]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
4b4e25f2 | 39 | #include "compat.h" |
39279cc3 CM |
40 | #include "ctree.h" |
41 | #include "disk-io.h" | |
42 | #include "transaction.h" | |
43 | #include "btrfs_inode.h" | |
44 | #include "ioctl.h" | |
45 | #include "print-tree.h" | |
0b86a832 | 46 | #include "volumes.h" |
e6dcd2dc | 47 | #include "ordered-data.h" |
95819c05 | 48 | #include "xattr.h" |
e02119d5 | 49 | #include "tree-log.h" |
c8b97818 | 50 | #include "compression.h" |
b4ce94de | 51 | #include "locking.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
54 | u64 ino; | |
55 | struct btrfs_root *root; | |
56 | }; | |
57 | ||
58 | static struct inode_operations btrfs_dir_inode_operations; | |
59 | static struct inode_operations btrfs_symlink_inode_operations; | |
60 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 61 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
62 | static struct inode_operations btrfs_file_inode_operations; |
63 | static struct address_space_operations btrfs_aops; | |
64 | static struct address_space_operations btrfs_symlink_aops; | |
65 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 66 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
67 | |
68 | static struct kmem_cache *btrfs_inode_cachep; | |
69 | struct kmem_cache *btrfs_trans_handle_cachep; | |
70 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
71 | struct kmem_cache *btrfs_path_cachep; |
72 | ||
73 | #define S_SHIFT 12 | |
74 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
75 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
76 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
77 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
78 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
79 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
80 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
81 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
82 | }; | |
83 | ||
7b128766 | 84 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 85 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
86 | static noinline int cow_file_range(struct inode *inode, |
87 | struct page *locked_page, | |
88 | u64 start, u64 end, int *page_started, | |
89 | unsigned long *nr_written, int unlock); | |
7b128766 | 90 | |
0279b4cd JO |
91 | static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) |
92 | { | |
93 | int err; | |
94 | ||
95 | err = btrfs_init_acl(inode, dir); | |
96 | if (!err) | |
97 | err = btrfs_xattr_security_init(inode, dir); | |
98 | return err; | |
99 | } | |
100 | ||
c8b97818 CM |
101 | /* |
102 | * this does all the hard work for inserting an inline extent into | |
103 | * the btree. The caller should have done a btrfs_drop_extents so that | |
104 | * no overlapping inline items exist in the btree | |
105 | */ | |
d397712b | 106 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
107 | struct btrfs_root *root, struct inode *inode, |
108 | u64 start, size_t size, size_t compressed_size, | |
109 | struct page **compressed_pages) | |
110 | { | |
111 | struct btrfs_key key; | |
112 | struct btrfs_path *path; | |
113 | struct extent_buffer *leaf; | |
114 | struct page *page = NULL; | |
115 | char *kaddr; | |
116 | unsigned long ptr; | |
117 | struct btrfs_file_extent_item *ei; | |
118 | int err = 0; | |
119 | int ret; | |
120 | size_t cur_size = size; | |
121 | size_t datasize; | |
122 | unsigned long offset; | |
123 | int use_compress = 0; | |
124 | ||
125 | if (compressed_size && compressed_pages) { | |
126 | use_compress = 1; | |
127 | cur_size = compressed_size; | |
128 | } | |
129 | ||
d397712b CM |
130 | path = btrfs_alloc_path(); |
131 | if (!path) | |
c8b97818 CM |
132 | return -ENOMEM; |
133 | ||
b9473439 | 134 | path->leave_spinning = 1; |
c8b97818 CM |
135 | btrfs_set_trans_block_group(trans, inode); |
136 | ||
137 | key.objectid = inode->i_ino; | |
138 | key.offset = start; | |
139 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
140 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
141 | ||
142 | inode_add_bytes(inode, size); | |
143 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
144 | datasize); | |
145 | BUG_ON(ret); | |
146 | if (ret) { | |
147 | err = ret; | |
c8b97818 CM |
148 | goto fail; |
149 | } | |
150 | leaf = path->nodes[0]; | |
151 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
152 | struct btrfs_file_extent_item); | |
153 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
154 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
155 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
156 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
157 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
158 | ptr = btrfs_file_extent_inline_start(ei); | |
159 | ||
160 | if (use_compress) { | |
161 | struct page *cpage; | |
162 | int i = 0; | |
d397712b | 163 | while (compressed_size > 0) { |
c8b97818 | 164 | cpage = compressed_pages[i]; |
5b050f04 | 165 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
166 | PAGE_CACHE_SIZE); |
167 | ||
b9473439 | 168 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 169 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 170 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
171 | |
172 | i++; | |
173 | ptr += cur_size; | |
174 | compressed_size -= cur_size; | |
175 | } | |
176 | btrfs_set_file_extent_compression(leaf, ei, | |
177 | BTRFS_COMPRESS_ZLIB); | |
178 | } else { | |
179 | page = find_get_page(inode->i_mapping, | |
180 | start >> PAGE_CACHE_SHIFT); | |
181 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
182 | kaddr = kmap_atomic(page, KM_USER0); | |
183 | offset = start & (PAGE_CACHE_SIZE - 1); | |
184 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
185 | kunmap_atomic(kaddr, KM_USER0); | |
186 | page_cache_release(page); | |
187 | } | |
188 | btrfs_mark_buffer_dirty(leaf); | |
189 | btrfs_free_path(path); | |
190 | ||
191 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
192 | btrfs_update_inode(trans, root, inode); | |
193 | return 0; | |
194 | fail: | |
195 | btrfs_free_path(path); | |
196 | return err; | |
197 | } | |
198 | ||
199 | ||
200 | /* | |
201 | * conditionally insert an inline extent into the file. This | |
202 | * does the checks required to make sure the data is small enough | |
203 | * to fit as an inline extent. | |
204 | */ | |
7f366cfe | 205 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
206 | struct btrfs_root *root, |
207 | struct inode *inode, u64 start, u64 end, | |
208 | size_t compressed_size, | |
209 | struct page **compressed_pages) | |
210 | { | |
211 | u64 isize = i_size_read(inode); | |
212 | u64 actual_end = min(end + 1, isize); | |
213 | u64 inline_len = actual_end - start; | |
214 | u64 aligned_end = (end + root->sectorsize - 1) & | |
215 | ~((u64)root->sectorsize - 1); | |
216 | u64 hint_byte; | |
217 | u64 data_len = inline_len; | |
218 | int ret; | |
219 | ||
220 | if (compressed_size) | |
221 | data_len = compressed_size; | |
222 | ||
223 | if (start > 0 || | |
70b99e69 | 224 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
225 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
226 | (!compressed_size && | |
227 | (actual_end & (root->sectorsize - 1)) == 0) || | |
228 | end + 1 < isize || | |
229 | data_len > root->fs_info->max_inline) { | |
230 | return 1; | |
231 | } | |
232 | ||
c8b97818 | 233 | ret = btrfs_drop_extents(trans, root, inode, start, |
a1ed835e CM |
234 | aligned_end, aligned_end, start, |
235 | &hint_byte, 1); | |
c8b97818 CM |
236 | BUG_ON(ret); |
237 | ||
238 | if (isize > actual_end) | |
239 | inline_len = min_t(u64, isize, actual_end); | |
240 | ret = insert_inline_extent(trans, root, inode, start, | |
241 | inline_len, compressed_size, | |
242 | compressed_pages); | |
243 | BUG_ON(ret); | |
a1ed835e | 244 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
245 | return 0; |
246 | } | |
247 | ||
771ed689 CM |
248 | struct async_extent { |
249 | u64 start; | |
250 | u64 ram_size; | |
251 | u64 compressed_size; | |
252 | struct page **pages; | |
253 | unsigned long nr_pages; | |
254 | struct list_head list; | |
255 | }; | |
256 | ||
257 | struct async_cow { | |
258 | struct inode *inode; | |
259 | struct btrfs_root *root; | |
260 | struct page *locked_page; | |
261 | u64 start; | |
262 | u64 end; | |
263 | struct list_head extents; | |
264 | struct btrfs_work work; | |
265 | }; | |
266 | ||
267 | static noinline int add_async_extent(struct async_cow *cow, | |
268 | u64 start, u64 ram_size, | |
269 | u64 compressed_size, | |
270 | struct page **pages, | |
271 | unsigned long nr_pages) | |
272 | { | |
273 | struct async_extent *async_extent; | |
274 | ||
275 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
276 | async_extent->start = start; | |
277 | async_extent->ram_size = ram_size; | |
278 | async_extent->compressed_size = compressed_size; | |
279 | async_extent->pages = pages; | |
280 | async_extent->nr_pages = nr_pages; | |
281 | list_add_tail(&async_extent->list, &cow->extents); | |
282 | return 0; | |
283 | } | |
284 | ||
d352ac68 | 285 | /* |
771ed689 CM |
286 | * we create compressed extents in two phases. The first |
287 | * phase compresses a range of pages that have already been | |
288 | * locked (both pages and state bits are locked). | |
c8b97818 | 289 | * |
771ed689 CM |
290 | * This is done inside an ordered work queue, and the compression |
291 | * is spread across many cpus. The actual IO submission is step | |
292 | * two, and the ordered work queue takes care of making sure that | |
293 | * happens in the same order things were put onto the queue by | |
294 | * writepages and friends. | |
c8b97818 | 295 | * |
771ed689 CM |
296 | * If this code finds it can't get good compression, it puts an |
297 | * entry onto the work queue to write the uncompressed bytes. This | |
298 | * makes sure that both compressed inodes and uncompressed inodes | |
299 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 300 | */ |
771ed689 CM |
301 | static noinline int compress_file_range(struct inode *inode, |
302 | struct page *locked_page, | |
303 | u64 start, u64 end, | |
304 | struct async_cow *async_cow, | |
305 | int *num_added) | |
b888db2b CM |
306 | { |
307 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
308 | struct btrfs_trans_handle *trans; | |
db94535d | 309 | u64 num_bytes; |
c8b97818 CM |
310 | u64 orig_start; |
311 | u64 disk_num_bytes; | |
db94535d | 312 | u64 blocksize = root->sectorsize; |
c8b97818 | 313 | u64 actual_end; |
42dc7bab | 314 | u64 isize = i_size_read(inode); |
e6dcd2dc | 315 | int ret = 0; |
c8b97818 CM |
316 | struct page **pages = NULL; |
317 | unsigned long nr_pages; | |
318 | unsigned long nr_pages_ret = 0; | |
319 | unsigned long total_compressed = 0; | |
320 | unsigned long total_in = 0; | |
321 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 322 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
323 | int i; |
324 | int will_compress; | |
b888db2b | 325 | |
c8b97818 CM |
326 | orig_start = start; |
327 | ||
42dc7bab | 328 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
329 | again: |
330 | will_compress = 0; | |
331 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
332 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 333 | |
f03d9301 CM |
334 | /* |
335 | * we don't want to send crud past the end of i_size through | |
336 | * compression, that's just a waste of CPU time. So, if the | |
337 | * end of the file is before the start of our current | |
338 | * requested range of bytes, we bail out to the uncompressed | |
339 | * cleanup code that can deal with all of this. | |
340 | * | |
341 | * It isn't really the fastest way to fix things, but this is a | |
342 | * very uncommon corner. | |
343 | */ | |
344 | if (actual_end <= start) | |
345 | goto cleanup_and_bail_uncompressed; | |
346 | ||
c8b97818 CM |
347 | total_compressed = actual_end - start; |
348 | ||
349 | /* we want to make sure that amount of ram required to uncompress | |
350 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
351 | * of a compressed extent to 128k. This is a crucial number |
352 | * because it also controls how easily we can spread reads across | |
353 | * cpus for decompression. | |
354 | * | |
355 | * We also want to make sure the amount of IO required to do | |
356 | * a random read is reasonably small, so we limit the size of | |
357 | * a compressed extent to 128k. | |
c8b97818 CM |
358 | */ |
359 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 360 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 361 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
362 | disk_num_bytes = num_bytes; |
363 | total_in = 0; | |
364 | ret = 0; | |
db94535d | 365 | |
771ed689 CM |
366 | /* |
367 | * we do compression for mount -o compress and when the | |
368 | * inode has not been flagged as nocompress. This flag can | |
369 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 370 | */ |
6cbff00f | 371 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
372 | btrfs_test_opt(root, COMPRESS)) { |
373 | WARN_ON(pages); | |
cfbc246e | 374 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 375 | |
c8b97818 CM |
376 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
377 | total_compressed, pages, | |
378 | nr_pages, &nr_pages_ret, | |
379 | &total_in, | |
380 | &total_compressed, | |
381 | max_compressed); | |
382 | ||
383 | if (!ret) { | |
384 | unsigned long offset = total_compressed & | |
385 | (PAGE_CACHE_SIZE - 1); | |
386 | struct page *page = pages[nr_pages_ret - 1]; | |
387 | char *kaddr; | |
388 | ||
389 | /* zero the tail end of the last page, we might be | |
390 | * sending it down to disk | |
391 | */ | |
392 | if (offset) { | |
393 | kaddr = kmap_atomic(page, KM_USER0); | |
394 | memset(kaddr + offset, 0, | |
395 | PAGE_CACHE_SIZE - offset); | |
396 | kunmap_atomic(kaddr, KM_USER0); | |
397 | } | |
398 | will_compress = 1; | |
399 | } | |
400 | } | |
401 | if (start == 0) { | |
771ed689 CM |
402 | trans = btrfs_join_transaction(root, 1); |
403 | BUG_ON(!trans); | |
404 | btrfs_set_trans_block_group(trans, inode); | |
405 | ||
c8b97818 | 406 | /* lets try to make an inline extent */ |
771ed689 | 407 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 408 | /* we didn't compress the entire range, try |
771ed689 | 409 | * to make an uncompressed inline extent. |
c8b97818 CM |
410 | */ |
411 | ret = cow_file_range_inline(trans, root, inode, | |
412 | start, end, 0, NULL); | |
413 | } else { | |
771ed689 | 414 | /* try making a compressed inline extent */ |
c8b97818 CM |
415 | ret = cow_file_range_inline(trans, root, inode, |
416 | start, end, | |
417 | total_compressed, pages); | |
418 | } | |
771ed689 | 419 | btrfs_end_transaction(trans, root); |
c8b97818 | 420 | if (ret == 0) { |
771ed689 CM |
421 | /* |
422 | * inline extent creation worked, we don't need | |
423 | * to create any more async work items. Unlock | |
424 | * and free up our temp pages. | |
425 | */ | |
c8b97818 | 426 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
427 | &BTRFS_I(inode)->io_tree, |
428 | start, end, NULL, | |
429 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 430 | EXTENT_CLEAR_DELALLOC | |
32c00aff | 431 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e | 432 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c8b97818 CM |
433 | ret = 0; |
434 | goto free_pages_out; | |
435 | } | |
436 | } | |
437 | ||
438 | if (will_compress) { | |
439 | /* | |
440 | * we aren't doing an inline extent round the compressed size | |
441 | * up to a block size boundary so the allocator does sane | |
442 | * things | |
443 | */ | |
444 | total_compressed = (total_compressed + blocksize - 1) & | |
445 | ~(blocksize - 1); | |
446 | ||
447 | /* | |
448 | * one last check to make sure the compression is really a | |
449 | * win, compare the page count read with the blocks on disk | |
450 | */ | |
451 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
452 | ~(PAGE_CACHE_SIZE - 1); | |
453 | if (total_compressed >= total_in) { | |
454 | will_compress = 0; | |
455 | } else { | |
456 | disk_num_bytes = total_compressed; | |
457 | num_bytes = total_in; | |
458 | } | |
459 | } | |
460 | if (!will_compress && pages) { | |
461 | /* | |
462 | * the compression code ran but failed to make things smaller, | |
463 | * free any pages it allocated and our page pointer array | |
464 | */ | |
465 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 466 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
467 | page_cache_release(pages[i]); |
468 | } | |
469 | kfree(pages); | |
470 | pages = NULL; | |
471 | total_compressed = 0; | |
472 | nr_pages_ret = 0; | |
473 | ||
474 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 475 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 476 | } |
771ed689 CM |
477 | if (will_compress) { |
478 | *num_added += 1; | |
c8b97818 | 479 | |
771ed689 CM |
480 | /* the async work queues will take care of doing actual |
481 | * allocation on disk for these compressed pages, | |
482 | * and will submit them to the elevator. | |
483 | */ | |
484 | add_async_extent(async_cow, start, num_bytes, | |
485 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 486 | |
42dc7bab | 487 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
488 | start += num_bytes; |
489 | pages = NULL; | |
490 | cond_resched(); | |
491 | goto again; | |
492 | } | |
493 | } else { | |
f03d9301 | 494 | cleanup_and_bail_uncompressed: |
771ed689 CM |
495 | /* |
496 | * No compression, but we still need to write the pages in | |
497 | * the file we've been given so far. redirty the locked | |
498 | * page if it corresponds to our extent and set things up | |
499 | * for the async work queue to run cow_file_range to do | |
500 | * the normal delalloc dance | |
501 | */ | |
502 | if (page_offset(locked_page) >= start && | |
503 | page_offset(locked_page) <= end) { | |
504 | __set_page_dirty_nobuffers(locked_page); | |
505 | /* unlocked later on in the async handlers */ | |
506 | } | |
507 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
508 | *num_added += 1; | |
509 | } | |
3b951516 | 510 | |
771ed689 CM |
511 | out: |
512 | return 0; | |
513 | ||
514 | free_pages_out: | |
515 | for (i = 0; i < nr_pages_ret; i++) { | |
516 | WARN_ON(pages[i]->mapping); | |
517 | page_cache_release(pages[i]); | |
518 | } | |
d397712b | 519 | kfree(pages); |
771ed689 CM |
520 | |
521 | goto out; | |
522 | } | |
523 | ||
524 | /* | |
525 | * phase two of compressed writeback. This is the ordered portion | |
526 | * of the code, which only gets called in the order the work was | |
527 | * queued. We walk all the async extents created by compress_file_range | |
528 | * and send them down to the disk. | |
529 | */ | |
530 | static noinline int submit_compressed_extents(struct inode *inode, | |
531 | struct async_cow *async_cow) | |
532 | { | |
533 | struct async_extent *async_extent; | |
534 | u64 alloc_hint = 0; | |
535 | struct btrfs_trans_handle *trans; | |
536 | struct btrfs_key ins; | |
537 | struct extent_map *em; | |
538 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
539 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
540 | struct extent_io_tree *io_tree; | |
541 | int ret; | |
542 | ||
543 | if (list_empty(&async_cow->extents)) | |
544 | return 0; | |
545 | ||
546 | trans = btrfs_join_transaction(root, 1); | |
547 | ||
d397712b | 548 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
549 | async_extent = list_entry(async_cow->extents.next, |
550 | struct async_extent, list); | |
551 | list_del(&async_extent->list); | |
c8b97818 | 552 | |
771ed689 CM |
553 | io_tree = &BTRFS_I(inode)->io_tree; |
554 | ||
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 */ | |
565 | 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); | |
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 | */ | |
577 | if (!page_started) | |
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); | |
605 | BUG_ON(ret); | |
606 | em = alloc_extent_map(GFP_NOFS); | |
607 | em->start = async_extent->start; | |
608 | em->len = async_extent->ram_size; | |
445a6944 | 609 | em->orig_start = em->start; |
c8b97818 | 610 | |
771ed689 CM |
611 | em->block_start = ins.objectid; |
612 | em->block_len = ins.offset; | |
613 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
614 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
615 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
616 | ||
d397712b | 617 | while (1) { |
890871be | 618 | write_lock(&em_tree->lock); |
771ed689 | 619 | ret = add_extent_mapping(em_tree, em); |
890871be | 620 | write_unlock(&em_tree->lock); |
771ed689 CM |
621 | if (ret != -EEXIST) { |
622 | free_extent_map(em); | |
623 | break; | |
624 | } | |
625 | btrfs_drop_extent_cache(inode, async_extent->start, | |
626 | async_extent->start + | |
627 | async_extent->ram_size - 1, 0); | |
628 | } | |
629 | ||
630 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
631 | ins.objectid, | |
632 | async_extent->ram_size, | |
633 | ins.offset, | |
634 | BTRFS_ORDERED_COMPRESSED); | |
635 | BUG_ON(ret); | |
636 | ||
637 | btrfs_end_transaction(trans, root); | |
638 | ||
639 | /* | |
640 | * clear dirty, set writeback and unlock the pages. | |
641 | */ | |
642 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
643 | &BTRFS_I(inode)->io_tree, |
644 | async_extent->start, | |
645 | async_extent->start + | |
646 | async_extent->ram_size - 1, | |
647 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
648 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 649 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 650 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
651 | |
652 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
653 | async_extent->start, |
654 | async_extent->ram_size, | |
655 | ins.objectid, | |
656 | ins.offset, async_extent->pages, | |
657 | async_extent->nr_pages); | |
771ed689 CM |
658 | |
659 | BUG_ON(ret); | |
660 | trans = btrfs_join_transaction(root, 1); | |
661 | alloc_hint = ins.objectid + ins.offset; | |
662 | kfree(async_extent); | |
663 | cond_resched(); | |
664 | } | |
665 | ||
666 | btrfs_end_transaction(trans, root); | |
667 | return 0; | |
668 | } | |
669 | ||
670 | /* | |
671 | * when extent_io.c finds a delayed allocation range in the file, | |
672 | * the call backs end up in this code. The basic idea is to | |
673 | * allocate extents on disk for the range, and create ordered data structs | |
674 | * in ram to track those extents. | |
675 | * | |
676 | * locked_page is the page that writepage had locked already. We use | |
677 | * it to make sure we don't do extra locks or unlocks. | |
678 | * | |
679 | * *page_started is set to one if we unlock locked_page and do everything | |
680 | * required to start IO on it. It may be clean and already done with | |
681 | * IO when we return. | |
682 | */ | |
683 | static noinline int cow_file_range(struct inode *inode, | |
684 | struct page *locked_page, | |
685 | u64 start, u64 end, int *page_started, | |
686 | unsigned long *nr_written, | |
687 | int unlock) | |
688 | { | |
689 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
690 | struct btrfs_trans_handle *trans; | |
691 | u64 alloc_hint = 0; | |
692 | u64 num_bytes; | |
693 | unsigned long ram_size; | |
694 | u64 disk_num_bytes; | |
695 | u64 cur_alloc_size; | |
696 | u64 blocksize = root->sectorsize; | |
697 | u64 actual_end; | |
42dc7bab | 698 | u64 isize = i_size_read(inode); |
771ed689 CM |
699 | struct btrfs_key ins; |
700 | struct extent_map *em; | |
701 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
702 | int ret = 0; | |
703 | ||
704 | trans = btrfs_join_transaction(root, 1); | |
705 | BUG_ON(!trans); | |
706 | btrfs_set_trans_block_group(trans, inode); | |
707 | ||
42dc7bab | 708 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
709 | |
710 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
711 | num_bytes = max(blocksize, num_bytes); | |
712 | disk_num_bytes = num_bytes; | |
713 | ret = 0; | |
714 | ||
715 | if (start == 0) { | |
716 | /* lets try to make an inline extent */ | |
717 | ret = cow_file_range_inline(trans, root, inode, | |
718 | start, end, 0, NULL); | |
719 | if (ret == 0) { | |
720 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
721 | &BTRFS_I(inode)->io_tree, |
722 | start, end, NULL, | |
723 | EXTENT_CLEAR_UNLOCK_PAGE | | |
724 | EXTENT_CLEAR_UNLOCK | | |
725 | EXTENT_CLEAR_DELALLOC | | |
32c00aff | 726 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e CM |
727 | EXTENT_CLEAR_DIRTY | |
728 | EXTENT_SET_WRITEBACK | | |
729 | EXTENT_END_WRITEBACK); | |
771ed689 CM |
730 | *nr_written = *nr_written + |
731 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
732 | *page_started = 1; | |
733 | ret = 0; | |
734 | goto out; | |
735 | } | |
736 | } | |
737 | ||
738 | BUG_ON(disk_num_bytes > | |
739 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
740 | ||
b917b7c3 CM |
741 | |
742 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
743 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
744 | start, num_bytes); | |
745 | if (em) { | |
746 | alloc_hint = em->block_start; | |
747 | free_extent_map(em); | |
748 | } | |
749 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
750 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
751 | ||
d397712b | 752 | while (disk_num_bytes > 0) { |
a791e35e CM |
753 | unsigned long op; |
754 | ||
c8b97818 | 755 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 756 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 757 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 758 | (u64)-1, &ins, 1); |
d397712b CM |
759 | BUG_ON(ret); |
760 | ||
e6dcd2dc CM |
761 | em = alloc_extent_map(GFP_NOFS); |
762 | em->start = start; | |
445a6944 | 763 | em->orig_start = em->start; |
771ed689 CM |
764 | ram_size = ins.offset; |
765 | em->len = ins.offset; | |
c8b97818 | 766 | |
e6dcd2dc | 767 | em->block_start = ins.objectid; |
c8b97818 | 768 | em->block_len = ins.offset; |
e6dcd2dc | 769 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 770 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 771 | |
d397712b | 772 | while (1) { |
890871be | 773 | write_lock(&em_tree->lock); |
e6dcd2dc | 774 | ret = add_extent_mapping(em_tree, em); |
890871be | 775 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
776 | if (ret != -EEXIST) { |
777 | free_extent_map(em); | |
778 | break; | |
779 | } | |
780 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 781 | start + ram_size - 1, 0); |
e6dcd2dc CM |
782 | } |
783 | ||
98d20f67 | 784 | cur_alloc_size = ins.offset; |
e6dcd2dc | 785 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 786 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 787 | BUG_ON(ret); |
c8b97818 | 788 | |
17d217fe YZ |
789 | if (root->root_key.objectid == |
790 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
791 | ret = btrfs_reloc_clone_csums(inode, start, | |
792 | cur_alloc_size); | |
793 | BUG_ON(ret); | |
794 | } | |
795 | ||
d397712b | 796 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 797 | break; |
d397712b | 798 | |
c8b97818 CM |
799 | /* we're not doing compressed IO, don't unlock the first |
800 | * page (which the caller expects to stay locked), don't | |
801 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
802 | * |
803 | * Do set the Private2 bit so we know this page was properly | |
804 | * setup for writepage | |
c8b97818 | 805 | */ |
a791e35e CM |
806 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
807 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
808 | EXTENT_SET_PRIVATE2; | |
809 | ||
c8b97818 CM |
810 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
811 | start, start + ram_size - 1, | |
a791e35e | 812 | locked_page, op); |
c8b97818 | 813 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
814 | num_bytes -= cur_alloc_size; |
815 | alloc_hint = ins.objectid + ins.offset; | |
816 | start += cur_alloc_size; | |
b888db2b | 817 | } |
b888db2b | 818 | out: |
771ed689 | 819 | ret = 0; |
b888db2b | 820 | btrfs_end_transaction(trans, root); |
c8b97818 | 821 | |
be20aa9d | 822 | return ret; |
771ed689 | 823 | } |
c8b97818 | 824 | |
771ed689 CM |
825 | /* |
826 | * work queue call back to started compression on a file and pages | |
827 | */ | |
828 | static noinline void async_cow_start(struct btrfs_work *work) | |
829 | { | |
830 | struct async_cow *async_cow; | |
831 | int num_added = 0; | |
832 | async_cow = container_of(work, struct async_cow, work); | |
833 | ||
834 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
835 | async_cow->start, async_cow->end, async_cow, | |
836 | &num_added); | |
837 | if (num_added == 0) | |
838 | async_cow->inode = NULL; | |
839 | } | |
840 | ||
841 | /* | |
842 | * work queue call back to submit previously compressed pages | |
843 | */ | |
844 | static noinline void async_cow_submit(struct btrfs_work *work) | |
845 | { | |
846 | struct async_cow *async_cow; | |
847 | struct btrfs_root *root; | |
848 | unsigned long nr_pages; | |
849 | ||
850 | async_cow = container_of(work, struct async_cow, work); | |
851 | ||
852 | root = async_cow->root; | |
853 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
854 | PAGE_CACHE_SHIFT; | |
855 | ||
856 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
857 | ||
858 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
859 | 5 * 1042 * 1024 && | |
860 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
861 | wake_up(&root->fs_info->async_submit_wait); | |
862 | ||
d397712b | 863 | if (async_cow->inode) |
771ed689 | 864 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 865 | } |
c8b97818 | 866 | |
771ed689 CM |
867 | static noinline void async_cow_free(struct btrfs_work *work) |
868 | { | |
869 | struct async_cow *async_cow; | |
870 | async_cow = container_of(work, struct async_cow, work); | |
871 | kfree(async_cow); | |
872 | } | |
873 | ||
874 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
875 | u64 start, u64 end, int *page_started, | |
876 | unsigned long *nr_written) | |
877 | { | |
878 | struct async_cow *async_cow; | |
879 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
880 | unsigned long nr_pages; | |
881 | u64 cur_end; | |
882 | int limit = 10 * 1024 * 1042; | |
883 | ||
a3429ab7 CM |
884 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
885 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 886 | while (start < end) { |
771ed689 CM |
887 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
888 | async_cow->inode = inode; | |
889 | async_cow->root = root; | |
890 | async_cow->locked_page = locked_page; | |
891 | async_cow->start = start; | |
892 | ||
6cbff00f | 893 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
894 | cur_end = end; |
895 | else | |
896 | cur_end = min(end, start + 512 * 1024 - 1); | |
897 | ||
898 | async_cow->end = cur_end; | |
899 | INIT_LIST_HEAD(&async_cow->extents); | |
900 | ||
901 | async_cow->work.func = async_cow_start; | |
902 | async_cow->work.ordered_func = async_cow_submit; | |
903 | async_cow->work.ordered_free = async_cow_free; | |
904 | async_cow->work.flags = 0; | |
905 | ||
771ed689 CM |
906 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
907 | PAGE_CACHE_SHIFT; | |
908 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
909 | ||
910 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
911 | &async_cow->work); | |
912 | ||
913 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
914 | wait_event(root->fs_info->async_submit_wait, | |
915 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
916 | limit)); | |
917 | } | |
918 | ||
d397712b | 919 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
920 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
921 | wait_event(root->fs_info->async_submit_wait, | |
922 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
923 | 0)); | |
924 | } | |
925 | ||
926 | *nr_written += nr_pages; | |
927 | start = cur_end + 1; | |
928 | } | |
929 | *page_started = 1; | |
930 | return 0; | |
be20aa9d CM |
931 | } |
932 | ||
d397712b | 933 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
934 | u64 bytenr, u64 num_bytes) |
935 | { | |
936 | int ret; | |
937 | struct btrfs_ordered_sum *sums; | |
938 | LIST_HEAD(list); | |
939 | ||
07d400a6 YZ |
940 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
941 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
942 | if (ret == 0 && list_empty(&list)) |
943 | return 0; | |
944 | ||
945 | while (!list_empty(&list)) { | |
946 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
947 | list_del(&sums->list); | |
948 | kfree(sums); | |
949 | } | |
950 | return 1; | |
951 | } | |
952 | ||
d352ac68 CM |
953 | /* |
954 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
955 | * of the extents that exist in the file, and COWs the file as required. | |
956 | * | |
957 | * If no cow copies or snapshots exist, we write directly to the existing | |
958 | * blocks on disk | |
959 | */ | |
7f366cfe CM |
960 | static noinline int run_delalloc_nocow(struct inode *inode, |
961 | struct page *locked_page, | |
771ed689 CM |
962 | u64 start, u64 end, int *page_started, int force, |
963 | unsigned long *nr_written) | |
be20aa9d | 964 | { |
be20aa9d | 965 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 966 | struct btrfs_trans_handle *trans; |
be20aa9d | 967 | struct extent_buffer *leaf; |
be20aa9d | 968 | struct btrfs_path *path; |
80ff3856 | 969 | struct btrfs_file_extent_item *fi; |
be20aa9d | 970 | struct btrfs_key found_key; |
80ff3856 YZ |
971 | u64 cow_start; |
972 | u64 cur_offset; | |
973 | u64 extent_end; | |
5d4f98a2 | 974 | u64 extent_offset; |
80ff3856 YZ |
975 | u64 disk_bytenr; |
976 | u64 num_bytes; | |
977 | int extent_type; | |
978 | int ret; | |
d899e052 | 979 | int type; |
80ff3856 YZ |
980 | int nocow; |
981 | int check_prev = 1; | |
be20aa9d CM |
982 | |
983 | path = btrfs_alloc_path(); | |
984 | BUG_ON(!path); | |
7ea394f1 YZ |
985 | trans = btrfs_join_transaction(root, 1); |
986 | BUG_ON(!trans); | |
be20aa9d | 987 | |
80ff3856 YZ |
988 | cow_start = (u64)-1; |
989 | cur_offset = start; | |
990 | while (1) { | |
991 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
992 | cur_offset, 0); | |
993 | BUG_ON(ret < 0); | |
994 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
995 | leaf = path->nodes[0]; | |
996 | btrfs_item_key_to_cpu(leaf, &found_key, | |
997 | path->slots[0] - 1); | |
998 | if (found_key.objectid == inode->i_ino && | |
999 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1000 | path->slots[0]--; | |
1001 | } | |
1002 | check_prev = 0; | |
1003 | next_slot: | |
1004 | leaf = path->nodes[0]; | |
1005 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1006 | ret = btrfs_next_leaf(root, path); | |
1007 | if (ret < 0) | |
1008 | BUG_ON(1); | |
1009 | if (ret > 0) | |
1010 | break; | |
1011 | leaf = path->nodes[0]; | |
1012 | } | |
be20aa9d | 1013 | |
80ff3856 YZ |
1014 | nocow = 0; |
1015 | disk_bytenr = 0; | |
17d217fe | 1016 | num_bytes = 0; |
80ff3856 YZ |
1017 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1018 | ||
1019 | if (found_key.objectid > inode->i_ino || | |
1020 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1021 | found_key.offset > end) | |
1022 | break; | |
1023 | ||
1024 | if (found_key.offset > cur_offset) { | |
1025 | extent_end = found_key.offset; | |
e9061e21 | 1026 | extent_type = 0; |
80ff3856 YZ |
1027 | goto out_check; |
1028 | } | |
1029 | ||
1030 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1031 | struct btrfs_file_extent_item); | |
1032 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1033 | ||
d899e052 YZ |
1034 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1035 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1036 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1037 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1038 | extent_end = found_key.offset + |
1039 | btrfs_file_extent_num_bytes(leaf, fi); | |
1040 | if (extent_end <= start) { | |
1041 | path->slots[0]++; | |
1042 | goto next_slot; | |
1043 | } | |
17d217fe YZ |
1044 | if (disk_bytenr == 0) |
1045 | goto out_check; | |
80ff3856 YZ |
1046 | if (btrfs_file_extent_compression(leaf, fi) || |
1047 | btrfs_file_extent_encryption(leaf, fi) || | |
1048 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1049 | goto out_check; | |
d899e052 YZ |
1050 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1051 | goto out_check; | |
d2fb3437 | 1052 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1053 | goto out_check; |
17d217fe | 1054 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1055 | found_key.offset - |
1056 | extent_offset, disk_bytenr)) | |
17d217fe | 1057 | goto out_check; |
5d4f98a2 | 1058 | disk_bytenr += extent_offset; |
17d217fe YZ |
1059 | disk_bytenr += cur_offset - found_key.offset; |
1060 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1061 | /* | |
1062 | * force cow if csum exists in the range. | |
1063 | * this ensure that csum for a given extent are | |
1064 | * either valid or do not exist. | |
1065 | */ | |
1066 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1067 | goto out_check; | |
80ff3856 YZ |
1068 | nocow = 1; |
1069 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1070 | extent_end = found_key.offset + | |
1071 | btrfs_file_extent_inline_len(leaf, fi); | |
1072 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1073 | } else { | |
1074 | BUG_ON(1); | |
1075 | } | |
1076 | out_check: | |
1077 | if (extent_end <= start) { | |
1078 | path->slots[0]++; | |
1079 | goto next_slot; | |
1080 | } | |
1081 | if (!nocow) { | |
1082 | if (cow_start == (u64)-1) | |
1083 | cow_start = cur_offset; | |
1084 | cur_offset = extent_end; | |
1085 | if (cur_offset > end) | |
1086 | break; | |
1087 | path->slots[0]++; | |
1088 | goto next_slot; | |
7ea394f1 YZ |
1089 | } |
1090 | ||
1091 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1092 | if (cow_start != (u64)-1) { |
1093 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1094 | found_key.offset - 1, page_started, |
1095 | nr_written, 1); | |
80ff3856 YZ |
1096 | BUG_ON(ret); |
1097 | cow_start = (u64)-1; | |
7ea394f1 | 1098 | } |
80ff3856 | 1099 | |
d899e052 YZ |
1100 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1101 | struct extent_map *em; | |
1102 | struct extent_map_tree *em_tree; | |
1103 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1104 | em = alloc_extent_map(GFP_NOFS); | |
1105 | em->start = cur_offset; | |
445a6944 | 1106 | em->orig_start = em->start; |
d899e052 YZ |
1107 | em->len = num_bytes; |
1108 | em->block_len = num_bytes; | |
1109 | em->block_start = disk_bytenr; | |
1110 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1111 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1112 | while (1) { | |
890871be | 1113 | write_lock(&em_tree->lock); |
d899e052 | 1114 | ret = add_extent_mapping(em_tree, em); |
890871be | 1115 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1116 | if (ret != -EEXIST) { |
1117 | free_extent_map(em); | |
1118 | break; | |
1119 | } | |
1120 | btrfs_drop_extent_cache(inode, em->start, | |
1121 | em->start + em->len - 1, 0); | |
1122 | } | |
1123 | type = BTRFS_ORDERED_PREALLOC; | |
1124 | } else { | |
1125 | type = BTRFS_ORDERED_NOCOW; | |
1126 | } | |
80ff3856 YZ |
1127 | |
1128 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1129 | num_bytes, num_bytes, type); |
1130 | BUG_ON(ret); | |
771ed689 | 1131 | |
d899e052 | 1132 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1133 | cur_offset, cur_offset + num_bytes - 1, |
1134 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1135 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1136 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1137 | cur_offset = extent_end; |
1138 | if (cur_offset > end) | |
1139 | break; | |
be20aa9d | 1140 | } |
80ff3856 YZ |
1141 | btrfs_release_path(root, path); |
1142 | ||
1143 | if (cur_offset <= end && cow_start == (u64)-1) | |
1144 | cow_start = cur_offset; | |
1145 | if (cow_start != (u64)-1) { | |
1146 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1147 | page_started, nr_written, 1); |
80ff3856 YZ |
1148 | BUG_ON(ret); |
1149 | } | |
1150 | ||
1151 | ret = btrfs_end_transaction(trans, root); | |
1152 | BUG_ON(ret); | |
7ea394f1 | 1153 | btrfs_free_path(path); |
80ff3856 | 1154 | return 0; |
be20aa9d CM |
1155 | } |
1156 | ||
d352ac68 CM |
1157 | /* |
1158 | * extent_io.c call back to do delayed allocation processing | |
1159 | */ | |
c8b97818 | 1160 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1161 | u64 start, u64 end, int *page_started, |
1162 | unsigned long *nr_written) | |
be20aa9d | 1163 | { |
be20aa9d | 1164 | int ret; |
7f366cfe | 1165 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1166 | |
6cbff00f | 1167 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1168 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1169 | page_started, 1, nr_written); |
6cbff00f | 1170 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1171 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1172 | page_started, 0, nr_written); |
7f366cfe CM |
1173 | else if (!btrfs_test_opt(root, COMPRESS)) |
1174 | ret = cow_file_range(inode, locked_page, start, end, | |
1175 | page_started, nr_written, 1); | |
be20aa9d | 1176 | else |
771ed689 | 1177 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1178 | page_started, nr_written); |
b888db2b CM |
1179 | return ret; |
1180 | } | |
1181 | ||
9ed74f2d JB |
1182 | static int btrfs_split_extent_hook(struct inode *inode, |
1183 | struct extent_state *orig, u64 split) | |
1184 | { | |
1185 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1186 | u64 size; | |
1187 | ||
1188 | if (!(orig->state & EXTENT_DELALLOC)) | |
1189 | return 0; | |
1190 | ||
1191 | size = orig->end - orig->start + 1; | |
1192 | if (size > root->fs_info->max_extent) { | |
1193 | u64 num_extents; | |
1194 | u64 new_size; | |
1195 | ||
1196 | new_size = orig->end - split + 1; | |
1197 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1198 | root->fs_info->max_extent); | |
1199 | ||
1200 | /* | |
32c00aff JB |
1201 | * if we break a large extent up then leave oustanding_extents |
1202 | * be, since we've already accounted for the large extent. | |
9ed74f2d JB |
1203 | */ |
1204 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1205 | root->fs_info->max_extent) < num_extents) | |
1206 | return 0; | |
1207 | } | |
1208 | ||
32c00aff JB |
1209 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1210 | BTRFS_I(inode)->outstanding_extents++; | |
1211 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1212 | |
1213 | return 0; | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1218 | * extents so we can keep track of new extents that are just merged onto old | |
1219 | * extents, such as when we are doing sequential writes, so we can properly | |
1220 | * account for the metadata space we'll need. | |
1221 | */ | |
1222 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1223 | struct extent_state *new, | |
1224 | struct extent_state *other) | |
1225 | { | |
1226 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1227 | u64 new_size, old_size; | |
1228 | u64 num_extents; | |
1229 | ||
1230 | /* not delalloc, ignore it */ | |
1231 | if (!(other->state & EXTENT_DELALLOC)) | |
1232 | return 0; | |
1233 | ||
1234 | old_size = other->end - other->start + 1; | |
1235 | if (new->start < other->start) | |
1236 | new_size = other->end - new->start + 1; | |
1237 | else | |
1238 | new_size = new->end - other->start + 1; | |
1239 | ||
1240 | /* we're not bigger than the max, unreserve the space and go */ | |
1241 | if (new_size <= root->fs_info->max_extent) { | |
32c00aff JB |
1242 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1243 | BTRFS_I(inode)->outstanding_extents--; | |
1244 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1245 | return 0; |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * If we grew by another max_extent, just return, we want to keep that | |
1250 | * reserved amount. | |
1251 | */ | |
1252 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1253 | root->fs_info->max_extent); | |
1254 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1255 | root->fs_info->max_extent) > num_extents) | |
1256 | return 0; | |
1257 | ||
32c00aff JB |
1258 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1259 | BTRFS_I(inode)->outstanding_extents--; | |
1260 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1261 | |
1262 | return 0; | |
1263 | } | |
1264 | ||
d352ac68 CM |
1265 | /* |
1266 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1267 | * bytes in this file, and to maintain the list of inodes that | |
1268 | * have pending delalloc work to be done. | |
1269 | */ | |
b2950863 | 1270 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1271 | unsigned long old, unsigned long bits) |
291d673e | 1272 | { |
9ed74f2d | 1273 | |
75eff68e CM |
1274 | /* |
1275 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1276 | * but in this case, we are only testeing for the DELALLOC | |
1277 | * bit, which is only set or cleared with irqs on | |
1278 | */ | |
b0c68f8b | 1279 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1280 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d | 1281 | |
32c00aff JB |
1282 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1283 | BTRFS_I(inode)->outstanding_extents++; | |
1284 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
6a63209f | 1285 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1286 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1287 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1288 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1289 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1290 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1291 | &root->fs_info->delalloc_inodes); | |
1292 | } | |
75eff68e | 1293 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1294 | } |
1295 | return 0; | |
1296 | } | |
1297 | ||
d352ac68 CM |
1298 | /* |
1299 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1300 | */ | |
9ed74f2d JB |
1301 | static int btrfs_clear_bit_hook(struct inode *inode, |
1302 | struct extent_state *state, unsigned long bits) | |
291d673e | 1303 | { |
75eff68e CM |
1304 | /* |
1305 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1306 | * but in this case, we are only testeing for the DELALLOC | |
1307 | * bit, which is only set or cleared with irqs on | |
1308 | */ | |
9ed74f2d | 1309 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1310 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1311 | |
32c00aff JB |
1312 | if (bits & EXTENT_DO_ACCOUNTING) { |
1313 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
1314 | BTRFS_I(inode)->outstanding_extents--; | |
1315 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
1316 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1317 | } | |
9ed74f2d | 1318 | |
75eff68e | 1319 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1320 | if (state->end - state->start + 1 > |
1321 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1322 | printk(KERN_INFO "btrfs warning: delalloc account " |
1323 | "%llu %llu\n", | |
9ed74f2d JB |
1324 | (unsigned long long) |
1325 | state->end - state->start + 1, | |
d397712b CM |
1326 | (unsigned long long) |
1327 | root->fs_info->delalloc_bytes); | |
6a63209f | 1328 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1329 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1330 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1331 | } else { |
6a63209f | 1332 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1333 | state->end - |
1334 | state->start + 1); | |
1335 | root->fs_info->delalloc_bytes -= state->end - | |
1336 | state->start + 1; | |
1337 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1338 | state->start + 1; | |
b0c68f8b | 1339 | } |
ea8c2819 CM |
1340 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1341 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1342 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1343 | } | |
75eff68e | 1344 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1345 | } |
1346 | return 0; | |
1347 | } | |
1348 | ||
d352ac68 CM |
1349 | /* |
1350 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1351 | * we don't create bios that span stripes or chunks | |
1352 | */ | |
239b14b3 | 1353 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1354 | size_t size, struct bio *bio, |
1355 | unsigned long bio_flags) | |
239b14b3 CM |
1356 | { |
1357 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1358 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1359 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1360 | u64 length = 0; |
1361 | u64 map_length; | |
239b14b3 CM |
1362 | int ret; |
1363 | ||
771ed689 CM |
1364 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1365 | return 0; | |
1366 | ||
f2d8d74d | 1367 | length = bio->bi_size; |
239b14b3 CM |
1368 | map_tree = &root->fs_info->mapping_tree; |
1369 | map_length = length; | |
cea9e445 | 1370 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1371 | &map_length, NULL, 0); |
cea9e445 | 1372 | |
d397712b | 1373 | if (map_length < length + size) |
239b14b3 | 1374 | return 1; |
239b14b3 CM |
1375 | return 0; |
1376 | } | |
1377 | ||
d352ac68 CM |
1378 | /* |
1379 | * in order to insert checksums into the metadata in large chunks, | |
1380 | * we wait until bio submission time. All the pages in the bio are | |
1381 | * checksummed and sums are attached onto the ordered extent record. | |
1382 | * | |
1383 | * At IO completion time the cums attached on the ordered extent record | |
1384 | * are inserted into the btree | |
1385 | */ | |
d397712b CM |
1386 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1387 | struct bio *bio, int mirror_num, | |
1388 | unsigned long bio_flags) | |
065631f6 | 1389 | { |
065631f6 | 1390 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1391 | int ret = 0; |
e015640f | 1392 | |
d20f7043 | 1393 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1394 | BUG_ON(ret); |
4a69a410 CM |
1395 | return 0; |
1396 | } | |
e015640f | 1397 | |
4a69a410 CM |
1398 | /* |
1399 | * in order to insert checksums into the metadata in large chunks, | |
1400 | * we wait until bio submission time. All the pages in the bio are | |
1401 | * checksummed and sums are attached onto the ordered extent record. | |
1402 | * | |
1403 | * At IO completion time the cums attached on the ordered extent record | |
1404 | * are inserted into the btree | |
1405 | */ | |
b2950863 | 1406 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1407 | int mirror_num, unsigned long bio_flags) |
1408 | { | |
1409 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1410 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1411 | } |
1412 | ||
d352ac68 | 1413 | /* |
cad321ad CM |
1414 | * extent_io.c submission hook. This does the right thing for csum calculation |
1415 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1416 | */ |
b2950863 | 1417 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1418 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1419 | { |
1420 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1421 | int ret = 0; | |
19b9bdb0 | 1422 | int skip_sum; |
44b8bd7e | 1423 | |
6cbff00f | 1424 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1425 | |
e6dcd2dc CM |
1426 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1427 | BUG_ON(ret); | |
065631f6 | 1428 | |
4d1b5fb4 | 1429 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1430 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1431 | return btrfs_submit_compressed_read(inode, bio, |
1432 | mirror_num, bio_flags); | |
d20f7043 CM |
1433 | } else if (!skip_sum) |
1434 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1435 | goto mapit; |
19b9bdb0 | 1436 | } else if (!skip_sum) { |
17d217fe YZ |
1437 | /* csum items have already been cloned */ |
1438 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1439 | goto mapit; | |
19b9bdb0 CM |
1440 | /* we're doing a write, do the async checksumming */ |
1441 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1442 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1443 | bio_flags, __btrfs_submit_bio_start, |
1444 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1445 | } |
1446 | ||
0b86a832 | 1447 | mapit: |
8b712842 | 1448 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1449 | } |
6885f308 | 1450 | |
d352ac68 CM |
1451 | /* |
1452 | * given a list of ordered sums record them in the inode. This happens | |
1453 | * at IO completion time based on sums calculated at bio submission time. | |
1454 | */ | |
ba1da2f4 | 1455 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1456 | struct inode *inode, u64 file_offset, |
1457 | struct list_head *list) | |
1458 | { | |
e6dcd2dc CM |
1459 | struct btrfs_ordered_sum *sum; |
1460 | ||
1461 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1462 | |
1463 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1464 | btrfs_csum_file_blocks(trans, |
1465 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1466 | } |
1467 | return 0; | |
1468 | } | |
1469 | ||
ea8c2819 CM |
1470 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1471 | { | |
d397712b | 1472 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1473 | WARN_ON(1); |
ea8c2819 CM |
1474 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1475 | GFP_NOFS); | |
1476 | } | |
1477 | ||
d352ac68 | 1478 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1479 | struct btrfs_writepage_fixup { |
1480 | struct page *page; | |
1481 | struct btrfs_work work; | |
1482 | }; | |
1483 | ||
b2950863 | 1484 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1485 | { |
1486 | struct btrfs_writepage_fixup *fixup; | |
1487 | struct btrfs_ordered_extent *ordered; | |
1488 | struct page *page; | |
1489 | struct inode *inode; | |
1490 | u64 page_start; | |
1491 | u64 page_end; | |
1492 | ||
1493 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1494 | page = fixup->page; | |
4a096752 | 1495 | again: |
247e743c CM |
1496 | lock_page(page); |
1497 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1498 | ClearPageChecked(page); | |
1499 | goto out_page; | |
1500 | } | |
1501 | ||
1502 | inode = page->mapping->host; | |
1503 | page_start = page_offset(page); | |
1504 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1505 | ||
1506 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1507 | |
1508 | /* already ordered? We're done */ | |
8b62b72b | 1509 | if (PagePrivate2(page)) |
247e743c | 1510 | goto out; |
4a096752 CM |
1511 | |
1512 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1513 | if (ordered) { | |
1514 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1515 | page_end, GFP_NOFS); | |
1516 | unlock_page(page); | |
1517 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1518 | goto again; | |
1519 | } | |
247e743c | 1520 | |
ea8c2819 | 1521 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1522 | ClearPageChecked(page); |
1523 | out: | |
1524 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1525 | out_page: | |
1526 | unlock_page(page); | |
1527 | page_cache_release(page); | |
1528 | } | |
1529 | ||
1530 | /* | |
1531 | * There are a few paths in the higher layers of the kernel that directly | |
1532 | * set the page dirty bit without asking the filesystem if it is a | |
1533 | * good idea. This causes problems because we want to make sure COW | |
1534 | * properly happens and the data=ordered rules are followed. | |
1535 | * | |
c8b97818 | 1536 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1537 | * hasn't been properly setup for IO. We kick off an async process |
1538 | * to fix it up. The async helper will wait for ordered extents, set | |
1539 | * the delalloc bit and make it safe to write the page. | |
1540 | */ | |
b2950863 | 1541 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1542 | { |
1543 | struct inode *inode = page->mapping->host; | |
1544 | struct btrfs_writepage_fixup *fixup; | |
1545 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1546 | |
8b62b72b CM |
1547 | /* this page is properly in the ordered list */ |
1548 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1549 | return 0; |
1550 | ||
1551 | if (PageChecked(page)) | |
1552 | return -EAGAIN; | |
1553 | ||
1554 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1555 | if (!fixup) | |
1556 | return -EAGAIN; | |
f421950f | 1557 | |
247e743c CM |
1558 | SetPageChecked(page); |
1559 | page_cache_get(page); | |
1560 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1561 | fixup->page = page; | |
1562 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1563 | return -EAGAIN; | |
1564 | } | |
1565 | ||
d899e052 YZ |
1566 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1567 | struct inode *inode, u64 file_pos, | |
1568 | u64 disk_bytenr, u64 disk_num_bytes, | |
1569 | u64 num_bytes, u64 ram_bytes, | |
e980b50c | 1570 | u64 locked_end, |
d899e052 YZ |
1571 | u8 compression, u8 encryption, |
1572 | u16 other_encoding, int extent_type) | |
1573 | { | |
1574 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1575 | struct btrfs_file_extent_item *fi; | |
1576 | struct btrfs_path *path; | |
1577 | struct extent_buffer *leaf; | |
1578 | struct btrfs_key ins; | |
1579 | u64 hint; | |
1580 | int ret; | |
1581 | ||
1582 | path = btrfs_alloc_path(); | |
1583 | BUG_ON(!path); | |
1584 | ||
b9473439 | 1585 | path->leave_spinning = 1; |
a1ed835e CM |
1586 | |
1587 | /* | |
1588 | * we may be replacing one extent in the tree with another. | |
1589 | * The new extent is pinned in the extent map, and we don't want | |
1590 | * to drop it from the cache until it is completely in the btree. | |
1591 | * | |
1592 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1593 | * the caller is expected to unpin it and allow it to be merged | |
1594 | * with the others. | |
1595 | */ | |
d899e052 | 1596 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
e980b50c | 1597 | file_pos + num_bytes, locked_end, |
a1ed835e | 1598 | file_pos, &hint, 0); |
d899e052 YZ |
1599 | BUG_ON(ret); |
1600 | ||
1601 | ins.objectid = inode->i_ino; | |
1602 | ins.offset = file_pos; | |
1603 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1604 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1605 | BUG_ON(ret); | |
1606 | leaf = path->nodes[0]; | |
1607 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1608 | struct btrfs_file_extent_item); | |
1609 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1610 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1611 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1612 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1613 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1614 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1615 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1616 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1617 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1618 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1619 | |
1620 | btrfs_unlock_up_safe(path, 1); | |
1621 | btrfs_set_lock_blocking(leaf); | |
1622 | ||
d899e052 YZ |
1623 | btrfs_mark_buffer_dirty(leaf); |
1624 | ||
1625 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1626 | |
1627 | ins.objectid = disk_bytenr; | |
1628 | ins.offset = disk_num_bytes; | |
1629 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1630 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1631 | root->root_key.objectid, | |
1632 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1633 | BUG_ON(ret); |
d899e052 | 1634 | btrfs_free_path(path); |
b9473439 | 1635 | |
d899e052 YZ |
1636 | return 0; |
1637 | } | |
1638 | ||
5d13a98f CM |
1639 | /* |
1640 | * helper function for btrfs_finish_ordered_io, this | |
1641 | * just reads in some of the csum leaves to prime them into ram | |
1642 | * before we start the transaction. It limits the amount of btree | |
1643 | * reads required while inside the transaction. | |
1644 | */ | |
1645 | static noinline void reada_csum(struct btrfs_root *root, | |
1646 | struct btrfs_path *path, | |
1647 | struct btrfs_ordered_extent *ordered_extent) | |
1648 | { | |
1649 | struct btrfs_ordered_sum *sum; | |
1650 | u64 bytenr; | |
1651 | ||
1652 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1653 | list); | |
1654 | bytenr = sum->sums[0].bytenr; | |
1655 | ||
1656 | /* | |
1657 | * we don't care about the results, the point of this search is | |
1658 | * just to get the btree leaves into ram | |
1659 | */ | |
1660 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1661 | } | |
1662 | ||
d352ac68 CM |
1663 | /* as ordered data IO finishes, this gets called so we can finish |
1664 | * an ordered extent if the range of bytes in the file it covers are | |
1665 | * fully written. | |
1666 | */ | |
211f90e6 | 1667 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1668 | { |
e6dcd2dc CM |
1669 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1670 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1671 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1672 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1673 | struct btrfs_path *path; |
d899e052 | 1674 | int compressed = 0; |
e6dcd2dc CM |
1675 | int ret; |
1676 | ||
1677 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1678 | if (!ret) |
e6dcd2dc | 1679 | return 0; |
e6dcd2dc | 1680 | |
b7ec40d7 CM |
1681 | /* |
1682 | * before we join the transaction, try to do some of our IO. | |
1683 | * This will limit the amount of IO that we have to do with | |
1684 | * the transaction running. We're unlikely to need to do any | |
1685 | * IO if the file extents are new, the disk_i_size checks | |
1686 | * covers the most common case. | |
1687 | */ | |
1688 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1689 | path = btrfs_alloc_path(); | |
1690 | if (path) { | |
1691 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1692 | inode->i_ino, | |
1693 | start, 0); | |
5d13a98f CM |
1694 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1695 | start); | |
1696 | if (!list_empty(&ordered_extent->list)) { | |
1697 | btrfs_release_path(root, path); | |
1698 | reada_csum(root, path, ordered_extent); | |
1699 | } | |
b7ec40d7 CM |
1700 | btrfs_free_path(path); |
1701 | } | |
1702 | } | |
1703 | ||
f9295749 | 1704 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc | 1705 | |
5d13a98f CM |
1706 | if (!ordered_extent) |
1707 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1708 | BUG_ON(!ordered_extent); |
7ea394f1 YZ |
1709 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1710 | goto nocow; | |
e6dcd2dc CM |
1711 | |
1712 | lock_extent(io_tree, ordered_extent->file_offset, | |
1713 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1714 | GFP_NOFS); | |
1715 | ||
c8b97818 | 1716 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1717 | compressed = 1; |
1718 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1719 | BUG_ON(compressed); | |
1720 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1721 | ordered_extent->file_offset, | |
1722 | ordered_extent->file_offset + | |
1723 | ordered_extent->len); | |
1724 | BUG_ON(ret); | |
1725 | } else { | |
1726 | ret = insert_reserved_file_extent(trans, inode, | |
1727 | ordered_extent->file_offset, | |
1728 | ordered_extent->start, | |
1729 | ordered_extent->disk_len, | |
1730 | ordered_extent->len, | |
1731 | ordered_extent->len, | |
e980b50c CM |
1732 | ordered_extent->file_offset + |
1733 | ordered_extent->len, | |
d899e052 YZ |
1734 | compressed, 0, 0, |
1735 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1736 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1737 | ordered_extent->file_offset, | |
1738 | ordered_extent->len); | |
d899e052 YZ |
1739 | BUG_ON(ret); |
1740 | } | |
e6dcd2dc CM |
1741 | unlock_extent(io_tree, ordered_extent->file_offset, |
1742 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1743 | GFP_NOFS); | |
7ea394f1 | 1744 | nocow: |
e6dcd2dc CM |
1745 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1746 | &ordered_extent->list); | |
1747 | ||
34353029 | 1748 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1749 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1750 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1751 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1752 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1753 | |
e6dcd2dc CM |
1754 | /* once for us */ |
1755 | btrfs_put_ordered_extent(ordered_extent); | |
1756 | /* once for the tree */ | |
1757 | btrfs_put_ordered_extent(ordered_extent); | |
1758 | ||
e6dcd2dc CM |
1759 | btrfs_end_transaction(trans, root); |
1760 | return 0; | |
1761 | } | |
1762 | ||
b2950863 | 1763 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1764 | struct extent_state *state, int uptodate) |
1765 | { | |
8b62b72b | 1766 | ClearPagePrivate2(page); |
211f90e6 CM |
1767 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1768 | } | |
1769 | ||
d352ac68 CM |
1770 | /* |
1771 | * When IO fails, either with EIO or csum verification fails, we | |
1772 | * try other mirrors that might have a good copy of the data. This | |
1773 | * io_failure_record is used to record state as we go through all the | |
1774 | * mirrors. If another mirror has good data, the page is set up to date | |
1775 | * and things continue. If a good mirror can't be found, the original | |
1776 | * bio end_io callback is called to indicate things have failed. | |
1777 | */ | |
7e38326f CM |
1778 | struct io_failure_record { |
1779 | struct page *page; | |
1780 | u64 start; | |
1781 | u64 len; | |
1782 | u64 logical; | |
d20f7043 | 1783 | unsigned long bio_flags; |
7e38326f CM |
1784 | int last_mirror; |
1785 | }; | |
1786 | ||
b2950863 | 1787 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1788 | struct page *page, u64 start, u64 end, |
1789 | struct extent_state *state) | |
7e38326f CM |
1790 | { |
1791 | struct io_failure_record *failrec = NULL; | |
1792 | u64 private; | |
1793 | struct extent_map *em; | |
1794 | struct inode *inode = page->mapping->host; | |
1795 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1796 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1797 | struct bio *bio; |
1798 | int num_copies; | |
1799 | int ret; | |
1259ab75 | 1800 | int rw; |
7e38326f CM |
1801 | u64 logical; |
1802 | ||
1803 | ret = get_state_private(failure_tree, start, &private); | |
1804 | if (ret) { | |
7e38326f CM |
1805 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1806 | if (!failrec) | |
1807 | return -ENOMEM; | |
1808 | failrec->start = start; | |
1809 | failrec->len = end - start + 1; | |
1810 | failrec->last_mirror = 0; | |
d20f7043 | 1811 | failrec->bio_flags = 0; |
7e38326f | 1812 | |
890871be | 1813 | read_lock(&em_tree->lock); |
3b951516 CM |
1814 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1815 | if (em->start > start || em->start + em->len < start) { | |
1816 | free_extent_map(em); | |
1817 | em = NULL; | |
1818 | } | |
890871be | 1819 | read_unlock(&em_tree->lock); |
7e38326f CM |
1820 | |
1821 | if (!em || IS_ERR(em)) { | |
1822 | kfree(failrec); | |
1823 | return -EIO; | |
1824 | } | |
1825 | logical = start - em->start; | |
1826 | logical = em->block_start + logical; | |
d20f7043 CM |
1827 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1828 | logical = em->block_start; | |
1829 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1830 | } | |
7e38326f CM |
1831 | failrec->logical = logical; |
1832 | free_extent_map(em); | |
1833 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1834 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1835 | set_state_private(failure_tree, start, |
1836 | (u64)(unsigned long)failrec); | |
7e38326f | 1837 | } else { |
587f7704 | 1838 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1839 | } |
1840 | num_copies = btrfs_num_copies( | |
1841 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1842 | failrec->logical, failrec->len); | |
1843 | failrec->last_mirror++; | |
1844 | if (!state) { | |
cad321ad | 1845 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1846 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1847 | failrec->start, | |
1848 | EXTENT_LOCKED); | |
1849 | if (state && state->start != failrec->start) | |
1850 | state = NULL; | |
cad321ad | 1851 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1852 | } |
1853 | if (!state || failrec->last_mirror > num_copies) { | |
1854 | set_state_private(failure_tree, failrec->start, 0); | |
1855 | clear_extent_bits(failure_tree, failrec->start, | |
1856 | failrec->start + failrec->len - 1, | |
1857 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1858 | kfree(failrec); | |
1859 | return -EIO; | |
1860 | } | |
1861 | bio = bio_alloc(GFP_NOFS, 1); | |
1862 | bio->bi_private = state; | |
1863 | bio->bi_end_io = failed_bio->bi_end_io; | |
1864 | bio->bi_sector = failrec->logical >> 9; | |
1865 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1866 | bio->bi_size = 0; |
d20f7043 | 1867 | |
7e38326f | 1868 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1869 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1870 | rw = WRITE; | |
1871 | else | |
1872 | rw = READ; | |
1873 | ||
1874 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1875 | failrec->last_mirror, |
d20f7043 | 1876 | failrec->bio_flags); |
1259ab75 CM |
1877 | return 0; |
1878 | } | |
1879 | ||
d352ac68 CM |
1880 | /* |
1881 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1882 | * to see if we need to process or clean up an io_failure_record | |
1883 | */ | |
b2950863 | 1884 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1885 | { |
1886 | u64 private; | |
1887 | u64 private_failure; | |
1888 | struct io_failure_record *failure; | |
1889 | int ret; | |
1890 | ||
1891 | private = 0; | |
1892 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1893 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1894 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1895 | start, &private_failure); | |
1896 | if (ret == 0) { | |
1897 | failure = (struct io_failure_record *)(unsigned long) | |
1898 | private_failure; | |
1899 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1900 | failure->start, 0); | |
1901 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1902 | failure->start, | |
1903 | failure->start + failure->len - 1, | |
1904 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1905 | GFP_NOFS); | |
1906 | kfree(failure); | |
1907 | } | |
1908 | } | |
7e38326f CM |
1909 | return 0; |
1910 | } | |
1911 | ||
d352ac68 CM |
1912 | /* |
1913 | * when reads are done, we need to check csums to verify the data is correct | |
1914 | * if there's a match, we allow the bio to finish. If not, we go through | |
1915 | * the io_failure_record routines to find good copies | |
1916 | */ | |
b2950863 | 1917 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1918 | struct extent_state *state) |
07157aac | 1919 | { |
35ebb934 | 1920 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1921 | struct inode *inode = page->mapping->host; |
d1310b2e | 1922 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1923 | char *kaddr; |
aadfeb6e | 1924 | u64 private = ~(u32)0; |
07157aac | 1925 | int ret; |
ff79f819 CM |
1926 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1927 | u32 csum = ~(u32)0; | |
d1310b2e | 1928 | |
d20f7043 CM |
1929 | if (PageChecked(page)) { |
1930 | ClearPageChecked(page); | |
1931 | goto good; | |
1932 | } | |
6cbff00f CH |
1933 | |
1934 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1935 | return 0; |
1936 | ||
1937 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1938 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1939 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1940 | GFP_NOFS); | |
b6cda9bc | 1941 | return 0; |
17d217fe | 1942 | } |
d20f7043 | 1943 | |
c2e639f0 | 1944 | if (state && state->start == start) { |
70dec807 CM |
1945 | private = state->private; |
1946 | ret = 0; | |
1947 | } else { | |
1948 | ret = get_state_private(io_tree, start, &private); | |
1949 | } | |
9ab86c8e | 1950 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1951 | if (ret) |
07157aac | 1952 | goto zeroit; |
d397712b | 1953 | |
ff79f819 CM |
1954 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1955 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1956 | if (csum != private) |
07157aac | 1957 | goto zeroit; |
d397712b | 1958 | |
9ab86c8e | 1959 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1960 | good: |
7e38326f CM |
1961 | /* if the io failure tree for this inode is non-empty, |
1962 | * check to see if we've recovered from a failed IO | |
1963 | */ | |
1259ab75 | 1964 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1965 | return 0; |
1966 | ||
1967 | zeroit: | |
193f284d CM |
1968 | if (printk_ratelimit()) { |
1969 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1970 | "private %llu\n", page->mapping->host->i_ino, | |
1971 | (unsigned long long)start, csum, | |
1972 | (unsigned long long)private); | |
1973 | } | |
db94535d CM |
1974 | memset(kaddr + offset, 1, end - start + 1); |
1975 | flush_dcache_page(page); | |
9ab86c8e | 1976 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1977 | if (private == 0) |
1978 | return 0; | |
7e38326f | 1979 | return -EIO; |
07157aac | 1980 | } |
b888db2b | 1981 | |
7b128766 JB |
1982 | /* |
1983 | * This creates an orphan entry for the given inode in case something goes | |
1984 | * wrong in the middle of an unlink/truncate. | |
1985 | */ | |
1986 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1987 | { | |
1988 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1989 | int ret = 0; | |
1990 | ||
bcc63abb | 1991 | spin_lock(&root->list_lock); |
7b128766 JB |
1992 | |
1993 | /* already on the orphan list, we're good */ | |
1994 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1995 | spin_unlock(&root->list_lock); |
7b128766 JB |
1996 | return 0; |
1997 | } | |
1998 | ||
1999 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2000 | ||
bcc63abb | 2001 | spin_unlock(&root->list_lock); |
7b128766 JB |
2002 | |
2003 | /* | |
2004 | * insert an orphan item to track this unlinked/truncated file | |
2005 | */ | |
2006 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2007 | ||
2008 | return ret; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2013 | * item for this particular inode. | |
2014 | */ | |
2015 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2016 | { | |
2017 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2018 | int ret = 0; | |
2019 | ||
bcc63abb | 2020 | spin_lock(&root->list_lock); |
7b128766 JB |
2021 | |
2022 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2023 | spin_unlock(&root->list_lock); |
7b128766 JB |
2024 | return 0; |
2025 | } | |
2026 | ||
2027 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2028 | if (!trans) { | |
bcc63abb | 2029 | spin_unlock(&root->list_lock); |
7b128766 JB |
2030 | return 0; |
2031 | } | |
2032 | ||
bcc63abb | 2033 | spin_unlock(&root->list_lock); |
7b128766 JB |
2034 | |
2035 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2036 | ||
2037 | return ret; | |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * this cleans up any orphans that may be left on the list from the last use | |
2042 | * of this root. | |
2043 | */ | |
2044 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2045 | { | |
2046 | struct btrfs_path *path; | |
2047 | struct extent_buffer *leaf; | |
2048 | struct btrfs_item *item; | |
2049 | struct btrfs_key key, found_key; | |
2050 | struct btrfs_trans_handle *trans; | |
2051 | struct inode *inode; | |
2052 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2053 | ||
7b128766 JB |
2054 | path = btrfs_alloc_path(); |
2055 | if (!path) | |
2056 | return; | |
2057 | path->reada = -1; | |
2058 | ||
2059 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2060 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2061 | key.offset = (u64)-1; | |
2062 | ||
7b128766 JB |
2063 | |
2064 | while (1) { | |
2065 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2066 | if (ret < 0) { | |
2067 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2068 | "\n", ret); | |
2069 | break; | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * if ret == 0 means we found what we were searching for, which | |
2074 | * is weird, but possible, so only screw with path if we didnt | |
2075 | * find the key and see if we have stuff that matches | |
2076 | */ | |
2077 | if (ret > 0) { | |
2078 | if (path->slots[0] == 0) | |
2079 | break; | |
2080 | path->slots[0]--; | |
2081 | } | |
2082 | ||
2083 | /* pull out the item */ | |
2084 | leaf = path->nodes[0]; | |
2085 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2086 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2087 | ||
2088 | /* make sure the item matches what we want */ | |
2089 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2090 | break; | |
2091 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2092 | break; | |
2093 | ||
2094 | /* release the path since we're done with it */ | |
2095 | btrfs_release_path(root, path); | |
2096 | ||
2097 | /* | |
2098 | * this is where we are basically btrfs_lookup, without the | |
2099 | * crossing root thing. we store the inode number in the | |
2100 | * offset of the orphan item. | |
2101 | */ | |
5d4f98a2 YZ |
2102 | found_key.objectid = found_key.offset; |
2103 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2104 | found_key.offset = 0; | |
2105 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2106 | if (IS_ERR(inode)) | |
7b128766 JB |
2107 | break; |
2108 | ||
7b128766 JB |
2109 | /* |
2110 | * add this inode to the orphan list so btrfs_orphan_del does | |
2111 | * the proper thing when we hit it | |
2112 | */ | |
bcc63abb | 2113 | spin_lock(&root->list_lock); |
7b128766 | 2114 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2115 | spin_unlock(&root->list_lock); |
7b128766 JB |
2116 | |
2117 | /* | |
2118 | * if this is a bad inode, means we actually succeeded in | |
2119 | * removing the inode, but not the orphan record, which means | |
2120 | * we need to manually delete the orphan since iput will just | |
2121 | * do a destroy_inode | |
2122 | */ | |
2123 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2124 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2125 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2126 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2127 | iput(inode); |
2128 | continue; | |
2129 | } | |
2130 | ||
2131 | /* if we have links, this was a truncate, lets do that */ | |
2132 | if (inode->i_nlink) { | |
2133 | nr_truncate++; | |
2134 | btrfs_truncate(inode); | |
2135 | } else { | |
2136 | nr_unlink++; | |
2137 | } | |
2138 | ||
2139 | /* this will do delete_inode and everything for us */ | |
2140 | iput(inode); | |
2141 | } | |
2142 | ||
2143 | if (nr_unlink) | |
2144 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2145 | if (nr_truncate) | |
2146 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2147 | ||
2148 | btrfs_free_path(path); | |
7b128766 JB |
2149 | } |
2150 | ||
46a53cca CM |
2151 | /* |
2152 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2153 | * don't find any xattrs, we know there can't be any acls. | |
2154 | * | |
2155 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2156 | */ | |
2157 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2158 | int slot, u64 objectid) | |
2159 | { | |
2160 | u32 nritems = btrfs_header_nritems(leaf); | |
2161 | struct btrfs_key found_key; | |
2162 | int scanned = 0; | |
2163 | ||
2164 | slot++; | |
2165 | while (slot < nritems) { | |
2166 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2167 | ||
2168 | /* we found a different objectid, there must not be acls */ | |
2169 | if (found_key.objectid != objectid) | |
2170 | return 0; | |
2171 | ||
2172 | /* we found an xattr, assume we've got an acl */ | |
2173 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2174 | return 1; | |
2175 | ||
2176 | /* | |
2177 | * we found a key greater than an xattr key, there can't | |
2178 | * be any acls later on | |
2179 | */ | |
2180 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2181 | return 0; | |
2182 | ||
2183 | slot++; | |
2184 | scanned++; | |
2185 | ||
2186 | /* | |
2187 | * it goes inode, inode backrefs, xattrs, extents, | |
2188 | * so if there are a ton of hard links to an inode there can | |
2189 | * be a lot of backrefs. Don't waste time searching too hard, | |
2190 | * this is just an optimization | |
2191 | */ | |
2192 | if (scanned >= 8) | |
2193 | break; | |
2194 | } | |
2195 | /* we hit the end of the leaf before we found an xattr or | |
2196 | * something larger than an xattr. We have to assume the inode | |
2197 | * has acls | |
2198 | */ | |
2199 | return 1; | |
2200 | } | |
2201 | ||
d352ac68 CM |
2202 | /* |
2203 | * read an inode from the btree into the in-memory inode | |
2204 | */ | |
5d4f98a2 | 2205 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2206 | { |
2207 | struct btrfs_path *path; | |
5f39d397 | 2208 | struct extent_buffer *leaf; |
39279cc3 | 2209 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2210 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2211 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2212 | struct btrfs_key location; | |
46a53cca | 2213 | int maybe_acls; |
39279cc3 | 2214 | u64 alloc_group_block; |
618e21d5 | 2215 | u32 rdev; |
39279cc3 CM |
2216 | int ret; |
2217 | ||
2218 | path = btrfs_alloc_path(); | |
2219 | BUG_ON(!path); | |
39279cc3 | 2220 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2221 | |
39279cc3 | 2222 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2223 | if (ret) |
39279cc3 | 2224 | goto make_bad; |
39279cc3 | 2225 | |
5f39d397 CM |
2226 | leaf = path->nodes[0]; |
2227 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2228 | struct btrfs_inode_item); | |
2229 | ||
2230 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2231 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2232 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2233 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2234 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2235 | |
2236 | tspec = btrfs_inode_atime(inode_item); | |
2237 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2238 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2239 | ||
2240 | tspec = btrfs_inode_mtime(inode_item); | |
2241 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2242 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2243 | ||
2244 | tspec = btrfs_inode_ctime(inode_item); | |
2245 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2246 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2247 | ||
a76a3cd4 | 2248 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2249 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2250 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2251 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2252 | inode->i_rdev = 0; |
5f39d397 CM |
2253 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2254 | ||
aec7477b | 2255 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2256 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2257 | |
5f39d397 | 2258 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2259 | |
46a53cca CM |
2260 | /* |
2261 | * try to precache a NULL acl entry for files that don't have | |
2262 | * any xattrs or acls | |
2263 | */ | |
2264 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2265 | if (!maybe_acls) |
2266 | cache_no_acl(inode); | |
46a53cca | 2267 | |
d2fb3437 YZ |
2268 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2269 | alloc_group_block, 0); | |
39279cc3 CM |
2270 | btrfs_free_path(path); |
2271 | inode_item = NULL; | |
2272 | ||
39279cc3 | 2273 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2274 | case S_IFREG: |
2275 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2276 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2277 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2278 | inode->i_fop = &btrfs_file_operations; |
2279 | inode->i_op = &btrfs_file_inode_operations; | |
2280 | break; | |
2281 | case S_IFDIR: | |
2282 | inode->i_fop = &btrfs_dir_file_operations; | |
2283 | if (root == root->fs_info->tree_root) | |
2284 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2285 | else | |
2286 | inode->i_op = &btrfs_dir_inode_operations; | |
2287 | break; | |
2288 | case S_IFLNK: | |
2289 | inode->i_op = &btrfs_symlink_inode_operations; | |
2290 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2291 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2292 | break; |
618e21d5 | 2293 | default: |
0279b4cd | 2294 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2295 | init_special_inode(inode, inode->i_mode, rdev); |
2296 | break; | |
39279cc3 | 2297 | } |
6cbff00f CH |
2298 | |
2299 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2300 | return; |
2301 | ||
2302 | make_bad: | |
39279cc3 | 2303 | btrfs_free_path(path); |
39279cc3 CM |
2304 | make_bad_inode(inode); |
2305 | } | |
2306 | ||
d352ac68 CM |
2307 | /* |
2308 | * given a leaf and an inode, copy the inode fields into the leaf | |
2309 | */ | |
e02119d5 CM |
2310 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2311 | struct extent_buffer *leaf, | |
5f39d397 | 2312 | struct btrfs_inode_item *item, |
39279cc3 CM |
2313 | struct inode *inode) |
2314 | { | |
5f39d397 CM |
2315 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2316 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2317 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2318 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2319 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2320 | ||
2321 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2322 | inode->i_atime.tv_sec); | |
2323 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2324 | inode->i_atime.tv_nsec); | |
2325 | ||
2326 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2327 | inode->i_mtime.tv_sec); | |
2328 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2329 | inode->i_mtime.tv_nsec); | |
2330 | ||
2331 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2332 | inode->i_ctime.tv_sec); | |
2333 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2334 | inode->i_ctime.tv_nsec); | |
2335 | ||
a76a3cd4 | 2336 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2337 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2338 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2339 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2340 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2341 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2342 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2343 | } |
2344 | ||
d352ac68 CM |
2345 | /* |
2346 | * copy everything in the in-memory inode into the btree. | |
2347 | */ | |
d397712b CM |
2348 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2349 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2350 | { |
2351 | struct btrfs_inode_item *inode_item; | |
2352 | struct btrfs_path *path; | |
5f39d397 | 2353 | struct extent_buffer *leaf; |
39279cc3 CM |
2354 | int ret; |
2355 | ||
2356 | path = btrfs_alloc_path(); | |
2357 | BUG_ON(!path); | |
b9473439 | 2358 | path->leave_spinning = 1; |
39279cc3 CM |
2359 | ret = btrfs_lookup_inode(trans, root, path, |
2360 | &BTRFS_I(inode)->location, 1); | |
2361 | if (ret) { | |
2362 | if (ret > 0) | |
2363 | ret = -ENOENT; | |
2364 | goto failed; | |
2365 | } | |
2366 | ||
b4ce94de | 2367 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2368 | leaf = path->nodes[0]; |
2369 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2370 | struct btrfs_inode_item); |
2371 | ||
e02119d5 | 2372 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2373 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2374 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2375 | ret = 0; |
2376 | failed: | |
39279cc3 CM |
2377 | btrfs_free_path(path); |
2378 | return ret; | |
2379 | } | |
2380 | ||
2381 | ||
d352ac68 CM |
2382 | /* |
2383 | * unlink helper that gets used here in inode.c and in the tree logging | |
2384 | * recovery code. It remove a link in a directory with a given name, and | |
2385 | * also drops the back refs in the inode to the directory | |
2386 | */ | |
e02119d5 CM |
2387 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2388 | struct btrfs_root *root, | |
2389 | struct inode *dir, struct inode *inode, | |
2390 | const char *name, int name_len) | |
39279cc3 CM |
2391 | { |
2392 | struct btrfs_path *path; | |
39279cc3 | 2393 | int ret = 0; |
5f39d397 | 2394 | struct extent_buffer *leaf; |
39279cc3 | 2395 | struct btrfs_dir_item *di; |
5f39d397 | 2396 | struct btrfs_key key; |
aec7477b | 2397 | u64 index; |
39279cc3 CM |
2398 | |
2399 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2400 | if (!path) { |
2401 | ret = -ENOMEM; | |
2402 | goto err; | |
2403 | } | |
2404 | ||
b9473439 | 2405 | path->leave_spinning = 1; |
39279cc3 CM |
2406 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2407 | name, name_len, -1); | |
2408 | if (IS_ERR(di)) { | |
2409 | ret = PTR_ERR(di); | |
2410 | goto err; | |
2411 | } | |
2412 | if (!di) { | |
2413 | ret = -ENOENT; | |
2414 | goto err; | |
2415 | } | |
5f39d397 CM |
2416 | leaf = path->nodes[0]; |
2417 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2418 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2419 | if (ret) |
2420 | goto err; | |
39279cc3 CM |
2421 | btrfs_release_path(root, path); |
2422 | ||
aec7477b | 2423 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2424 | inode->i_ino, |
2425 | dir->i_ino, &index); | |
aec7477b | 2426 | if (ret) { |
d397712b | 2427 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2428 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2429 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2430 | goto err; |
2431 | } | |
2432 | ||
39279cc3 | 2433 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2434 | index, name, name_len, -1); |
39279cc3 CM |
2435 | if (IS_ERR(di)) { |
2436 | ret = PTR_ERR(di); | |
2437 | goto err; | |
2438 | } | |
2439 | if (!di) { | |
2440 | ret = -ENOENT; | |
2441 | goto err; | |
2442 | } | |
2443 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2444 | btrfs_release_path(root, path); |
39279cc3 | 2445 | |
e02119d5 CM |
2446 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2447 | inode, dir->i_ino); | |
49eb7e46 | 2448 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2449 | |
2450 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2451 | dir, index); | |
2452 | BUG_ON(ret); | |
39279cc3 CM |
2453 | err: |
2454 | btrfs_free_path(path); | |
e02119d5 CM |
2455 | if (ret) |
2456 | goto out; | |
2457 | ||
2458 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2459 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2460 | btrfs_update_inode(trans, root, dir); | |
2461 | btrfs_drop_nlink(inode); | |
2462 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2463 | out: |
39279cc3 CM |
2464 | return ret; |
2465 | } | |
2466 | ||
2467 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2468 | { | |
2469 | struct btrfs_root *root; | |
2470 | struct btrfs_trans_handle *trans; | |
7b128766 | 2471 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2472 | int ret; |
1832a6d5 | 2473 | unsigned long nr = 0; |
39279cc3 CM |
2474 | |
2475 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2476 | |
39279cc3 | 2477 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2478 | |
39279cc3 | 2479 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2480 | |
2481 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2482 | ||
e02119d5 CM |
2483 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2484 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2485 | |
2486 | if (inode->i_nlink == 0) | |
2487 | ret = btrfs_orphan_add(trans, inode); | |
2488 | ||
d3c2fdcf | 2489 | nr = trans->blocks_used; |
5f39d397 | 2490 | |
89ce8a63 | 2491 | btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2492 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2493 | return ret; |
2494 | } | |
2495 | ||
4df27c4d YZ |
2496 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2497 | struct btrfs_root *root, | |
2498 | struct inode *dir, u64 objectid, | |
2499 | const char *name, int name_len) | |
2500 | { | |
2501 | struct btrfs_path *path; | |
2502 | struct extent_buffer *leaf; | |
2503 | struct btrfs_dir_item *di; | |
2504 | struct btrfs_key key; | |
2505 | u64 index; | |
2506 | int ret; | |
2507 | ||
2508 | path = btrfs_alloc_path(); | |
2509 | if (!path) | |
2510 | return -ENOMEM; | |
2511 | ||
2512 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2513 | name, name_len, -1); | |
2514 | BUG_ON(!di || IS_ERR(di)); | |
2515 | ||
2516 | leaf = path->nodes[0]; | |
2517 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2518 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2519 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2520 | BUG_ON(ret); | |
2521 | btrfs_release_path(root, path); | |
2522 | ||
2523 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2524 | objectid, root->root_key.objectid, | |
2525 | dir->i_ino, &index, name, name_len); | |
2526 | if (ret < 0) { | |
2527 | BUG_ON(ret != -ENOENT); | |
2528 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2529 | name, name_len); | |
2530 | BUG_ON(!di || IS_ERR(di)); | |
2531 | ||
2532 | leaf = path->nodes[0]; | |
2533 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2534 | btrfs_release_path(root, path); | |
2535 | index = key.offset; | |
2536 | } | |
2537 | ||
2538 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2539 | index, name, name_len, -1); | |
2540 | BUG_ON(!di || IS_ERR(di)); | |
2541 | ||
2542 | leaf = path->nodes[0]; | |
2543 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2544 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2545 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2546 | BUG_ON(ret); | |
2547 | btrfs_release_path(root, path); | |
2548 | ||
2549 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2550 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2551 | ret = btrfs_update_inode(trans, root, dir); | |
2552 | BUG_ON(ret); | |
2553 | dir->i_sb->s_dirt = 1; | |
2554 | ||
2555 | btrfs_free_path(path); | |
2556 | return 0; | |
2557 | } | |
2558 | ||
39279cc3 CM |
2559 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2560 | { | |
2561 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2562 | int err = 0; |
39279cc3 CM |
2563 | int ret; |
2564 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2565 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2566 | unsigned long nr = 0; |
39279cc3 | 2567 | |
3394e160 | 2568 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2569 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2570 | return -ENOTEMPTY; |
2571 | ||
39279cc3 CM |
2572 | trans = btrfs_start_transaction(root, 1); |
2573 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2574 | |
4df27c4d YZ |
2575 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2576 | err = btrfs_unlink_subvol(trans, root, dir, | |
2577 | BTRFS_I(inode)->location.objectid, | |
2578 | dentry->d_name.name, | |
2579 | dentry->d_name.len); | |
2580 | goto out; | |
2581 | } | |
2582 | ||
7b128766 JB |
2583 | err = btrfs_orphan_add(trans, inode); |
2584 | if (err) | |
4df27c4d | 2585 | goto out; |
7b128766 | 2586 | |
39279cc3 | 2587 | /* now the directory is empty */ |
e02119d5 CM |
2588 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2589 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2590 | if (!err) |
dbe674a9 | 2591 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2592 | out: |
d3c2fdcf | 2593 | nr = trans->blocks_used; |
89ce8a63 | 2594 | ret = btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2595 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2596 | |
39279cc3 CM |
2597 | if (ret && !err) |
2598 | err = ret; | |
2599 | return err; | |
2600 | } | |
2601 | ||
d20f7043 | 2602 | #if 0 |
323ac95b CM |
2603 | /* |
2604 | * when truncating bytes in a file, it is possible to avoid reading | |
2605 | * the leaves that contain only checksum items. This can be the | |
2606 | * majority of the IO required to delete a large file, but it must | |
2607 | * be done carefully. | |
2608 | * | |
2609 | * The keys in the level just above the leaves are checked to make sure | |
2610 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2611 | * after the new size. | |
2612 | * | |
2613 | * Then the key for the next leaf is checked to make sure it also has | |
2614 | * a checksum item for the same file. If it does, we know our target leaf | |
2615 | * contains only checksum items, and it can be safely freed without reading | |
2616 | * it. | |
2617 | * | |
2618 | * This is just an optimization targeted at large files. It may do | |
2619 | * nothing. It will return 0 unless things went badly. | |
2620 | */ | |
2621 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2622 | struct btrfs_root *root, | |
2623 | struct btrfs_path *path, | |
2624 | struct inode *inode, u64 new_size) | |
2625 | { | |
2626 | struct btrfs_key key; | |
2627 | int ret; | |
2628 | int nritems; | |
2629 | struct btrfs_key found_key; | |
2630 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2631 | struct btrfs_leaf_ref *ref; |
2632 | u64 leaf_gen; | |
2633 | u64 leaf_start; | |
323ac95b CM |
2634 | |
2635 | path->lowest_level = 1; | |
2636 | key.objectid = inode->i_ino; | |
2637 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2638 | key.offset = new_size; | |
2639 | again: | |
2640 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2641 | if (ret < 0) | |
2642 | goto out; | |
2643 | ||
2644 | if (path->nodes[1] == NULL) { | |
2645 | ret = 0; | |
2646 | goto out; | |
2647 | } | |
2648 | ret = 0; | |
2649 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2650 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2651 | ||
2652 | if (!nritems) | |
2653 | goto out; | |
2654 | ||
2655 | if (path->slots[1] >= nritems) | |
2656 | goto next_node; | |
2657 | ||
2658 | /* did we find a key greater than anything we want to delete? */ | |
2659 | if (found_key.objectid > inode->i_ino || | |
2660 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2661 | goto out; | |
2662 | ||
2663 | /* we check the next key in the node to make sure the leave contains | |
2664 | * only checksum items. This comparison doesn't work if our | |
2665 | * leaf is the last one in the node | |
2666 | */ | |
2667 | if (path->slots[1] + 1 >= nritems) { | |
2668 | next_node: | |
2669 | /* search forward from the last key in the node, this | |
2670 | * will bring us into the next node in the tree | |
2671 | */ | |
2672 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2673 | ||
2674 | /* unlikely, but we inc below, so check to be safe */ | |
2675 | if (found_key.offset == (u64)-1) | |
2676 | goto out; | |
2677 | ||
2678 | /* search_forward needs a path with locks held, do the | |
2679 | * search again for the original key. It is possible | |
2680 | * this will race with a balance and return a path that | |
2681 | * we could modify, but this drop is just an optimization | |
2682 | * and is allowed to miss some leaves. | |
2683 | */ | |
2684 | btrfs_release_path(root, path); | |
2685 | found_key.offset++; | |
2686 | ||
2687 | /* setup a max key for search_forward */ | |
2688 | other_key.offset = (u64)-1; | |
2689 | other_key.type = key.type; | |
2690 | other_key.objectid = key.objectid; | |
2691 | ||
2692 | path->keep_locks = 1; | |
2693 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2694 | path, 0, 0); | |
2695 | path->keep_locks = 0; | |
2696 | if (ret || found_key.objectid != key.objectid || | |
2697 | found_key.type != key.type) { | |
2698 | ret = 0; | |
2699 | goto out; | |
2700 | } | |
2701 | ||
2702 | key.offset = found_key.offset; | |
2703 | btrfs_release_path(root, path); | |
2704 | cond_resched(); | |
2705 | goto again; | |
2706 | } | |
2707 | ||
2708 | /* we know there's one more slot after us in the tree, | |
2709 | * read that key so we can verify it is also a checksum item | |
2710 | */ | |
2711 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2712 | ||
2713 | if (found_key.objectid < inode->i_ino) | |
2714 | goto next_key; | |
2715 | ||
2716 | if (found_key.type != key.type || found_key.offset < new_size) | |
2717 | goto next_key; | |
2718 | ||
2719 | /* | |
2720 | * if the key for the next leaf isn't a csum key from this objectid, | |
2721 | * we can't be sure there aren't good items inside this leaf. | |
2722 | * Bail out | |
2723 | */ | |
2724 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2725 | goto out; | |
2726 | ||
5b84e8d6 YZ |
2727 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2728 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2729 | /* |
2730 | * it is safe to delete this leaf, it contains only | |
2731 | * csum items from this inode at an offset >= new_size | |
2732 | */ | |
5b84e8d6 | 2733 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2734 | BUG_ON(ret); |
2735 | ||
5b84e8d6 YZ |
2736 | if (root->ref_cows && leaf_gen < trans->transid) { |
2737 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2738 | if (ref) { | |
2739 | ref->root_gen = root->root_key.offset; | |
2740 | ref->bytenr = leaf_start; | |
2741 | ref->owner = 0; | |
2742 | ref->generation = leaf_gen; | |
2743 | ref->nritems = 0; | |
2744 | ||
bd56b302 CM |
2745 | btrfs_sort_leaf_ref(ref); |
2746 | ||
5b84e8d6 YZ |
2747 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2748 | WARN_ON(ret); | |
2749 | btrfs_free_leaf_ref(root, ref); | |
2750 | } else { | |
2751 | WARN_ON(1); | |
2752 | } | |
2753 | } | |
323ac95b CM |
2754 | next_key: |
2755 | btrfs_release_path(root, path); | |
2756 | ||
2757 | if (other_key.objectid == inode->i_ino && | |
2758 | other_key.type == key.type && other_key.offset > key.offset) { | |
2759 | key.offset = other_key.offset; | |
2760 | cond_resched(); | |
2761 | goto again; | |
2762 | } | |
2763 | ret = 0; | |
2764 | out: | |
2765 | /* fixup any changes we've made to the path */ | |
2766 | path->lowest_level = 0; | |
2767 | path->keep_locks = 0; | |
2768 | btrfs_release_path(root, path); | |
2769 | return ret; | |
2770 | } | |
2771 | ||
d20f7043 CM |
2772 | #endif |
2773 | ||
39279cc3 CM |
2774 | /* |
2775 | * this can truncate away extent items, csum items and directory items. | |
2776 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2777 | * any higher than new_size |
39279cc3 CM |
2778 | * |
2779 | * csum items that cross the new i_size are truncated to the new size | |
2780 | * as well. | |
7b128766 JB |
2781 | * |
2782 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2783 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2784 | */ |
e02119d5 CM |
2785 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2786 | struct btrfs_root *root, | |
2787 | struct inode *inode, | |
2788 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2789 | { |
2790 | int ret; | |
2791 | struct btrfs_path *path; | |
2792 | struct btrfs_key key; | |
5f39d397 | 2793 | struct btrfs_key found_key; |
06d9a8d7 | 2794 | u32 found_type = (u8)-1; |
5f39d397 | 2795 | struct extent_buffer *leaf; |
39279cc3 CM |
2796 | struct btrfs_file_extent_item *fi; |
2797 | u64 extent_start = 0; | |
db94535d | 2798 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2799 | u64 extent_offset = 0; |
39279cc3 CM |
2800 | u64 item_end = 0; |
2801 | int found_extent; | |
2802 | int del_item; | |
85e21bac CM |
2803 | int pending_del_nr = 0; |
2804 | int pending_del_slot = 0; | |
179e29e4 | 2805 | int extent_type = -1; |
771ed689 | 2806 | int encoding; |
3b951516 | 2807 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2808 | |
e02119d5 | 2809 | if (root->ref_cows) |
5b21f2ed | 2810 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2811 | path = btrfs_alloc_path(); |
2812 | BUG_ON(!path); | |
33c17ad5 | 2813 | path->reada = -1; |
5f39d397 | 2814 | |
39279cc3 CM |
2815 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2816 | key.objectid = inode->i_ino; | |
2817 | key.offset = (u64)-1; | |
5f39d397 CM |
2818 | key.type = (u8)-1; |
2819 | ||
85e21bac | 2820 | search_again: |
b9473439 | 2821 | path->leave_spinning = 1; |
85e21bac | 2822 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2823 | if (ret < 0) |
85e21bac | 2824 | goto error; |
d397712b | 2825 | |
85e21bac | 2826 | if (ret > 0) { |
e02119d5 CM |
2827 | /* there are no items in the tree for us to truncate, we're |
2828 | * done | |
2829 | */ | |
2830 | if (path->slots[0] == 0) { | |
2831 | ret = 0; | |
2832 | goto error; | |
2833 | } | |
85e21bac CM |
2834 | path->slots[0]--; |
2835 | } | |
2836 | ||
d397712b | 2837 | while (1) { |
39279cc3 | 2838 | fi = NULL; |
5f39d397 CM |
2839 | leaf = path->nodes[0]; |
2840 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2841 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2842 | encoding = 0; |
39279cc3 | 2843 | |
5f39d397 | 2844 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2845 | break; |
5f39d397 | 2846 | |
85e21bac | 2847 | if (found_type < min_type) |
39279cc3 CM |
2848 | break; |
2849 | ||
5f39d397 | 2850 | item_end = found_key.offset; |
39279cc3 | 2851 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2852 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2853 | struct btrfs_file_extent_item); |
179e29e4 | 2854 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2855 | encoding = btrfs_file_extent_compression(leaf, fi); |
2856 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2857 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2858 | ||
179e29e4 | 2859 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2860 | item_end += |
db94535d | 2861 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2862 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2863 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2864 | fi); |
39279cc3 | 2865 | } |
008630c1 | 2866 | item_end--; |
39279cc3 | 2867 | } |
e02119d5 | 2868 | if (item_end < new_size) { |
d397712b | 2869 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2870 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2871 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2872 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2873 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2874 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2875 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2876 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2877 | else if (found_type) |
b888db2b | 2878 | found_type--; |
d397712b | 2879 | else |
b888db2b | 2880 | break; |
a61721d5 | 2881 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2882 | goto next; |
39279cc3 | 2883 | } |
e02119d5 | 2884 | if (found_key.offset >= new_size) |
39279cc3 CM |
2885 | del_item = 1; |
2886 | else | |
2887 | del_item = 0; | |
2888 | found_extent = 0; | |
2889 | ||
2890 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2891 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2892 | goto delete; | |
2893 | ||
2894 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2895 | u64 num_dec; |
db94535d | 2896 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2897 | if (!del_item && !encoding) { |
db94535d CM |
2898 | u64 orig_num_bytes = |
2899 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2900 | extent_num_bytes = new_size - |
5f39d397 | 2901 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2902 | extent_num_bytes = extent_num_bytes & |
2903 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2904 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2905 | extent_num_bytes); | |
2906 | num_dec = (orig_num_bytes - | |
9069218d | 2907 | extent_num_bytes); |
e02119d5 | 2908 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2909 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2910 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2911 | } else { |
db94535d CM |
2912 | extent_num_bytes = |
2913 | btrfs_file_extent_disk_num_bytes(leaf, | |
2914 | fi); | |
5d4f98a2 YZ |
2915 | extent_offset = found_key.offset - |
2916 | btrfs_file_extent_offset(leaf, fi); | |
2917 | ||
39279cc3 | 2918 | /* FIXME blocksize != 4096 */ |
9069218d | 2919 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2920 | if (extent_start != 0) { |
2921 | found_extent = 1; | |
e02119d5 | 2922 | if (root->ref_cows) |
a76a3cd4 | 2923 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2924 | } |
39279cc3 | 2925 | } |
9069218d | 2926 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2927 | /* |
2928 | * we can't truncate inline items that have had | |
2929 | * special encodings | |
2930 | */ | |
2931 | if (!del_item && | |
2932 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2933 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2934 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2935 | u32 size = new_size - found_key.offset; |
2936 | ||
2937 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2938 | inode_sub_bytes(inode, item_end + 1 - |
2939 | new_size); | |
e02119d5 CM |
2940 | } |
2941 | size = | |
2942 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2943 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2944 | size, 1); |
9069218d | 2945 | BUG_ON(ret); |
e02119d5 | 2946 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2947 | inode_sub_bytes(inode, item_end + 1 - |
2948 | found_key.offset); | |
9069218d | 2949 | } |
39279cc3 | 2950 | } |
179e29e4 | 2951 | delete: |
39279cc3 | 2952 | if (del_item) { |
85e21bac CM |
2953 | if (!pending_del_nr) { |
2954 | /* no pending yet, add ourselves */ | |
2955 | pending_del_slot = path->slots[0]; | |
2956 | pending_del_nr = 1; | |
2957 | } else if (pending_del_nr && | |
2958 | path->slots[0] + 1 == pending_del_slot) { | |
2959 | /* hop on the pending chunk */ | |
2960 | pending_del_nr++; | |
2961 | pending_del_slot = path->slots[0]; | |
2962 | } else { | |
d397712b | 2963 | BUG(); |
85e21bac | 2964 | } |
39279cc3 CM |
2965 | } else { |
2966 | break; | |
2967 | } | |
5d4f98a2 | 2968 | if (found_extent && root->ref_cows) { |
b9473439 | 2969 | btrfs_set_path_blocking(path); |
39279cc3 | 2970 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
2971 | extent_num_bytes, 0, |
2972 | btrfs_header_owner(leaf), | |
2973 | inode->i_ino, extent_offset); | |
39279cc3 CM |
2974 | BUG_ON(ret); |
2975 | } | |
85e21bac CM |
2976 | next: |
2977 | if (path->slots[0] == 0) { | |
2978 | if (pending_del_nr) | |
2979 | goto del_pending; | |
2980 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2981 | if (found_type == BTRFS_INODE_ITEM_KEY) |
2982 | break; | |
85e21bac CM |
2983 | goto search_again; |
2984 | } | |
2985 | ||
2986 | path->slots[0]--; | |
2987 | if (pending_del_nr && | |
2988 | path->slots[0] + 1 != pending_del_slot) { | |
2989 | struct btrfs_key debug; | |
2990 | del_pending: | |
2991 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2992 | pending_del_slot); | |
2993 | ret = btrfs_del_items(trans, root, path, | |
2994 | pending_del_slot, | |
2995 | pending_del_nr); | |
2996 | BUG_ON(ret); | |
2997 | pending_del_nr = 0; | |
2998 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2999 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3000 | break; | |
85e21bac CM |
3001 | goto search_again; |
3002 | } | |
39279cc3 CM |
3003 | } |
3004 | ret = 0; | |
3005 | error: | |
85e21bac CM |
3006 | if (pending_del_nr) { |
3007 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3008 | pending_del_nr); | |
3009 | } | |
39279cc3 | 3010 | btrfs_free_path(path); |
39279cc3 CM |
3011 | return ret; |
3012 | } | |
3013 | ||
3014 | /* | |
3015 | * taken from block_truncate_page, but does cow as it zeros out | |
3016 | * any bytes left in the last page in the file. | |
3017 | */ | |
3018 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3019 | { | |
3020 | struct inode *inode = mapping->host; | |
db94535d | 3021 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3022 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3023 | struct btrfs_ordered_extent *ordered; | |
3024 | char *kaddr; | |
db94535d | 3025 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3026 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3027 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3028 | struct page *page; | |
39279cc3 | 3029 | int ret = 0; |
a52d9a80 | 3030 | u64 page_start; |
e6dcd2dc | 3031 | u64 page_end; |
39279cc3 CM |
3032 | |
3033 | if ((offset & (blocksize - 1)) == 0) | |
3034 | goto out; | |
5d5e103a JB |
3035 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
3036 | if (ret) | |
3037 | goto out; | |
3038 | ||
3039 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); | |
3040 | if (ret) | |
3041 | goto out; | |
39279cc3 CM |
3042 | |
3043 | ret = -ENOMEM; | |
211c17f5 | 3044 | again: |
39279cc3 | 3045 | page = grab_cache_page(mapping, index); |
5d5e103a JB |
3046 | if (!page) { |
3047 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3048 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 | 3049 | goto out; |
5d5e103a | 3050 | } |
e6dcd2dc CM |
3051 | |
3052 | page_start = page_offset(page); | |
3053 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3054 | ||
39279cc3 | 3055 | if (!PageUptodate(page)) { |
9ebefb18 | 3056 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3057 | lock_page(page); |
211c17f5 CM |
3058 | if (page->mapping != mapping) { |
3059 | unlock_page(page); | |
3060 | page_cache_release(page); | |
3061 | goto again; | |
3062 | } | |
39279cc3 CM |
3063 | if (!PageUptodate(page)) { |
3064 | ret = -EIO; | |
89642229 | 3065 | goto out_unlock; |
39279cc3 CM |
3066 | } |
3067 | } | |
211c17f5 | 3068 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3069 | |
3070 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3071 | set_page_extent_mapped(page); | |
3072 | ||
3073 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3074 | if (ordered) { | |
3075 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3076 | unlock_page(page); | |
3077 | page_cache_release(page); | |
eb84ae03 | 3078 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3079 | btrfs_put_ordered_extent(ordered); |
3080 | goto again; | |
3081 | } | |
3082 | ||
5d5e103a JB |
3083 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
3084 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, | |
3085 | GFP_NOFS); | |
3086 | ||
9ed74f2d JB |
3087 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3088 | if (ret) { | |
3089 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3090 | goto out_unlock; | |
3091 | } | |
3092 | ||
e6dcd2dc CM |
3093 | ret = 0; |
3094 | if (offset != PAGE_CACHE_SIZE) { | |
3095 | kaddr = kmap(page); | |
3096 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3097 | flush_dcache_page(page); | |
3098 | kunmap(page); | |
3099 | } | |
247e743c | 3100 | ClearPageChecked(page); |
e6dcd2dc CM |
3101 | set_page_dirty(page); |
3102 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3103 | |
89642229 | 3104 | out_unlock: |
5d5e103a JB |
3105 | if (ret) |
3106 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3107 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 CM |
3108 | unlock_page(page); |
3109 | page_cache_release(page); | |
3110 | out: | |
3111 | return ret; | |
3112 | } | |
3113 | ||
9036c102 | 3114 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3115 | { |
9036c102 YZ |
3116 | struct btrfs_trans_handle *trans; |
3117 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3118 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3119 | struct extent_map *em; | |
3120 | u64 mask = root->sectorsize - 1; | |
3121 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3122 | u64 block_end = (size + mask) & ~mask; | |
3123 | u64 last_byte; | |
3124 | u64 cur_offset; | |
3125 | u64 hole_size; | |
9ed74f2d | 3126 | int err = 0; |
39279cc3 | 3127 | |
9036c102 YZ |
3128 | if (size <= hole_start) |
3129 | return 0; | |
3130 | ||
5d5e103a JB |
3131 | err = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
3132 | if (err) | |
3133 | return err; | |
2bf5a725 | 3134 | |
9036c102 YZ |
3135 | while (1) { |
3136 | struct btrfs_ordered_extent *ordered; | |
3137 | btrfs_wait_ordered_range(inode, hole_start, | |
3138 | block_end - hole_start); | |
3139 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3140 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3141 | if (!ordered) | |
3142 | break; | |
3143 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3144 | btrfs_put_ordered_extent(ordered); | |
3145 | } | |
39279cc3 | 3146 | |
9036c102 YZ |
3147 | trans = btrfs_start_transaction(root, 1); |
3148 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3149 | |
9036c102 YZ |
3150 | cur_offset = hole_start; |
3151 | while (1) { | |
3152 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3153 | block_end - cur_offset, 0); | |
3154 | BUG_ON(IS_ERR(em) || !em); | |
3155 | last_byte = min(extent_map_end(em), block_end); | |
3156 | last_byte = (last_byte + mask) & ~mask; | |
3157 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3158 | u64 hint_byte = 0; |
9036c102 | 3159 | hole_size = last_byte - cur_offset; |
771ed689 CM |
3160 | err = btrfs_drop_extents(trans, root, inode, |
3161 | cur_offset, | |
3162 | cur_offset + hole_size, | |
e980b50c | 3163 | block_end, |
a1ed835e | 3164 | cur_offset, &hint_byte, 1); |
771ed689 CM |
3165 | if (err) |
3166 | break; | |
9ed74f2d JB |
3167 | |
3168 | err = btrfs_reserve_metadata_space(root, 1); | |
3169 | if (err) | |
3170 | break; | |
3171 | ||
9036c102 YZ |
3172 | err = btrfs_insert_file_extent(trans, root, |
3173 | inode->i_ino, cur_offset, 0, | |
3174 | 0, hole_size, 0, hole_size, | |
3175 | 0, 0, 0); | |
3176 | btrfs_drop_extent_cache(inode, hole_start, | |
3177 | last_byte - 1, 0); | |
9ed74f2d | 3178 | btrfs_unreserve_metadata_space(root, 1); |
9036c102 YZ |
3179 | } |
3180 | free_extent_map(em); | |
3181 | cur_offset = last_byte; | |
3182 | if (err || cur_offset >= block_end) | |
3183 | break; | |
3184 | } | |
1832a6d5 | 3185 | |
9036c102 YZ |
3186 | btrfs_end_transaction(trans, root); |
3187 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3188 | return err; | |
3189 | } | |
39279cc3 | 3190 | |
9036c102 YZ |
3191 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3192 | { | |
3193 | struct inode *inode = dentry->d_inode; | |
3194 | int err; | |
39279cc3 | 3195 | |
9036c102 YZ |
3196 | err = inode_change_ok(inode, attr); |
3197 | if (err) | |
3198 | return err; | |
2bf5a725 | 3199 | |
5a3f23d5 CM |
3200 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3201 | if (attr->ia_size > inode->i_size) { | |
3202 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3203 | if (err) | |
3204 | return err; | |
3205 | } else if (inode->i_size > 0 && | |
3206 | attr->ia_size == 0) { | |
3207 | ||
3208 | /* we're truncating a file that used to have good | |
3209 | * data down to zero. Make sure it gets into | |
3210 | * the ordered flush list so that any new writes | |
3211 | * get down to disk quickly. | |
3212 | */ | |
3213 | BTRFS_I(inode)->ordered_data_close = 1; | |
3214 | } | |
39279cc3 | 3215 | } |
9036c102 | 3216 | |
39279cc3 | 3217 | err = inode_setattr(inode, attr); |
33268eaf JB |
3218 | |
3219 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3220 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3221 | return err; |
3222 | } | |
61295eb8 | 3223 | |
39279cc3 CM |
3224 | void btrfs_delete_inode(struct inode *inode) |
3225 | { | |
3226 | struct btrfs_trans_handle *trans; | |
3227 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3228 | unsigned long nr; |
39279cc3 CM |
3229 | int ret; |
3230 | ||
3231 | truncate_inode_pages(&inode->i_data, 0); | |
3232 | if (is_bad_inode(inode)) { | |
7b128766 | 3233 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3234 | goto no_delete; |
3235 | } | |
4a096752 | 3236 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3237 | |
76dda93c YZ |
3238 | if (inode->i_nlink > 0) { |
3239 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3240 | goto no_delete; | |
3241 | } | |
3242 | ||
dbe674a9 | 3243 | btrfs_i_size_write(inode, 0); |
180591bc | 3244 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3245 | |
39279cc3 | 3246 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3247 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3248 | if (ret) { |
3249 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3250 | goto no_delete_lock; |
7b128766 JB |
3251 | } |
3252 | ||
3253 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3254 | |
d3c2fdcf | 3255 | nr = trans->blocks_used; |
85e21bac | 3256 | clear_inode(inode); |
5f39d397 | 3257 | |
39279cc3 | 3258 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3259 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3260 | return; |
54aa1f4d CM |
3261 | |
3262 | no_delete_lock: | |
d3c2fdcf | 3263 | nr = trans->blocks_used; |
54aa1f4d | 3264 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3265 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3266 | no_delete: |
3267 | clear_inode(inode); | |
3268 | } | |
3269 | ||
3270 | /* | |
3271 | * this returns the key found in the dir entry in the location pointer. | |
3272 | * If no dir entries were found, location->objectid is 0. | |
3273 | */ | |
3274 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3275 | struct btrfs_key *location) | |
3276 | { | |
3277 | const char *name = dentry->d_name.name; | |
3278 | int namelen = dentry->d_name.len; | |
3279 | struct btrfs_dir_item *di; | |
3280 | struct btrfs_path *path; | |
3281 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3282 | int ret = 0; |
39279cc3 CM |
3283 | |
3284 | path = btrfs_alloc_path(); | |
3285 | BUG_ON(!path); | |
3954401f | 3286 | |
39279cc3 CM |
3287 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3288 | namelen, 0); | |
0d9f7f3e Y |
3289 | if (IS_ERR(di)) |
3290 | ret = PTR_ERR(di); | |
d397712b CM |
3291 | |
3292 | if (!di || IS_ERR(di)) | |
3954401f | 3293 | goto out_err; |
d397712b | 3294 | |
5f39d397 | 3295 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3296 | out: |
39279cc3 CM |
3297 | btrfs_free_path(path); |
3298 | return ret; | |
3954401f CM |
3299 | out_err: |
3300 | location->objectid = 0; | |
3301 | goto out; | |
39279cc3 CM |
3302 | } |
3303 | ||
3304 | /* | |
3305 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3306 | * needs to be changed to reflect the root directory of the tree root. This | |
3307 | * is kind of like crossing a mount point. | |
3308 | */ | |
3309 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3310 | struct inode *dir, |
3311 | struct dentry *dentry, | |
3312 | struct btrfs_key *location, | |
3313 | struct btrfs_root **sub_root) | |
39279cc3 | 3314 | { |
4df27c4d YZ |
3315 | struct btrfs_path *path; |
3316 | struct btrfs_root *new_root; | |
3317 | struct btrfs_root_ref *ref; | |
3318 | struct extent_buffer *leaf; | |
3319 | int ret; | |
3320 | int err = 0; | |
39279cc3 | 3321 | |
4df27c4d YZ |
3322 | path = btrfs_alloc_path(); |
3323 | if (!path) { | |
3324 | err = -ENOMEM; | |
3325 | goto out; | |
3326 | } | |
39279cc3 | 3327 | |
4df27c4d YZ |
3328 | err = -ENOENT; |
3329 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3330 | BTRFS_I(dir)->root->root_key.objectid, | |
3331 | location->objectid); | |
3332 | if (ret) { | |
3333 | if (ret < 0) | |
3334 | err = ret; | |
3335 | goto out; | |
3336 | } | |
39279cc3 | 3337 | |
4df27c4d YZ |
3338 | leaf = path->nodes[0]; |
3339 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3340 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3341 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3342 | goto out; | |
39279cc3 | 3343 | |
4df27c4d YZ |
3344 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3345 | (unsigned long)(ref + 1), | |
3346 | dentry->d_name.len); | |
3347 | if (ret) | |
3348 | goto out; | |
3349 | ||
3350 | btrfs_release_path(root->fs_info->tree_root, path); | |
3351 | ||
3352 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3353 | if (IS_ERR(new_root)) { | |
3354 | err = PTR_ERR(new_root); | |
3355 | goto out; | |
3356 | } | |
3357 | ||
3358 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3359 | err = -ENOENT; | |
3360 | goto out; | |
3361 | } | |
3362 | ||
3363 | *sub_root = new_root; | |
3364 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3365 | location->type = BTRFS_INODE_ITEM_KEY; | |
3366 | location->offset = 0; | |
3367 | err = 0; | |
3368 | out: | |
3369 | btrfs_free_path(path); | |
3370 | return err; | |
39279cc3 CM |
3371 | } |
3372 | ||
5d4f98a2 YZ |
3373 | static void inode_tree_add(struct inode *inode) |
3374 | { | |
3375 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3376 | struct btrfs_inode *entry; | |
03e860bd NP |
3377 | struct rb_node **p; |
3378 | struct rb_node *parent; | |
03e860bd NP |
3379 | again: |
3380 | p = &root->inode_tree.rb_node; | |
3381 | parent = NULL; | |
5d4f98a2 | 3382 | |
76dda93c YZ |
3383 | if (hlist_unhashed(&inode->i_hash)) |
3384 | return; | |
3385 | ||
5d4f98a2 YZ |
3386 | spin_lock(&root->inode_lock); |
3387 | while (*p) { | |
3388 | parent = *p; | |
3389 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3390 | ||
3391 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3392 | p = &parent->rb_left; |
5d4f98a2 | 3393 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3394 | p = &parent->rb_right; |
5d4f98a2 YZ |
3395 | else { |
3396 | WARN_ON(!(entry->vfs_inode.i_state & | |
3397 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd NP |
3398 | rb_erase(parent, &root->inode_tree); |
3399 | RB_CLEAR_NODE(parent); | |
3400 | spin_unlock(&root->inode_lock); | |
3401 | goto again; | |
5d4f98a2 YZ |
3402 | } |
3403 | } | |
3404 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3405 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3406 | spin_unlock(&root->inode_lock); | |
3407 | } | |
3408 | ||
3409 | static void inode_tree_del(struct inode *inode) | |
3410 | { | |
3411 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3412 | int empty = 0; |
5d4f98a2 | 3413 | |
03e860bd | 3414 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3415 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3416 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3417 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3418 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3419 | } |
03e860bd | 3420 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3421 | |
3422 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3423 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3424 | spin_lock(&root->inode_lock); | |
3425 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3426 | spin_unlock(&root->inode_lock); | |
3427 | if (empty) | |
3428 | btrfs_add_dead_root(root); | |
3429 | } | |
3430 | } | |
3431 | ||
3432 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3433 | { | |
3434 | struct rb_node *node; | |
3435 | struct rb_node *prev; | |
3436 | struct btrfs_inode *entry; | |
3437 | struct inode *inode; | |
3438 | u64 objectid = 0; | |
3439 | ||
3440 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3441 | ||
3442 | spin_lock(&root->inode_lock); | |
3443 | again: | |
3444 | node = root->inode_tree.rb_node; | |
3445 | prev = NULL; | |
3446 | while (node) { | |
3447 | prev = node; | |
3448 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3449 | ||
3450 | if (objectid < entry->vfs_inode.i_ino) | |
3451 | node = node->rb_left; | |
3452 | else if (objectid > entry->vfs_inode.i_ino) | |
3453 | node = node->rb_right; | |
3454 | else | |
3455 | break; | |
3456 | } | |
3457 | if (!node) { | |
3458 | while (prev) { | |
3459 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3460 | if (objectid <= entry->vfs_inode.i_ino) { | |
3461 | node = prev; | |
3462 | break; | |
3463 | } | |
3464 | prev = rb_next(prev); | |
3465 | } | |
3466 | } | |
3467 | while (node) { | |
3468 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3469 | objectid = entry->vfs_inode.i_ino + 1; | |
3470 | inode = igrab(&entry->vfs_inode); | |
3471 | if (inode) { | |
3472 | spin_unlock(&root->inode_lock); | |
3473 | if (atomic_read(&inode->i_count) > 1) | |
3474 | d_prune_aliases(inode); | |
3475 | /* | |
3476 | * btrfs_drop_inode will remove it from | |
3477 | * the inode cache when its usage count | |
3478 | * hits zero. | |
3479 | */ | |
3480 | iput(inode); | |
3481 | cond_resched(); | |
3482 | spin_lock(&root->inode_lock); | |
3483 | goto again; | |
3484 | } | |
3485 | ||
3486 | if (cond_resched_lock(&root->inode_lock)) | |
3487 | goto again; | |
3488 | ||
3489 | node = rb_next(node); | |
3490 | } | |
3491 | spin_unlock(&root->inode_lock); | |
3492 | return 0; | |
5d4f98a2 YZ |
3493 | } |
3494 | ||
e02119d5 | 3495 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3496 | { |
e02119d5 CM |
3497 | struct btrfs_inode *bi = BTRFS_I(inode); |
3498 | ||
e02119d5 | 3499 | bi->generation = 0; |
c3027eb5 | 3500 | bi->sequence = 0; |
e02119d5 | 3501 | bi->last_trans = 0; |
257c62e1 | 3502 | bi->last_sub_trans = 0; |
e02119d5 CM |
3503 | bi->logged_trans = 0; |
3504 | bi->delalloc_bytes = 0; | |
6a63209f | 3505 | bi->reserved_bytes = 0; |
e02119d5 CM |
3506 | bi->disk_i_size = 0; |
3507 | bi->flags = 0; | |
3508 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3509 | bi->last_unlink_trans = 0; |
2757495c | 3510 | bi->ordered_data_close = 0; |
d1310b2e CM |
3511 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3512 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3513 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3514 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3515 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3516 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3517 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3518 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3519 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3520 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3521 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3522 | } | |
3523 | ||
3524 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3525 | { | |
3526 | struct btrfs_iget_args *args = p; | |
3527 | inode->i_ino = args->ino; | |
3528 | init_btrfs_i(inode); | |
3529 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3530 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3531 | return 0; |
3532 | } | |
3533 | ||
3534 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3535 | { | |
3536 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3537 | return args->ino == inode->i_ino && |
3538 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3539 | } |
3540 | ||
5d4f98a2 YZ |
3541 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3542 | u64 objectid, | |
3543 | struct btrfs_root *root) | |
39279cc3 CM |
3544 | { |
3545 | struct inode *inode; | |
3546 | struct btrfs_iget_args args; | |
3547 | args.ino = objectid; | |
3548 | args.root = root; | |
3549 | ||
3550 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3551 | btrfs_init_locked_inode, | |
3552 | (void *)&args); | |
3553 | return inode; | |
3554 | } | |
3555 | ||
1a54ef8c BR |
3556 | /* Get an inode object given its location and corresponding root. |
3557 | * Returns in *is_new if the inode was read from disk | |
3558 | */ | |
3559 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3560 | struct btrfs_root *root) |
1a54ef8c BR |
3561 | { |
3562 | struct inode *inode; | |
3563 | ||
3564 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3565 | if (!inode) | |
5d4f98a2 | 3566 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3567 | |
3568 | if (inode->i_state & I_NEW) { | |
3569 | BTRFS_I(inode)->root = root; | |
3570 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3571 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3572 | |
3573 | inode_tree_add(inode); | |
1a54ef8c | 3574 | unlock_new_inode(inode); |
1a54ef8c BR |
3575 | } |
3576 | ||
3577 | return inode; | |
3578 | } | |
3579 | ||
4df27c4d YZ |
3580 | static struct inode *new_simple_dir(struct super_block *s, |
3581 | struct btrfs_key *key, | |
3582 | struct btrfs_root *root) | |
3583 | { | |
3584 | struct inode *inode = new_inode(s); | |
3585 | ||
3586 | if (!inode) | |
3587 | return ERR_PTR(-ENOMEM); | |
3588 | ||
3589 | init_btrfs_i(inode); | |
3590 | ||
3591 | BTRFS_I(inode)->root = root; | |
3592 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3593 | BTRFS_I(inode)->dummy_inode = 1; | |
3594 | ||
3595 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3596 | inode->i_op = &simple_dir_inode_operations; | |
3597 | inode->i_fop = &simple_dir_operations; | |
3598 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3599 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3600 | ||
3601 | return inode; | |
3602 | } | |
3603 | ||
3de4586c | 3604 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3605 | { |
d397712b | 3606 | struct inode *inode; |
4df27c4d | 3607 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3608 | struct btrfs_root *sub_root = root; |
3609 | struct btrfs_key location; | |
76dda93c | 3610 | int index; |
5d4f98a2 | 3611 | int ret; |
39279cc3 | 3612 | |
76dda93c YZ |
3613 | dentry->d_op = &btrfs_dentry_operations; |
3614 | ||
39279cc3 CM |
3615 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3616 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3617 | |
39279cc3 | 3618 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3619 | |
39279cc3 CM |
3620 | if (ret < 0) |
3621 | return ERR_PTR(ret); | |
5f39d397 | 3622 | |
4df27c4d YZ |
3623 | if (location.objectid == 0) |
3624 | return NULL; | |
3625 | ||
3626 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3627 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3628 | return inode; | |
3629 | } | |
3630 | ||
3631 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3632 | ||
76dda93c | 3633 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3634 | ret = fixup_tree_root_location(root, dir, dentry, |
3635 | &location, &sub_root); | |
3636 | if (ret < 0) { | |
3637 | if (ret != -ENOENT) | |
3638 | inode = ERR_PTR(ret); | |
3639 | else | |
3640 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3641 | } else { | |
5d4f98a2 | 3642 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3643 | } |
76dda93c YZ |
3644 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3645 | ||
3de4586c CM |
3646 | return inode; |
3647 | } | |
3648 | ||
76dda93c YZ |
3649 | static int btrfs_dentry_delete(struct dentry *dentry) |
3650 | { | |
3651 | struct btrfs_root *root; | |
3652 | ||
efefb143 YZ |
3653 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3654 | dentry = dentry->d_parent; | |
76dda93c | 3655 | |
efefb143 YZ |
3656 | if (dentry->d_inode) { |
3657 | root = BTRFS_I(dentry->d_inode)->root; | |
3658 | if (btrfs_root_refs(&root->root_item) == 0) | |
3659 | return 1; | |
3660 | } | |
76dda93c YZ |
3661 | return 0; |
3662 | } | |
3663 | ||
3de4586c CM |
3664 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3665 | struct nameidata *nd) | |
3666 | { | |
3667 | struct inode *inode; | |
3668 | ||
3de4586c CM |
3669 | inode = btrfs_lookup_dentry(dir, dentry); |
3670 | if (IS_ERR(inode)) | |
3671 | return ERR_CAST(inode); | |
7b128766 | 3672 | |
39279cc3 CM |
3673 | return d_splice_alias(inode, dentry); |
3674 | } | |
3675 | ||
39279cc3 CM |
3676 | static unsigned char btrfs_filetype_table[] = { |
3677 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3678 | }; | |
3679 | ||
cbdf5a24 DW |
3680 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3681 | filldir_t filldir) | |
39279cc3 | 3682 | { |
6da6abae | 3683 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3684 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3685 | struct btrfs_item *item; | |
3686 | struct btrfs_dir_item *di; | |
3687 | struct btrfs_key key; | |
5f39d397 | 3688 | struct btrfs_key found_key; |
39279cc3 CM |
3689 | struct btrfs_path *path; |
3690 | int ret; | |
3691 | u32 nritems; | |
5f39d397 | 3692 | struct extent_buffer *leaf; |
39279cc3 CM |
3693 | int slot; |
3694 | int advance; | |
3695 | unsigned char d_type; | |
3696 | int over = 0; | |
3697 | u32 di_cur; | |
3698 | u32 di_total; | |
3699 | u32 di_len; | |
3700 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3701 | char tmp_name[32]; |
3702 | char *name_ptr; | |
3703 | int name_len; | |
39279cc3 CM |
3704 | |
3705 | /* FIXME, use a real flag for deciding about the key type */ | |
3706 | if (root->fs_info->tree_root == root) | |
3707 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3708 | |
3954401f CM |
3709 | /* special case for "." */ |
3710 | if (filp->f_pos == 0) { | |
3711 | over = filldir(dirent, ".", 1, | |
3712 | 1, inode->i_ino, | |
3713 | DT_DIR); | |
3714 | if (over) | |
3715 | return 0; | |
3716 | filp->f_pos = 1; | |
3717 | } | |
3954401f CM |
3718 | /* special case for .., just use the back ref */ |
3719 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3720 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3721 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3722 | 2, pino, DT_DIR); |
3954401f | 3723 | if (over) |
49593bfa | 3724 | return 0; |
3954401f CM |
3725 | filp->f_pos = 2; |
3726 | } | |
49593bfa DW |
3727 | path = btrfs_alloc_path(); |
3728 | path->reada = 2; | |
3729 | ||
39279cc3 CM |
3730 | btrfs_set_key_type(&key, key_type); |
3731 | key.offset = filp->f_pos; | |
49593bfa | 3732 | key.objectid = inode->i_ino; |
5f39d397 | 3733 | |
39279cc3 CM |
3734 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3735 | if (ret < 0) | |
3736 | goto err; | |
3737 | advance = 0; | |
49593bfa DW |
3738 | |
3739 | while (1) { | |
5f39d397 CM |
3740 | leaf = path->nodes[0]; |
3741 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3742 | slot = path->slots[0]; |
3743 | if (advance || slot >= nritems) { | |
49593bfa | 3744 | if (slot >= nritems - 1) { |
39279cc3 CM |
3745 | ret = btrfs_next_leaf(root, path); |
3746 | if (ret) | |
3747 | break; | |
5f39d397 CM |
3748 | leaf = path->nodes[0]; |
3749 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3750 | slot = path->slots[0]; |
3751 | } else { | |
3752 | slot++; | |
3753 | path->slots[0]++; | |
3754 | } | |
3755 | } | |
3de4586c | 3756 | |
39279cc3 | 3757 | advance = 1; |
5f39d397 CM |
3758 | item = btrfs_item_nr(leaf, slot); |
3759 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3760 | ||
3761 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3762 | break; |
5f39d397 | 3763 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3764 | break; |
5f39d397 | 3765 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3766 | continue; |
5f39d397 CM |
3767 | |
3768 | filp->f_pos = found_key.offset; | |
49593bfa | 3769 | |
39279cc3 CM |
3770 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3771 | di_cur = 0; | |
5f39d397 | 3772 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3773 | |
3774 | while (di_cur < di_total) { | |
5f39d397 CM |
3775 | struct btrfs_key location; |
3776 | ||
3777 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3778 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3779 | name_ptr = tmp_name; |
3780 | } else { | |
3781 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3782 | if (!name_ptr) { |
3783 | ret = -ENOMEM; | |
3784 | goto err; | |
3785 | } | |
5f39d397 CM |
3786 | } |
3787 | read_extent_buffer(leaf, name_ptr, | |
3788 | (unsigned long)(di + 1), name_len); | |
3789 | ||
3790 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3791 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3792 | |
3793 | /* is this a reference to our own snapshot? If so | |
3794 | * skip it | |
3795 | */ | |
3796 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3797 | location.objectid == root->root_key.objectid) { | |
3798 | over = 0; | |
3799 | goto skip; | |
3800 | } | |
5f39d397 | 3801 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3802 | found_key.offset, location.objectid, |
39279cc3 | 3803 | d_type); |
5f39d397 | 3804 | |
3de4586c | 3805 | skip: |
5f39d397 CM |
3806 | if (name_ptr != tmp_name) |
3807 | kfree(name_ptr); | |
3808 | ||
39279cc3 CM |
3809 | if (over) |
3810 | goto nopos; | |
5103e947 | 3811 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3812 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3813 | di_cur += di_len; |
3814 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3815 | } | |
3816 | } | |
49593bfa DW |
3817 | |
3818 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3819 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3820 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3821 | else |
3822 | filp->f_pos++; | |
39279cc3 CM |
3823 | nopos: |
3824 | ret = 0; | |
3825 | err: | |
39279cc3 | 3826 | btrfs_free_path(path); |
39279cc3 CM |
3827 | return ret; |
3828 | } | |
3829 | ||
3830 | int btrfs_write_inode(struct inode *inode, int wait) | |
3831 | { | |
3832 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3833 | struct btrfs_trans_handle *trans; | |
3834 | int ret = 0; | |
3835 | ||
c146afad | 3836 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3837 | return 0; |
3838 | ||
39279cc3 | 3839 | if (wait) { |
f9295749 | 3840 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3841 | btrfs_set_trans_block_group(trans, inode); |
3842 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3843 | } |
3844 | return ret; | |
3845 | } | |
3846 | ||
3847 | /* | |
54aa1f4d | 3848 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3849 | * inode changes. But, it is most likely to find the inode in cache. |
3850 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3851 | * to keep or drop this code. | |
3852 | */ | |
3853 | void btrfs_dirty_inode(struct inode *inode) | |
3854 | { | |
3855 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3856 | struct btrfs_trans_handle *trans; | |
3857 | ||
f9295749 | 3858 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3859 | btrfs_set_trans_block_group(trans, inode); |
3860 | btrfs_update_inode(trans, root, inode); | |
3861 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3862 | } |
3863 | ||
d352ac68 CM |
3864 | /* |
3865 | * find the highest existing sequence number in a directory | |
3866 | * and then set the in-memory index_cnt variable to reflect | |
3867 | * free sequence numbers | |
3868 | */ | |
aec7477b JB |
3869 | static int btrfs_set_inode_index_count(struct inode *inode) |
3870 | { | |
3871 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3872 | struct btrfs_key key, found_key; | |
3873 | struct btrfs_path *path; | |
3874 | struct extent_buffer *leaf; | |
3875 | int ret; | |
3876 | ||
3877 | key.objectid = inode->i_ino; | |
3878 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3879 | key.offset = (u64)-1; | |
3880 | ||
3881 | path = btrfs_alloc_path(); | |
3882 | if (!path) | |
3883 | return -ENOMEM; | |
3884 | ||
3885 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3886 | if (ret < 0) | |
3887 | goto out; | |
3888 | /* FIXME: we should be able to handle this */ | |
3889 | if (ret == 0) | |
3890 | goto out; | |
3891 | ret = 0; | |
3892 | ||
3893 | /* | |
3894 | * MAGIC NUMBER EXPLANATION: | |
3895 | * since we search a directory based on f_pos we have to start at 2 | |
3896 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3897 | * else has to start at 2 | |
3898 | */ | |
3899 | if (path->slots[0] == 0) { | |
3900 | BTRFS_I(inode)->index_cnt = 2; | |
3901 | goto out; | |
3902 | } | |
3903 | ||
3904 | path->slots[0]--; | |
3905 | ||
3906 | leaf = path->nodes[0]; | |
3907 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3908 | ||
3909 | if (found_key.objectid != inode->i_ino || | |
3910 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3911 | BTRFS_I(inode)->index_cnt = 2; | |
3912 | goto out; | |
3913 | } | |
3914 | ||
3915 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3916 | out: | |
3917 | btrfs_free_path(path); | |
3918 | return ret; | |
3919 | } | |
3920 | ||
d352ac68 CM |
3921 | /* |
3922 | * helper to find a free sequence number in a given directory. This current | |
3923 | * code is very simple, later versions will do smarter things in the btree | |
3924 | */ | |
3de4586c | 3925 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3926 | { |
3927 | int ret = 0; | |
3928 | ||
3929 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3930 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3931 | if (ret) |
aec7477b JB |
3932 | return ret; |
3933 | } | |
3934 | ||
00e4e6b3 | 3935 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3936 | BTRFS_I(dir)->index_cnt++; |
3937 | ||
3938 | return ret; | |
3939 | } | |
3940 | ||
39279cc3 CM |
3941 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3942 | struct btrfs_root *root, | |
aec7477b | 3943 | struct inode *dir, |
9c58309d | 3944 | const char *name, int name_len, |
d2fb3437 YZ |
3945 | u64 ref_objectid, u64 objectid, |
3946 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3947 | { |
3948 | struct inode *inode; | |
5f39d397 | 3949 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3950 | struct btrfs_key *location; |
5f39d397 | 3951 | struct btrfs_path *path; |
9c58309d CM |
3952 | struct btrfs_inode_ref *ref; |
3953 | struct btrfs_key key[2]; | |
3954 | u32 sizes[2]; | |
3955 | unsigned long ptr; | |
39279cc3 CM |
3956 | int ret; |
3957 | int owner; | |
3958 | ||
5f39d397 CM |
3959 | path = btrfs_alloc_path(); |
3960 | BUG_ON(!path); | |
3961 | ||
39279cc3 CM |
3962 | inode = new_inode(root->fs_info->sb); |
3963 | if (!inode) | |
3964 | return ERR_PTR(-ENOMEM); | |
3965 | ||
aec7477b | 3966 | if (dir) { |
3de4586c | 3967 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
3968 | if (ret) { |
3969 | iput(inode); | |
aec7477b | 3970 | return ERR_PTR(ret); |
09771430 | 3971 | } |
aec7477b JB |
3972 | } |
3973 | /* | |
3974 | * index_cnt is ignored for everything but a dir, | |
3975 | * btrfs_get_inode_index_count has an explanation for the magic | |
3976 | * number | |
3977 | */ | |
e02119d5 | 3978 | init_btrfs_i(inode); |
aec7477b | 3979 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3980 | BTRFS_I(inode)->root = root; |
e02119d5 | 3981 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 3982 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 3983 | |
39279cc3 CM |
3984 | if (mode & S_IFDIR) |
3985 | owner = 0; | |
3986 | else | |
3987 | owner = 1; | |
d2fb3437 YZ |
3988 | BTRFS_I(inode)->block_group = |
3989 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
3990 | |
3991 | key[0].objectid = objectid; | |
3992 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3993 | key[0].offset = 0; | |
3994 | ||
3995 | key[1].objectid = objectid; | |
3996 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3997 | key[1].offset = ref_objectid; | |
3998 | ||
3999 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4000 | sizes[1] = name_len + sizeof(*ref); | |
4001 | ||
b9473439 | 4002 | path->leave_spinning = 1; |
9c58309d CM |
4003 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4004 | if (ret != 0) | |
5f39d397 CM |
4005 | goto fail; |
4006 | ||
79683f2d | 4007 | inode->i_uid = current_fsuid(); |
8c087b51 | 4008 | |
42f15d77 | 4009 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4010 | inode->i_gid = dir->i_gid; |
4011 | if (S_ISDIR(mode)) | |
4012 | mode |= S_ISGID; | |
4013 | } else | |
4014 | inode->i_gid = current_fsgid(); | |
4015 | ||
39279cc3 CM |
4016 | inode->i_mode = mode; |
4017 | inode->i_ino = objectid; | |
a76a3cd4 | 4018 | inode_set_bytes(inode, 0); |
39279cc3 | 4019 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4020 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4021 | struct btrfs_inode_item); | |
e02119d5 | 4022 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4023 | |
4024 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4025 | struct btrfs_inode_ref); | |
4026 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4027 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4028 | ptr = (unsigned long)(ref + 1); |
4029 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4030 | ||
5f39d397 CM |
4031 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4032 | btrfs_free_path(path); | |
4033 | ||
39279cc3 CM |
4034 | location = &BTRFS_I(inode)->location; |
4035 | location->objectid = objectid; | |
39279cc3 CM |
4036 | location->offset = 0; |
4037 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4038 | ||
6cbff00f CH |
4039 | btrfs_inherit_iflags(inode, dir); |
4040 | ||
94272164 CM |
4041 | if ((mode & S_IFREG)) { |
4042 | if (btrfs_test_opt(root, NODATASUM)) | |
4043 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4044 | if (btrfs_test_opt(root, NODATACOW)) | |
4045 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4046 | } | |
4047 | ||
39279cc3 | 4048 | insert_inode_hash(inode); |
5d4f98a2 | 4049 | inode_tree_add(inode); |
39279cc3 | 4050 | return inode; |
5f39d397 | 4051 | fail: |
aec7477b JB |
4052 | if (dir) |
4053 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4054 | btrfs_free_path(path); |
09771430 | 4055 | iput(inode); |
5f39d397 | 4056 | return ERR_PTR(ret); |
39279cc3 CM |
4057 | } |
4058 | ||
4059 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4060 | { | |
4061 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4062 | } | |
4063 | ||
d352ac68 CM |
4064 | /* |
4065 | * utility function to add 'inode' into 'parent_inode' with | |
4066 | * a give name and a given sequence number. | |
4067 | * if 'add_backref' is true, also insert a backref from the | |
4068 | * inode to the parent directory. | |
4069 | */ | |
e02119d5 CM |
4070 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4071 | struct inode *parent_inode, struct inode *inode, | |
4072 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4073 | { |
4df27c4d | 4074 | int ret = 0; |
39279cc3 | 4075 | struct btrfs_key key; |
e02119d5 | 4076 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4077 | |
4df27c4d YZ |
4078 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4079 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4080 | } else { | |
4081 | key.objectid = inode->i_ino; | |
4082 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4083 | key.offset = 0; | |
4084 | } | |
4085 | ||
4086 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4087 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4088 | key.objectid, root->root_key.objectid, | |
4089 | parent_inode->i_ino, | |
4090 | index, name, name_len); | |
4091 | } else if (add_backref) { | |
4092 | ret = btrfs_insert_inode_ref(trans, root, | |
4093 | name, name_len, inode->i_ino, | |
4094 | parent_inode->i_ino, index); | |
4095 | } | |
39279cc3 | 4096 | |
39279cc3 | 4097 | if (ret == 0) { |
4df27c4d YZ |
4098 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4099 | parent_inode->i_ino, &key, | |
4100 | btrfs_inode_type(inode), index); | |
4101 | BUG_ON(ret); | |
4102 | ||
dbe674a9 | 4103 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4104 | name_len * 2); |
79c44584 | 4105 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4106 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4107 | } |
4108 | return ret; | |
4109 | } | |
4110 | ||
4111 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4112 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4113 | int backref, u64 index) |
39279cc3 | 4114 | { |
e02119d5 CM |
4115 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4116 | inode, dentry->d_name.name, | |
4117 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4118 | if (!err) { |
4119 | d_instantiate(dentry, inode); | |
4120 | return 0; | |
4121 | } | |
4122 | if (err > 0) | |
4123 | err = -EEXIST; | |
4124 | return err; | |
4125 | } | |
4126 | ||
618e21d5 JB |
4127 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4128 | int mode, dev_t rdev) | |
4129 | { | |
4130 | struct btrfs_trans_handle *trans; | |
4131 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4132 | struct inode *inode = NULL; |
618e21d5 JB |
4133 | int err; |
4134 | int drop_inode = 0; | |
4135 | u64 objectid; | |
1832a6d5 | 4136 | unsigned long nr = 0; |
00e4e6b3 | 4137 | u64 index = 0; |
618e21d5 JB |
4138 | |
4139 | if (!new_valid_dev(rdev)) | |
4140 | return -EINVAL; | |
4141 | ||
9ed74f2d JB |
4142 | /* |
4143 | * 2 for inode item and ref | |
4144 | * 2 for dir items | |
4145 | * 1 for xattr if selinux is on | |
4146 | */ | |
4147 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4148 | if (err) |
9ed74f2d | 4149 | return err; |
1832a6d5 | 4150 | |
618e21d5 | 4151 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4152 | if (!trans) |
4153 | goto fail; | |
618e21d5 JB |
4154 | btrfs_set_trans_block_group(trans, dir); |
4155 | ||
4156 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4157 | if (err) { | |
4158 | err = -ENOSPC; | |
4159 | goto out_unlock; | |
4160 | } | |
4161 | ||
aec7477b | 4162 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4163 | dentry->d_name.len, |
4164 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4165 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4166 | err = PTR_ERR(inode); |
4167 | if (IS_ERR(inode)) | |
4168 | goto out_unlock; | |
4169 | ||
0279b4cd | 4170 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4171 | if (err) { |
4172 | drop_inode = 1; | |
4173 | goto out_unlock; | |
4174 | } | |
4175 | ||
618e21d5 | 4176 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4177 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4178 | if (err) |
4179 | drop_inode = 1; | |
4180 | else { | |
4181 | inode->i_op = &btrfs_special_inode_operations; | |
4182 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4183 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4184 | } |
618e21d5 JB |
4185 | btrfs_update_inode_block_group(trans, inode); |
4186 | btrfs_update_inode_block_group(trans, dir); | |
4187 | out_unlock: | |
d3c2fdcf | 4188 | nr = trans->blocks_used; |
89ce8a63 | 4189 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4190 | fail: |
9ed74f2d | 4191 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4192 | if (drop_inode) { |
4193 | inode_dec_link_count(inode); | |
4194 | iput(inode); | |
4195 | } | |
d3c2fdcf | 4196 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4197 | return err; |
4198 | } | |
4199 | ||
39279cc3 CM |
4200 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4201 | int mode, struct nameidata *nd) | |
4202 | { | |
4203 | struct btrfs_trans_handle *trans; | |
4204 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4205 | struct inode *inode = NULL; |
39279cc3 CM |
4206 | int err; |
4207 | int drop_inode = 0; | |
1832a6d5 | 4208 | unsigned long nr = 0; |
39279cc3 | 4209 | u64 objectid; |
00e4e6b3 | 4210 | u64 index = 0; |
39279cc3 | 4211 | |
9ed74f2d JB |
4212 | /* |
4213 | * 2 for inode item and ref | |
4214 | * 2 for dir items | |
4215 | * 1 for xattr if selinux is on | |
4216 | */ | |
4217 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4218 | if (err) |
9ed74f2d JB |
4219 | return err; |
4220 | ||
39279cc3 | 4221 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4222 | if (!trans) |
4223 | goto fail; | |
39279cc3 CM |
4224 | btrfs_set_trans_block_group(trans, dir); |
4225 | ||
4226 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4227 | if (err) { | |
4228 | err = -ENOSPC; | |
4229 | goto out_unlock; | |
4230 | } | |
4231 | ||
aec7477b | 4232 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4233 | dentry->d_name.len, |
4234 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4235 | objectid, BTRFS_I(dir)->block_group, mode, |
4236 | &index); | |
39279cc3 CM |
4237 | err = PTR_ERR(inode); |
4238 | if (IS_ERR(inode)) | |
4239 | goto out_unlock; | |
4240 | ||
0279b4cd | 4241 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4242 | if (err) { |
4243 | drop_inode = 1; | |
4244 | goto out_unlock; | |
4245 | } | |
4246 | ||
39279cc3 | 4247 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4248 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4249 | if (err) |
4250 | drop_inode = 1; | |
4251 | else { | |
4252 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4253 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4254 | inode->i_fop = &btrfs_file_operations; |
4255 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4256 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4257 | } |
39279cc3 CM |
4258 | btrfs_update_inode_block_group(trans, inode); |
4259 | btrfs_update_inode_block_group(trans, dir); | |
4260 | out_unlock: | |
d3c2fdcf | 4261 | nr = trans->blocks_used; |
ab78c84d | 4262 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4263 | fail: |
9ed74f2d | 4264 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4265 | if (drop_inode) { |
4266 | inode_dec_link_count(inode); | |
4267 | iput(inode); | |
4268 | } | |
d3c2fdcf | 4269 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4270 | return err; |
4271 | } | |
4272 | ||
4273 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4274 | struct dentry *dentry) | |
4275 | { | |
4276 | struct btrfs_trans_handle *trans; | |
4277 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4278 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4279 | u64 index; |
1832a6d5 | 4280 | unsigned long nr = 0; |
39279cc3 CM |
4281 | int err; |
4282 | int drop_inode = 0; | |
4283 | ||
4284 | if (inode->i_nlink == 0) | |
4285 | return -ENOENT; | |
4286 | ||
9ed74f2d JB |
4287 | /* |
4288 | * 1 item for inode ref | |
4289 | * 2 items for dir items | |
4290 | */ | |
4291 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4292 | if (err) |
9ed74f2d JB |
4293 | return err; |
4294 | ||
4295 | btrfs_inc_nlink(inode); | |
4296 | ||
3de4586c | 4297 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4298 | if (err) |
4299 | goto fail; | |
4300 | ||
39279cc3 | 4301 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4302 | |
39279cc3 CM |
4303 | btrfs_set_trans_block_group(trans, dir); |
4304 | atomic_inc(&inode->i_count); | |
aec7477b | 4305 | |
00e4e6b3 | 4306 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4307 | |
a5719521 | 4308 | if (err) { |
54aa1f4d | 4309 | drop_inode = 1; |
a5719521 YZ |
4310 | } else { |
4311 | btrfs_update_inode_block_group(trans, dir); | |
4312 | err = btrfs_update_inode(trans, root, inode); | |
4313 | BUG_ON(err); | |
4314 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4315 | } | |
39279cc3 | 4316 | |
d3c2fdcf | 4317 | nr = trans->blocks_used; |
ab78c84d | 4318 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4319 | fail: |
9ed74f2d | 4320 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4321 | if (drop_inode) { |
4322 | inode_dec_link_count(inode); | |
4323 | iput(inode); | |
4324 | } | |
d3c2fdcf | 4325 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4326 | return err; |
4327 | } | |
4328 | ||
39279cc3 CM |
4329 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4330 | { | |
b9d86667 | 4331 | struct inode *inode = NULL; |
39279cc3 CM |
4332 | struct btrfs_trans_handle *trans; |
4333 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4334 | int err = 0; | |
4335 | int drop_on_err = 0; | |
b9d86667 | 4336 | u64 objectid = 0; |
00e4e6b3 | 4337 | u64 index = 0; |
d3c2fdcf | 4338 | unsigned long nr = 1; |
39279cc3 | 4339 | |
9ed74f2d JB |
4340 | /* |
4341 | * 2 items for inode and ref | |
4342 | * 2 items for dir items | |
4343 | * 1 for xattr if selinux is on | |
4344 | */ | |
4345 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4346 | if (err) |
9ed74f2d | 4347 | return err; |
1832a6d5 | 4348 | |
39279cc3 | 4349 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4350 | if (!trans) { |
4351 | err = -ENOMEM; | |
39279cc3 CM |
4352 | goto out_unlock; |
4353 | } | |
9ed74f2d | 4354 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4355 | |
4356 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4357 | if (err) { | |
4358 | err = -ENOSPC; | |
4359 | goto out_unlock; | |
4360 | } | |
4361 | ||
aec7477b | 4362 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4363 | dentry->d_name.len, |
4364 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4365 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4366 | &index); | |
39279cc3 CM |
4367 | if (IS_ERR(inode)) { |
4368 | err = PTR_ERR(inode); | |
4369 | goto out_fail; | |
4370 | } | |
5f39d397 | 4371 | |
39279cc3 | 4372 | drop_on_err = 1; |
33268eaf | 4373 | |
0279b4cd | 4374 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4375 | if (err) |
4376 | goto out_fail; | |
4377 | ||
39279cc3 CM |
4378 | inode->i_op = &btrfs_dir_inode_operations; |
4379 | inode->i_fop = &btrfs_dir_file_operations; | |
4380 | btrfs_set_trans_block_group(trans, inode); | |
4381 | ||
dbe674a9 | 4382 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4383 | err = btrfs_update_inode(trans, root, inode); |
4384 | if (err) | |
4385 | goto out_fail; | |
5f39d397 | 4386 | |
e02119d5 CM |
4387 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4388 | inode, dentry->d_name.name, | |
4389 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4390 | if (err) |
4391 | goto out_fail; | |
5f39d397 | 4392 | |
39279cc3 CM |
4393 | d_instantiate(dentry, inode); |
4394 | drop_on_err = 0; | |
39279cc3 CM |
4395 | btrfs_update_inode_block_group(trans, inode); |
4396 | btrfs_update_inode_block_group(trans, dir); | |
4397 | ||
4398 | out_fail: | |
d3c2fdcf | 4399 | nr = trans->blocks_used; |
ab78c84d | 4400 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4401 | |
39279cc3 | 4402 | out_unlock: |
9ed74f2d | 4403 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4404 | if (drop_on_err) |
4405 | iput(inode); | |
d3c2fdcf | 4406 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4407 | return err; |
4408 | } | |
4409 | ||
d352ac68 CM |
4410 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4411 | * and an extent that you want to insert, deal with overlap and insert | |
4412 | * the new extent into the tree. | |
4413 | */ | |
3b951516 CM |
4414 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4415 | struct extent_map *existing, | |
e6dcd2dc CM |
4416 | struct extent_map *em, |
4417 | u64 map_start, u64 map_len) | |
3b951516 CM |
4418 | { |
4419 | u64 start_diff; | |
3b951516 | 4420 | |
e6dcd2dc CM |
4421 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4422 | start_diff = map_start - em->start; | |
4423 | em->start = map_start; | |
4424 | em->len = map_len; | |
c8b97818 CM |
4425 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4426 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4427 | em->block_start += start_diff; |
c8b97818 CM |
4428 | em->block_len -= start_diff; |
4429 | } | |
e6dcd2dc | 4430 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4431 | } |
4432 | ||
c8b97818 CM |
4433 | static noinline int uncompress_inline(struct btrfs_path *path, |
4434 | struct inode *inode, struct page *page, | |
4435 | size_t pg_offset, u64 extent_offset, | |
4436 | struct btrfs_file_extent_item *item) | |
4437 | { | |
4438 | int ret; | |
4439 | struct extent_buffer *leaf = path->nodes[0]; | |
4440 | char *tmp; | |
4441 | size_t max_size; | |
4442 | unsigned long inline_size; | |
4443 | unsigned long ptr; | |
4444 | ||
4445 | WARN_ON(pg_offset != 0); | |
4446 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4447 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4448 | btrfs_item_nr(leaf, path->slots[0])); | |
4449 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4450 | ptr = btrfs_file_extent_inline_start(item); | |
4451 | ||
4452 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4453 | ||
5b050f04 | 4454 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4455 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4456 | inline_size, max_size); | |
4457 | if (ret) { | |
4458 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4459 | unsigned long copy_size = min_t(u64, | |
4460 | PAGE_CACHE_SIZE - pg_offset, | |
4461 | max_size - extent_offset); | |
4462 | memset(kaddr + pg_offset, 0, copy_size); | |
4463 | kunmap_atomic(kaddr, KM_USER0); | |
4464 | } | |
4465 | kfree(tmp); | |
4466 | return 0; | |
4467 | } | |
4468 | ||
d352ac68 CM |
4469 | /* |
4470 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4471 | * the ugly parts come from merging extents from the disk with the in-ram |
4472 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4473 | * where the in-ram extents might be locked pending data=ordered completion. |
4474 | * | |
4475 | * This also copies inline extents directly into the page. | |
4476 | */ | |
d397712b | 4477 | |
a52d9a80 | 4478 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4479 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4480 | int create) |
4481 | { | |
4482 | int ret; | |
4483 | int err = 0; | |
db94535d | 4484 | u64 bytenr; |
a52d9a80 CM |
4485 | u64 extent_start = 0; |
4486 | u64 extent_end = 0; | |
4487 | u64 objectid = inode->i_ino; | |
4488 | u32 found_type; | |
f421950f | 4489 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4490 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4491 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4492 | struct extent_buffer *leaf; |
4493 | struct btrfs_key found_key; | |
a52d9a80 CM |
4494 | struct extent_map *em = NULL; |
4495 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4496 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4497 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4498 | int compressed; |
a52d9a80 | 4499 | |
a52d9a80 | 4500 | again: |
890871be | 4501 | read_lock(&em_tree->lock); |
d1310b2e | 4502 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4503 | if (em) |
4504 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4505 | read_unlock(&em_tree->lock); |
d1310b2e | 4506 | |
a52d9a80 | 4507 | if (em) { |
e1c4b745 CM |
4508 | if (em->start > start || em->start + em->len <= start) |
4509 | free_extent_map(em); | |
4510 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4511 | free_extent_map(em); |
4512 | else | |
4513 | goto out; | |
a52d9a80 | 4514 | } |
d1310b2e | 4515 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4516 | if (!em) { |
d1310b2e CM |
4517 | err = -ENOMEM; |
4518 | goto out; | |
a52d9a80 | 4519 | } |
e6dcd2dc | 4520 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4521 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4522 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4523 | em->len = (u64)-1; |
c8b97818 | 4524 | em->block_len = (u64)-1; |
f421950f CM |
4525 | |
4526 | if (!path) { | |
4527 | path = btrfs_alloc_path(); | |
4528 | BUG_ON(!path); | |
4529 | } | |
4530 | ||
179e29e4 CM |
4531 | ret = btrfs_lookup_file_extent(trans, root, path, |
4532 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4533 | if (ret < 0) { |
4534 | err = ret; | |
4535 | goto out; | |
4536 | } | |
4537 | ||
4538 | if (ret != 0) { | |
4539 | if (path->slots[0] == 0) | |
4540 | goto not_found; | |
4541 | path->slots[0]--; | |
4542 | } | |
4543 | ||
5f39d397 CM |
4544 | leaf = path->nodes[0]; |
4545 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4546 | struct btrfs_file_extent_item); |
a52d9a80 | 4547 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4548 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4549 | found_type = btrfs_key_type(&found_key); | |
4550 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4551 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4552 | goto not_found; | |
4553 | } | |
4554 | ||
5f39d397 CM |
4555 | found_type = btrfs_file_extent_type(leaf, item); |
4556 | extent_start = found_key.offset; | |
c8b97818 | 4557 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4558 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4559 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4560 | extent_end = extent_start + |
db94535d | 4561 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4562 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4563 | size_t size; | |
4564 | size = btrfs_file_extent_inline_len(leaf, item); | |
4565 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4566 | ~((u64)root->sectorsize - 1); | |
4567 | } | |
4568 | ||
4569 | if (start >= extent_end) { | |
4570 | path->slots[0]++; | |
4571 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4572 | ret = btrfs_next_leaf(root, path); | |
4573 | if (ret < 0) { | |
4574 | err = ret; | |
4575 | goto out; | |
a52d9a80 | 4576 | } |
9036c102 YZ |
4577 | if (ret > 0) |
4578 | goto not_found; | |
4579 | leaf = path->nodes[0]; | |
a52d9a80 | 4580 | } |
9036c102 YZ |
4581 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4582 | if (found_key.objectid != objectid || | |
4583 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4584 | goto not_found; | |
4585 | if (start + len <= found_key.offset) | |
4586 | goto not_found; | |
4587 | em->start = start; | |
4588 | em->len = found_key.offset - start; | |
4589 | goto not_found_em; | |
4590 | } | |
4591 | ||
d899e052 YZ |
4592 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4593 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4594 | em->start = extent_start; |
4595 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4596 | em->orig_start = extent_start - |
4597 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4598 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4599 | if (bytenr == 0) { | |
5f39d397 | 4600 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4601 | goto insert; |
4602 | } | |
c8b97818 CM |
4603 | if (compressed) { |
4604 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4605 | em->block_start = bytenr; | |
4606 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4607 | item); | |
4608 | } else { | |
4609 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4610 | em->block_start = bytenr; | |
4611 | em->block_len = em->len; | |
d899e052 YZ |
4612 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4613 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4614 | } |
a52d9a80 CM |
4615 | goto insert; |
4616 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4617 | unsigned long ptr; |
a52d9a80 | 4618 | char *map; |
3326d1b0 CM |
4619 | size_t size; |
4620 | size_t extent_offset; | |
4621 | size_t copy_size; | |
a52d9a80 | 4622 | |
689f9346 | 4623 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4624 | if (!page || create) { |
689f9346 | 4625 | em->start = extent_start; |
9036c102 | 4626 | em->len = extent_end - extent_start; |
689f9346 Y |
4627 | goto out; |
4628 | } | |
5f39d397 | 4629 | |
9036c102 YZ |
4630 | size = btrfs_file_extent_inline_len(leaf, item); |
4631 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4632 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4633 | size - extent_offset); |
3326d1b0 | 4634 | em->start = extent_start + extent_offset; |
70dec807 CM |
4635 | em->len = (copy_size + root->sectorsize - 1) & |
4636 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4637 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4638 | if (compressed) |
4639 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4640 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4641 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4642 | if (btrfs_file_extent_compression(leaf, item) == |
4643 | BTRFS_COMPRESS_ZLIB) { | |
4644 | ret = uncompress_inline(path, inode, page, | |
4645 | pg_offset, | |
4646 | extent_offset, item); | |
4647 | BUG_ON(ret); | |
4648 | } else { | |
4649 | map = kmap(page); | |
4650 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4651 | copy_size); | |
93c82d57 CM |
4652 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4653 | memset(map + pg_offset + copy_size, 0, | |
4654 | PAGE_CACHE_SIZE - pg_offset - | |
4655 | copy_size); | |
4656 | } | |
c8b97818 CM |
4657 | kunmap(page); |
4658 | } | |
179e29e4 CM |
4659 | flush_dcache_page(page); |
4660 | } else if (create && PageUptodate(page)) { | |
4661 | if (!trans) { | |
4662 | kunmap(page); | |
4663 | free_extent_map(em); | |
4664 | em = NULL; | |
4665 | btrfs_release_path(root, path); | |
f9295749 | 4666 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4667 | goto again; |
4668 | } | |
c8b97818 | 4669 | map = kmap(page); |
70dec807 | 4670 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4671 | copy_size); |
c8b97818 | 4672 | kunmap(page); |
179e29e4 | 4673 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4674 | } |
d1310b2e CM |
4675 | set_extent_uptodate(io_tree, em->start, |
4676 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4677 | goto insert; |
4678 | } else { | |
d397712b | 4679 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4680 | WARN_ON(1); |
4681 | } | |
4682 | not_found: | |
4683 | em->start = start; | |
d1310b2e | 4684 | em->len = len; |
a52d9a80 | 4685 | not_found_em: |
5f39d397 | 4686 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4687 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4688 | insert: |
4689 | btrfs_release_path(root, path); | |
d1310b2e | 4690 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4691 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4692 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4693 | (unsigned long long)em->len, | |
4694 | (unsigned long long)start, | |
4695 | (unsigned long long)len); | |
a52d9a80 CM |
4696 | err = -EIO; |
4697 | goto out; | |
4698 | } | |
d1310b2e CM |
4699 | |
4700 | err = 0; | |
890871be | 4701 | write_lock(&em_tree->lock); |
a52d9a80 | 4702 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4703 | /* it is possible that someone inserted the extent into the tree |
4704 | * while we had the lock dropped. It is also possible that | |
4705 | * an overlapping map exists in the tree | |
4706 | */ | |
a52d9a80 | 4707 | if (ret == -EEXIST) { |
3b951516 | 4708 | struct extent_map *existing; |
e6dcd2dc CM |
4709 | |
4710 | ret = 0; | |
4711 | ||
3b951516 | 4712 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4713 | if (existing && (existing->start > start || |
4714 | existing->start + existing->len <= start)) { | |
4715 | free_extent_map(existing); | |
4716 | existing = NULL; | |
4717 | } | |
3b951516 CM |
4718 | if (!existing) { |
4719 | existing = lookup_extent_mapping(em_tree, em->start, | |
4720 | em->len); | |
4721 | if (existing) { | |
4722 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4723 | em, start, |
4724 | root->sectorsize); | |
3b951516 CM |
4725 | free_extent_map(existing); |
4726 | if (err) { | |
4727 | free_extent_map(em); | |
4728 | em = NULL; | |
4729 | } | |
4730 | } else { | |
4731 | err = -EIO; | |
3b951516 CM |
4732 | free_extent_map(em); |
4733 | em = NULL; | |
4734 | } | |
4735 | } else { | |
4736 | free_extent_map(em); | |
4737 | em = existing; | |
e6dcd2dc | 4738 | err = 0; |
a52d9a80 | 4739 | } |
a52d9a80 | 4740 | } |
890871be | 4741 | write_unlock(&em_tree->lock); |
a52d9a80 | 4742 | out: |
f421950f CM |
4743 | if (path) |
4744 | btrfs_free_path(path); | |
a52d9a80 CM |
4745 | if (trans) { |
4746 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4747 | if (!err) |
a52d9a80 CM |
4748 | err = ret; |
4749 | } | |
a52d9a80 CM |
4750 | if (err) { |
4751 | free_extent_map(em); | |
a52d9a80 CM |
4752 | return ERR_PTR(err); |
4753 | } | |
4754 | return em; | |
4755 | } | |
4756 | ||
16432985 CM |
4757 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4758 | const struct iovec *iov, loff_t offset, | |
4759 | unsigned long nr_segs) | |
4760 | { | |
e1c4b745 | 4761 | return -EINVAL; |
16432985 CM |
4762 | } |
4763 | ||
1506fcc8 YS |
4764 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4765 | __u64 start, __u64 len) | |
4766 | { | |
4767 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4768 | } | |
4769 | ||
a52d9a80 | 4770 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4771 | { |
d1310b2e CM |
4772 | struct extent_io_tree *tree; |
4773 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4774 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4775 | } |
1832a6d5 | 4776 | |
a52d9a80 | 4777 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4778 | { |
d1310b2e | 4779 | struct extent_io_tree *tree; |
b888db2b CM |
4780 | |
4781 | ||
4782 | if (current->flags & PF_MEMALLOC) { | |
4783 | redirty_page_for_writepage(wbc, page); | |
4784 | unlock_page(page); | |
4785 | return 0; | |
4786 | } | |
d1310b2e | 4787 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4788 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4789 | } |
4790 | ||
f421950f CM |
4791 | int btrfs_writepages(struct address_space *mapping, |
4792 | struct writeback_control *wbc) | |
b293f02e | 4793 | { |
d1310b2e | 4794 | struct extent_io_tree *tree; |
771ed689 | 4795 | |
d1310b2e | 4796 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4797 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4798 | } | |
4799 | ||
3ab2fb5a CM |
4800 | static int |
4801 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4802 | struct list_head *pages, unsigned nr_pages) | |
4803 | { | |
d1310b2e CM |
4804 | struct extent_io_tree *tree; |
4805 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4806 | return extent_readpages(tree, mapping, pages, nr_pages, |
4807 | btrfs_get_extent); | |
4808 | } | |
e6dcd2dc | 4809 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4810 | { |
d1310b2e CM |
4811 | struct extent_io_tree *tree; |
4812 | struct extent_map_tree *map; | |
a52d9a80 | 4813 | int ret; |
8c2383c3 | 4814 | |
d1310b2e CM |
4815 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4816 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4817 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4818 | if (ret == 1) { |
4819 | ClearPagePrivate(page); | |
4820 | set_page_private(page, 0); | |
4821 | page_cache_release(page); | |
39279cc3 | 4822 | } |
a52d9a80 | 4823 | return ret; |
39279cc3 CM |
4824 | } |
4825 | ||
e6dcd2dc CM |
4826 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4827 | { | |
98509cfc CM |
4828 | if (PageWriteback(page) || PageDirty(page)) |
4829 | return 0; | |
b335b003 | 4830 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4831 | } |
4832 | ||
a52d9a80 | 4833 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4834 | { |
d1310b2e | 4835 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4836 | struct btrfs_ordered_extent *ordered; |
4837 | u64 page_start = page_offset(page); | |
4838 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4839 | |
8b62b72b CM |
4840 | |
4841 | /* | |
4842 | * we have the page locked, so new writeback can't start, | |
4843 | * and the dirty bit won't be cleared while we are here. | |
4844 | * | |
4845 | * Wait for IO on this page so that we can safely clear | |
4846 | * the PagePrivate2 bit and do ordered accounting | |
4847 | */ | |
e6dcd2dc | 4848 | wait_on_page_writeback(page); |
8b62b72b | 4849 | |
d1310b2e | 4850 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4851 | if (offset) { |
4852 | btrfs_releasepage(page, GFP_NOFS); | |
4853 | return; | |
4854 | } | |
e6dcd2dc CM |
4855 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4856 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4857 | page_offset(page)); | |
4858 | if (ordered) { | |
eb84ae03 CM |
4859 | /* |
4860 | * IO on this page will never be started, so we need | |
4861 | * to account for any ordered extents now | |
4862 | */ | |
e6dcd2dc CM |
4863 | clear_extent_bit(tree, page_start, page_end, |
4864 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff JB |
4865 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
4866 | NULL, GFP_NOFS); | |
8b62b72b CM |
4867 | /* |
4868 | * whoever cleared the private bit is responsible | |
4869 | * for the finish_ordered_io | |
4870 | */ | |
4871 | if (TestClearPagePrivate2(page)) { | |
4872 | btrfs_finish_ordered_io(page->mapping->host, | |
4873 | page_start, page_end); | |
4874 | } | |
e6dcd2dc CM |
4875 | btrfs_put_ordered_extent(ordered); |
4876 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4877 | } | |
4878 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff JB |
4879 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4880 | EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); | |
e6dcd2dc CM |
4881 | __btrfs_releasepage(page, GFP_NOFS); |
4882 | ||
4a096752 | 4883 | ClearPageChecked(page); |
9ad6b7bc | 4884 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4885 | ClearPagePrivate(page); |
4886 | set_page_private(page, 0); | |
4887 | page_cache_release(page); | |
4888 | } | |
39279cc3 CM |
4889 | } |
4890 | ||
9ebefb18 CM |
4891 | /* |
4892 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4893 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4894 | * be careful to check for EOF conditions here. We set the page up correctly | |
4895 | * for a written page which means we get ENOSPC checking when writing into | |
4896 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4897 | * support these features. | |
4898 | * | |
4899 | * We are not allowed to take the i_mutex here so we have to play games to | |
4900 | * protect against truncate races as the page could now be beyond EOF. Because | |
4901 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4902 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4903 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4904 | * unlock the page. | |
4905 | */ | |
c2ec175c | 4906 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4907 | { |
c2ec175c | 4908 | struct page *page = vmf->page; |
6da6abae | 4909 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4910 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4911 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4912 | struct btrfs_ordered_extent *ordered; | |
4913 | char *kaddr; | |
4914 | unsigned long zero_start; | |
9ebefb18 | 4915 | loff_t size; |
1832a6d5 | 4916 | int ret; |
a52d9a80 | 4917 | u64 page_start; |
e6dcd2dc | 4918 | u64 page_end; |
9ebefb18 | 4919 | |
6a63209f | 4920 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4921 | if (ret) { |
4922 | if (ret == -ENOMEM) | |
4923 | ret = VM_FAULT_OOM; | |
4924 | else /* -ENOSPC, -EIO, etc */ | |
4925 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4926 | goto out; |
56a76f82 | 4927 | } |
1832a6d5 | 4928 | |
9ed74f2d JB |
4929 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
4930 | if (ret) { | |
4931 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
4932 | ret = VM_FAULT_SIGBUS; | |
4933 | goto out; | |
4934 | } | |
4935 | ||
56a76f82 | 4936 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4937 | again: |
9ebefb18 | 4938 | lock_page(page); |
9ebefb18 | 4939 | size = i_size_read(inode); |
e6dcd2dc CM |
4940 | page_start = page_offset(page); |
4941 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4942 | |
9ebefb18 | 4943 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4944 | (page_start >= size)) { |
6a63209f | 4945 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
4946 | /* page got truncated out from underneath us */ |
4947 | goto out_unlock; | |
4948 | } | |
e6dcd2dc CM |
4949 | wait_on_page_writeback(page); |
4950 | ||
4951 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4952 | set_page_extent_mapped(page); | |
4953 | ||
eb84ae03 CM |
4954 | /* |
4955 | * we can't set the delalloc bits if there are pending ordered | |
4956 | * extents. Drop our locks and wait for them to finish | |
4957 | */ | |
e6dcd2dc CM |
4958 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4959 | if (ordered) { | |
4960 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4961 | unlock_page(page); | |
eb84ae03 | 4962 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4963 | btrfs_put_ordered_extent(ordered); |
4964 | goto again; | |
4965 | } | |
4966 | ||
fbf19087 JB |
4967 | /* |
4968 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
4969 | * if it was already dirty, so for space accounting reasons we need to | |
4970 | * clear any delalloc bits for the range we are fixing to save. There | |
4971 | * is probably a better way to do this, but for now keep consistent with | |
4972 | * prepare_pages in the normal write path. | |
4973 | */ | |
4974 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
32c00aff JB |
4975 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
4976 | GFP_NOFS); | |
fbf19087 | 4977 | |
9ed74f2d JB |
4978 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
4979 | if (ret) { | |
4980 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4981 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 4982 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
4983 | goto out_unlock; |
4984 | } | |
e6dcd2dc | 4985 | ret = 0; |
9ebefb18 CM |
4986 | |
4987 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4988 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4989 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4990 | else |
e6dcd2dc | 4991 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4992 | |
e6dcd2dc CM |
4993 | if (zero_start != PAGE_CACHE_SIZE) { |
4994 | kaddr = kmap(page); | |
4995 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4996 | flush_dcache_page(page); | |
4997 | kunmap(page); | |
4998 | } | |
247e743c | 4999 | ClearPageChecked(page); |
e6dcd2dc | 5000 | set_page_dirty(page); |
50a9b214 | 5001 | SetPageUptodate(page); |
5a3f23d5 | 5002 | |
257c62e1 CM |
5003 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5004 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5005 | ||
e6dcd2dc | 5006 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
5007 | |
5008 | out_unlock: | |
9ed74f2d | 5009 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
5010 | if (!ret) |
5011 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5012 | unlock_page(page); |
1832a6d5 | 5013 | out: |
9ebefb18 CM |
5014 | return ret; |
5015 | } | |
5016 | ||
39279cc3 CM |
5017 | static void btrfs_truncate(struct inode *inode) |
5018 | { | |
5019 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5020 | int ret; | |
5021 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5022 | unsigned long nr; |
dbe674a9 | 5023 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
5024 | |
5025 | if (!S_ISREG(inode->i_mode)) | |
5026 | return; | |
5027 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
5028 | return; | |
5029 | ||
5d5e103a JB |
5030 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5031 | if (ret) | |
5032 | return; | |
4a096752 | 5033 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 5034 | |
39279cc3 | 5035 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
5036 | |
5037 | /* | |
5038 | * setattr is responsible for setting the ordered_data_close flag, | |
5039 | * but that is only tested during the last file release. That | |
5040 | * could happen well after the next commit, leaving a great big | |
5041 | * window where new writes may get lost if someone chooses to write | |
5042 | * to this file after truncating to zero | |
5043 | * | |
5044 | * The inode doesn't have any dirty data here, and so if we commit | |
5045 | * this is a noop. If someone immediately starts writing to the inode | |
5046 | * it is very likely we'll catch some of their writes in this | |
5047 | * transaction, and the commit will find this file on the ordered | |
5048 | * data list with good things to send down. | |
5049 | * | |
5050 | * This is a best effort solution, there is still a window where | |
5051 | * using truncate to replace the contents of the file will | |
5052 | * end up with a zero length file after a crash. | |
5053 | */ | |
5054 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5055 | btrfs_add_ordered_operation(trans, root, inode); | |
5056 | ||
39279cc3 | 5057 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 5058 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 5059 | |
7b128766 JB |
5060 | ret = btrfs_orphan_add(trans, inode); |
5061 | if (ret) | |
5062 | goto out; | |
39279cc3 | 5063 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 5064 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 5065 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 5066 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 5067 | |
7b128766 JB |
5068 | ret = btrfs_orphan_del(trans, inode); |
5069 | BUG_ON(ret); | |
5070 | ||
5071 | out: | |
5072 | nr = trans->blocks_used; | |
89ce8a63 | 5073 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5074 | BUG_ON(ret); |
d3c2fdcf | 5075 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5076 | } |
5077 | ||
d352ac68 CM |
5078 | /* |
5079 | * create a new subvolume directory/inode (helper for the ioctl). | |
5080 | */ | |
d2fb3437 | 5081 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5082 | struct btrfs_root *new_root, |
d2fb3437 | 5083 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5084 | { |
39279cc3 | 5085 | struct inode *inode; |
76dda93c | 5086 | int err; |
00e4e6b3 | 5087 | u64 index = 0; |
39279cc3 | 5088 | |
aec7477b | 5089 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5090 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5091 | if (IS_ERR(inode)) |
f46b5a66 | 5092 | return PTR_ERR(inode); |
39279cc3 CM |
5093 | inode->i_op = &btrfs_dir_inode_operations; |
5094 | inode->i_fop = &btrfs_dir_file_operations; | |
5095 | ||
39279cc3 | 5096 | inode->i_nlink = 1; |
dbe674a9 | 5097 | btrfs_i_size_write(inode, 0); |
3b96362c | 5098 | |
76dda93c YZ |
5099 | err = btrfs_update_inode(trans, new_root, inode); |
5100 | BUG_ON(err); | |
cb8e7090 | 5101 | |
76dda93c | 5102 | iput(inode); |
cb8e7090 | 5103 | return 0; |
39279cc3 CM |
5104 | } |
5105 | ||
d352ac68 CM |
5106 | /* helper function for file defrag and space balancing. This |
5107 | * forces readahead on a given range of bytes in an inode | |
5108 | */ | |
edbd8d4e | 5109 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5110 | struct file_ra_state *ra, struct file *file, |
5111 | pgoff_t offset, pgoff_t last_index) | |
5112 | { | |
8e7bf94f | 5113 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5114 | |
86479a04 CM |
5115 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5116 | return offset + req_size; | |
86479a04 CM |
5117 | } |
5118 | ||
39279cc3 CM |
5119 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5120 | { | |
5121 | struct btrfs_inode *ei; | |
5122 | ||
5123 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5124 | if (!ei) | |
5125 | return NULL; | |
15ee9bc7 | 5126 | ei->last_trans = 0; |
257c62e1 | 5127 | ei->last_sub_trans = 0; |
e02119d5 | 5128 | ei->logged_trans = 0; |
32c00aff JB |
5129 | ei->outstanding_extents = 0; |
5130 | ei->reserved_extents = 0; | |
5131 | spin_lock_init(&ei->accounting_lock); | |
e6dcd2dc | 5132 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5133 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5134 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5135 | return &ei->vfs_inode; |
5136 | } | |
5137 | ||
5138 | void btrfs_destroy_inode(struct inode *inode) | |
5139 | { | |
e6dcd2dc | 5140 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5141 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5142 | ||
39279cc3 CM |
5143 | WARN_ON(!list_empty(&inode->i_dentry)); |
5144 | WARN_ON(inode->i_data.nrpages); | |
5145 | ||
5a3f23d5 CM |
5146 | /* |
5147 | * Make sure we're properly removed from the ordered operation | |
5148 | * lists. | |
5149 | */ | |
5150 | smp_mb(); | |
5151 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5152 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5153 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5154 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5155 | } | |
5156 | ||
5157 | spin_lock(&root->list_lock); | |
7b128766 JB |
5158 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
5159 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
5160 | " list\n", inode->i_ino); | |
5161 | dump_stack(); | |
5162 | } | |
5a3f23d5 | 5163 | spin_unlock(&root->list_lock); |
7b128766 | 5164 | |
d397712b | 5165 | while (1) { |
e6dcd2dc CM |
5166 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5167 | if (!ordered) | |
5168 | break; | |
5169 | else { | |
d397712b CM |
5170 | printk(KERN_ERR "btrfs found ordered " |
5171 | "extent %llu %llu on inode cleanup\n", | |
5172 | (unsigned long long)ordered->file_offset, | |
5173 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5174 | btrfs_remove_ordered_extent(inode, ordered); |
5175 | btrfs_put_ordered_extent(ordered); | |
5176 | btrfs_put_ordered_extent(ordered); | |
5177 | } | |
5178 | } | |
5d4f98a2 | 5179 | inode_tree_del(inode); |
5b21f2ed | 5180 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
5181 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5182 | } | |
5183 | ||
76dda93c YZ |
5184 | void btrfs_drop_inode(struct inode *inode) |
5185 | { | |
5186 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5187 | ||
5188 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5189 | generic_delete_inode(inode); | |
5190 | else | |
5191 | generic_drop_inode(inode); | |
5192 | } | |
5193 | ||
0ee0fda0 | 5194 | static void init_once(void *foo) |
39279cc3 CM |
5195 | { |
5196 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5197 | ||
5198 | inode_init_once(&ei->vfs_inode); | |
5199 | } | |
5200 | ||
5201 | void btrfs_destroy_cachep(void) | |
5202 | { | |
5203 | if (btrfs_inode_cachep) | |
5204 | kmem_cache_destroy(btrfs_inode_cachep); | |
5205 | if (btrfs_trans_handle_cachep) | |
5206 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5207 | if (btrfs_transaction_cachep) | |
5208 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5209 | if (btrfs_path_cachep) |
5210 | kmem_cache_destroy(btrfs_path_cachep); | |
5211 | } | |
5212 | ||
5213 | int btrfs_init_cachep(void) | |
5214 | { | |
9601e3f6 CH |
5215 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5216 | sizeof(struct btrfs_inode), 0, | |
5217 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5218 | if (!btrfs_inode_cachep) |
5219 | goto fail; | |
9601e3f6 CH |
5220 | |
5221 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5222 | sizeof(struct btrfs_trans_handle), 0, | |
5223 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5224 | if (!btrfs_trans_handle_cachep) |
5225 | goto fail; | |
9601e3f6 CH |
5226 | |
5227 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5228 | sizeof(struct btrfs_transaction), 0, | |
5229 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5230 | if (!btrfs_transaction_cachep) |
5231 | goto fail; | |
9601e3f6 CH |
5232 | |
5233 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5234 | sizeof(struct btrfs_path), 0, | |
5235 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5236 | if (!btrfs_path_cachep) |
5237 | goto fail; | |
9601e3f6 | 5238 | |
39279cc3 CM |
5239 | return 0; |
5240 | fail: | |
5241 | btrfs_destroy_cachep(); | |
5242 | return -ENOMEM; | |
5243 | } | |
5244 | ||
5245 | static int btrfs_getattr(struct vfsmount *mnt, | |
5246 | struct dentry *dentry, struct kstat *stat) | |
5247 | { | |
5248 | struct inode *inode = dentry->d_inode; | |
5249 | generic_fillattr(inode, stat); | |
3394e160 | 5250 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5251 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5252 | stat->blocks = (inode_get_bytes(inode) + |
5253 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5254 | return 0; |
5255 | } | |
5256 | ||
d397712b CM |
5257 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5258 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5259 | { |
5260 | struct btrfs_trans_handle *trans; | |
5261 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5262 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5263 | struct inode *new_inode = new_dentry->d_inode; |
5264 | struct inode *old_inode = old_dentry->d_inode; | |
5265 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5266 | u64 index = 0; |
4df27c4d | 5267 | u64 root_objectid; |
39279cc3 CM |
5268 | int ret; |
5269 | ||
f679a840 YZ |
5270 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5271 | return -EPERM; | |
5272 | ||
4df27c4d YZ |
5273 | /* we only allow rename subvolume link between subvolumes */ |
5274 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5275 | return -EXDEV; |
5276 | ||
4df27c4d YZ |
5277 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5278 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5279 | return -ENOTEMPTY; |
5f39d397 | 5280 | |
4df27c4d YZ |
5281 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5282 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5283 | return -ENOTEMPTY; | |
0660b5af | 5284 | |
9ed74f2d JB |
5285 | /* |
5286 | * 2 items for dir items | |
5287 | * 1 item for orphan entry | |
5288 | * 1 item for ref | |
5289 | */ | |
5290 | ret = btrfs_reserve_metadata_space(root, 4); | |
1832a6d5 | 5291 | if (ret) |
4df27c4d | 5292 | return ret; |
1832a6d5 | 5293 | |
5a3f23d5 CM |
5294 | /* |
5295 | * we're using rename to replace one file with another. | |
5296 | * and the replacement file is large. Start IO on it now so | |
5297 | * we don't add too much work to the end of the transaction | |
5298 | */ | |
4baf8c92 | 5299 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5300 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5301 | filemap_flush(old_inode->i_mapping); | |
5302 | ||
76dda93c YZ |
5303 | /* close the racy window with snapshot create/destroy ioctl */ |
5304 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5305 | down_read(&root->fs_info->subvol_sem); | |
5306 | ||
39279cc3 | 5307 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5308 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5309 | |
4df27c4d YZ |
5310 | if (dest != root) |
5311 | btrfs_record_root_in_trans(trans, dest); | |
5312 | ||
a5719521 YZ |
5313 | ret = btrfs_set_inode_index(new_dir, &index); |
5314 | if (ret) | |
5315 | goto out_fail; | |
5a3f23d5 | 5316 | |
a5719521 | 5317 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5318 | /* force full log commit if subvolume involved. */ |
5319 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5320 | } else { | |
a5719521 YZ |
5321 | ret = btrfs_insert_inode_ref(trans, dest, |
5322 | new_dentry->d_name.name, | |
5323 | new_dentry->d_name.len, | |
5324 | old_inode->i_ino, | |
5325 | new_dir->i_ino, index); | |
5326 | if (ret) | |
5327 | goto out_fail; | |
4df27c4d YZ |
5328 | /* |
5329 | * this is an ugly little race, but the rename is required | |
5330 | * to make sure that if we crash, the inode is either at the | |
5331 | * old name or the new one. pinning the log transaction lets | |
5332 | * us make sure we don't allow a log commit to come in after | |
5333 | * we unlink the name but before we add the new name back in. | |
5334 | */ | |
5335 | btrfs_pin_log_trans(root); | |
5336 | } | |
a5719521 YZ |
5337 | /* |
5338 | * make sure the inode gets flushed if it is replacing | |
5339 | * something. | |
5340 | */ | |
5341 | if (new_inode && new_inode->i_size && | |
5342 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5343 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5344 | } | |
39279cc3 | 5345 | |
39279cc3 CM |
5346 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5347 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5348 | old_inode->i_ctime = ctime; | |
5f39d397 | 5349 | |
12fcfd22 CM |
5350 | if (old_dentry->d_parent != new_dentry->d_parent) |
5351 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5352 | ||
4df27c4d YZ |
5353 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5354 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5355 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5356 | old_dentry->d_name.name, | |
5357 | old_dentry->d_name.len); | |
5358 | } else { | |
5359 | btrfs_inc_nlink(old_dentry->d_inode); | |
5360 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5361 | old_dentry->d_inode, | |
5362 | old_dentry->d_name.name, | |
5363 | old_dentry->d_name.len); | |
5364 | } | |
5365 | BUG_ON(ret); | |
39279cc3 CM |
5366 | |
5367 | if (new_inode) { | |
5368 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5369 | if (unlikely(new_inode->i_ino == |
5370 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5371 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5372 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5373 | root_objectid, | |
5374 | new_dentry->d_name.name, | |
5375 | new_dentry->d_name.len); | |
5376 | BUG_ON(new_inode->i_nlink == 0); | |
5377 | } else { | |
5378 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5379 | new_dentry->d_inode, | |
5380 | new_dentry->d_name.name, | |
5381 | new_dentry->d_name.len); | |
5382 | } | |
5383 | BUG_ON(ret); | |
7b128766 | 5384 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5385 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5386 | BUG_ON(ret); |
7b128766 | 5387 | } |
39279cc3 | 5388 | } |
aec7477b | 5389 | |
4df27c4d YZ |
5390 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5391 | new_dentry->d_name.name, | |
a5719521 | 5392 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5393 | BUG_ON(ret); |
39279cc3 | 5394 | |
4df27c4d YZ |
5395 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5396 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5397 | new_dentry->d_parent); | |
5398 | btrfs_end_log_trans(root); | |
5399 | } | |
a5719521 | 5400 | out_fail: |
ab78c84d | 5401 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5402 | |
76dda93c YZ |
5403 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5404 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d JB |
5405 | |
5406 | btrfs_unreserve_metadata_space(root, 4); | |
39279cc3 CM |
5407 | return ret; |
5408 | } | |
5409 | ||
d352ac68 CM |
5410 | /* |
5411 | * some fairly slow code that needs optimization. This walks the list | |
5412 | * of all the inodes with pending delalloc and forces them to disk. | |
5413 | */ | |
ea8c2819 CM |
5414 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5415 | { | |
5416 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5417 | struct btrfs_inode *binode; | |
5b21f2ed | 5418 | struct inode *inode; |
ea8c2819 | 5419 | |
c146afad YZ |
5420 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5421 | return -EROFS; | |
5422 | ||
75eff68e | 5423 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5424 | while (!list_empty(head)) { |
ea8c2819 CM |
5425 | binode = list_entry(head->next, struct btrfs_inode, |
5426 | delalloc_inodes); | |
5b21f2ed ZY |
5427 | inode = igrab(&binode->vfs_inode); |
5428 | if (!inode) | |
5429 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5430 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5431 | if (inode) { |
8c8bee1d | 5432 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5433 | iput(inode); |
5434 | } | |
5435 | cond_resched(); | |
75eff68e | 5436 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5437 | } |
75eff68e | 5438 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5439 | |
5440 | /* the filemap_flush will queue IO into the worker threads, but | |
5441 | * we have to make sure the IO is actually started and that | |
5442 | * ordered extents get created before we return | |
5443 | */ | |
5444 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5445 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5446 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5447 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5448 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5449 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5450 | } |
5451 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5452 | return 0; |
5453 | } | |
5454 | ||
39279cc3 CM |
5455 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5456 | const char *symname) | |
5457 | { | |
5458 | struct btrfs_trans_handle *trans; | |
5459 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5460 | struct btrfs_path *path; | |
5461 | struct btrfs_key key; | |
1832a6d5 | 5462 | struct inode *inode = NULL; |
39279cc3 CM |
5463 | int err; |
5464 | int drop_inode = 0; | |
5465 | u64 objectid; | |
00e4e6b3 | 5466 | u64 index = 0 ; |
39279cc3 CM |
5467 | int name_len; |
5468 | int datasize; | |
5f39d397 | 5469 | unsigned long ptr; |
39279cc3 | 5470 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5471 | struct extent_buffer *leaf; |
1832a6d5 | 5472 | unsigned long nr = 0; |
39279cc3 CM |
5473 | |
5474 | name_len = strlen(symname) + 1; | |
5475 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5476 | return -ENAMETOOLONG; | |
1832a6d5 | 5477 | |
9ed74f2d JB |
5478 | /* |
5479 | * 2 items for inode item and ref | |
5480 | * 2 items for dir items | |
5481 | * 1 item for xattr if selinux is on | |
5482 | */ | |
5483 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5484 | if (err) |
9ed74f2d | 5485 | return err; |
1832a6d5 | 5486 | |
39279cc3 | 5487 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5488 | if (!trans) |
5489 | goto out_fail; | |
39279cc3 CM |
5490 | btrfs_set_trans_block_group(trans, dir); |
5491 | ||
5492 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5493 | if (err) { | |
5494 | err = -ENOSPC; | |
5495 | goto out_unlock; | |
5496 | } | |
5497 | ||
aec7477b | 5498 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5499 | dentry->d_name.len, |
5500 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5501 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5502 | &index); | |
39279cc3 CM |
5503 | err = PTR_ERR(inode); |
5504 | if (IS_ERR(inode)) | |
5505 | goto out_unlock; | |
5506 | ||
0279b4cd | 5507 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5508 | if (err) { |
5509 | drop_inode = 1; | |
5510 | goto out_unlock; | |
5511 | } | |
5512 | ||
39279cc3 | 5513 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5514 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5515 | if (err) |
5516 | drop_inode = 1; | |
5517 | else { | |
5518 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5519 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5520 | inode->i_fop = &btrfs_file_operations; |
5521 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5522 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5523 | } |
39279cc3 CM |
5524 | btrfs_update_inode_block_group(trans, inode); |
5525 | btrfs_update_inode_block_group(trans, dir); | |
5526 | if (drop_inode) | |
5527 | goto out_unlock; | |
5528 | ||
5529 | path = btrfs_alloc_path(); | |
5530 | BUG_ON(!path); | |
5531 | key.objectid = inode->i_ino; | |
5532 | key.offset = 0; | |
39279cc3 CM |
5533 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5534 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5535 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5536 | datasize); | |
54aa1f4d CM |
5537 | if (err) { |
5538 | drop_inode = 1; | |
5539 | goto out_unlock; | |
5540 | } | |
5f39d397 CM |
5541 | leaf = path->nodes[0]; |
5542 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5543 | struct btrfs_file_extent_item); | |
5544 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5545 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5546 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5547 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5548 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5549 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5550 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5551 | ||
39279cc3 | 5552 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5553 | write_extent_buffer(leaf, symname, ptr, name_len); |
5554 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5555 | btrfs_free_path(path); |
5f39d397 | 5556 | |
39279cc3 CM |
5557 | inode->i_op = &btrfs_symlink_inode_operations; |
5558 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5559 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5560 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5561 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5562 | err = btrfs_update_inode(trans, root, inode); |
5563 | if (err) | |
5564 | drop_inode = 1; | |
39279cc3 CM |
5565 | |
5566 | out_unlock: | |
d3c2fdcf | 5567 | nr = trans->blocks_used; |
ab78c84d | 5568 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5569 | out_fail: |
9ed74f2d | 5570 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5571 | if (drop_inode) { |
5572 | inode_dec_link_count(inode); | |
5573 | iput(inode); | |
5574 | } | |
d3c2fdcf | 5575 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5576 | return err; |
5577 | } | |
16432985 | 5578 | |
546888da CM |
5579 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5580 | struct inode *inode, u64 start, u64 end, | |
e980b50c | 5581 | u64 locked_end, u64 alloc_hint, int mode) |
d899e052 | 5582 | { |
d899e052 YZ |
5583 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5584 | struct btrfs_key ins; | |
5585 | u64 alloc_size; | |
5586 | u64 cur_offset = start; | |
5587 | u64 num_bytes = end - start; | |
5588 | int ret = 0; | |
5589 | ||
d899e052 YZ |
5590 | while (num_bytes > 0) { |
5591 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d JB |
5592 | |
5593 | ret = btrfs_reserve_metadata_space(root, 1); | |
5594 | if (ret) | |
5595 | goto out; | |
5596 | ||
d899e052 YZ |
5597 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5598 | root->sectorsize, 0, alloc_hint, | |
5599 | (u64)-1, &ins, 1); | |
5600 | if (ret) { | |
5601 | WARN_ON(1); | |
5602 | goto out; | |
5603 | } | |
5604 | ret = insert_reserved_file_extent(trans, inode, | |
5605 | cur_offset, ins.objectid, | |
5606 | ins.offset, ins.offset, | |
e980b50c CM |
5607 | ins.offset, locked_end, |
5608 | 0, 0, 0, | |
d899e052 YZ |
5609 | BTRFS_FILE_EXTENT_PREALLOC); |
5610 | BUG_ON(ret); | |
a1ed835e CM |
5611 | btrfs_drop_extent_cache(inode, cur_offset, |
5612 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5613 | num_bytes -= ins.offset; |
5614 | cur_offset += ins.offset; | |
5615 | alloc_hint = ins.objectid + ins.offset; | |
9ed74f2d | 5616 | btrfs_unreserve_metadata_space(root, 1); |
d899e052 YZ |
5617 | } |
5618 | out: | |
5619 | if (cur_offset > start) { | |
5620 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5621 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5622 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5623 | cur_offset > i_size_read(inode)) | |
5624 | btrfs_i_size_write(inode, cur_offset); | |
5625 | ret = btrfs_update_inode(trans, root, inode); | |
5626 | BUG_ON(ret); | |
5627 | } | |
5628 | ||
d899e052 YZ |
5629 | return ret; |
5630 | } | |
5631 | ||
5632 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5633 | loff_t offset, loff_t len) | |
5634 | { | |
5635 | u64 cur_offset; | |
5636 | u64 last_byte; | |
5637 | u64 alloc_start; | |
5638 | u64 alloc_end; | |
5639 | u64 alloc_hint = 0; | |
e980b50c | 5640 | u64 locked_end; |
d899e052 YZ |
5641 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5642 | struct extent_map *em; | |
546888da | 5643 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5644 | struct btrfs_root *root; |
d899e052 YZ |
5645 | int ret; |
5646 | ||
5647 | alloc_start = offset & ~mask; | |
5648 | alloc_end = (offset + len + mask) & ~mask; | |
5649 | ||
546888da CM |
5650 | /* |
5651 | * wait for ordered IO before we have any locks. We'll loop again | |
5652 | * below with the locks held. | |
5653 | */ | |
5654 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5655 | ||
d899e052 YZ |
5656 | mutex_lock(&inode->i_mutex); |
5657 | if (alloc_start > inode->i_size) { | |
5658 | ret = btrfs_cont_expand(inode, alloc_start); | |
5659 | if (ret) | |
5660 | goto out; | |
5661 | } | |
5662 | ||
a970b0a1 JB |
5663 | root = BTRFS_I(inode)->root; |
5664 | ||
5665 | ret = btrfs_check_data_free_space(root, inode, | |
5666 | alloc_end - alloc_start); | |
5667 | if (ret) | |
5668 | goto out; | |
5669 | ||
e980b50c | 5670 | locked_end = alloc_end - 1; |
d899e052 YZ |
5671 | while (1) { |
5672 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5673 | |
5674 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5675 | if (!trans) { | |
5676 | ret = -EIO; | |
a970b0a1 | 5677 | goto out_free; |
546888da CM |
5678 | } |
5679 | ||
5680 | /* the extent lock is ordered inside the running | |
5681 | * transaction | |
5682 | */ | |
e980b50c CM |
5683 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5684 | GFP_NOFS); | |
d899e052 YZ |
5685 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5686 | alloc_end - 1); | |
5687 | if (ordered && | |
5688 | ordered->file_offset + ordered->len > alloc_start && | |
5689 | ordered->file_offset < alloc_end) { | |
5690 | btrfs_put_ordered_extent(ordered); | |
5691 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5692 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5693 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5694 | ||
5695 | /* | |
5696 | * we can't wait on the range with the transaction | |
5697 | * running or with the extent lock held | |
5698 | */ | |
d899e052 YZ |
5699 | btrfs_wait_ordered_range(inode, alloc_start, |
5700 | alloc_end - alloc_start); | |
5701 | } else { | |
5702 | if (ordered) | |
5703 | btrfs_put_ordered_extent(ordered); | |
5704 | break; | |
5705 | } | |
5706 | } | |
5707 | ||
5708 | cur_offset = alloc_start; | |
5709 | while (1) { | |
5710 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5711 | alloc_end - cur_offset, 0); | |
5712 | BUG_ON(IS_ERR(em) || !em); | |
5713 | last_byte = min(extent_map_end(em), alloc_end); | |
5714 | last_byte = (last_byte + mask) & ~mask; | |
5715 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5716 | ret = prealloc_file_range(trans, inode, cur_offset, |
e980b50c CM |
5717 | last_byte, locked_end + 1, |
5718 | alloc_hint, mode); | |
d899e052 YZ |
5719 | if (ret < 0) { |
5720 | free_extent_map(em); | |
5721 | break; | |
5722 | } | |
5723 | } | |
5724 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5725 | alloc_hint = em->block_start; | |
5726 | free_extent_map(em); | |
5727 | ||
5728 | cur_offset = last_byte; | |
5729 | if (cur_offset >= alloc_end) { | |
5730 | ret = 0; | |
5731 | break; | |
5732 | } | |
5733 | } | |
e980b50c | 5734 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5735 | GFP_NOFS); |
546888da CM |
5736 | |
5737 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5738 | out_free: |
5739 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5740 | out: |
5741 | mutex_unlock(&inode->i_mutex); | |
5742 | return ret; | |
5743 | } | |
5744 | ||
e6dcd2dc CM |
5745 | static int btrfs_set_page_dirty(struct page *page) |
5746 | { | |
e6dcd2dc CM |
5747 | return __set_page_dirty_nobuffers(page); |
5748 | } | |
5749 | ||
0ee0fda0 | 5750 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5751 | { |
6cbff00f | 5752 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5753 | return -EACCES; |
33268eaf | 5754 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5755 | } |
39279cc3 CM |
5756 | |
5757 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 5758 | .getattr = btrfs_getattr, |
39279cc3 CM |
5759 | .lookup = btrfs_lookup, |
5760 | .create = btrfs_create, | |
5761 | .unlink = btrfs_unlink, | |
5762 | .link = btrfs_link, | |
5763 | .mkdir = btrfs_mkdir, | |
5764 | .rmdir = btrfs_rmdir, | |
5765 | .rename = btrfs_rename, | |
5766 | .symlink = btrfs_symlink, | |
5767 | .setattr = btrfs_setattr, | |
618e21d5 | 5768 | .mknod = btrfs_mknod, |
95819c05 CH |
5769 | .setxattr = btrfs_setxattr, |
5770 | .getxattr = btrfs_getxattr, | |
5103e947 | 5771 | .listxattr = btrfs_listxattr, |
95819c05 | 5772 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5773 | .permission = btrfs_permission, |
39279cc3 | 5774 | }; |
39279cc3 CM |
5775 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
5776 | .lookup = btrfs_lookup, | |
fdebe2bd | 5777 | .permission = btrfs_permission, |
39279cc3 | 5778 | }; |
76dda93c | 5779 | |
39279cc3 CM |
5780 | static struct file_operations btrfs_dir_file_operations = { |
5781 | .llseek = generic_file_llseek, | |
5782 | .read = generic_read_dir, | |
cbdf5a24 | 5783 | .readdir = btrfs_real_readdir, |
34287aa3 | 5784 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5785 | #ifdef CONFIG_COMPAT |
34287aa3 | 5786 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5787 | #endif |
6bf13c0c | 5788 | .release = btrfs_release_file, |
e02119d5 | 5789 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5790 | }; |
5791 | ||
d1310b2e | 5792 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5793 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5794 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5795 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5796 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5797 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5798 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5799 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5800 | .set_bit_hook = btrfs_set_bit_hook, |
5801 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
5802 | .merge_extent_hook = btrfs_merge_extent_hook, |
5803 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
5804 | }; |
5805 | ||
35054394 CM |
5806 | /* |
5807 | * btrfs doesn't support the bmap operation because swapfiles | |
5808 | * use bmap to make a mapping of extents in the file. They assume | |
5809 | * these extents won't change over the life of the file and they | |
5810 | * use the bmap result to do IO directly to the drive. | |
5811 | * | |
5812 | * the btrfs bmap call would return logical addresses that aren't | |
5813 | * suitable for IO and they also will change frequently as COW | |
5814 | * operations happen. So, swapfile + btrfs == corruption. | |
5815 | * | |
5816 | * For now we're avoiding this by dropping bmap. | |
5817 | */ | |
39279cc3 CM |
5818 | static struct address_space_operations btrfs_aops = { |
5819 | .readpage = btrfs_readpage, | |
5820 | .writepage = btrfs_writepage, | |
b293f02e | 5821 | .writepages = btrfs_writepages, |
3ab2fb5a | 5822 | .readpages = btrfs_readpages, |
39279cc3 | 5823 | .sync_page = block_sync_page, |
16432985 | 5824 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5825 | .invalidatepage = btrfs_invalidatepage, |
5826 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5827 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5828 | }; |
5829 | ||
5830 | static struct address_space_operations btrfs_symlink_aops = { | |
5831 | .readpage = btrfs_readpage, | |
5832 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5833 | .invalidatepage = btrfs_invalidatepage, |
5834 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5835 | }; |
5836 | ||
5837 | static struct inode_operations btrfs_file_inode_operations = { | |
5838 | .truncate = btrfs_truncate, | |
5839 | .getattr = btrfs_getattr, | |
5840 | .setattr = btrfs_setattr, | |
95819c05 CH |
5841 | .setxattr = btrfs_setxattr, |
5842 | .getxattr = btrfs_getxattr, | |
5103e947 | 5843 | .listxattr = btrfs_listxattr, |
95819c05 | 5844 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5845 | .permission = btrfs_permission, |
d899e052 | 5846 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5847 | .fiemap = btrfs_fiemap, |
39279cc3 | 5848 | }; |
618e21d5 JB |
5849 | static struct inode_operations btrfs_special_inode_operations = { |
5850 | .getattr = btrfs_getattr, | |
5851 | .setattr = btrfs_setattr, | |
fdebe2bd | 5852 | .permission = btrfs_permission, |
95819c05 CH |
5853 | .setxattr = btrfs_setxattr, |
5854 | .getxattr = btrfs_getxattr, | |
33268eaf | 5855 | .listxattr = btrfs_listxattr, |
95819c05 | 5856 | .removexattr = btrfs_removexattr, |
618e21d5 | 5857 | }; |
39279cc3 CM |
5858 | static struct inode_operations btrfs_symlink_inode_operations = { |
5859 | .readlink = generic_readlink, | |
5860 | .follow_link = page_follow_link_light, | |
5861 | .put_link = page_put_link, | |
fdebe2bd | 5862 | .permission = btrfs_permission, |
0279b4cd JO |
5863 | .setxattr = btrfs_setxattr, |
5864 | .getxattr = btrfs_getxattr, | |
5865 | .listxattr = btrfs_listxattr, | |
5866 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5867 | }; |
76dda93c | 5868 | |
82d339d9 | 5869 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
5870 | .d_delete = btrfs_dentry_delete, |
5871 | }; |