_file, |
e02119d5 | 5845 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5846 | }; |
5847 | ||
d1310b2e | 5848 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5849 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5850 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5851 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5852 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5853 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5854 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5855 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5856 | .set_bit_hook = btrfs_set_bit_hook, |
5857 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
5858 | .merge_extent_hook = btrfs_merge_extent_hook, |
5859 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
5860 | }; |
5861 | ||
35054394 CM |
5862 | /* |
5863 | * btrfs doesn't support the bmap operation because swapfiles | |
5864 | * use bmap to make a mapping of extents in the file. They assume | |
5865 | * these extents won't change over the life of the file and they | |
5866 | * use the bmap result to do IO directly to the drive. | |
5867 | * | |
5868 | * the btrfs bmap call would return logical addresses that aren't | |
5869 | * suitable for IO and they also will change frequently as COW | |
5870 | * operations happen. So, swapfile + btrfs == corruption. | |
5871 | * | |
5872 | * For now we're avoiding this by dropping bmap. | |
5873 | */ | |
7f09410b | 5874 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
5875 | .readpage = btrfs_readpage, |
5876 | .writepage = btrfs_writepage, | |
b293f02e | 5877 | .writepages = btrfs_writepages, |
3ab2fb5a | 5878 | .readpages = btrfs_readpages, |
39279cc3 | 5879 | .sync_page = block_sync_page, |
16432985 | 5880 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5881 | .invalidatepage = btrfs_invalidatepage, |
5882 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5883 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 5884 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
5885 | }; |
5886 | ||
7f09410b | 5887 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
5888 | .readpage = btrfs_readpage, |
5889 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5890 | .invalidatepage = btrfs_invalidatepage, |
5891 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5892 | }; |
5893 | ||
6e1d5dcc | 5894 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
5895 | .truncate = btrfs_truncate, |
5896 | .getattr = btrfs_getattr, | |
5897 | .setattr = btrfs_setattr, | |
95819c05 CH |
5898 | .setxattr = btrfs_setxattr, |
5899 | .getxattr = btrfs_getxattr, | |
5103e947 | 5900 | .listxattr = btrfs_listxattr, |
95819c05 | 5901 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5902 | .permission = btrfs_permission, |
d899e052 | 5903 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5904 | .fiemap = btrfs_fiemap, |
39279cc3 | 5905 | }; |
6e1d5dcc | 5906 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
5907 | .getattr = btrfs_getattr, |
5908 | .setattr = btrfs_setattr, | |
fdebe2bd | 5909 | .permission = btrfs_permission, |
95819c05 CH |
5910 | .setxattr = btrfs_setxattr, |
5911 | .getxattr = btrfs_getxattr, | |
33268eaf | 5912 | .listxattr = btrfs_listxattr, |
95819c05 | 5913 | .removexattr = btrfs_removexattr, |
618e21d5 | 5914 | }; |
6e1d5dcc | 5915 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
5916 | .readlink = generic_readlink, |
5917 | .follow_link = page_follow_link_light, | |
5918 | .put_link = page_put_link, | |
fdebe2bd | 5919 | .permission = btrfs_permission, |
0279b4cd JO |
5920 | .setxattr = btrfs_setxattr, |
5921 | .getxattr = btrfs_getxattr, | |
5922 | .listxattr = btrfs_listxattr, | |
5923 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5924 | }; |
76dda93c | 5925 | |
82d339d9 | 5926 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
5927 | .d_delete = btrfs_dentry_delete, |
5928 | }; |