]>
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> | |
29 | #include <linux/smp_lock.h> | |
30 | #include <linux/backing-dev.h> | |
31 | #include <linux/mpage.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/writeback.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/compat.h> | |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
92fee66d | 37 | #include <linux/version.h> |
5103e947 | 38 | #include <linux/xattr.h> |
33268eaf | 39 | #include <linux/posix_acl.h> |
d899e052 | 40 | #include <linux/falloc.h> |
4b4e25f2 | 41 | #include "compat.h" |
39279cc3 CM |
42 | #include "ctree.h" |
43 | #include "disk-io.h" | |
44 | #include "transaction.h" | |
45 | #include "btrfs_inode.h" | |
46 | #include "ioctl.h" | |
47 | #include "print-tree.h" | |
0b86a832 | 48 | #include "volumes.h" |
e6dcd2dc | 49 | #include "ordered-data.h" |
95819c05 | 50 | #include "xattr.h" |
e02119d5 | 51 | #include "tree-log.h" |
5b84e8d6 | 52 | #include "ref-cache.h" |
c8b97818 | 53 | #include "compression.h" |
39279cc3 CM |
54 | |
55 | struct btrfs_iget_args { | |
56 | u64 ino; | |
57 | struct btrfs_root *root; | |
58 | }; | |
59 | ||
60 | static struct inode_operations btrfs_dir_inode_operations; | |
61 | static struct inode_operations btrfs_symlink_inode_operations; | |
62 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 63 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
64 | static struct inode_operations btrfs_file_inode_operations; |
65 | static struct address_space_operations btrfs_aops; | |
66 | static struct address_space_operations btrfs_symlink_aops; | |
67 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 68 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
69 | |
70 | static struct kmem_cache *btrfs_inode_cachep; | |
71 | struct kmem_cache *btrfs_trans_handle_cachep; | |
72 | struct kmem_cache *btrfs_transaction_cachep; | |
73 | struct kmem_cache *btrfs_bit_radix_cachep; | |
74 | struct kmem_cache *btrfs_path_cachep; | |
75 | ||
76 | #define S_SHIFT 12 | |
77 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
78 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
79 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
80 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
81 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
82 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
83 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
84 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
85 | }; | |
86 | ||
7b128766 | 87 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 88 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
89 | static noinline int cow_file_range(struct inode *inode, |
90 | struct page *locked_page, | |
91 | u64 start, u64 end, int *page_started, | |
92 | unsigned long *nr_written, int unlock); | |
7b128766 | 93 | |
d352ac68 CM |
94 | /* |
95 | * a very lame attempt at stopping writes when the FS is 85% full. There | |
96 | * are countless ways this is incorrect, but it is better than nothing. | |
97 | */ | |
1832a6d5 CM |
98 | int btrfs_check_free_space(struct btrfs_root *root, u64 num_required, |
99 | int for_del) | |
100 | { | |
a2135011 CM |
101 | u64 total; |
102 | u64 used; | |
1832a6d5 CM |
103 | u64 thresh; |
104 | int ret = 0; | |
105 | ||
75eff68e | 106 | spin_lock(&root->fs_info->delalloc_lock); |
a2135011 CM |
107 | total = btrfs_super_total_bytes(&root->fs_info->super_copy); |
108 | used = btrfs_super_bytes_used(&root->fs_info->super_copy); | |
1832a6d5 | 109 | if (for_del) |
f9ef6604 | 110 | thresh = total * 90; |
1832a6d5 | 111 | else |
f9ef6604 CM |
112 | thresh = total * 85; |
113 | ||
114 | do_div(thresh, 100); | |
1832a6d5 | 115 | |
1832a6d5 CM |
116 | if (used + root->fs_info->delalloc_bytes + num_required > thresh) |
117 | ret = -ENOSPC; | |
75eff68e | 118 | spin_unlock(&root->fs_info->delalloc_lock); |
1832a6d5 CM |
119 | return ret; |
120 | } | |
121 | ||
c8b97818 CM |
122 | /* |
123 | * this does all the hard work for inserting an inline extent into | |
124 | * the btree. The caller should have done a btrfs_drop_extents so that | |
125 | * no overlapping inline items exist in the btree | |
126 | */ | |
d397712b | 127 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
128 | struct btrfs_root *root, struct inode *inode, |
129 | u64 start, size_t size, size_t compressed_size, | |
130 | struct page **compressed_pages) | |
131 | { | |
132 | struct btrfs_key key; | |
133 | struct btrfs_path *path; | |
134 | struct extent_buffer *leaf; | |
135 | struct page *page = NULL; | |
136 | char *kaddr; | |
137 | unsigned long ptr; | |
138 | struct btrfs_file_extent_item *ei; | |
139 | int err = 0; | |
140 | int ret; | |
141 | size_t cur_size = size; | |
142 | size_t datasize; | |
143 | unsigned long offset; | |
144 | int use_compress = 0; | |
145 | ||
146 | if (compressed_size && compressed_pages) { | |
147 | use_compress = 1; | |
148 | cur_size = compressed_size; | |
149 | } | |
150 | ||
d397712b CM |
151 | path = btrfs_alloc_path(); |
152 | if (!path) | |
c8b97818 CM |
153 | return -ENOMEM; |
154 | ||
155 | btrfs_set_trans_block_group(trans, inode); | |
156 | ||
157 | key.objectid = inode->i_ino; | |
158 | key.offset = start; | |
159 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
160 | inode_add_bytes(inode, size); | |
161 | datasize = btrfs_file_extent_calc_inline_size(cur_size); | |
162 | ||
163 | inode_add_bytes(inode, size); | |
164 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
165 | datasize); | |
166 | BUG_ON(ret); | |
167 | if (ret) { | |
168 | err = ret; | |
c8b97818 CM |
169 | goto fail; |
170 | } | |
171 | leaf = path->nodes[0]; | |
172 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
173 | struct btrfs_file_extent_item); | |
174 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
175 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
176 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
177 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
178 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
179 | ptr = btrfs_file_extent_inline_start(ei); | |
180 | ||
181 | if (use_compress) { | |
182 | struct page *cpage; | |
183 | int i = 0; | |
d397712b | 184 | while (compressed_size > 0) { |
c8b97818 | 185 | cpage = compressed_pages[i]; |
5b050f04 | 186 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
187 | PAGE_CACHE_SIZE); |
188 | ||
189 | kaddr = kmap(cpage); | |
190 | write_extent_buffer(leaf, kaddr, ptr, cur_size); | |
191 | kunmap(cpage); | |
192 | ||
193 | i++; | |
194 | ptr += cur_size; | |
195 | compressed_size -= cur_size; | |
196 | } | |
197 | btrfs_set_file_extent_compression(leaf, ei, | |
198 | BTRFS_COMPRESS_ZLIB); | |
199 | } else { | |
200 | page = find_get_page(inode->i_mapping, | |
201 | start >> PAGE_CACHE_SHIFT); | |
202 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
203 | kaddr = kmap_atomic(page, KM_USER0); | |
204 | offset = start & (PAGE_CACHE_SIZE - 1); | |
205 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
206 | kunmap_atomic(kaddr, KM_USER0); | |
207 | page_cache_release(page); | |
208 | } | |
209 | btrfs_mark_buffer_dirty(leaf); | |
210 | btrfs_free_path(path); | |
211 | ||
212 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
213 | btrfs_update_inode(trans, root, inode); | |
214 | return 0; | |
215 | fail: | |
216 | btrfs_free_path(path); | |
217 | return err; | |
218 | } | |
219 | ||
220 | ||
221 | /* | |
222 | * conditionally insert an inline extent into the file. This | |
223 | * does the checks required to make sure the data is small enough | |
224 | * to fit as an inline extent. | |
225 | */ | |
226 | static int cow_file_range_inline(struct btrfs_trans_handle *trans, | |
227 | struct btrfs_root *root, | |
228 | struct inode *inode, u64 start, u64 end, | |
229 | size_t compressed_size, | |
230 | struct page **compressed_pages) | |
231 | { | |
232 | u64 isize = i_size_read(inode); | |
233 | u64 actual_end = min(end + 1, isize); | |
234 | u64 inline_len = actual_end - start; | |
235 | u64 aligned_end = (end + root->sectorsize - 1) & | |
236 | ~((u64)root->sectorsize - 1); | |
237 | u64 hint_byte; | |
238 | u64 data_len = inline_len; | |
239 | int ret; | |
240 | ||
241 | if (compressed_size) | |
242 | data_len = compressed_size; | |
243 | ||
244 | if (start > 0 || | |
70b99e69 | 245 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
246 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
247 | (!compressed_size && | |
248 | (actual_end & (root->sectorsize - 1)) == 0) || | |
249 | end + 1 < isize || | |
250 | data_len > root->fs_info->max_inline) { | |
251 | return 1; | |
252 | } | |
253 | ||
c8b97818 | 254 | ret = btrfs_drop_extents(trans, root, inode, start, |
70b99e69 | 255 | aligned_end, start, &hint_byte); |
c8b97818 CM |
256 | BUG_ON(ret); |
257 | ||
258 | if (isize > actual_end) | |
259 | inline_len = min_t(u64, isize, actual_end); | |
260 | ret = insert_inline_extent(trans, root, inode, start, | |
261 | inline_len, compressed_size, | |
262 | compressed_pages); | |
263 | BUG_ON(ret); | |
264 | btrfs_drop_extent_cache(inode, start, aligned_end, 0); | |
c8b97818 CM |
265 | return 0; |
266 | } | |
267 | ||
771ed689 CM |
268 | struct async_extent { |
269 | u64 start; | |
270 | u64 ram_size; | |
271 | u64 compressed_size; | |
272 | struct page **pages; | |
273 | unsigned long nr_pages; | |
274 | struct list_head list; | |
275 | }; | |
276 | ||
277 | struct async_cow { | |
278 | struct inode *inode; | |
279 | struct btrfs_root *root; | |
280 | struct page *locked_page; | |
281 | u64 start; | |
282 | u64 end; | |
283 | struct list_head extents; | |
284 | struct btrfs_work work; | |
285 | }; | |
286 | ||
287 | static noinline int add_async_extent(struct async_cow *cow, | |
288 | u64 start, u64 ram_size, | |
289 | u64 compressed_size, | |
290 | struct page **pages, | |
291 | unsigned long nr_pages) | |
292 | { | |
293 | struct async_extent *async_extent; | |
294 | ||
295 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
296 | async_extent->start = start; | |
297 | async_extent->ram_size = ram_size; | |
298 | async_extent->compressed_size = compressed_size; | |
299 | async_extent->pages = pages; | |
300 | async_extent->nr_pages = nr_pages; | |
301 | list_add_tail(&async_extent->list, &cow->extents); | |
302 | return 0; | |
303 | } | |
304 | ||
d352ac68 | 305 | /* |
771ed689 CM |
306 | * we create compressed extents in two phases. The first |
307 | * phase compresses a range of pages that have already been | |
308 | * locked (both pages and state bits are locked). | |
c8b97818 | 309 | * |
771ed689 CM |
310 | * This is done inside an ordered work queue, and the compression |
311 | * is spread across many cpus. The actual IO submission is step | |
312 | * two, and the ordered work queue takes care of making sure that | |
313 | * happens in the same order things were put onto the queue by | |
314 | * writepages and friends. | |
c8b97818 | 315 | * |
771ed689 CM |
316 | * If this code finds it can't get good compression, it puts an |
317 | * entry onto the work queue to write the uncompressed bytes. This | |
318 | * makes sure that both compressed inodes and uncompressed inodes | |
319 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 320 | */ |
771ed689 CM |
321 | static noinline int compress_file_range(struct inode *inode, |
322 | struct page *locked_page, | |
323 | u64 start, u64 end, | |
324 | struct async_cow *async_cow, | |
325 | int *num_added) | |
b888db2b CM |
326 | { |
327 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
328 | struct btrfs_trans_handle *trans; | |
db94535d | 329 | u64 num_bytes; |
c8b97818 CM |
330 | u64 orig_start; |
331 | u64 disk_num_bytes; | |
db94535d | 332 | u64 blocksize = root->sectorsize; |
c8b97818 | 333 | u64 actual_end; |
42dc7bab | 334 | u64 isize = i_size_read(inode); |
e6dcd2dc | 335 | int ret = 0; |
c8b97818 CM |
336 | struct page **pages = NULL; |
337 | unsigned long nr_pages; | |
338 | unsigned long nr_pages_ret = 0; | |
339 | unsigned long total_compressed = 0; | |
340 | unsigned long total_in = 0; | |
341 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 342 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
343 | int i; |
344 | int will_compress; | |
b888db2b | 345 | |
c8b97818 CM |
346 | orig_start = start; |
347 | ||
42dc7bab | 348 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
349 | again: |
350 | will_compress = 0; | |
351 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
352 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 353 | |
c8b97818 CM |
354 | total_compressed = actual_end - start; |
355 | ||
356 | /* we want to make sure that amount of ram required to uncompress | |
357 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
358 | * of a compressed extent to 128k. This is a crucial number |
359 | * because it also controls how easily we can spread reads across | |
360 | * cpus for decompression. | |
361 | * | |
362 | * We also want to make sure the amount of IO required to do | |
363 | * a random read is reasonably small, so we limit the size of | |
364 | * a compressed extent to 128k. | |
c8b97818 CM |
365 | */ |
366 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 367 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 368 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
369 | disk_num_bytes = num_bytes; |
370 | total_in = 0; | |
371 | ret = 0; | |
db94535d | 372 | |
771ed689 CM |
373 | /* |
374 | * we do compression for mount -o compress and when the | |
375 | * inode has not been flagged as nocompress. This flag can | |
376 | * change at any time if we discover bad compression ratios. | |
c8b97818 CM |
377 | */ |
378 | if (!btrfs_test_flag(inode, NOCOMPRESS) && | |
379 | btrfs_test_opt(root, COMPRESS)) { | |
380 | WARN_ON(pages); | |
cfbc246e | 381 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 382 | |
c8b97818 CM |
383 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
384 | total_compressed, pages, | |
385 | nr_pages, &nr_pages_ret, | |
386 | &total_in, | |
387 | &total_compressed, | |
388 | max_compressed); | |
389 | ||
390 | if (!ret) { | |
391 | unsigned long offset = total_compressed & | |
392 | (PAGE_CACHE_SIZE - 1); | |
393 | struct page *page = pages[nr_pages_ret - 1]; | |
394 | char *kaddr; | |
395 | ||
396 | /* zero the tail end of the last page, we might be | |
397 | * sending it down to disk | |
398 | */ | |
399 | if (offset) { | |
400 | kaddr = kmap_atomic(page, KM_USER0); | |
401 | memset(kaddr + offset, 0, | |
402 | PAGE_CACHE_SIZE - offset); | |
403 | kunmap_atomic(kaddr, KM_USER0); | |
404 | } | |
405 | will_compress = 1; | |
406 | } | |
407 | } | |
408 | if (start == 0) { | |
771ed689 CM |
409 | trans = btrfs_join_transaction(root, 1); |
410 | BUG_ON(!trans); | |
411 | btrfs_set_trans_block_group(trans, inode); | |
412 | ||
c8b97818 | 413 | /* lets try to make an inline extent */ |
771ed689 | 414 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 415 | /* we didn't compress the entire range, try |
771ed689 | 416 | * to make an uncompressed inline extent. |
c8b97818 CM |
417 | */ |
418 | ret = cow_file_range_inline(trans, root, inode, | |
419 | start, end, 0, NULL); | |
420 | } else { | |
771ed689 | 421 | /* try making a compressed inline extent */ |
c8b97818 CM |
422 | ret = cow_file_range_inline(trans, root, inode, |
423 | start, end, | |
424 | total_compressed, pages); | |
425 | } | |
771ed689 | 426 | btrfs_end_transaction(trans, root); |
c8b97818 | 427 | if (ret == 0) { |
771ed689 CM |
428 | /* |
429 | * inline extent creation worked, we don't need | |
430 | * to create any more async work items. Unlock | |
431 | * and free up our temp pages. | |
432 | */ | |
c8b97818 CM |
433 | extent_clear_unlock_delalloc(inode, |
434 | &BTRFS_I(inode)->io_tree, | |
771ed689 CM |
435 | start, end, NULL, 1, 0, |
436 | 0, 1, 1, 1); | |
c8b97818 CM |
437 | ret = 0; |
438 | goto free_pages_out; | |
439 | } | |
440 | } | |
441 | ||
442 | if (will_compress) { | |
443 | /* | |
444 | * we aren't doing an inline extent round the compressed size | |
445 | * up to a block size boundary so the allocator does sane | |
446 | * things | |
447 | */ | |
448 | total_compressed = (total_compressed + blocksize - 1) & | |
449 | ~(blocksize - 1); | |
450 | ||
451 | /* | |
452 | * one last check to make sure the compression is really a | |
453 | * win, compare the page count read with the blocks on disk | |
454 | */ | |
455 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
456 | ~(PAGE_CACHE_SIZE - 1); | |
457 | if (total_compressed >= total_in) { | |
458 | will_compress = 0; | |
459 | } else { | |
460 | disk_num_bytes = total_compressed; | |
461 | num_bytes = total_in; | |
462 | } | |
463 | } | |
464 | if (!will_compress && pages) { | |
465 | /* | |
466 | * the compression code ran but failed to make things smaller, | |
467 | * free any pages it allocated and our page pointer array | |
468 | */ | |
469 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 470 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
471 | page_cache_release(pages[i]); |
472 | } | |
473 | kfree(pages); | |
474 | pages = NULL; | |
475 | total_compressed = 0; | |
476 | nr_pages_ret = 0; | |
477 | ||
478 | /* flag the file so we don't compress in the future */ | |
479 | btrfs_set_flag(inode, NOCOMPRESS); | |
480 | } | |
771ed689 CM |
481 | if (will_compress) { |
482 | *num_added += 1; | |
c8b97818 | 483 | |
771ed689 CM |
484 | /* the async work queues will take care of doing actual |
485 | * allocation on disk for these compressed pages, | |
486 | * and will submit them to the elevator. | |
487 | */ | |
488 | add_async_extent(async_cow, start, num_bytes, | |
489 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 490 | |
42dc7bab | 491 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
492 | start += num_bytes; |
493 | pages = NULL; | |
494 | cond_resched(); | |
495 | goto again; | |
496 | } | |
497 | } else { | |
498 | /* | |
499 | * No compression, but we still need to write the pages in | |
500 | * the file we've been given so far. redirty the locked | |
501 | * page if it corresponds to our extent and set things up | |
502 | * for the async work queue to run cow_file_range to do | |
503 | * the normal delalloc dance | |
504 | */ | |
505 | if (page_offset(locked_page) >= start && | |
506 | page_offset(locked_page) <= end) { | |
507 | __set_page_dirty_nobuffers(locked_page); | |
508 | /* unlocked later on in the async handlers */ | |
509 | } | |
510 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
511 | *num_added += 1; | |
512 | } | |
3b951516 | 513 | |
771ed689 CM |
514 | out: |
515 | return 0; | |
516 | ||
517 | free_pages_out: | |
518 | for (i = 0; i < nr_pages_ret; i++) { | |
519 | WARN_ON(pages[i]->mapping); | |
520 | page_cache_release(pages[i]); | |
521 | } | |
d397712b | 522 | kfree(pages); |
771ed689 CM |
523 | |
524 | goto out; | |
525 | } | |
526 | ||
527 | /* | |
528 | * phase two of compressed writeback. This is the ordered portion | |
529 | * of the code, which only gets called in the order the work was | |
530 | * queued. We walk all the async extents created by compress_file_range | |
531 | * and send them down to the disk. | |
532 | */ | |
533 | static noinline int submit_compressed_extents(struct inode *inode, | |
534 | struct async_cow *async_cow) | |
535 | { | |
536 | struct async_extent *async_extent; | |
537 | u64 alloc_hint = 0; | |
538 | struct btrfs_trans_handle *trans; | |
539 | struct btrfs_key ins; | |
540 | struct extent_map *em; | |
541 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
542 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
543 | struct extent_io_tree *io_tree; | |
544 | int ret; | |
545 | ||
546 | if (list_empty(&async_cow->extents)) | |
547 | return 0; | |
548 | ||
549 | trans = btrfs_join_transaction(root, 1); | |
550 | ||
d397712b | 551 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
552 | async_extent = list_entry(async_cow->extents.next, |
553 | struct async_extent, list); | |
554 | list_del(&async_extent->list); | |
c8b97818 | 555 | |
771ed689 CM |
556 | io_tree = &BTRFS_I(inode)->io_tree; |
557 | ||
558 | /* did the compression code fall back to uncompressed IO? */ | |
559 | if (!async_extent->pages) { | |
560 | int page_started = 0; | |
561 | unsigned long nr_written = 0; | |
562 | ||
563 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
564 | async_extent->start + |
565 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
566 | |
567 | /* allocate blocks */ | |
568 | cow_file_range(inode, async_cow->locked_page, | |
569 | async_extent->start, | |
570 | async_extent->start + | |
571 | async_extent->ram_size - 1, | |
572 | &page_started, &nr_written, 0); | |
573 | ||
574 | /* | |
575 | * if page_started, cow_file_range inserted an | |
576 | * inline extent and took care of all the unlocking | |
577 | * and IO for us. Otherwise, we need to submit | |
578 | * all those pages down to the drive. | |
579 | */ | |
580 | if (!page_started) | |
581 | extent_write_locked_range(io_tree, | |
582 | inode, async_extent->start, | |
d397712b | 583 | async_extent->start + |
771ed689 CM |
584 | async_extent->ram_size - 1, |
585 | btrfs_get_extent, | |
586 | WB_SYNC_ALL); | |
587 | kfree(async_extent); | |
588 | cond_resched(); | |
589 | continue; | |
590 | } | |
591 | ||
592 | lock_extent(io_tree, async_extent->start, | |
593 | async_extent->start + async_extent->ram_size - 1, | |
594 | GFP_NOFS); | |
c8b97818 | 595 | /* |
771ed689 CM |
596 | * here we're doing allocation and writeback of the |
597 | * compressed pages | |
c8b97818 | 598 | */ |
771ed689 CM |
599 | btrfs_drop_extent_cache(inode, async_extent->start, |
600 | async_extent->start + | |
601 | async_extent->ram_size - 1, 0); | |
602 | ||
603 | ret = btrfs_reserve_extent(trans, root, | |
604 | async_extent->compressed_size, | |
605 | async_extent->compressed_size, | |
606 | 0, alloc_hint, | |
607 | (u64)-1, &ins, 1); | |
608 | BUG_ON(ret); | |
609 | em = alloc_extent_map(GFP_NOFS); | |
610 | em->start = async_extent->start; | |
611 | em->len = async_extent->ram_size; | |
445a6944 | 612 | em->orig_start = em->start; |
c8b97818 | 613 | |
771ed689 CM |
614 | em->block_start = ins.objectid; |
615 | em->block_len = ins.offset; | |
616 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
617 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
618 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
619 | ||
d397712b | 620 | while (1) { |
771ed689 CM |
621 | spin_lock(&em_tree->lock); |
622 | ret = add_extent_mapping(em_tree, em); | |
623 | spin_unlock(&em_tree->lock); | |
624 | if (ret != -EEXIST) { | |
625 | free_extent_map(em); | |
626 | break; | |
627 | } | |
628 | btrfs_drop_extent_cache(inode, async_extent->start, | |
629 | async_extent->start + | |
630 | async_extent->ram_size - 1, 0); | |
631 | } | |
632 | ||
633 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
634 | ins.objectid, | |
635 | async_extent->ram_size, | |
636 | ins.offset, | |
637 | BTRFS_ORDERED_COMPRESSED); | |
638 | BUG_ON(ret); | |
639 | ||
640 | btrfs_end_transaction(trans, root); | |
641 | ||
642 | /* | |
643 | * clear dirty, set writeback and unlock the pages. | |
644 | */ | |
645 | extent_clear_unlock_delalloc(inode, | |
646 | &BTRFS_I(inode)->io_tree, | |
647 | async_extent->start, | |
648 | async_extent->start + | |
649 | async_extent->ram_size - 1, | |
650 | NULL, 1, 1, 0, 1, 1, 0); | |
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, | |
721 | &BTRFS_I(inode)->io_tree, | |
722 | start, end, NULL, 1, 1, | |
723 | 1, 1, 1, 1); | |
724 | *nr_written = *nr_written + | |
725 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
726 | *page_started = 1; | |
727 | ret = 0; | |
728 | goto out; | |
729 | } | |
730 | } | |
731 | ||
732 | BUG_ON(disk_num_bytes > | |
733 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
734 | ||
735 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
736 | ||
d397712b | 737 | while (disk_num_bytes > 0) { |
c8b97818 | 738 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 739 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 740 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 741 | (u64)-1, &ins, 1); |
d397712b CM |
742 | BUG_ON(ret); |
743 | ||
e6dcd2dc CM |
744 | em = alloc_extent_map(GFP_NOFS); |
745 | em->start = start; | |
445a6944 | 746 | em->orig_start = em->start; |
c8b97818 | 747 | |
771ed689 CM |
748 | ram_size = ins.offset; |
749 | em->len = ins.offset; | |
c8b97818 | 750 | |
e6dcd2dc | 751 | em->block_start = ins.objectid; |
c8b97818 | 752 | em->block_len = ins.offset; |
e6dcd2dc | 753 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 754 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 755 | |
d397712b | 756 | while (1) { |
e6dcd2dc CM |
757 | spin_lock(&em_tree->lock); |
758 | ret = add_extent_mapping(em_tree, em); | |
759 | spin_unlock(&em_tree->lock); | |
760 | if (ret != -EEXIST) { | |
761 | free_extent_map(em); | |
762 | break; | |
763 | } | |
764 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 765 | start + ram_size - 1, 0); |
e6dcd2dc CM |
766 | } |
767 | ||
98d20f67 | 768 | cur_alloc_size = ins.offset; |
e6dcd2dc | 769 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 770 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 771 | BUG_ON(ret); |
c8b97818 | 772 | |
17d217fe YZ |
773 | if (root->root_key.objectid == |
774 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
775 | ret = btrfs_reloc_clone_csums(inode, start, | |
776 | cur_alloc_size); | |
777 | BUG_ON(ret); | |
778 | } | |
779 | ||
d397712b | 780 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 781 | break; |
d397712b | 782 | |
c8b97818 CM |
783 | /* we're not doing compressed IO, don't unlock the first |
784 | * page (which the caller expects to stay locked), don't | |
785 | * clear any dirty bits and don't set any writeback bits | |
786 | */ | |
787 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
788 | start, start + ram_size - 1, | |
771ed689 CM |
789 | locked_page, unlock, 1, |
790 | 1, 0, 0, 0); | |
c8b97818 | 791 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
792 | num_bytes -= cur_alloc_size; |
793 | alloc_hint = ins.objectid + ins.offset; | |
794 | start += cur_alloc_size; | |
b888db2b | 795 | } |
b888db2b | 796 | out: |
771ed689 | 797 | ret = 0; |
b888db2b | 798 | btrfs_end_transaction(trans, root); |
c8b97818 | 799 | |
be20aa9d | 800 | return ret; |
771ed689 | 801 | } |
c8b97818 | 802 | |
771ed689 CM |
803 | /* |
804 | * work queue call back to started compression on a file and pages | |
805 | */ | |
806 | static noinline void async_cow_start(struct btrfs_work *work) | |
807 | { | |
808 | struct async_cow *async_cow; | |
809 | int num_added = 0; | |
810 | async_cow = container_of(work, struct async_cow, work); | |
811 | ||
812 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
813 | async_cow->start, async_cow->end, async_cow, | |
814 | &num_added); | |
815 | if (num_added == 0) | |
816 | async_cow->inode = NULL; | |
817 | } | |
818 | ||
819 | /* | |
820 | * work queue call back to submit previously compressed pages | |
821 | */ | |
822 | static noinline void async_cow_submit(struct btrfs_work *work) | |
823 | { | |
824 | struct async_cow *async_cow; | |
825 | struct btrfs_root *root; | |
826 | unsigned long nr_pages; | |
827 | ||
828 | async_cow = container_of(work, struct async_cow, work); | |
829 | ||
830 | root = async_cow->root; | |
831 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
832 | PAGE_CACHE_SHIFT; | |
833 | ||
834 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
835 | ||
836 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
837 | 5 * 1042 * 1024 && | |
838 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
839 | wake_up(&root->fs_info->async_submit_wait); | |
840 | ||
d397712b | 841 | if (async_cow->inode) |
771ed689 | 842 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 843 | } |
c8b97818 | 844 | |
771ed689 CM |
845 | static noinline void async_cow_free(struct btrfs_work *work) |
846 | { | |
847 | struct async_cow *async_cow; | |
848 | async_cow = container_of(work, struct async_cow, work); | |
849 | kfree(async_cow); | |
850 | } | |
851 | ||
852 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
853 | u64 start, u64 end, int *page_started, | |
854 | unsigned long *nr_written) | |
855 | { | |
856 | struct async_cow *async_cow; | |
857 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
858 | unsigned long nr_pages; | |
859 | u64 cur_end; | |
860 | int limit = 10 * 1024 * 1042; | |
861 | ||
862 | if (!btrfs_test_opt(root, COMPRESS)) { | |
863 | return cow_file_range(inode, locked_page, start, end, | |
864 | page_started, nr_written, 1); | |
865 | } | |
866 | ||
867 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | | |
868 | EXTENT_DELALLOC, 1, 0, GFP_NOFS); | |
d397712b | 869 | while (start < end) { |
771ed689 CM |
870 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
871 | async_cow->inode = inode; | |
872 | async_cow->root = root; | |
873 | async_cow->locked_page = locked_page; | |
874 | async_cow->start = start; | |
875 | ||
876 | if (btrfs_test_flag(inode, NOCOMPRESS)) | |
877 | cur_end = end; | |
878 | else | |
879 | cur_end = min(end, start + 512 * 1024 - 1); | |
880 | ||
881 | async_cow->end = cur_end; | |
882 | INIT_LIST_HEAD(&async_cow->extents); | |
883 | ||
884 | async_cow->work.func = async_cow_start; | |
885 | async_cow->work.ordered_func = async_cow_submit; | |
886 | async_cow->work.ordered_free = async_cow_free; | |
887 | async_cow->work.flags = 0; | |
888 | ||
771ed689 CM |
889 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
890 | PAGE_CACHE_SHIFT; | |
891 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
892 | ||
893 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
894 | &async_cow->work); | |
895 | ||
896 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
897 | wait_event(root->fs_info->async_submit_wait, | |
898 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
899 | limit)); | |
900 | } | |
901 | ||
d397712b | 902 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
903 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
904 | wait_event(root->fs_info->async_submit_wait, | |
905 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
906 | 0)); | |
907 | } | |
908 | ||
909 | *nr_written += nr_pages; | |
910 | start = cur_end + 1; | |
911 | } | |
912 | *page_started = 1; | |
913 | return 0; | |
be20aa9d CM |
914 | } |
915 | ||
d397712b | 916 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
917 | u64 bytenr, u64 num_bytes) |
918 | { | |
919 | int ret; | |
920 | struct btrfs_ordered_sum *sums; | |
921 | LIST_HEAD(list); | |
922 | ||
923 | ret = btrfs_lookup_csums_range(root, bytenr, bytenr + num_bytes - 1, | |
924 | &list); | |
925 | if (ret == 0 && list_empty(&list)) | |
926 | return 0; | |
927 | ||
928 | while (!list_empty(&list)) { | |
929 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
930 | list_del(&sums->list); | |
931 | kfree(sums); | |
932 | } | |
933 | return 1; | |
934 | } | |
935 | ||
d352ac68 CM |
936 | /* |
937 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
938 | * of the extents that exist in the file, and COWs the file as required. | |
939 | * | |
940 | * If no cow copies or snapshots exist, we write directly to the existing | |
941 | * blocks on disk | |
942 | */ | |
c8b97818 | 943 | static int run_delalloc_nocow(struct inode *inode, struct page *locked_page, |
771ed689 CM |
944 | u64 start, u64 end, int *page_started, int force, |
945 | unsigned long *nr_written) | |
be20aa9d | 946 | { |
be20aa9d | 947 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 948 | struct btrfs_trans_handle *trans; |
be20aa9d | 949 | struct extent_buffer *leaf; |
be20aa9d | 950 | struct btrfs_path *path; |
80ff3856 | 951 | struct btrfs_file_extent_item *fi; |
be20aa9d | 952 | struct btrfs_key found_key; |
80ff3856 YZ |
953 | u64 cow_start; |
954 | u64 cur_offset; | |
955 | u64 extent_end; | |
956 | u64 disk_bytenr; | |
957 | u64 num_bytes; | |
958 | int extent_type; | |
959 | int ret; | |
d899e052 | 960 | int type; |
80ff3856 YZ |
961 | int nocow; |
962 | int check_prev = 1; | |
be20aa9d CM |
963 | |
964 | path = btrfs_alloc_path(); | |
965 | BUG_ON(!path); | |
7ea394f1 YZ |
966 | trans = btrfs_join_transaction(root, 1); |
967 | BUG_ON(!trans); | |
be20aa9d | 968 | |
80ff3856 YZ |
969 | cow_start = (u64)-1; |
970 | cur_offset = start; | |
971 | while (1) { | |
972 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
973 | cur_offset, 0); | |
974 | BUG_ON(ret < 0); | |
975 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
976 | leaf = path->nodes[0]; | |
977 | btrfs_item_key_to_cpu(leaf, &found_key, | |
978 | path->slots[0] - 1); | |
979 | if (found_key.objectid == inode->i_ino && | |
980 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
981 | path->slots[0]--; | |
982 | } | |
983 | check_prev = 0; | |
984 | next_slot: | |
985 | leaf = path->nodes[0]; | |
986 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
987 | ret = btrfs_next_leaf(root, path); | |
988 | if (ret < 0) | |
989 | BUG_ON(1); | |
990 | if (ret > 0) | |
991 | break; | |
992 | leaf = path->nodes[0]; | |
993 | } | |
be20aa9d | 994 | |
80ff3856 YZ |
995 | nocow = 0; |
996 | disk_bytenr = 0; | |
17d217fe | 997 | num_bytes = 0; |
80ff3856 YZ |
998 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
999 | ||
1000 | if (found_key.objectid > inode->i_ino || | |
1001 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1002 | found_key.offset > end) | |
1003 | break; | |
1004 | ||
1005 | if (found_key.offset > cur_offset) { | |
1006 | extent_end = found_key.offset; | |
1007 | goto out_check; | |
1008 | } | |
1009 | ||
1010 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1011 | struct btrfs_file_extent_item); | |
1012 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1013 | ||
d899e052 YZ |
1014 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1015 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 YZ |
1016 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
1017 | extent_end = found_key.offset + | |
1018 | btrfs_file_extent_num_bytes(leaf, fi); | |
1019 | if (extent_end <= start) { | |
1020 | path->slots[0]++; | |
1021 | goto next_slot; | |
1022 | } | |
17d217fe YZ |
1023 | if (disk_bytenr == 0) |
1024 | goto out_check; | |
80ff3856 YZ |
1025 | if (btrfs_file_extent_compression(leaf, fi) || |
1026 | btrfs_file_extent_encryption(leaf, fi) || | |
1027 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1028 | goto out_check; | |
d899e052 YZ |
1029 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1030 | goto out_check; | |
d2fb3437 | 1031 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1032 | goto out_check; |
17d217fe YZ |
1033 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
1034 | disk_bytenr)) | |
1035 | goto out_check; | |
80ff3856 | 1036 | disk_bytenr += btrfs_file_extent_offset(leaf, fi); |
17d217fe YZ |
1037 | disk_bytenr += cur_offset - found_key.offset; |
1038 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1039 | /* | |
1040 | * force cow if csum exists in the range. | |
1041 | * this ensure that csum for a given extent are | |
1042 | * either valid or do not exist. | |
1043 | */ | |
1044 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1045 | goto out_check; | |
80ff3856 YZ |
1046 | nocow = 1; |
1047 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1048 | extent_end = found_key.offset + | |
1049 | btrfs_file_extent_inline_len(leaf, fi); | |
1050 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1051 | } else { | |
1052 | BUG_ON(1); | |
1053 | } | |
1054 | out_check: | |
1055 | if (extent_end <= start) { | |
1056 | path->slots[0]++; | |
1057 | goto next_slot; | |
1058 | } | |
1059 | if (!nocow) { | |
1060 | if (cow_start == (u64)-1) | |
1061 | cow_start = cur_offset; | |
1062 | cur_offset = extent_end; | |
1063 | if (cur_offset > end) | |
1064 | break; | |
1065 | path->slots[0]++; | |
1066 | goto next_slot; | |
7ea394f1 YZ |
1067 | } |
1068 | ||
1069 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1070 | if (cow_start != (u64)-1) { |
1071 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1072 | found_key.offset - 1, page_started, |
1073 | nr_written, 1); | |
80ff3856 YZ |
1074 | BUG_ON(ret); |
1075 | cow_start = (u64)-1; | |
7ea394f1 | 1076 | } |
80ff3856 | 1077 | |
d899e052 YZ |
1078 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1079 | struct extent_map *em; | |
1080 | struct extent_map_tree *em_tree; | |
1081 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1082 | em = alloc_extent_map(GFP_NOFS); | |
1083 | em->start = cur_offset; | |
445a6944 | 1084 | em->orig_start = em->start; |
d899e052 YZ |
1085 | em->len = num_bytes; |
1086 | em->block_len = num_bytes; | |
1087 | em->block_start = disk_bytenr; | |
1088 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1089 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1090 | while (1) { | |
1091 | spin_lock(&em_tree->lock); | |
1092 | ret = add_extent_mapping(em_tree, em); | |
1093 | spin_unlock(&em_tree->lock); | |
1094 | if (ret != -EEXIST) { | |
1095 | free_extent_map(em); | |
1096 | break; | |
1097 | } | |
1098 | btrfs_drop_extent_cache(inode, em->start, | |
1099 | em->start + em->len - 1, 0); | |
1100 | } | |
1101 | type = BTRFS_ORDERED_PREALLOC; | |
1102 | } else { | |
1103 | type = BTRFS_ORDERED_NOCOW; | |
1104 | } | |
80ff3856 YZ |
1105 | |
1106 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1107 | num_bytes, num_bytes, type); |
1108 | BUG_ON(ret); | |
771ed689 | 1109 | |
d899e052 YZ |
1110 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
1111 | cur_offset, cur_offset + num_bytes - 1, | |
771ed689 | 1112 | locked_page, 1, 1, 1, 0, 0, 0); |
80ff3856 YZ |
1113 | cur_offset = extent_end; |
1114 | if (cur_offset > end) | |
1115 | break; | |
be20aa9d | 1116 | } |
80ff3856 YZ |
1117 | btrfs_release_path(root, path); |
1118 | ||
1119 | if (cur_offset <= end && cow_start == (u64)-1) | |
1120 | cow_start = cur_offset; | |
1121 | if (cow_start != (u64)-1) { | |
1122 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1123 | page_started, nr_written, 1); |
80ff3856 YZ |
1124 | BUG_ON(ret); |
1125 | } | |
1126 | ||
1127 | ret = btrfs_end_transaction(trans, root); | |
1128 | BUG_ON(ret); | |
7ea394f1 | 1129 | btrfs_free_path(path); |
80ff3856 | 1130 | return 0; |
be20aa9d CM |
1131 | } |
1132 | ||
d352ac68 CM |
1133 | /* |
1134 | * extent_io.c call back to do delayed allocation processing | |
1135 | */ | |
c8b97818 | 1136 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1137 | u64 start, u64 end, int *page_started, |
1138 | unsigned long *nr_written) | |
be20aa9d | 1139 | { |
be20aa9d | 1140 | int ret; |
a2135011 | 1141 | |
17d217fe | 1142 | if (btrfs_test_flag(inode, NODATACOW)) |
c8b97818 | 1143 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1144 | page_started, 1, nr_written); |
d899e052 YZ |
1145 | else if (btrfs_test_flag(inode, PREALLOC)) |
1146 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1147 | page_started, 0, nr_written); |
be20aa9d | 1148 | else |
771ed689 | 1149 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1150 | page_started, nr_written); |
1832a6d5 | 1151 | |
b888db2b CM |
1152 | return ret; |
1153 | } | |
1154 | ||
d352ac68 CM |
1155 | /* |
1156 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1157 | * bytes in this file, and to maintain the list of inodes that | |
1158 | * have pending delalloc work to be done. | |
1159 | */ | |
b2950863 | 1160 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1161 | unsigned long old, unsigned long bits) |
291d673e | 1162 | { |
75eff68e CM |
1163 | /* |
1164 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1165 | * but in this case, we are only testeing for the DELALLOC | |
1166 | * bit, which is only set or cleared with irqs on | |
1167 | */ | |
b0c68f8b | 1168 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1169 | struct btrfs_root *root = BTRFS_I(inode)->root; |
75eff68e | 1170 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1171 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1172 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1173 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1174 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1175 | &root->fs_info->delalloc_inodes); | |
1176 | } | |
75eff68e | 1177 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1178 | } |
1179 | return 0; | |
1180 | } | |
1181 | ||
d352ac68 CM |
1182 | /* |
1183 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1184 | */ | |
b2950863 | 1185 | static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1186 | unsigned long old, unsigned long bits) |
291d673e | 1187 | { |
75eff68e CM |
1188 | /* |
1189 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1190 | * but in this case, we are only testeing for the DELALLOC | |
1191 | * bit, which is only set or cleared with irqs on | |
1192 | */ | |
b0c68f8b | 1193 | if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1194 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1195 | |
75eff68e | 1196 | spin_lock(&root->fs_info->delalloc_lock); |
b0c68f8b | 1197 | if (end - start + 1 > root->fs_info->delalloc_bytes) { |
d397712b CM |
1198 | printk(KERN_INFO "btrfs warning: delalloc account " |
1199 | "%llu %llu\n", | |
1200 | (unsigned long long)end - start + 1, | |
1201 | (unsigned long long) | |
1202 | root->fs_info->delalloc_bytes); | |
b0c68f8b | 1203 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1204 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b CM |
1205 | } else { |
1206 | root->fs_info->delalloc_bytes -= end - start + 1; | |
9069218d | 1207 | BTRFS_I(inode)->delalloc_bytes -= end - start + 1; |
b0c68f8b | 1208 | } |
ea8c2819 CM |
1209 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1210 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1211 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1212 | } | |
75eff68e | 1213 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1214 | } |
1215 | return 0; | |
1216 | } | |
1217 | ||
d352ac68 CM |
1218 | /* |
1219 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1220 | * we don't create bios that span stripes or chunks | |
1221 | */ | |
239b14b3 | 1222 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1223 | size_t size, struct bio *bio, |
1224 | unsigned long bio_flags) | |
239b14b3 CM |
1225 | { |
1226 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1227 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1228 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1229 | u64 length = 0; |
1230 | u64 map_length; | |
239b14b3 CM |
1231 | int ret; |
1232 | ||
771ed689 CM |
1233 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1234 | return 0; | |
1235 | ||
f2d8d74d | 1236 | length = bio->bi_size; |
239b14b3 CM |
1237 | map_tree = &root->fs_info->mapping_tree; |
1238 | map_length = length; | |
cea9e445 | 1239 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1240 | &map_length, NULL, 0); |
cea9e445 | 1241 | |
d397712b | 1242 | if (map_length < length + size) |
239b14b3 | 1243 | return 1; |
239b14b3 CM |
1244 | return 0; |
1245 | } | |
1246 | ||
d352ac68 CM |
1247 | /* |
1248 | * in order to insert checksums into the metadata in large chunks, | |
1249 | * we wait until bio submission time. All the pages in the bio are | |
1250 | * checksummed and sums are attached onto the ordered extent record. | |
1251 | * | |
1252 | * At IO completion time the cums attached on the ordered extent record | |
1253 | * are inserted into the btree | |
1254 | */ | |
d397712b CM |
1255 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1256 | struct bio *bio, int mirror_num, | |
1257 | unsigned long bio_flags) | |
065631f6 | 1258 | { |
065631f6 | 1259 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1260 | int ret = 0; |
e015640f | 1261 | |
d20f7043 | 1262 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1263 | BUG_ON(ret); |
4a69a410 CM |
1264 | return 0; |
1265 | } | |
e015640f | 1266 | |
4a69a410 CM |
1267 | /* |
1268 | * in order to insert checksums into the metadata in large chunks, | |
1269 | * we wait until bio submission time. All the pages in the bio are | |
1270 | * checksummed and sums are attached onto the ordered extent record. | |
1271 | * | |
1272 | * At IO completion time the cums attached on the ordered extent record | |
1273 | * are inserted into the btree | |
1274 | */ | |
b2950863 | 1275 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1276 | int mirror_num, unsigned long bio_flags) |
1277 | { | |
1278 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1279 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1280 | } |
1281 | ||
d352ac68 | 1282 | /* |
cad321ad CM |
1283 | * extent_io.c submission hook. This does the right thing for csum calculation |
1284 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1285 | */ |
b2950863 | 1286 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1287 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1288 | { |
1289 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1290 | int ret = 0; | |
19b9bdb0 | 1291 | int skip_sum; |
44b8bd7e | 1292 | |
cad321ad CM |
1293 | skip_sum = btrfs_test_flag(inode, NODATASUM); |
1294 | ||
e6dcd2dc CM |
1295 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1296 | BUG_ON(ret); | |
065631f6 | 1297 | |
4d1b5fb4 | 1298 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1299 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1300 | return btrfs_submit_compressed_read(inode, bio, |
1301 | mirror_num, bio_flags); | |
d20f7043 CM |
1302 | } else if (!skip_sum) |
1303 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1304 | goto mapit; |
19b9bdb0 | 1305 | } else if (!skip_sum) { |
17d217fe YZ |
1306 | /* csum items have already been cloned */ |
1307 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1308 | goto mapit; | |
19b9bdb0 CM |
1309 | /* we're doing a write, do the async checksumming */ |
1310 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1311 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1312 | bio_flags, __btrfs_submit_bio_start, |
1313 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1314 | } |
1315 | ||
0b86a832 | 1316 | mapit: |
8b712842 | 1317 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1318 | } |
6885f308 | 1319 | |
d352ac68 CM |
1320 | /* |
1321 | * given a list of ordered sums record them in the inode. This happens | |
1322 | * at IO completion time based on sums calculated at bio submission time. | |
1323 | */ | |
ba1da2f4 | 1324 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1325 | struct inode *inode, u64 file_offset, |
1326 | struct list_head *list) | |
1327 | { | |
1328 | struct list_head *cur; | |
1329 | struct btrfs_ordered_sum *sum; | |
1330 | ||
1331 | btrfs_set_trans_block_group(trans, inode); | |
ba1da2f4 | 1332 | list_for_each(cur, list) { |
e6dcd2dc | 1333 | sum = list_entry(cur, struct btrfs_ordered_sum, list); |
d20f7043 CM |
1334 | btrfs_csum_file_blocks(trans, |
1335 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1336 | } |
1337 | return 0; | |
1338 | } | |
1339 | ||
ea8c2819 CM |
1340 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1341 | { | |
d397712b | 1342 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1343 | WARN_ON(1); |
ea8c2819 CM |
1344 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1345 | GFP_NOFS); | |
1346 | } | |
1347 | ||
d352ac68 | 1348 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1349 | struct btrfs_writepage_fixup { |
1350 | struct page *page; | |
1351 | struct btrfs_work work; | |
1352 | }; | |
1353 | ||
b2950863 | 1354 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1355 | { |
1356 | struct btrfs_writepage_fixup *fixup; | |
1357 | struct btrfs_ordered_extent *ordered; | |
1358 | struct page *page; | |
1359 | struct inode *inode; | |
1360 | u64 page_start; | |
1361 | u64 page_end; | |
1362 | ||
1363 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1364 | page = fixup->page; | |
4a096752 | 1365 | again: |
247e743c CM |
1366 | lock_page(page); |
1367 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1368 | ClearPageChecked(page); | |
1369 | goto out_page; | |
1370 | } | |
1371 | ||
1372 | inode = page->mapping->host; | |
1373 | page_start = page_offset(page); | |
1374 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1375 | ||
1376 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1377 | |
1378 | /* already ordered? We're done */ | |
1379 | if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
1380 | EXTENT_ORDERED, 0)) { | |
247e743c | 1381 | goto out; |
4a096752 CM |
1382 | } |
1383 | ||
1384 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1385 | if (ordered) { | |
1386 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1387 | page_end, GFP_NOFS); | |
1388 | unlock_page(page); | |
1389 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1390 | goto again; | |
1391 | } | |
247e743c | 1392 | |
ea8c2819 | 1393 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1394 | ClearPageChecked(page); |
1395 | out: | |
1396 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1397 | out_page: | |
1398 | unlock_page(page); | |
1399 | page_cache_release(page); | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * There are a few paths in the higher layers of the kernel that directly | |
1404 | * set the page dirty bit without asking the filesystem if it is a | |
1405 | * good idea. This causes problems because we want to make sure COW | |
1406 | * properly happens and the data=ordered rules are followed. | |
1407 | * | |
c8b97818 | 1408 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1409 | * hasn't been properly setup for IO. We kick off an async process |
1410 | * to fix it up. The async helper will wait for ordered extents, set | |
1411 | * the delalloc bit and make it safe to write the page. | |
1412 | */ | |
b2950863 | 1413 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1414 | { |
1415 | struct inode *inode = page->mapping->host; | |
1416 | struct btrfs_writepage_fixup *fixup; | |
1417 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1418 | int ret; | |
1419 | ||
1420 | ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1421 | EXTENT_ORDERED, 0); | |
1422 | if (ret) | |
1423 | return 0; | |
1424 | ||
1425 | if (PageChecked(page)) | |
1426 | return -EAGAIN; | |
1427 | ||
1428 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1429 | if (!fixup) | |
1430 | return -EAGAIN; | |
f421950f | 1431 | |
247e743c CM |
1432 | SetPageChecked(page); |
1433 | page_cache_get(page); | |
1434 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1435 | fixup->page = page; | |
1436 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1437 | return -EAGAIN; | |
1438 | } | |
1439 | ||
d899e052 YZ |
1440 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1441 | struct inode *inode, u64 file_pos, | |
1442 | u64 disk_bytenr, u64 disk_num_bytes, | |
1443 | u64 num_bytes, u64 ram_bytes, | |
1444 | u8 compression, u8 encryption, | |
1445 | u16 other_encoding, int extent_type) | |
1446 | { | |
1447 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1448 | struct btrfs_file_extent_item *fi; | |
1449 | struct btrfs_path *path; | |
1450 | struct extent_buffer *leaf; | |
1451 | struct btrfs_key ins; | |
1452 | u64 hint; | |
1453 | int ret; | |
1454 | ||
1455 | path = btrfs_alloc_path(); | |
1456 | BUG_ON(!path); | |
1457 | ||
1458 | ret = btrfs_drop_extents(trans, root, inode, file_pos, | |
1459 | file_pos + num_bytes, file_pos, &hint); | |
1460 | BUG_ON(ret); | |
1461 | ||
1462 | ins.objectid = inode->i_ino; | |
1463 | ins.offset = file_pos; | |
1464 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1465 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1466 | BUG_ON(ret); | |
1467 | leaf = path->nodes[0]; | |
1468 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1469 | struct btrfs_file_extent_item); | |
1470 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1471 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1472 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1473 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1474 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1475 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1476 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1477 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1478 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1479 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
1480 | btrfs_mark_buffer_dirty(leaf); | |
1481 | ||
1482 | inode_add_bytes(inode, num_bytes); | |
1483 | btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0); | |
1484 | ||
1485 | ins.objectid = disk_bytenr; | |
1486 | ins.offset = disk_num_bytes; | |
1487 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1488 | ret = btrfs_alloc_reserved_extent(trans, root, leaf->start, | |
1489 | root->root_key.objectid, | |
1490 | trans->transid, inode->i_ino, &ins); | |
1491 | BUG_ON(ret); | |
1492 | ||
1493 | btrfs_free_path(path); | |
1494 | return 0; | |
1495 | } | |
1496 | ||
d352ac68 CM |
1497 | /* as ordered data IO finishes, this gets called so we can finish |
1498 | * an ordered extent if the range of bytes in the file it covers are | |
1499 | * fully written. | |
1500 | */ | |
211f90e6 | 1501 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1502 | { |
e6dcd2dc CM |
1503 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1504 | struct btrfs_trans_handle *trans; | |
1505 | struct btrfs_ordered_extent *ordered_extent; | |
1506 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
d899e052 | 1507 | int compressed = 0; |
e6dcd2dc CM |
1508 | int ret; |
1509 | ||
1510 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1511 | if (!ret) |
e6dcd2dc | 1512 | return 0; |
e6dcd2dc | 1513 | |
f9295749 | 1514 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc CM |
1515 | |
1516 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
1517 | BUG_ON(!ordered_extent); | |
7ea394f1 YZ |
1518 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1519 | goto nocow; | |
e6dcd2dc CM |
1520 | |
1521 | lock_extent(io_tree, ordered_extent->file_offset, | |
1522 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1523 | GFP_NOFS); | |
1524 | ||
c8b97818 | 1525 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1526 | compressed = 1; |
1527 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1528 | BUG_ON(compressed); | |
1529 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1530 | ordered_extent->file_offset, | |
1531 | ordered_extent->file_offset + | |
1532 | ordered_extent->len); | |
1533 | BUG_ON(ret); | |
1534 | } else { | |
1535 | ret = insert_reserved_file_extent(trans, inode, | |
1536 | ordered_extent->file_offset, | |
1537 | ordered_extent->start, | |
1538 | ordered_extent->disk_len, | |
1539 | ordered_extent->len, | |
1540 | ordered_extent->len, | |
1541 | compressed, 0, 0, | |
1542 | BTRFS_FILE_EXTENT_REG); | |
1543 | BUG_ON(ret); | |
1544 | } | |
e6dcd2dc CM |
1545 | unlock_extent(io_tree, ordered_extent->file_offset, |
1546 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1547 | GFP_NOFS); | |
7ea394f1 | 1548 | nocow: |
e6dcd2dc CM |
1549 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1550 | &ordered_extent->list); | |
1551 | ||
34353029 | 1552 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1553 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1554 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1555 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1556 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1557 | |
e6dcd2dc CM |
1558 | /* once for us */ |
1559 | btrfs_put_ordered_extent(ordered_extent); | |
1560 | /* once for the tree */ | |
1561 | btrfs_put_ordered_extent(ordered_extent); | |
1562 | ||
e6dcd2dc CM |
1563 | btrfs_end_transaction(trans, root); |
1564 | return 0; | |
1565 | } | |
1566 | ||
b2950863 | 1567 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1568 | struct extent_state *state, int uptodate) |
1569 | { | |
1570 | return btrfs_finish_ordered_io(page->mapping->host, start, end); | |
1571 | } | |
1572 | ||
d352ac68 CM |
1573 | /* |
1574 | * When IO fails, either with EIO or csum verification fails, we | |
1575 | * try other mirrors that might have a good copy of the data. This | |
1576 | * io_failure_record is used to record state as we go through all the | |
1577 | * mirrors. If another mirror has good data, the page is set up to date | |
1578 | * and things continue. If a good mirror can't be found, the original | |
1579 | * bio end_io callback is called to indicate things have failed. | |
1580 | */ | |
7e38326f CM |
1581 | struct io_failure_record { |
1582 | struct page *page; | |
1583 | u64 start; | |
1584 | u64 len; | |
1585 | u64 logical; | |
d20f7043 | 1586 | unsigned long bio_flags; |
7e38326f CM |
1587 | int last_mirror; |
1588 | }; | |
1589 | ||
b2950863 | 1590 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1591 | struct page *page, u64 start, u64 end, |
1592 | struct extent_state *state) | |
7e38326f CM |
1593 | { |
1594 | struct io_failure_record *failrec = NULL; | |
1595 | u64 private; | |
1596 | struct extent_map *em; | |
1597 | struct inode *inode = page->mapping->host; | |
1598 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1599 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1600 | struct bio *bio; |
1601 | int num_copies; | |
1602 | int ret; | |
1259ab75 | 1603 | int rw; |
7e38326f CM |
1604 | u64 logical; |
1605 | ||
1606 | ret = get_state_private(failure_tree, start, &private); | |
1607 | if (ret) { | |
7e38326f CM |
1608 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1609 | if (!failrec) | |
1610 | return -ENOMEM; | |
1611 | failrec->start = start; | |
1612 | failrec->len = end - start + 1; | |
1613 | failrec->last_mirror = 0; | |
d20f7043 | 1614 | failrec->bio_flags = 0; |
7e38326f | 1615 | |
3b951516 CM |
1616 | spin_lock(&em_tree->lock); |
1617 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
1618 | if (em->start > start || em->start + em->len < start) { | |
1619 | free_extent_map(em); | |
1620 | em = NULL; | |
1621 | } | |
1622 | spin_unlock(&em_tree->lock); | |
7e38326f CM |
1623 | |
1624 | if (!em || IS_ERR(em)) { | |
1625 | kfree(failrec); | |
1626 | return -EIO; | |
1627 | } | |
1628 | logical = start - em->start; | |
1629 | logical = em->block_start + logical; | |
d20f7043 CM |
1630 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1631 | logical = em->block_start; | |
1632 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1633 | } | |
7e38326f CM |
1634 | failrec->logical = logical; |
1635 | free_extent_map(em); | |
1636 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1637 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1638 | set_state_private(failure_tree, start, |
1639 | (u64)(unsigned long)failrec); | |
7e38326f | 1640 | } else { |
587f7704 | 1641 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1642 | } |
1643 | num_copies = btrfs_num_copies( | |
1644 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1645 | failrec->logical, failrec->len); | |
1646 | failrec->last_mirror++; | |
1647 | if (!state) { | |
cad321ad | 1648 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1649 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1650 | failrec->start, | |
1651 | EXTENT_LOCKED); | |
1652 | if (state && state->start != failrec->start) | |
1653 | state = NULL; | |
cad321ad | 1654 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1655 | } |
1656 | if (!state || failrec->last_mirror > num_copies) { | |
1657 | set_state_private(failure_tree, failrec->start, 0); | |
1658 | clear_extent_bits(failure_tree, failrec->start, | |
1659 | failrec->start + failrec->len - 1, | |
1660 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1661 | kfree(failrec); | |
1662 | return -EIO; | |
1663 | } | |
1664 | bio = bio_alloc(GFP_NOFS, 1); | |
1665 | bio->bi_private = state; | |
1666 | bio->bi_end_io = failed_bio->bi_end_io; | |
1667 | bio->bi_sector = failrec->logical >> 9; | |
1668 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1669 | bio->bi_size = 0; |
d20f7043 | 1670 | |
7e38326f | 1671 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1672 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1673 | rw = WRITE; | |
1674 | else | |
1675 | rw = READ; | |
1676 | ||
1677 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1678 | failrec->last_mirror, |
d20f7043 | 1679 | failrec->bio_flags); |
1259ab75 CM |
1680 | return 0; |
1681 | } | |
1682 | ||
d352ac68 CM |
1683 | /* |
1684 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1685 | * to see if we need to process or clean up an io_failure_record | |
1686 | */ | |
b2950863 | 1687 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1688 | { |
1689 | u64 private; | |
1690 | u64 private_failure; | |
1691 | struct io_failure_record *failure; | |
1692 | int ret; | |
1693 | ||
1694 | private = 0; | |
1695 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1696 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1697 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1698 | start, &private_failure); | |
1699 | if (ret == 0) { | |
1700 | failure = (struct io_failure_record *)(unsigned long) | |
1701 | private_failure; | |
1702 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1703 | failure->start, 0); | |
1704 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1705 | failure->start, | |
1706 | failure->start + failure->len - 1, | |
1707 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1708 | GFP_NOFS); | |
1709 | kfree(failure); | |
1710 | } | |
1711 | } | |
7e38326f CM |
1712 | return 0; |
1713 | } | |
1714 | ||
d352ac68 CM |
1715 | /* |
1716 | * when reads are done, we need to check csums to verify the data is correct | |
1717 | * if there's a match, we allow the bio to finish. If not, we go through | |
1718 | * the io_failure_record routines to find good copies | |
1719 | */ | |
b2950863 | 1720 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1721 | struct extent_state *state) |
07157aac | 1722 | { |
35ebb934 | 1723 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1724 | struct inode *inode = page->mapping->host; |
d1310b2e | 1725 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1726 | char *kaddr; |
aadfeb6e | 1727 | u64 private = ~(u32)0; |
07157aac | 1728 | int ret; |
ff79f819 CM |
1729 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1730 | u32 csum = ~(u32)0; | |
bbf0d006 | 1731 | unsigned long flags; |
d1310b2e | 1732 | |
d20f7043 CM |
1733 | if (PageChecked(page)) { |
1734 | ClearPageChecked(page); | |
1735 | goto good; | |
1736 | } | |
17d217fe YZ |
1737 | if (btrfs_test_flag(inode, NODATASUM)) |
1738 | return 0; | |
1739 | ||
1740 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
1741 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) { | |
1742 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, | |
1743 | GFP_NOFS); | |
b6cda9bc | 1744 | return 0; |
17d217fe | 1745 | } |
d20f7043 | 1746 | |
c2e639f0 | 1747 | if (state && state->start == start) { |
70dec807 CM |
1748 | private = state->private; |
1749 | ret = 0; | |
1750 | } else { | |
1751 | ret = get_state_private(io_tree, start, &private); | |
1752 | } | |
bbf0d006 | 1753 | local_irq_save(flags); |
07157aac | 1754 | kaddr = kmap_atomic(page, KM_IRQ0); |
d397712b | 1755 | if (ret) |
07157aac | 1756 | goto zeroit; |
d397712b | 1757 | |
ff79f819 CM |
1758 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1759 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1760 | if (csum != private) |
07157aac | 1761 | goto zeroit; |
d397712b | 1762 | |
07157aac | 1763 | kunmap_atomic(kaddr, KM_IRQ0); |
bbf0d006 | 1764 | local_irq_restore(flags); |
d20f7043 | 1765 | good: |
7e38326f CM |
1766 | /* if the io failure tree for this inode is non-empty, |
1767 | * check to see if we've recovered from a failed IO | |
1768 | */ | |
1259ab75 | 1769 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1770 | return 0; |
1771 | ||
1772 | zeroit: | |
d397712b CM |
1773 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " |
1774 | "private %llu\n", page->mapping->host->i_ino, | |
1775 | (unsigned long long)start, csum, | |
1776 | (unsigned long long)private); | |
db94535d CM |
1777 | memset(kaddr + offset, 1, end - start + 1); |
1778 | flush_dcache_page(page); | |
07157aac | 1779 | kunmap_atomic(kaddr, KM_IRQ0); |
bbf0d006 | 1780 | local_irq_restore(flags); |
3b951516 CM |
1781 | if (private == 0) |
1782 | return 0; | |
7e38326f | 1783 | return -EIO; |
07157aac | 1784 | } |
b888db2b | 1785 | |
7b128766 JB |
1786 | /* |
1787 | * This creates an orphan entry for the given inode in case something goes | |
1788 | * wrong in the middle of an unlink/truncate. | |
1789 | */ | |
1790 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1791 | { | |
1792 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1793 | int ret = 0; | |
1794 | ||
bcc63abb | 1795 | spin_lock(&root->list_lock); |
7b128766 JB |
1796 | |
1797 | /* already on the orphan list, we're good */ | |
1798 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1799 | spin_unlock(&root->list_lock); |
7b128766 JB |
1800 | return 0; |
1801 | } | |
1802 | ||
1803 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
1804 | ||
bcc63abb | 1805 | spin_unlock(&root->list_lock); |
7b128766 JB |
1806 | |
1807 | /* | |
1808 | * insert an orphan item to track this unlinked/truncated file | |
1809 | */ | |
1810 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
1811 | ||
1812 | return ret; | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
1817 | * item for this particular inode. | |
1818 | */ | |
1819 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
1820 | { | |
1821 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1822 | int ret = 0; | |
1823 | ||
bcc63abb | 1824 | spin_lock(&root->list_lock); |
7b128766 JB |
1825 | |
1826 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1827 | spin_unlock(&root->list_lock); |
7b128766 JB |
1828 | return 0; |
1829 | } | |
1830 | ||
1831 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
1832 | if (!trans) { | |
bcc63abb | 1833 | spin_unlock(&root->list_lock); |
7b128766 JB |
1834 | return 0; |
1835 | } | |
1836 | ||
bcc63abb | 1837 | spin_unlock(&root->list_lock); |
7b128766 JB |
1838 | |
1839 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
1840 | ||
1841 | return ret; | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * this cleans up any orphans that may be left on the list from the last use | |
1846 | * of this root. | |
1847 | */ | |
1848 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
1849 | { | |
1850 | struct btrfs_path *path; | |
1851 | struct extent_buffer *leaf; | |
1852 | struct btrfs_item *item; | |
1853 | struct btrfs_key key, found_key; | |
1854 | struct btrfs_trans_handle *trans; | |
1855 | struct inode *inode; | |
1856 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
1857 | ||
7b128766 JB |
1858 | path = btrfs_alloc_path(); |
1859 | if (!path) | |
1860 | return; | |
1861 | path->reada = -1; | |
1862 | ||
1863 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
1864 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
1865 | key.offset = (u64)-1; | |
1866 | ||
7b128766 JB |
1867 | |
1868 | while (1) { | |
1869 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1870 | if (ret < 0) { | |
1871 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
1872 | "\n", ret); | |
1873 | break; | |
1874 | } | |
1875 | ||
1876 | /* | |
1877 | * if ret == 0 means we found what we were searching for, which | |
1878 | * is weird, but possible, so only screw with path if we didnt | |
1879 | * find the key and see if we have stuff that matches | |
1880 | */ | |
1881 | if (ret > 0) { | |
1882 | if (path->slots[0] == 0) | |
1883 | break; | |
1884 | path->slots[0]--; | |
1885 | } | |
1886 | ||
1887 | /* pull out the item */ | |
1888 | leaf = path->nodes[0]; | |
1889 | item = btrfs_item_nr(leaf, path->slots[0]); | |
1890 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1891 | ||
1892 | /* make sure the item matches what we want */ | |
1893 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
1894 | break; | |
1895 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
1896 | break; | |
1897 | ||
1898 | /* release the path since we're done with it */ | |
1899 | btrfs_release_path(root, path); | |
1900 | ||
1901 | /* | |
1902 | * this is where we are basically btrfs_lookup, without the | |
1903 | * crossing root thing. we store the inode number in the | |
1904 | * offset of the orphan item. | |
1905 | */ | |
5b21f2ed | 1906 | inode = btrfs_iget_locked(root->fs_info->sb, |
7b128766 JB |
1907 | found_key.offset, root); |
1908 | if (!inode) | |
1909 | break; | |
1910 | ||
1911 | if (inode->i_state & I_NEW) { | |
1912 | BTRFS_I(inode)->root = root; | |
1913 | ||
1914 | /* have to set the location manually */ | |
1915 | BTRFS_I(inode)->location.objectid = inode->i_ino; | |
1916 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
1917 | BTRFS_I(inode)->location.offset = 0; | |
1918 | ||
1919 | btrfs_read_locked_inode(inode); | |
1920 | unlock_new_inode(inode); | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * add this inode to the orphan list so btrfs_orphan_del does | |
1925 | * the proper thing when we hit it | |
1926 | */ | |
bcc63abb | 1927 | spin_lock(&root->list_lock); |
7b128766 | 1928 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 1929 | spin_unlock(&root->list_lock); |
7b128766 JB |
1930 | |
1931 | /* | |
1932 | * if this is a bad inode, means we actually succeeded in | |
1933 | * removing the inode, but not the orphan record, which means | |
1934 | * we need to manually delete the orphan since iput will just | |
1935 | * do a destroy_inode | |
1936 | */ | |
1937 | if (is_bad_inode(inode)) { | |
5b21f2ed | 1938 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 1939 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 1940 | btrfs_end_transaction(trans, root); |
7b128766 JB |
1941 | iput(inode); |
1942 | continue; | |
1943 | } | |
1944 | ||
1945 | /* if we have links, this was a truncate, lets do that */ | |
1946 | if (inode->i_nlink) { | |
1947 | nr_truncate++; | |
1948 | btrfs_truncate(inode); | |
1949 | } else { | |
1950 | nr_unlink++; | |
1951 | } | |
1952 | ||
1953 | /* this will do delete_inode and everything for us */ | |
1954 | iput(inode); | |
1955 | } | |
1956 | ||
1957 | if (nr_unlink) | |
1958 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
1959 | if (nr_truncate) | |
1960 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
1961 | ||
1962 | btrfs_free_path(path); | |
7b128766 JB |
1963 | } |
1964 | ||
d352ac68 CM |
1965 | /* |
1966 | * read an inode from the btree into the in-memory inode | |
1967 | */ | |
39279cc3 CM |
1968 | void btrfs_read_locked_inode(struct inode *inode) |
1969 | { | |
1970 | struct btrfs_path *path; | |
5f39d397 | 1971 | struct extent_buffer *leaf; |
39279cc3 | 1972 | struct btrfs_inode_item *inode_item; |
0b86a832 | 1973 | struct btrfs_timespec *tspec; |
39279cc3 CM |
1974 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1975 | struct btrfs_key location; | |
1976 | u64 alloc_group_block; | |
618e21d5 | 1977 | u32 rdev; |
39279cc3 CM |
1978 | int ret; |
1979 | ||
1980 | path = btrfs_alloc_path(); | |
1981 | BUG_ON(!path); | |
39279cc3 | 1982 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 1983 | |
39279cc3 | 1984 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 1985 | if (ret) |
39279cc3 | 1986 | goto make_bad; |
39279cc3 | 1987 | |
5f39d397 CM |
1988 | leaf = path->nodes[0]; |
1989 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
1990 | struct btrfs_inode_item); | |
1991 | ||
1992 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
1993 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
1994 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
1995 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 1996 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
1997 | |
1998 | tspec = btrfs_inode_atime(inode_item); | |
1999 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2000 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2001 | ||
2002 | tspec = btrfs_inode_mtime(inode_item); | |
2003 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2004 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2005 | ||
2006 | tspec = btrfs_inode_ctime(inode_item); | |
2007 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2008 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2009 | ||
a76a3cd4 | 2010 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2011 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2012 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2013 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2014 | inode->i_rdev = 0; |
5f39d397 CM |
2015 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2016 | ||
aec7477b | 2017 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2018 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2019 | |
5f39d397 | 2020 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
d2fb3437 YZ |
2021 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2022 | alloc_group_block, 0); | |
39279cc3 CM |
2023 | btrfs_free_path(path); |
2024 | inode_item = NULL; | |
2025 | ||
39279cc3 | 2026 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2027 | case S_IFREG: |
2028 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2029 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2030 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2031 | inode->i_fop = &btrfs_file_operations; |
2032 | inode->i_op = &btrfs_file_inode_operations; | |
2033 | break; | |
2034 | case S_IFDIR: | |
2035 | inode->i_fop = &btrfs_dir_file_operations; | |
2036 | if (root == root->fs_info->tree_root) | |
2037 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2038 | else | |
2039 | inode->i_op = &btrfs_dir_inode_operations; | |
2040 | break; | |
2041 | case S_IFLNK: | |
2042 | inode->i_op = &btrfs_symlink_inode_operations; | |
2043 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2044 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2045 | break; |
618e21d5 JB |
2046 | default: |
2047 | init_special_inode(inode, inode->i_mode, rdev); | |
2048 | break; | |
39279cc3 CM |
2049 | } |
2050 | return; | |
2051 | ||
2052 | make_bad: | |
39279cc3 | 2053 | btrfs_free_path(path); |
39279cc3 CM |
2054 | make_bad_inode(inode); |
2055 | } | |
2056 | ||
d352ac68 CM |
2057 | /* |
2058 | * given a leaf and an inode, copy the inode fields into the leaf | |
2059 | */ | |
e02119d5 CM |
2060 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2061 | struct extent_buffer *leaf, | |
5f39d397 | 2062 | struct btrfs_inode_item *item, |
39279cc3 CM |
2063 | struct inode *inode) |
2064 | { | |
5f39d397 CM |
2065 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2066 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2067 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2068 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2069 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2070 | ||
2071 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2072 | inode->i_atime.tv_sec); | |
2073 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2074 | inode->i_atime.tv_nsec); | |
2075 | ||
2076 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2077 | inode->i_mtime.tv_sec); | |
2078 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2079 | inode->i_mtime.tv_nsec); | |
2080 | ||
2081 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2082 | inode->i_ctime.tv_sec); | |
2083 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2084 | inode->i_ctime.tv_nsec); | |
2085 | ||
a76a3cd4 | 2086 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2087 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2088 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2089 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2090 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2091 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2092 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2093 | } |
2094 | ||
d352ac68 CM |
2095 | /* |
2096 | * copy everything in the in-memory inode into the btree. | |
2097 | */ | |
d397712b CM |
2098 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2099 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2100 | { |
2101 | struct btrfs_inode_item *inode_item; | |
2102 | struct btrfs_path *path; | |
5f39d397 | 2103 | struct extent_buffer *leaf; |
39279cc3 CM |
2104 | int ret; |
2105 | ||
2106 | path = btrfs_alloc_path(); | |
2107 | BUG_ON(!path); | |
39279cc3 CM |
2108 | ret = btrfs_lookup_inode(trans, root, path, |
2109 | &BTRFS_I(inode)->location, 1); | |
2110 | if (ret) { | |
2111 | if (ret > 0) | |
2112 | ret = -ENOENT; | |
2113 | goto failed; | |
2114 | } | |
2115 | ||
5f39d397 CM |
2116 | leaf = path->nodes[0]; |
2117 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2118 | struct btrfs_inode_item); |
2119 | ||
e02119d5 | 2120 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2121 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2122 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2123 | ret = 0; |
2124 | failed: | |
39279cc3 CM |
2125 | btrfs_free_path(path); |
2126 | return ret; | |
2127 | } | |
2128 | ||
2129 | ||
d352ac68 CM |
2130 | /* |
2131 | * unlink helper that gets used here in inode.c and in the tree logging | |
2132 | * recovery code. It remove a link in a directory with a given name, and | |
2133 | * also drops the back refs in the inode to the directory | |
2134 | */ | |
e02119d5 CM |
2135 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2136 | struct btrfs_root *root, | |
2137 | struct inode *dir, struct inode *inode, | |
2138 | const char *name, int name_len) | |
39279cc3 CM |
2139 | { |
2140 | struct btrfs_path *path; | |
39279cc3 | 2141 | int ret = 0; |
5f39d397 | 2142 | struct extent_buffer *leaf; |
39279cc3 | 2143 | struct btrfs_dir_item *di; |
5f39d397 | 2144 | struct btrfs_key key; |
aec7477b | 2145 | u64 index; |
39279cc3 CM |
2146 | |
2147 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2148 | if (!path) { |
2149 | ret = -ENOMEM; | |
2150 | goto err; | |
2151 | } | |
2152 | ||
39279cc3 CM |
2153 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2154 | name, name_len, -1); | |
2155 | if (IS_ERR(di)) { | |
2156 | ret = PTR_ERR(di); | |
2157 | goto err; | |
2158 | } | |
2159 | if (!di) { | |
2160 | ret = -ENOENT; | |
2161 | goto err; | |
2162 | } | |
5f39d397 CM |
2163 | leaf = path->nodes[0]; |
2164 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2165 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2166 | if (ret) |
2167 | goto err; | |
39279cc3 CM |
2168 | btrfs_release_path(root, path); |
2169 | ||
aec7477b | 2170 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2171 | inode->i_ino, |
2172 | dir->i_ino, &index); | |
aec7477b | 2173 | if (ret) { |
d397712b | 2174 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2175 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2176 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2177 | goto err; |
2178 | } | |
2179 | ||
39279cc3 | 2180 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2181 | index, name, name_len, -1); |
39279cc3 CM |
2182 | if (IS_ERR(di)) { |
2183 | ret = PTR_ERR(di); | |
2184 | goto err; | |
2185 | } | |
2186 | if (!di) { | |
2187 | ret = -ENOENT; | |
2188 | goto err; | |
2189 | } | |
2190 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2191 | btrfs_release_path(root, path); |
39279cc3 | 2192 | |
e02119d5 CM |
2193 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2194 | inode, dir->i_ino); | |
49eb7e46 CM |
2195 | BUG_ON(ret != 0 && ret != -ENOENT); |
2196 | if (ret != -ENOENT) | |
2197 | BTRFS_I(dir)->log_dirty_trans = trans->transid; | |
e02119d5 CM |
2198 | |
2199 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2200 | dir, index); | |
2201 | BUG_ON(ret); | |
39279cc3 CM |
2202 | err: |
2203 | btrfs_free_path(path); | |
e02119d5 CM |
2204 | if (ret) |
2205 | goto out; | |
2206 | ||
2207 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2208 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2209 | btrfs_update_inode(trans, root, dir); | |
2210 | btrfs_drop_nlink(inode); | |
2211 | ret = btrfs_update_inode(trans, root, inode); | |
2212 | dir->i_sb->s_dirt = 1; | |
2213 | out: | |
39279cc3 CM |
2214 | return ret; |
2215 | } | |
2216 | ||
2217 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2218 | { | |
2219 | struct btrfs_root *root; | |
2220 | struct btrfs_trans_handle *trans; | |
7b128766 | 2221 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2222 | int ret; |
1832a6d5 | 2223 | unsigned long nr = 0; |
39279cc3 CM |
2224 | |
2225 | root = BTRFS_I(dir)->root; | |
1832a6d5 CM |
2226 | |
2227 | ret = btrfs_check_free_space(root, 1, 1); | |
2228 | if (ret) | |
2229 | goto fail; | |
2230 | ||
39279cc3 | 2231 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2232 | |
39279cc3 | 2233 | btrfs_set_trans_block_group(trans, dir); |
e02119d5 CM |
2234 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2235 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2236 | |
2237 | if (inode->i_nlink == 0) | |
2238 | ret = btrfs_orphan_add(trans, inode); | |
2239 | ||
d3c2fdcf | 2240 | nr = trans->blocks_used; |
5f39d397 | 2241 | |
89ce8a63 | 2242 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2243 | fail: |
d3c2fdcf | 2244 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2245 | return ret; |
2246 | } | |
2247 | ||
2248 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2249 | { | |
2250 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2251 | int err = 0; |
39279cc3 CM |
2252 | int ret; |
2253 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2254 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2255 | unsigned long nr = 0; |
39279cc3 | 2256 | |
3394e160 CM |
2257 | /* |
2258 | * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir | |
2259 | * the root of a subvolume or snapshot | |
2260 | */ | |
2261 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || | |
2262 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
134d4512 | 2263 | return -ENOTEMPTY; |
925baedd | 2264 | } |
134d4512 | 2265 | |
1832a6d5 CM |
2266 | ret = btrfs_check_free_space(root, 1, 1); |
2267 | if (ret) | |
2268 | goto fail; | |
2269 | ||
39279cc3 CM |
2270 | trans = btrfs_start_transaction(root, 1); |
2271 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2272 | |
7b128766 JB |
2273 | err = btrfs_orphan_add(trans, inode); |
2274 | if (err) | |
2275 | goto fail_trans; | |
2276 | ||
39279cc3 | 2277 | /* now the directory is empty */ |
e02119d5 CM |
2278 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2279 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2280 | if (!err) |
dbe674a9 | 2281 | btrfs_i_size_write(inode, 0); |
3954401f | 2282 | |
7b128766 | 2283 | fail_trans: |
d3c2fdcf | 2284 | nr = trans->blocks_used; |
89ce8a63 | 2285 | ret = btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2286 | fail: |
d3c2fdcf | 2287 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2288 | |
39279cc3 CM |
2289 | if (ret && !err) |
2290 | err = ret; | |
2291 | return err; | |
2292 | } | |
2293 | ||
d20f7043 | 2294 | #if 0 |
323ac95b CM |
2295 | /* |
2296 | * when truncating bytes in a file, it is possible to avoid reading | |
2297 | * the leaves that contain only checksum items. This can be the | |
2298 | * majority of the IO required to delete a large file, but it must | |
2299 | * be done carefully. | |
2300 | * | |
2301 | * The keys in the level just above the leaves are checked to make sure | |
2302 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2303 | * after the new size. | |
2304 | * | |
2305 | * Then the key for the next leaf is checked to make sure it also has | |
2306 | * a checksum item for the same file. If it does, we know our target leaf | |
2307 | * contains only checksum items, and it can be safely freed without reading | |
2308 | * it. | |
2309 | * | |
2310 | * This is just an optimization targeted at large files. It may do | |
2311 | * nothing. It will return 0 unless things went badly. | |
2312 | */ | |
2313 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2314 | struct btrfs_root *root, | |
2315 | struct btrfs_path *path, | |
2316 | struct inode *inode, u64 new_size) | |
2317 | { | |
2318 | struct btrfs_key key; | |
2319 | int ret; | |
2320 | int nritems; | |
2321 | struct btrfs_key found_key; | |
2322 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2323 | struct btrfs_leaf_ref *ref; |
2324 | u64 leaf_gen; | |
2325 | u64 leaf_start; | |
323ac95b CM |
2326 | |
2327 | path->lowest_level = 1; | |
2328 | key.objectid = inode->i_ino; | |
2329 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2330 | key.offset = new_size; | |
2331 | again: | |
2332 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2333 | if (ret < 0) | |
2334 | goto out; | |
2335 | ||
2336 | if (path->nodes[1] == NULL) { | |
2337 | ret = 0; | |
2338 | goto out; | |
2339 | } | |
2340 | ret = 0; | |
2341 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2342 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2343 | ||
2344 | if (!nritems) | |
2345 | goto out; | |
2346 | ||
2347 | if (path->slots[1] >= nritems) | |
2348 | goto next_node; | |
2349 | ||
2350 | /* did we find a key greater than anything we want to delete? */ | |
2351 | if (found_key.objectid > inode->i_ino || | |
2352 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2353 | goto out; | |
2354 | ||
2355 | /* we check the next key in the node to make sure the leave contains | |
2356 | * only checksum items. This comparison doesn't work if our | |
2357 | * leaf is the last one in the node | |
2358 | */ | |
2359 | if (path->slots[1] + 1 >= nritems) { | |
2360 | next_node: | |
2361 | /* search forward from the last key in the node, this | |
2362 | * will bring us into the next node in the tree | |
2363 | */ | |
2364 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2365 | ||
2366 | /* unlikely, but we inc below, so check to be safe */ | |
2367 | if (found_key.offset == (u64)-1) | |
2368 | goto out; | |
2369 | ||
2370 | /* search_forward needs a path with locks held, do the | |
2371 | * search again for the original key. It is possible | |
2372 | * this will race with a balance and return a path that | |
2373 | * we could modify, but this drop is just an optimization | |
2374 | * and is allowed to miss some leaves. | |
2375 | */ | |
2376 | btrfs_release_path(root, path); | |
2377 | found_key.offset++; | |
2378 | ||
2379 | /* setup a max key for search_forward */ | |
2380 | other_key.offset = (u64)-1; | |
2381 | other_key.type = key.type; | |
2382 | other_key.objectid = key.objectid; | |
2383 | ||
2384 | path->keep_locks = 1; | |
2385 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2386 | path, 0, 0); | |
2387 | path->keep_locks = 0; | |
2388 | if (ret || found_key.objectid != key.objectid || | |
2389 | found_key.type != key.type) { | |
2390 | ret = 0; | |
2391 | goto out; | |
2392 | } | |
2393 | ||
2394 | key.offset = found_key.offset; | |
2395 | btrfs_release_path(root, path); | |
2396 | cond_resched(); | |
2397 | goto again; | |
2398 | } | |
2399 | ||
2400 | /* we know there's one more slot after us in the tree, | |
2401 | * read that key so we can verify it is also a checksum item | |
2402 | */ | |
2403 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2404 | ||
2405 | if (found_key.objectid < inode->i_ino) | |
2406 | goto next_key; | |
2407 | ||
2408 | if (found_key.type != key.type || found_key.offset < new_size) | |
2409 | goto next_key; | |
2410 | ||
2411 | /* | |
2412 | * if the key for the next leaf isn't a csum key from this objectid, | |
2413 | * we can't be sure there aren't good items inside this leaf. | |
2414 | * Bail out | |
2415 | */ | |
2416 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2417 | goto out; | |
2418 | ||
5b84e8d6 YZ |
2419 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2420 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2421 | /* |
2422 | * it is safe to delete this leaf, it contains only | |
2423 | * csum items from this inode at an offset >= new_size | |
2424 | */ | |
5b84e8d6 | 2425 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2426 | BUG_ON(ret); |
2427 | ||
5b84e8d6 YZ |
2428 | if (root->ref_cows && leaf_gen < trans->transid) { |
2429 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2430 | if (ref) { | |
2431 | ref->root_gen = root->root_key.offset; | |
2432 | ref->bytenr = leaf_start; | |
2433 | ref->owner = 0; | |
2434 | ref->generation = leaf_gen; | |
2435 | ref->nritems = 0; | |
2436 | ||
2437 | ret = btrfs_add_leaf_ref(root, ref, 0); | |
2438 | WARN_ON(ret); | |
2439 | btrfs_free_leaf_ref(root, ref); | |
2440 | } else { | |
2441 | WARN_ON(1); | |
2442 | } | |
2443 | } | |
323ac95b CM |
2444 | next_key: |
2445 | btrfs_release_path(root, path); | |
2446 | ||
2447 | if (other_key.objectid == inode->i_ino && | |
2448 | other_key.type == key.type && other_key.offset > key.offset) { | |
2449 | key.offset = other_key.offset; | |
2450 | cond_resched(); | |
2451 | goto again; | |
2452 | } | |
2453 | ret = 0; | |
2454 | out: | |
2455 | /* fixup any changes we've made to the path */ | |
2456 | path->lowest_level = 0; | |
2457 | path->keep_locks = 0; | |
2458 | btrfs_release_path(root, path); | |
2459 | return ret; | |
2460 | } | |
2461 | ||
d20f7043 CM |
2462 | #endif |
2463 | ||
39279cc3 CM |
2464 | /* |
2465 | * this can truncate away extent items, csum items and directory items. | |
2466 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2467 | * any higher than new_size |
39279cc3 CM |
2468 | * |
2469 | * csum items that cross the new i_size are truncated to the new size | |
2470 | * as well. | |
7b128766 JB |
2471 | * |
2472 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2473 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2474 | */ |
e02119d5 CM |
2475 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2476 | struct btrfs_root *root, | |
2477 | struct inode *inode, | |
2478 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2479 | { |
2480 | int ret; | |
2481 | struct btrfs_path *path; | |
2482 | struct btrfs_key key; | |
5f39d397 | 2483 | struct btrfs_key found_key; |
39279cc3 | 2484 | u32 found_type; |
5f39d397 | 2485 | struct extent_buffer *leaf; |
39279cc3 CM |
2486 | struct btrfs_file_extent_item *fi; |
2487 | u64 extent_start = 0; | |
db94535d | 2488 | u64 extent_num_bytes = 0; |
39279cc3 | 2489 | u64 item_end = 0; |
7bb86316 | 2490 | u64 root_gen = 0; |
d8d5f3e1 | 2491 | u64 root_owner = 0; |
39279cc3 CM |
2492 | int found_extent; |
2493 | int del_item; | |
85e21bac CM |
2494 | int pending_del_nr = 0; |
2495 | int pending_del_slot = 0; | |
179e29e4 | 2496 | int extent_type = -1; |
771ed689 | 2497 | int encoding; |
3b951516 | 2498 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2499 | |
e02119d5 | 2500 | if (root->ref_cows) |
5b21f2ed | 2501 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 | 2502 | path = btrfs_alloc_path(); |
3c69faec | 2503 | path->reada = -1; |
39279cc3 | 2504 | BUG_ON(!path); |
5f39d397 | 2505 | |
39279cc3 CM |
2506 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2507 | key.objectid = inode->i_ino; | |
2508 | key.offset = (u64)-1; | |
5f39d397 CM |
2509 | key.type = (u8)-1; |
2510 | ||
85e21bac | 2511 | btrfs_init_path(path); |
323ac95b | 2512 | |
85e21bac CM |
2513 | search_again: |
2514 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
d397712b | 2515 | if (ret < 0) |
85e21bac | 2516 | goto error; |
d397712b | 2517 | |
85e21bac | 2518 | if (ret > 0) { |
e02119d5 CM |
2519 | /* there are no items in the tree for us to truncate, we're |
2520 | * done | |
2521 | */ | |
2522 | if (path->slots[0] == 0) { | |
2523 | ret = 0; | |
2524 | goto error; | |
2525 | } | |
85e21bac CM |
2526 | path->slots[0]--; |
2527 | } | |
2528 | ||
d397712b | 2529 | while (1) { |
39279cc3 | 2530 | fi = NULL; |
5f39d397 CM |
2531 | leaf = path->nodes[0]; |
2532 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2533 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2534 | encoding = 0; |
39279cc3 | 2535 | |
5f39d397 | 2536 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2537 | break; |
5f39d397 | 2538 | |
85e21bac | 2539 | if (found_type < min_type) |
39279cc3 CM |
2540 | break; |
2541 | ||
5f39d397 | 2542 | item_end = found_key.offset; |
39279cc3 | 2543 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2544 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2545 | struct btrfs_file_extent_item); |
179e29e4 | 2546 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2547 | encoding = btrfs_file_extent_compression(leaf, fi); |
2548 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2549 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2550 | ||
179e29e4 | 2551 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2552 | item_end += |
db94535d | 2553 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2554 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2555 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2556 | fi); |
39279cc3 | 2557 | } |
008630c1 | 2558 | item_end--; |
39279cc3 | 2559 | } |
e02119d5 | 2560 | if (item_end < new_size) { |
d397712b | 2561 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2562 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2563 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2564 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2565 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2566 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2567 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2568 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2569 | else if (found_type) |
b888db2b | 2570 | found_type--; |
d397712b | 2571 | else |
b888db2b | 2572 | break; |
a61721d5 | 2573 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2574 | goto next; |
39279cc3 | 2575 | } |
e02119d5 | 2576 | if (found_key.offset >= new_size) |
39279cc3 CM |
2577 | del_item = 1; |
2578 | else | |
2579 | del_item = 0; | |
2580 | found_extent = 0; | |
2581 | ||
2582 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2583 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2584 | goto delete; | |
2585 | ||
2586 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2587 | u64 num_dec; |
db94535d | 2588 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2589 | if (!del_item && !encoding) { |
db94535d CM |
2590 | u64 orig_num_bytes = |
2591 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2592 | extent_num_bytes = new_size - |
5f39d397 | 2593 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2594 | extent_num_bytes = extent_num_bytes & |
2595 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2596 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2597 | extent_num_bytes); | |
2598 | num_dec = (orig_num_bytes - | |
9069218d | 2599 | extent_num_bytes); |
e02119d5 | 2600 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2601 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2602 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2603 | } else { |
db94535d CM |
2604 | extent_num_bytes = |
2605 | btrfs_file_extent_disk_num_bytes(leaf, | |
2606 | fi); | |
39279cc3 | 2607 | /* FIXME blocksize != 4096 */ |
9069218d | 2608 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2609 | if (extent_start != 0) { |
2610 | found_extent = 1; | |
e02119d5 | 2611 | if (root->ref_cows) |
a76a3cd4 | 2612 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2613 | } |
31840ae1 | 2614 | root_gen = btrfs_header_generation(leaf); |
d8d5f3e1 | 2615 | root_owner = btrfs_header_owner(leaf); |
39279cc3 | 2616 | } |
9069218d | 2617 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2618 | /* |
2619 | * we can't truncate inline items that have had | |
2620 | * special encodings | |
2621 | */ | |
2622 | if (!del_item && | |
2623 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2624 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2625 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2626 | u32 size = new_size - found_key.offset; |
2627 | ||
2628 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2629 | inode_sub_bytes(inode, item_end + 1 - |
2630 | new_size); | |
e02119d5 CM |
2631 | } |
2632 | size = | |
2633 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2634 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2635 | size, 1); |
9069218d | 2636 | BUG_ON(ret); |
e02119d5 | 2637 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2638 | inode_sub_bytes(inode, item_end + 1 - |
2639 | found_key.offset); | |
9069218d | 2640 | } |
39279cc3 | 2641 | } |
179e29e4 | 2642 | delete: |
39279cc3 | 2643 | if (del_item) { |
85e21bac CM |
2644 | if (!pending_del_nr) { |
2645 | /* no pending yet, add ourselves */ | |
2646 | pending_del_slot = path->slots[0]; | |
2647 | pending_del_nr = 1; | |
2648 | } else if (pending_del_nr && | |
2649 | path->slots[0] + 1 == pending_del_slot) { | |
2650 | /* hop on the pending chunk */ | |
2651 | pending_del_nr++; | |
2652 | pending_del_slot = path->slots[0]; | |
2653 | } else { | |
d397712b | 2654 | BUG(); |
85e21bac | 2655 | } |
39279cc3 CM |
2656 | } else { |
2657 | break; | |
2658 | } | |
39279cc3 CM |
2659 | if (found_extent) { |
2660 | ret = btrfs_free_extent(trans, root, extent_start, | |
7bb86316 | 2661 | extent_num_bytes, |
31840ae1 | 2662 | leaf->start, root_owner, |
3bb1a1bc | 2663 | root_gen, inode->i_ino, 0); |
39279cc3 CM |
2664 | BUG_ON(ret); |
2665 | } | |
85e21bac CM |
2666 | next: |
2667 | if (path->slots[0] == 0) { | |
2668 | if (pending_del_nr) | |
2669 | goto del_pending; | |
2670 | btrfs_release_path(root, path); | |
2671 | goto search_again; | |
2672 | } | |
2673 | ||
2674 | path->slots[0]--; | |
2675 | if (pending_del_nr && | |
2676 | path->slots[0] + 1 != pending_del_slot) { | |
2677 | struct btrfs_key debug; | |
2678 | del_pending: | |
2679 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2680 | pending_del_slot); | |
2681 | ret = btrfs_del_items(trans, root, path, | |
2682 | pending_del_slot, | |
2683 | pending_del_nr); | |
2684 | BUG_ON(ret); | |
2685 | pending_del_nr = 0; | |
2686 | btrfs_release_path(root, path); | |
2687 | goto search_again; | |
2688 | } | |
39279cc3 CM |
2689 | } |
2690 | ret = 0; | |
2691 | error: | |
85e21bac CM |
2692 | if (pending_del_nr) { |
2693 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
2694 | pending_del_nr); | |
2695 | } | |
39279cc3 CM |
2696 | btrfs_free_path(path); |
2697 | inode->i_sb->s_dirt = 1; | |
2698 | return ret; | |
2699 | } | |
2700 | ||
2701 | /* | |
2702 | * taken from block_truncate_page, but does cow as it zeros out | |
2703 | * any bytes left in the last page in the file. | |
2704 | */ | |
2705 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
2706 | { | |
2707 | struct inode *inode = mapping->host; | |
db94535d | 2708 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
2709 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2710 | struct btrfs_ordered_extent *ordered; | |
2711 | char *kaddr; | |
db94535d | 2712 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
2713 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
2714 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
2715 | struct page *page; | |
39279cc3 | 2716 | int ret = 0; |
a52d9a80 | 2717 | u64 page_start; |
e6dcd2dc | 2718 | u64 page_end; |
39279cc3 CM |
2719 | |
2720 | if ((offset & (blocksize - 1)) == 0) | |
2721 | goto out; | |
2722 | ||
2723 | ret = -ENOMEM; | |
211c17f5 | 2724 | again: |
39279cc3 CM |
2725 | page = grab_cache_page(mapping, index); |
2726 | if (!page) | |
2727 | goto out; | |
e6dcd2dc CM |
2728 | |
2729 | page_start = page_offset(page); | |
2730 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2731 | ||
39279cc3 | 2732 | if (!PageUptodate(page)) { |
9ebefb18 | 2733 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 2734 | lock_page(page); |
211c17f5 CM |
2735 | if (page->mapping != mapping) { |
2736 | unlock_page(page); | |
2737 | page_cache_release(page); | |
2738 | goto again; | |
2739 | } | |
39279cc3 CM |
2740 | if (!PageUptodate(page)) { |
2741 | ret = -EIO; | |
89642229 | 2742 | goto out_unlock; |
39279cc3 CM |
2743 | } |
2744 | } | |
211c17f5 | 2745 | wait_on_page_writeback(page); |
e6dcd2dc CM |
2746 | |
2747 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2748 | set_page_extent_mapped(page); | |
2749 | ||
2750 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
2751 | if (ordered) { | |
2752 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2753 | unlock_page(page); | |
2754 | page_cache_release(page); | |
eb84ae03 | 2755 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
2756 | btrfs_put_ordered_extent(ordered); |
2757 | goto again; | |
2758 | } | |
2759 | ||
ea8c2819 | 2760 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc CM |
2761 | ret = 0; |
2762 | if (offset != PAGE_CACHE_SIZE) { | |
2763 | kaddr = kmap(page); | |
2764 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
2765 | flush_dcache_page(page); | |
2766 | kunmap(page); | |
2767 | } | |
247e743c | 2768 | ClearPageChecked(page); |
e6dcd2dc CM |
2769 | set_page_dirty(page); |
2770 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 2771 | |
89642229 | 2772 | out_unlock: |
39279cc3 CM |
2773 | unlock_page(page); |
2774 | page_cache_release(page); | |
2775 | out: | |
2776 | return ret; | |
2777 | } | |
2778 | ||
9036c102 | 2779 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 2780 | { |
9036c102 YZ |
2781 | struct btrfs_trans_handle *trans; |
2782 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2783 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2784 | struct extent_map *em; | |
2785 | u64 mask = root->sectorsize - 1; | |
2786 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
2787 | u64 block_end = (size + mask) & ~mask; | |
2788 | u64 last_byte; | |
2789 | u64 cur_offset; | |
2790 | u64 hole_size; | |
39279cc3 CM |
2791 | int err; |
2792 | ||
9036c102 YZ |
2793 | if (size <= hole_start) |
2794 | return 0; | |
2795 | ||
2796 | err = btrfs_check_free_space(root, 1, 0); | |
39279cc3 CM |
2797 | if (err) |
2798 | return err; | |
2799 | ||
9036c102 | 2800 | btrfs_truncate_page(inode->i_mapping, inode->i_size); |
2bf5a725 | 2801 | |
9036c102 YZ |
2802 | while (1) { |
2803 | struct btrfs_ordered_extent *ordered; | |
2804 | btrfs_wait_ordered_range(inode, hole_start, | |
2805 | block_end - hole_start); | |
2806 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2807 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
2808 | if (!ordered) | |
2809 | break; | |
2810 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2811 | btrfs_put_ordered_extent(ordered); | |
2812 | } | |
39279cc3 | 2813 | |
9036c102 YZ |
2814 | trans = btrfs_start_transaction(root, 1); |
2815 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 2816 | |
9036c102 YZ |
2817 | cur_offset = hole_start; |
2818 | while (1) { | |
2819 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
2820 | block_end - cur_offset, 0); | |
2821 | BUG_ON(IS_ERR(em) || !em); | |
2822 | last_byte = min(extent_map_end(em), block_end); | |
2823 | last_byte = (last_byte + mask) & ~mask; | |
2824 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 2825 | u64 hint_byte = 0; |
9036c102 | 2826 | hole_size = last_byte - cur_offset; |
771ed689 CM |
2827 | err = btrfs_drop_extents(trans, root, inode, |
2828 | cur_offset, | |
2829 | cur_offset + hole_size, | |
2830 | cur_offset, &hint_byte); | |
2831 | if (err) | |
2832 | break; | |
9036c102 YZ |
2833 | err = btrfs_insert_file_extent(trans, root, |
2834 | inode->i_ino, cur_offset, 0, | |
2835 | 0, hole_size, 0, hole_size, | |
2836 | 0, 0, 0); | |
2837 | btrfs_drop_extent_cache(inode, hole_start, | |
2838 | last_byte - 1, 0); | |
2839 | } | |
2840 | free_extent_map(em); | |
2841 | cur_offset = last_byte; | |
2842 | if (err || cur_offset >= block_end) | |
2843 | break; | |
2844 | } | |
1832a6d5 | 2845 | |
9036c102 YZ |
2846 | btrfs_end_transaction(trans, root); |
2847 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2848 | return err; | |
2849 | } | |
39279cc3 | 2850 | |
9036c102 YZ |
2851 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
2852 | { | |
2853 | struct inode *inode = dentry->d_inode; | |
2854 | int err; | |
39279cc3 | 2855 | |
9036c102 YZ |
2856 | err = inode_change_ok(inode, attr); |
2857 | if (err) | |
2858 | return err; | |
2bf5a725 | 2859 | |
9036c102 YZ |
2860 | if (S_ISREG(inode->i_mode) && |
2861 | attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) { | |
2862 | err = btrfs_cont_expand(inode, attr->ia_size); | |
54aa1f4d CM |
2863 | if (err) |
2864 | return err; | |
39279cc3 | 2865 | } |
9036c102 | 2866 | |
39279cc3 | 2867 | err = inode_setattr(inode, attr); |
33268eaf JB |
2868 | |
2869 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
2870 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
2871 | return err; |
2872 | } | |
61295eb8 | 2873 | |
39279cc3 CM |
2874 | void btrfs_delete_inode(struct inode *inode) |
2875 | { | |
2876 | struct btrfs_trans_handle *trans; | |
2877 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 2878 | unsigned long nr; |
39279cc3 CM |
2879 | int ret; |
2880 | ||
2881 | truncate_inode_pages(&inode->i_data, 0); | |
2882 | if (is_bad_inode(inode)) { | |
7b128766 | 2883 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
2884 | goto no_delete; |
2885 | } | |
4a096752 | 2886 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 2887 | |
dbe674a9 | 2888 | btrfs_i_size_write(inode, 0); |
180591bc | 2889 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 2890 | |
39279cc3 | 2891 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 2892 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
2893 | if (ret) { |
2894 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 2895 | goto no_delete_lock; |
7b128766 JB |
2896 | } |
2897 | ||
2898 | btrfs_orphan_del(trans, inode); | |
85e21bac | 2899 | |
d3c2fdcf | 2900 | nr = trans->blocks_used; |
85e21bac | 2901 | clear_inode(inode); |
5f39d397 | 2902 | |
39279cc3 | 2903 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2904 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 2905 | return; |
54aa1f4d CM |
2906 | |
2907 | no_delete_lock: | |
d3c2fdcf | 2908 | nr = trans->blocks_used; |
54aa1f4d | 2909 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2910 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2911 | no_delete: |
2912 | clear_inode(inode); | |
2913 | } | |
2914 | ||
2915 | /* | |
2916 | * this returns the key found in the dir entry in the location pointer. | |
2917 | * If no dir entries were found, location->objectid is 0. | |
2918 | */ | |
2919 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
2920 | struct btrfs_key *location) | |
2921 | { | |
2922 | const char *name = dentry->d_name.name; | |
2923 | int namelen = dentry->d_name.len; | |
2924 | struct btrfs_dir_item *di; | |
2925 | struct btrfs_path *path; | |
2926 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 2927 | int ret = 0; |
39279cc3 CM |
2928 | |
2929 | path = btrfs_alloc_path(); | |
2930 | BUG_ON(!path); | |
3954401f | 2931 | |
39279cc3 CM |
2932 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
2933 | namelen, 0); | |
0d9f7f3e Y |
2934 | if (IS_ERR(di)) |
2935 | ret = PTR_ERR(di); | |
d397712b CM |
2936 | |
2937 | if (!di || IS_ERR(di)) | |
3954401f | 2938 | goto out_err; |
d397712b | 2939 | |
5f39d397 | 2940 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 2941 | out: |
39279cc3 CM |
2942 | btrfs_free_path(path); |
2943 | return ret; | |
3954401f CM |
2944 | out_err: |
2945 | location->objectid = 0; | |
2946 | goto out; | |
39279cc3 CM |
2947 | } |
2948 | ||
2949 | /* | |
2950 | * when we hit a tree root in a directory, the btrfs part of the inode | |
2951 | * needs to be changed to reflect the root directory of the tree root. This | |
2952 | * is kind of like crossing a mount point. | |
2953 | */ | |
2954 | static int fixup_tree_root_location(struct btrfs_root *root, | |
2955 | struct btrfs_key *location, | |
58176a96 JB |
2956 | struct btrfs_root **sub_root, |
2957 | struct dentry *dentry) | |
39279cc3 | 2958 | { |
39279cc3 CM |
2959 | struct btrfs_root_item *ri; |
2960 | ||
2961 | if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY) | |
2962 | return 0; | |
2963 | if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) | |
2964 | return 0; | |
2965 | ||
58176a96 JB |
2966 | *sub_root = btrfs_read_fs_root(root->fs_info, location, |
2967 | dentry->d_name.name, | |
2968 | dentry->d_name.len); | |
39279cc3 CM |
2969 | if (IS_ERR(*sub_root)) |
2970 | return PTR_ERR(*sub_root); | |
2971 | ||
2972 | ri = &(*sub_root)->root_item; | |
2973 | location->objectid = btrfs_root_dirid(ri); | |
39279cc3 CM |
2974 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); |
2975 | location->offset = 0; | |
2976 | ||
39279cc3 CM |
2977 | return 0; |
2978 | } | |
2979 | ||
e02119d5 | 2980 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 2981 | { |
e02119d5 CM |
2982 | struct btrfs_inode *bi = BTRFS_I(inode); |
2983 | ||
2984 | bi->i_acl = NULL; | |
2985 | bi->i_default_acl = NULL; | |
2986 | ||
2987 | bi->generation = 0; | |
c3027eb5 | 2988 | bi->sequence = 0; |
e02119d5 CM |
2989 | bi->last_trans = 0; |
2990 | bi->logged_trans = 0; | |
2991 | bi->delalloc_bytes = 0; | |
2992 | bi->disk_i_size = 0; | |
2993 | bi->flags = 0; | |
2994 | bi->index_cnt = (u64)-1; | |
49eb7e46 | 2995 | bi->log_dirty_trans = 0; |
d1310b2e CM |
2996 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
2997 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 2998 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
2999 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3000 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3001 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
ba1da2f4 | 3002 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3003 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3004 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3005 | } | |
3006 | ||
3007 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3008 | { | |
3009 | struct btrfs_iget_args *args = p; | |
3010 | inode->i_ino = args->ino; | |
3011 | init_btrfs_i(inode); | |
3012 | BTRFS_I(inode)->root = args->root; | |
39279cc3 CM |
3013 | return 0; |
3014 | } | |
3015 | ||
3016 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3017 | { | |
3018 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3019 | return args->ino == inode->i_ino && |
3020 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3021 | } |
3022 | ||
5b21f2ed ZY |
3023 | struct inode *btrfs_ilookup(struct super_block *s, u64 objectid, |
3024 | struct btrfs_root *root, int wait) | |
3025 | { | |
3026 | struct inode *inode; | |
3027 | struct btrfs_iget_args args; | |
3028 | args.ino = objectid; | |
3029 | args.root = root; | |
3030 | ||
3031 | if (wait) { | |
3032 | inode = ilookup5(s, objectid, btrfs_find_actor, | |
3033 | (void *)&args); | |
3034 | } else { | |
3035 | inode = ilookup5_nowait(s, objectid, btrfs_find_actor, | |
3036 | (void *)&args); | |
3037 | } | |
3038 | return inode; | |
3039 | } | |
3040 | ||
39279cc3 CM |
3041 | struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid, |
3042 | struct btrfs_root *root) | |
3043 | { | |
3044 | struct inode *inode; | |
3045 | struct btrfs_iget_args args; | |
3046 | args.ino = objectid; | |
3047 | args.root = root; | |
3048 | ||
3049 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3050 | btrfs_init_locked_inode, | |
3051 | (void *)&args); | |
3052 | return inode; | |
3053 | } | |
3054 | ||
1a54ef8c BR |
3055 | /* Get an inode object given its location and corresponding root. |
3056 | * Returns in *is_new if the inode was read from disk | |
3057 | */ | |
3058 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
3059 | struct btrfs_root *root, int *is_new) | |
3060 | { | |
3061 | struct inode *inode; | |
3062 | ||
3063 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3064 | if (!inode) | |
3065 | return ERR_PTR(-EACCES); | |
3066 | ||
3067 | if (inode->i_state & I_NEW) { | |
3068 | BTRFS_I(inode)->root = root; | |
3069 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3070 | btrfs_read_locked_inode(inode); | |
3071 | unlock_new_inode(inode); | |
3072 | if (is_new) | |
3073 | *is_new = 1; | |
3074 | } else { | |
3075 | if (is_new) | |
3076 | *is_new = 0; | |
3077 | } | |
3078 | ||
3079 | return inode; | |
3080 | } | |
3081 | ||
3de4586c | 3082 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3083 | { |
d397712b | 3084 | struct inode *inode; |
39279cc3 CM |
3085 | struct btrfs_inode *bi = BTRFS_I(dir); |
3086 | struct btrfs_root *root = bi->root; | |
3087 | struct btrfs_root *sub_root = root; | |
3088 | struct btrfs_key location; | |
c146afad | 3089 | int ret, new; |
39279cc3 CM |
3090 | |
3091 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3092 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3093 | |
39279cc3 | 3094 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3095 | |
39279cc3 CM |
3096 | if (ret < 0) |
3097 | return ERR_PTR(ret); | |
5f39d397 | 3098 | |
39279cc3 CM |
3099 | inode = NULL; |
3100 | if (location.objectid) { | |
58176a96 JB |
3101 | ret = fixup_tree_root_location(root, &location, &sub_root, |
3102 | dentry); | |
39279cc3 CM |
3103 | if (ret < 0) |
3104 | return ERR_PTR(ret); | |
3105 | if (ret > 0) | |
3106 | return ERR_PTR(-ENOENT); | |
1a54ef8c BR |
3107 | inode = btrfs_iget(dir->i_sb, &location, sub_root, &new); |
3108 | if (IS_ERR(inode)) | |
3109 | return ERR_CAST(inode); | |
39279cc3 | 3110 | } |
3de4586c CM |
3111 | return inode; |
3112 | } | |
3113 | ||
3114 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, | |
3115 | struct nameidata *nd) | |
3116 | { | |
3117 | struct inode *inode; | |
3118 | ||
3119 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3120 | return ERR_PTR(-ENAMETOOLONG); | |
3121 | ||
3122 | inode = btrfs_lookup_dentry(dir, dentry); | |
3123 | if (IS_ERR(inode)) | |
3124 | return ERR_CAST(inode); | |
7b128766 | 3125 | |
39279cc3 CM |
3126 | return d_splice_alias(inode, dentry); |
3127 | } | |
3128 | ||
39279cc3 CM |
3129 | static unsigned char btrfs_filetype_table[] = { |
3130 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3131 | }; | |
3132 | ||
cbdf5a24 DW |
3133 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3134 | filldir_t filldir) | |
39279cc3 | 3135 | { |
6da6abae | 3136 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3137 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3138 | struct btrfs_item *item; | |
3139 | struct btrfs_dir_item *di; | |
3140 | struct btrfs_key key; | |
5f39d397 | 3141 | struct btrfs_key found_key; |
39279cc3 CM |
3142 | struct btrfs_path *path; |
3143 | int ret; | |
3144 | u32 nritems; | |
5f39d397 | 3145 | struct extent_buffer *leaf; |
39279cc3 CM |
3146 | int slot; |
3147 | int advance; | |
3148 | unsigned char d_type; | |
3149 | int over = 0; | |
3150 | u32 di_cur; | |
3151 | u32 di_total; | |
3152 | u32 di_len; | |
3153 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3154 | char tmp_name[32]; |
3155 | char *name_ptr; | |
3156 | int name_len; | |
39279cc3 CM |
3157 | |
3158 | /* FIXME, use a real flag for deciding about the key type */ | |
3159 | if (root->fs_info->tree_root == root) | |
3160 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3161 | |
3954401f CM |
3162 | /* special case for "." */ |
3163 | if (filp->f_pos == 0) { | |
3164 | over = filldir(dirent, ".", 1, | |
3165 | 1, inode->i_ino, | |
3166 | DT_DIR); | |
3167 | if (over) | |
3168 | return 0; | |
3169 | filp->f_pos = 1; | |
3170 | } | |
3954401f CM |
3171 | /* special case for .., just use the back ref */ |
3172 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3173 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3174 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3175 | 2, pino, DT_DIR); |
3954401f | 3176 | if (over) |
49593bfa | 3177 | return 0; |
3954401f CM |
3178 | filp->f_pos = 2; |
3179 | } | |
49593bfa DW |
3180 | path = btrfs_alloc_path(); |
3181 | path->reada = 2; | |
3182 | ||
39279cc3 CM |
3183 | btrfs_set_key_type(&key, key_type); |
3184 | key.offset = filp->f_pos; | |
49593bfa | 3185 | key.objectid = inode->i_ino; |
5f39d397 | 3186 | |
39279cc3 CM |
3187 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3188 | if (ret < 0) | |
3189 | goto err; | |
3190 | advance = 0; | |
49593bfa DW |
3191 | |
3192 | while (1) { | |
5f39d397 CM |
3193 | leaf = path->nodes[0]; |
3194 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3195 | slot = path->slots[0]; |
3196 | if (advance || slot >= nritems) { | |
49593bfa | 3197 | if (slot >= nritems - 1) { |
39279cc3 CM |
3198 | ret = btrfs_next_leaf(root, path); |
3199 | if (ret) | |
3200 | break; | |
5f39d397 CM |
3201 | leaf = path->nodes[0]; |
3202 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3203 | slot = path->slots[0]; |
3204 | } else { | |
3205 | slot++; | |
3206 | path->slots[0]++; | |
3207 | } | |
3208 | } | |
3de4586c | 3209 | |
39279cc3 | 3210 | advance = 1; |
5f39d397 CM |
3211 | item = btrfs_item_nr(leaf, slot); |
3212 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3213 | ||
3214 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3215 | break; |
5f39d397 | 3216 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3217 | break; |
5f39d397 | 3218 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3219 | continue; |
5f39d397 CM |
3220 | |
3221 | filp->f_pos = found_key.offset; | |
49593bfa | 3222 | |
39279cc3 CM |
3223 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3224 | di_cur = 0; | |
5f39d397 | 3225 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3226 | |
3227 | while (di_cur < di_total) { | |
5f39d397 CM |
3228 | struct btrfs_key location; |
3229 | ||
3230 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3231 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3232 | name_ptr = tmp_name; |
3233 | } else { | |
3234 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3235 | if (!name_ptr) { |
3236 | ret = -ENOMEM; | |
3237 | goto err; | |
3238 | } | |
5f39d397 CM |
3239 | } |
3240 | read_extent_buffer(leaf, name_ptr, | |
3241 | (unsigned long)(di + 1), name_len); | |
3242 | ||
3243 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3244 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3245 | |
3246 | /* is this a reference to our own snapshot? If so | |
3247 | * skip it | |
3248 | */ | |
3249 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3250 | location.objectid == root->root_key.objectid) { | |
3251 | over = 0; | |
3252 | goto skip; | |
3253 | } | |
5f39d397 | 3254 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3255 | found_key.offset, location.objectid, |
39279cc3 | 3256 | d_type); |
5f39d397 | 3257 | |
3de4586c | 3258 | skip: |
5f39d397 CM |
3259 | if (name_ptr != tmp_name) |
3260 | kfree(name_ptr); | |
3261 | ||
39279cc3 CM |
3262 | if (over) |
3263 | goto nopos; | |
5103e947 | 3264 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3265 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3266 | di_cur += di_len; |
3267 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3268 | } | |
3269 | } | |
49593bfa DW |
3270 | |
3271 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 YZ |
3272 | if (key_type == BTRFS_DIR_INDEX_KEY) |
3273 | filp->f_pos = INT_LIMIT(typeof(filp->f_pos)); | |
3274 | else | |
3275 | filp->f_pos++; | |
39279cc3 CM |
3276 | nopos: |
3277 | ret = 0; | |
3278 | err: | |
39279cc3 | 3279 | btrfs_free_path(path); |
39279cc3 CM |
3280 | return ret; |
3281 | } | |
3282 | ||
3283 | int btrfs_write_inode(struct inode *inode, int wait) | |
3284 | { | |
3285 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3286 | struct btrfs_trans_handle *trans; | |
3287 | int ret = 0; | |
3288 | ||
c146afad | 3289 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3290 | return 0; |
3291 | ||
39279cc3 | 3292 | if (wait) { |
f9295749 | 3293 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3294 | btrfs_set_trans_block_group(trans, inode); |
3295 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3296 | } |
3297 | return ret; | |
3298 | } | |
3299 | ||
3300 | /* | |
54aa1f4d | 3301 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3302 | * inode changes. But, it is most likely to find the inode in cache. |
3303 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3304 | * to keep or drop this code. | |
3305 | */ | |
3306 | void btrfs_dirty_inode(struct inode *inode) | |
3307 | { | |
3308 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3309 | struct btrfs_trans_handle *trans; | |
3310 | ||
f9295749 | 3311 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3312 | btrfs_set_trans_block_group(trans, inode); |
3313 | btrfs_update_inode(trans, root, inode); | |
3314 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3315 | } |
3316 | ||
d352ac68 CM |
3317 | /* |
3318 | * find the highest existing sequence number in a directory | |
3319 | * and then set the in-memory index_cnt variable to reflect | |
3320 | * free sequence numbers | |
3321 | */ | |
aec7477b JB |
3322 | static int btrfs_set_inode_index_count(struct inode *inode) |
3323 | { | |
3324 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3325 | struct btrfs_key key, found_key; | |
3326 | struct btrfs_path *path; | |
3327 | struct extent_buffer *leaf; | |
3328 | int ret; | |
3329 | ||
3330 | key.objectid = inode->i_ino; | |
3331 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3332 | key.offset = (u64)-1; | |
3333 | ||
3334 | path = btrfs_alloc_path(); | |
3335 | if (!path) | |
3336 | return -ENOMEM; | |
3337 | ||
3338 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3339 | if (ret < 0) | |
3340 | goto out; | |
3341 | /* FIXME: we should be able to handle this */ | |
3342 | if (ret == 0) | |
3343 | goto out; | |
3344 | ret = 0; | |
3345 | ||
3346 | /* | |
3347 | * MAGIC NUMBER EXPLANATION: | |
3348 | * since we search a directory based on f_pos we have to start at 2 | |
3349 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3350 | * else has to start at 2 | |
3351 | */ | |
3352 | if (path->slots[0] == 0) { | |
3353 | BTRFS_I(inode)->index_cnt = 2; | |
3354 | goto out; | |
3355 | } | |
3356 | ||
3357 | path->slots[0]--; | |
3358 | ||
3359 | leaf = path->nodes[0]; | |
3360 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3361 | ||
3362 | if (found_key.objectid != inode->i_ino || | |
3363 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3364 | BTRFS_I(inode)->index_cnt = 2; | |
3365 | goto out; | |
3366 | } | |
3367 | ||
3368 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3369 | out: | |
3370 | btrfs_free_path(path); | |
3371 | return ret; | |
3372 | } | |
3373 | ||
d352ac68 CM |
3374 | /* |
3375 | * helper to find a free sequence number in a given directory. This current | |
3376 | * code is very simple, later versions will do smarter things in the btree | |
3377 | */ | |
3de4586c | 3378 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3379 | { |
3380 | int ret = 0; | |
3381 | ||
3382 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3383 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3384 | if (ret) |
aec7477b JB |
3385 | return ret; |
3386 | } | |
3387 | ||
00e4e6b3 | 3388 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3389 | BTRFS_I(dir)->index_cnt++; |
3390 | ||
3391 | return ret; | |
3392 | } | |
3393 | ||
39279cc3 CM |
3394 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3395 | struct btrfs_root *root, | |
aec7477b | 3396 | struct inode *dir, |
9c58309d | 3397 | const char *name, int name_len, |
d2fb3437 YZ |
3398 | u64 ref_objectid, u64 objectid, |
3399 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3400 | { |
3401 | struct inode *inode; | |
5f39d397 | 3402 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3403 | struct btrfs_key *location; |
5f39d397 | 3404 | struct btrfs_path *path; |
9c58309d CM |
3405 | struct btrfs_inode_ref *ref; |
3406 | struct btrfs_key key[2]; | |
3407 | u32 sizes[2]; | |
3408 | unsigned long ptr; | |
39279cc3 CM |
3409 | int ret; |
3410 | int owner; | |
3411 | ||
5f39d397 CM |
3412 | path = btrfs_alloc_path(); |
3413 | BUG_ON(!path); | |
3414 | ||
39279cc3 CM |
3415 | inode = new_inode(root->fs_info->sb); |
3416 | if (!inode) | |
3417 | return ERR_PTR(-ENOMEM); | |
3418 | ||
aec7477b | 3419 | if (dir) { |
3de4586c | 3420 | ret = btrfs_set_inode_index(dir, index); |
aec7477b JB |
3421 | if (ret) |
3422 | return ERR_PTR(ret); | |
aec7477b JB |
3423 | } |
3424 | /* | |
3425 | * index_cnt is ignored for everything but a dir, | |
3426 | * btrfs_get_inode_index_count has an explanation for the magic | |
3427 | * number | |
3428 | */ | |
e02119d5 | 3429 | init_btrfs_i(inode); |
aec7477b | 3430 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3431 | BTRFS_I(inode)->root = root; |
e02119d5 | 3432 | BTRFS_I(inode)->generation = trans->transid; |
b888db2b | 3433 | |
39279cc3 CM |
3434 | if (mode & S_IFDIR) |
3435 | owner = 0; | |
3436 | else | |
3437 | owner = 1; | |
d2fb3437 YZ |
3438 | BTRFS_I(inode)->block_group = |
3439 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
17d217fe YZ |
3440 | if ((mode & S_IFREG)) { |
3441 | if (btrfs_test_opt(root, NODATASUM)) | |
3442 | btrfs_set_flag(inode, NODATASUM); | |
3443 | if (btrfs_test_opt(root, NODATACOW)) | |
3444 | btrfs_set_flag(inode, NODATACOW); | |
3445 | } | |
9c58309d CM |
3446 | |
3447 | key[0].objectid = objectid; | |
3448 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3449 | key[0].offset = 0; | |
3450 | ||
3451 | key[1].objectid = objectid; | |
3452 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3453 | key[1].offset = ref_objectid; | |
3454 | ||
3455 | sizes[0] = sizeof(struct btrfs_inode_item); | |
3456 | sizes[1] = name_len + sizeof(*ref); | |
3457 | ||
3458 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); | |
3459 | if (ret != 0) | |
5f39d397 CM |
3460 | goto fail; |
3461 | ||
9c58309d CM |
3462 | if (objectid > root->highest_inode) |
3463 | root->highest_inode = objectid; | |
3464 | ||
79683f2d CM |
3465 | inode->i_uid = current_fsuid(); |
3466 | inode->i_gid = current_fsgid(); | |
39279cc3 CM |
3467 | inode->i_mode = mode; |
3468 | inode->i_ino = objectid; | |
a76a3cd4 | 3469 | inode_set_bytes(inode, 0); |
39279cc3 | 3470 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
3471 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3472 | struct btrfs_inode_item); | |
e02119d5 | 3473 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
3474 | |
3475 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3476 | struct btrfs_inode_ref); | |
3477 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 3478 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
3479 | ptr = (unsigned long)(ref + 1); |
3480 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
3481 | ||
5f39d397 CM |
3482 | btrfs_mark_buffer_dirty(path->nodes[0]); |
3483 | btrfs_free_path(path); | |
3484 | ||
39279cc3 CM |
3485 | location = &BTRFS_I(inode)->location; |
3486 | location->objectid = objectid; | |
39279cc3 CM |
3487 | location->offset = 0; |
3488 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
3489 | ||
39279cc3 CM |
3490 | insert_inode_hash(inode); |
3491 | return inode; | |
5f39d397 | 3492 | fail: |
aec7477b JB |
3493 | if (dir) |
3494 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 CM |
3495 | btrfs_free_path(path); |
3496 | return ERR_PTR(ret); | |
39279cc3 CM |
3497 | } |
3498 | ||
3499 | static inline u8 btrfs_inode_type(struct inode *inode) | |
3500 | { | |
3501 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
3502 | } | |
3503 | ||
d352ac68 CM |
3504 | /* |
3505 | * utility function to add 'inode' into 'parent_inode' with | |
3506 | * a give name and a given sequence number. | |
3507 | * if 'add_backref' is true, also insert a backref from the | |
3508 | * inode to the parent directory. | |
3509 | */ | |
e02119d5 CM |
3510 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
3511 | struct inode *parent_inode, struct inode *inode, | |
3512 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 CM |
3513 | { |
3514 | int ret; | |
3515 | struct btrfs_key key; | |
e02119d5 | 3516 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 3517 | |
39279cc3 | 3518 | key.objectid = inode->i_ino; |
39279cc3 CM |
3519 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
3520 | key.offset = 0; | |
3521 | ||
e02119d5 CM |
3522 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
3523 | parent_inode->i_ino, | |
aec7477b | 3524 | &key, btrfs_inode_type(inode), |
00e4e6b3 | 3525 | index); |
39279cc3 | 3526 | if (ret == 0) { |
9c58309d CM |
3527 | if (add_backref) { |
3528 | ret = btrfs_insert_inode_ref(trans, root, | |
e02119d5 CM |
3529 | name, name_len, |
3530 | inode->i_ino, | |
3531 | parent_inode->i_ino, | |
3532 | index); | |
9c58309d | 3533 | } |
dbe674a9 | 3534 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 3535 | name_len * 2); |
79c44584 | 3536 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 3537 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
3538 | } |
3539 | return ret; | |
3540 | } | |
3541 | ||
3542 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 3543 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 3544 | int backref, u64 index) |
39279cc3 | 3545 | { |
e02119d5 CM |
3546 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3547 | inode, dentry->d_name.name, | |
3548 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
3549 | if (!err) { |
3550 | d_instantiate(dentry, inode); | |
3551 | return 0; | |
3552 | } | |
3553 | if (err > 0) | |
3554 | err = -EEXIST; | |
3555 | return err; | |
3556 | } | |
3557 | ||
618e21d5 JB |
3558 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
3559 | int mode, dev_t rdev) | |
3560 | { | |
3561 | struct btrfs_trans_handle *trans; | |
3562 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3563 | struct inode *inode = NULL; |
618e21d5 JB |
3564 | int err; |
3565 | int drop_inode = 0; | |
3566 | u64 objectid; | |
1832a6d5 | 3567 | unsigned long nr = 0; |
00e4e6b3 | 3568 | u64 index = 0; |
618e21d5 JB |
3569 | |
3570 | if (!new_valid_dev(rdev)) | |
3571 | return -EINVAL; | |
3572 | ||
1832a6d5 CM |
3573 | err = btrfs_check_free_space(root, 1, 0); |
3574 | if (err) | |
3575 | goto fail; | |
3576 | ||
618e21d5 JB |
3577 | trans = btrfs_start_transaction(root, 1); |
3578 | btrfs_set_trans_block_group(trans, dir); | |
3579 | ||
3580 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3581 | if (err) { | |
3582 | err = -ENOSPC; | |
3583 | goto out_unlock; | |
3584 | } | |
3585 | ||
aec7477b | 3586 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3587 | dentry->d_name.len, |
3588 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 3589 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
3590 | err = PTR_ERR(inode); |
3591 | if (IS_ERR(inode)) | |
3592 | goto out_unlock; | |
3593 | ||
33268eaf JB |
3594 | err = btrfs_init_acl(inode, dir); |
3595 | if (err) { | |
3596 | drop_inode = 1; | |
3597 | goto out_unlock; | |
3598 | } | |
3599 | ||
618e21d5 | 3600 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3601 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
3602 | if (err) |
3603 | drop_inode = 1; | |
3604 | else { | |
3605 | inode->i_op = &btrfs_special_inode_operations; | |
3606 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 3607 | btrfs_update_inode(trans, root, inode); |
618e21d5 JB |
3608 | } |
3609 | dir->i_sb->s_dirt = 1; | |
3610 | btrfs_update_inode_block_group(trans, inode); | |
3611 | btrfs_update_inode_block_group(trans, dir); | |
3612 | out_unlock: | |
d3c2fdcf | 3613 | nr = trans->blocks_used; |
89ce8a63 | 3614 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3615 | fail: |
618e21d5 JB |
3616 | if (drop_inode) { |
3617 | inode_dec_link_count(inode); | |
3618 | iput(inode); | |
3619 | } | |
d3c2fdcf | 3620 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
3621 | return err; |
3622 | } | |
3623 | ||
39279cc3 CM |
3624 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
3625 | int mode, struct nameidata *nd) | |
3626 | { | |
3627 | struct btrfs_trans_handle *trans; | |
3628 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3629 | struct inode *inode = NULL; |
39279cc3 CM |
3630 | int err; |
3631 | int drop_inode = 0; | |
1832a6d5 | 3632 | unsigned long nr = 0; |
39279cc3 | 3633 | u64 objectid; |
00e4e6b3 | 3634 | u64 index = 0; |
39279cc3 | 3635 | |
1832a6d5 CM |
3636 | err = btrfs_check_free_space(root, 1, 0); |
3637 | if (err) | |
3638 | goto fail; | |
39279cc3 CM |
3639 | trans = btrfs_start_transaction(root, 1); |
3640 | btrfs_set_trans_block_group(trans, dir); | |
3641 | ||
3642 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3643 | if (err) { | |
3644 | err = -ENOSPC; | |
3645 | goto out_unlock; | |
3646 | } | |
3647 | ||
aec7477b | 3648 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3649 | dentry->d_name.len, |
3650 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
3651 | objectid, BTRFS_I(dir)->block_group, mode, |
3652 | &index); | |
39279cc3 CM |
3653 | err = PTR_ERR(inode); |
3654 | if (IS_ERR(inode)) | |
3655 | goto out_unlock; | |
3656 | ||
33268eaf JB |
3657 | err = btrfs_init_acl(inode, dir); |
3658 | if (err) { | |
3659 | drop_inode = 1; | |
3660 | goto out_unlock; | |
3661 | } | |
3662 | ||
39279cc3 | 3663 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3664 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
3665 | if (err) |
3666 | drop_inode = 1; | |
3667 | else { | |
3668 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3669 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
3670 | inode->i_fop = &btrfs_file_operations; |
3671 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 3672 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3673 | } |
3674 | dir->i_sb->s_dirt = 1; | |
3675 | btrfs_update_inode_block_group(trans, inode); | |
3676 | btrfs_update_inode_block_group(trans, dir); | |
3677 | out_unlock: | |
d3c2fdcf | 3678 | nr = trans->blocks_used; |
ab78c84d | 3679 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3680 | fail: |
39279cc3 CM |
3681 | if (drop_inode) { |
3682 | inode_dec_link_count(inode); | |
3683 | iput(inode); | |
3684 | } | |
d3c2fdcf | 3685 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3686 | return err; |
3687 | } | |
3688 | ||
3689 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
3690 | struct dentry *dentry) | |
3691 | { | |
3692 | struct btrfs_trans_handle *trans; | |
3693 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3694 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 3695 | u64 index; |
1832a6d5 | 3696 | unsigned long nr = 0; |
39279cc3 CM |
3697 | int err; |
3698 | int drop_inode = 0; | |
3699 | ||
3700 | if (inode->i_nlink == 0) | |
3701 | return -ENOENT; | |
3702 | ||
e02119d5 | 3703 | btrfs_inc_nlink(inode); |
1832a6d5 CM |
3704 | err = btrfs_check_free_space(root, 1, 0); |
3705 | if (err) | |
3706 | goto fail; | |
3de4586c | 3707 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
3708 | if (err) |
3709 | goto fail; | |
3710 | ||
39279cc3 | 3711 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 3712 | |
39279cc3 CM |
3713 | btrfs_set_trans_block_group(trans, dir); |
3714 | atomic_inc(&inode->i_count); | |
aec7477b | 3715 | |
00e4e6b3 | 3716 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 3717 | |
39279cc3 CM |
3718 | if (err) |
3719 | drop_inode = 1; | |
5f39d397 | 3720 | |
39279cc3 CM |
3721 | dir->i_sb->s_dirt = 1; |
3722 | btrfs_update_inode_block_group(trans, dir); | |
54aa1f4d | 3723 | err = btrfs_update_inode(trans, root, inode); |
5f39d397 | 3724 | |
54aa1f4d CM |
3725 | if (err) |
3726 | drop_inode = 1; | |
39279cc3 | 3727 | |
d3c2fdcf | 3728 | nr = trans->blocks_used; |
ab78c84d | 3729 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3730 | fail: |
39279cc3 CM |
3731 | if (drop_inode) { |
3732 | inode_dec_link_count(inode); | |
3733 | iput(inode); | |
3734 | } | |
d3c2fdcf | 3735 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3736 | return err; |
3737 | } | |
3738 | ||
39279cc3 CM |
3739 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
3740 | { | |
b9d86667 | 3741 | struct inode *inode = NULL; |
39279cc3 CM |
3742 | struct btrfs_trans_handle *trans; |
3743 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3744 | int err = 0; | |
3745 | int drop_on_err = 0; | |
b9d86667 | 3746 | u64 objectid = 0; |
00e4e6b3 | 3747 | u64 index = 0; |
d3c2fdcf | 3748 | unsigned long nr = 1; |
39279cc3 | 3749 | |
1832a6d5 CM |
3750 | err = btrfs_check_free_space(root, 1, 0); |
3751 | if (err) | |
3752 | goto out_unlock; | |
3753 | ||
39279cc3 CM |
3754 | trans = btrfs_start_transaction(root, 1); |
3755 | btrfs_set_trans_block_group(trans, dir); | |
5f39d397 | 3756 | |
39279cc3 CM |
3757 | if (IS_ERR(trans)) { |
3758 | err = PTR_ERR(trans); | |
3759 | goto out_unlock; | |
3760 | } | |
3761 | ||
3762 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3763 | if (err) { | |
3764 | err = -ENOSPC; | |
3765 | goto out_unlock; | |
3766 | } | |
3767 | ||
aec7477b | 3768 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3769 | dentry->d_name.len, |
3770 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
3771 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
3772 | &index); | |
39279cc3 CM |
3773 | if (IS_ERR(inode)) { |
3774 | err = PTR_ERR(inode); | |
3775 | goto out_fail; | |
3776 | } | |
5f39d397 | 3777 | |
39279cc3 | 3778 | drop_on_err = 1; |
33268eaf JB |
3779 | |
3780 | err = btrfs_init_acl(inode, dir); | |
3781 | if (err) | |
3782 | goto out_fail; | |
3783 | ||
39279cc3 CM |
3784 | inode->i_op = &btrfs_dir_inode_operations; |
3785 | inode->i_fop = &btrfs_dir_file_operations; | |
3786 | btrfs_set_trans_block_group(trans, inode); | |
3787 | ||
dbe674a9 | 3788 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
3789 | err = btrfs_update_inode(trans, root, inode); |
3790 | if (err) | |
3791 | goto out_fail; | |
5f39d397 | 3792 | |
e02119d5 CM |
3793 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3794 | inode, dentry->d_name.name, | |
3795 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
3796 | if (err) |
3797 | goto out_fail; | |
5f39d397 | 3798 | |
39279cc3 CM |
3799 | d_instantiate(dentry, inode); |
3800 | drop_on_err = 0; | |
3801 | dir->i_sb->s_dirt = 1; | |
3802 | btrfs_update_inode_block_group(trans, inode); | |
3803 | btrfs_update_inode_block_group(trans, dir); | |
3804 | ||
3805 | out_fail: | |
d3c2fdcf | 3806 | nr = trans->blocks_used; |
ab78c84d | 3807 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 3808 | |
39279cc3 | 3809 | out_unlock: |
39279cc3 CM |
3810 | if (drop_on_err) |
3811 | iput(inode); | |
d3c2fdcf | 3812 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3813 | return err; |
3814 | } | |
3815 | ||
d352ac68 CM |
3816 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
3817 | * and an extent that you want to insert, deal with overlap and insert | |
3818 | * the new extent into the tree. | |
3819 | */ | |
3b951516 CM |
3820 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
3821 | struct extent_map *existing, | |
e6dcd2dc CM |
3822 | struct extent_map *em, |
3823 | u64 map_start, u64 map_len) | |
3b951516 CM |
3824 | { |
3825 | u64 start_diff; | |
3b951516 | 3826 | |
e6dcd2dc CM |
3827 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
3828 | start_diff = map_start - em->start; | |
3829 | em->start = map_start; | |
3830 | em->len = map_len; | |
c8b97818 CM |
3831 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
3832 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 3833 | em->block_start += start_diff; |
c8b97818 CM |
3834 | em->block_len -= start_diff; |
3835 | } | |
e6dcd2dc | 3836 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
3837 | } |
3838 | ||
c8b97818 CM |
3839 | static noinline int uncompress_inline(struct btrfs_path *path, |
3840 | struct inode *inode, struct page *page, | |
3841 | size_t pg_offset, u64 extent_offset, | |
3842 | struct btrfs_file_extent_item *item) | |
3843 | { | |
3844 | int ret; | |
3845 | struct extent_buffer *leaf = path->nodes[0]; | |
3846 | char *tmp; | |
3847 | size_t max_size; | |
3848 | unsigned long inline_size; | |
3849 | unsigned long ptr; | |
3850 | ||
3851 | WARN_ON(pg_offset != 0); | |
3852 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
3853 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
3854 | btrfs_item_nr(leaf, path->slots[0])); | |
3855 | tmp = kmalloc(inline_size, GFP_NOFS); | |
3856 | ptr = btrfs_file_extent_inline_start(item); | |
3857 | ||
3858 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
3859 | ||
5b050f04 | 3860 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
3861 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
3862 | inline_size, max_size); | |
3863 | if (ret) { | |
3864 | char *kaddr = kmap_atomic(page, KM_USER0); | |
3865 | unsigned long copy_size = min_t(u64, | |
3866 | PAGE_CACHE_SIZE - pg_offset, | |
3867 | max_size - extent_offset); | |
3868 | memset(kaddr + pg_offset, 0, copy_size); | |
3869 | kunmap_atomic(kaddr, KM_USER0); | |
3870 | } | |
3871 | kfree(tmp); | |
3872 | return 0; | |
3873 | } | |
3874 | ||
d352ac68 CM |
3875 | /* |
3876 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
3877 | * the ugly parts come from merging extents from the disk with the in-ram |
3878 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
3879 | * where the in-ram extents might be locked pending data=ordered completion. |
3880 | * | |
3881 | * This also copies inline extents directly into the page. | |
3882 | */ | |
d397712b | 3883 | |
a52d9a80 | 3884 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 3885 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
3886 | int create) |
3887 | { | |
3888 | int ret; | |
3889 | int err = 0; | |
db94535d | 3890 | u64 bytenr; |
a52d9a80 CM |
3891 | u64 extent_start = 0; |
3892 | u64 extent_end = 0; | |
3893 | u64 objectid = inode->i_ino; | |
3894 | u32 found_type; | |
f421950f | 3895 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
3896 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3897 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
3898 | struct extent_buffer *leaf; |
3899 | struct btrfs_key found_key; | |
a52d9a80 CM |
3900 | struct extent_map *em = NULL; |
3901 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 3902 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 3903 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 3904 | int compressed; |
a52d9a80 | 3905 | |
a52d9a80 | 3906 | again: |
d1310b2e CM |
3907 | spin_lock(&em_tree->lock); |
3908 | em = lookup_extent_mapping(em_tree, start, len); | |
a061fc8d CM |
3909 | if (em) |
3910 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
d1310b2e CM |
3911 | spin_unlock(&em_tree->lock); |
3912 | ||
a52d9a80 | 3913 | if (em) { |
e1c4b745 CM |
3914 | if (em->start > start || em->start + em->len <= start) |
3915 | free_extent_map(em); | |
3916 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
3917 | free_extent_map(em); |
3918 | else | |
3919 | goto out; | |
a52d9a80 | 3920 | } |
d1310b2e | 3921 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 3922 | if (!em) { |
d1310b2e CM |
3923 | err = -ENOMEM; |
3924 | goto out; | |
a52d9a80 | 3925 | } |
e6dcd2dc | 3926 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 3927 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 3928 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 3929 | em->len = (u64)-1; |
c8b97818 | 3930 | em->block_len = (u64)-1; |
f421950f CM |
3931 | |
3932 | if (!path) { | |
3933 | path = btrfs_alloc_path(); | |
3934 | BUG_ON(!path); | |
3935 | } | |
3936 | ||
179e29e4 CM |
3937 | ret = btrfs_lookup_file_extent(trans, root, path, |
3938 | objectid, start, trans != NULL); | |
a52d9a80 CM |
3939 | if (ret < 0) { |
3940 | err = ret; | |
3941 | goto out; | |
3942 | } | |
3943 | ||
3944 | if (ret != 0) { | |
3945 | if (path->slots[0] == 0) | |
3946 | goto not_found; | |
3947 | path->slots[0]--; | |
3948 | } | |
3949 | ||
5f39d397 CM |
3950 | leaf = path->nodes[0]; |
3951 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 3952 | struct btrfs_file_extent_item); |
a52d9a80 | 3953 | /* are we inside the extent that was found? */ |
5f39d397 CM |
3954 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3955 | found_type = btrfs_key_type(&found_key); | |
3956 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
3957 | found_type != BTRFS_EXTENT_DATA_KEY) { |
3958 | goto not_found; | |
3959 | } | |
3960 | ||
5f39d397 CM |
3961 | found_type = btrfs_file_extent_type(leaf, item); |
3962 | extent_start = found_key.offset; | |
c8b97818 | 3963 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
3964 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3965 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 3966 | extent_end = extent_start + |
db94535d | 3967 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
3968 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
3969 | size_t size; | |
3970 | size = btrfs_file_extent_inline_len(leaf, item); | |
3971 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
3972 | ~((u64)root->sectorsize - 1); | |
3973 | } | |
3974 | ||
3975 | if (start >= extent_end) { | |
3976 | path->slots[0]++; | |
3977 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3978 | ret = btrfs_next_leaf(root, path); | |
3979 | if (ret < 0) { | |
3980 | err = ret; | |
3981 | goto out; | |
a52d9a80 | 3982 | } |
9036c102 YZ |
3983 | if (ret > 0) |
3984 | goto not_found; | |
3985 | leaf = path->nodes[0]; | |
a52d9a80 | 3986 | } |
9036c102 YZ |
3987 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3988 | if (found_key.objectid != objectid || | |
3989 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
3990 | goto not_found; | |
3991 | if (start + len <= found_key.offset) | |
3992 | goto not_found; | |
3993 | em->start = start; | |
3994 | em->len = found_key.offset - start; | |
3995 | goto not_found_em; | |
3996 | } | |
3997 | ||
d899e052 YZ |
3998 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3999 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4000 | em->start = extent_start; |
4001 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4002 | em->orig_start = extent_start - |
4003 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4004 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4005 | if (bytenr == 0) { | |
5f39d397 | 4006 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4007 | goto insert; |
4008 | } | |
c8b97818 CM |
4009 | if (compressed) { |
4010 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4011 | em->block_start = bytenr; | |
4012 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4013 | item); | |
4014 | } else { | |
4015 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4016 | em->block_start = bytenr; | |
4017 | em->block_len = em->len; | |
d899e052 YZ |
4018 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4019 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4020 | } |
a52d9a80 CM |
4021 | goto insert; |
4022 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4023 | unsigned long ptr; |
a52d9a80 | 4024 | char *map; |
3326d1b0 CM |
4025 | size_t size; |
4026 | size_t extent_offset; | |
4027 | size_t copy_size; | |
a52d9a80 | 4028 | |
689f9346 | 4029 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4030 | if (!page || create) { |
689f9346 | 4031 | em->start = extent_start; |
9036c102 | 4032 | em->len = extent_end - extent_start; |
689f9346 Y |
4033 | goto out; |
4034 | } | |
5f39d397 | 4035 | |
9036c102 YZ |
4036 | size = btrfs_file_extent_inline_len(leaf, item); |
4037 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4038 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4039 | size - extent_offset); |
3326d1b0 | 4040 | em->start = extent_start + extent_offset; |
70dec807 CM |
4041 | em->len = (copy_size + root->sectorsize - 1) & |
4042 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4043 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4044 | if (compressed) |
4045 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4046 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4047 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4048 | if (btrfs_file_extent_compression(leaf, item) == |
4049 | BTRFS_COMPRESS_ZLIB) { | |
4050 | ret = uncompress_inline(path, inode, page, | |
4051 | pg_offset, | |
4052 | extent_offset, item); | |
4053 | BUG_ON(ret); | |
4054 | } else { | |
4055 | map = kmap(page); | |
4056 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4057 | copy_size); | |
4058 | kunmap(page); | |
4059 | } | |
179e29e4 CM |
4060 | flush_dcache_page(page); |
4061 | } else if (create && PageUptodate(page)) { | |
4062 | if (!trans) { | |
4063 | kunmap(page); | |
4064 | free_extent_map(em); | |
4065 | em = NULL; | |
4066 | btrfs_release_path(root, path); | |
f9295749 | 4067 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4068 | goto again; |
4069 | } | |
c8b97818 | 4070 | map = kmap(page); |
70dec807 | 4071 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4072 | copy_size); |
c8b97818 | 4073 | kunmap(page); |
179e29e4 | 4074 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4075 | } |
d1310b2e CM |
4076 | set_extent_uptodate(io_tree, em->start, |
4077 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4078 | goto insert; |
4079 | } else { | |
d397712b | 4080 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4081 | WARN_ON(1); |
4082 | } | |
4083 | not_found: | |
4084 | em->start = start; | |
d1310b2e | 4085 | em->len = len; |
a52d9a80 | 4086 | not_found_em: |
5f39d397 | 4087 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4088 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4089 | insert: |
4090 | btrfs_release_path(root, path); | |
d1310b2e | 4091 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4092 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4093 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4094 | (unsigned long long)em->len, | |
4095 | (unsigned long long)start, | |
4096 | (unsigned long long)len); | |
a52d9a80 CM |
4097 | err = -EIO; |
4098 | goto out; | |
4099 | } | |
d1310b2e CM |
4100 | |
4101 | err = 0; | |
4102 | spin_lock(&em_tree->lock); | |
a52d9a80 | 4103 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4104 | /* it is possible that someone inserted the extent into the tree |
4105 | * while we had the lock dropped. It is also possible that | |
4106 | * an overlapping map exists in the tree | |
4107 | */ | |
a52d9a80 | 4108 | if (ret == -EEXIST) { |
3b951516 | 4109 | struct extent_map *existing; |
e6dcd2dc CM |
4110 | |
4111 | ret = 0; | |
4112 | ||
3b951516 | 4113 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4114 | if (existing && (existing->start > start || |
4115 | existing->start + existing->len <= start)) { | |
4116 | free_extent_map(existing); | |
4117 | existing = NULL; | |
4118 | } | |
3b951516 CM |
4119 | if (!existing) { |
4120 | existing = lookup_extent_mapping(em_tree, em->start, | |
4121 | em->len); | |
4122 | if (existing) { | |
4123 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4124 | em, start, |
4125 | root->sectorsize); | |
3b951516 CM |
4126 | free_extent_map(existing); |
4127 | if (err) { | |
4128 | free_extent_map(em); | |
4129 | em = NULL; | |
4130 | } | |
4131 | } else { | |
4132 | err = -EIO; | |
3b951516 CM |
4133 | free_extent_map(em); |
4134 | em = NULL; | |
4135 | } | |
4136 | } else { | |
4137 | free_extent_map(em); | |
4138 | em = existing; | |
e6dcd2dc | 4139 | err = 0; |
a52d9a80 | 4140 | } |
a52d9a80 | 4141 | } |
d1310b2e | 4142 | spin_unlock(&em_tree->lock); |
a52d9a80 | 4143 | out: |
f421950f CM |
4144 | if (path) |
4145 | btrfs_free_path(path); | |
a52d9a80 CM |
4146 | if (trans) { |
4147 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4148 | if (!err) |
a52d9a80 CM |
4149 | err = ret; |
4150 | } | |
a52d9a80 CM |
4151 | if (err) { |
4152 | free_extent_map(em); | |
4153 | WARN_ON(1); | |
4154 | return ERR_PTR(err); | |
4155 | } | |
4156 | return em; | |
4157 | } | |
4158 | ||
16432985 CM |
4159 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4160 | const struct iovec *iov, loff_t offset, | |
4161 | unsigned long nr_segs) | |
4162 | { | |
e1c4b745 | 4163 | return -EINVAL; |
16432985 CM |
4164 | } |
4165 | ||
d396c6f5 | 4166 | static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock) |
39279cc3 | 4167 | { |
d396c6f5 | 4168 | return extent_bmap(mapping, iblock, btrfs_get_extent); |
39279cc3 CM |
4169 | } |
4170 | ||
a52d9a80 | 4171 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4172 | { |
d1310b2e CM |
4173 | struct extent_io_tree *tree; |
4174 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4175 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4176 | } |
1832a6d5 | 4177 | |
a52d9a80 | 4178 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4179 | { |
d1310b2e | 4180 | struct extent_io_tree *tree; |
b888db2b CM |
4181 | |
4182 | ||
4183 | if (current->flags & PF_MEMALLOC) { | |
4184 | redirty_page_for_writepage(wbc, page); | |
4185 | unlock_page(page); | |
4186 | return 0; | |
4187 | } | |
d1310b2e | 4188 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4189 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4190 | } |
4191 | ||
f421950f CM |
4192 | int btrfs_writepages(struct address_space *mapping, |
4193 | struct writeback_control *wbc) | |
b293f02e | 4194 | { |
d1310b2e | 4195 | struct extent_io_tree *tree; |
771ed689 | 4196 | |
d1310b2e | 4197 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4198 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4199 | } | |
4200 | ||
3ab2fb5a CM |
4201 | static int |
4202 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4203 | struct list_head *pages, unsigned nr_pages) | |
4204 | { | |
d1310b2e CM |
4205 | struct extent_io_tree *tree; |
4206 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4207 | return extent_readpages(tree, mapping, pages, nr_pages, |
4208 | btrfs_get_extent); | |
4209 | } | |
e6dcd2dc | 4210 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4211 | { |
d1310b2e CM |
4212 | struct extent_io_tree *tree; |
4213 | struct extent_map_tree *map; | |
a52d9a80 | 4214 | int ret; |
8c2383c3 | 4215 | |
d1310b2e CM |
4216 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4217 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4218 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4219 | if (ret == 1) { |
4220 | ClearPagePrivate(page); | |
4221 | set_page_private(page, 0); | |
4222 | page_cache_release(page); | |
39279cc3 | 4223 | } |
a52d9a80 | 4224 | return ret; |
39279cc3 CM |
4225 | } |
4226 | ||
e6dcd2dc CM |
4227 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4228 | { | |
98509cfc CM |
4229 | if (PageWriteback(page) || PageDirty(page)) |
4230 | return 0; | |
e6dcd2dc CM |
4231 | return __btrfs_releasepage(page, gfp_flags); |
4232 | } | |
4233 | ||
a52d9a80 | 4234 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4235 | { |
d1310b2e | 4236 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4237 | struct btrfs_ordered_extent *ordered; |
4238 | u64 page_start = page_offset(page); | |
4239 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4240 | |
e6dcd2dc | 4241 | wait_on_page_writeback(page); |
d1310b2e | 4242 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4243 | if (offset) { |
4244 | btrfs_releasepage(page, GFP_NOFS); | |
4245 | return; | |
4246 | } | |
4247 | ||
4248 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4249 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4250 | page_offset(page)); | |
4251 | if (ordered) { | |
eb84ae03 CM |
4252 | /* |
4253 | * IO on this page will never be started, so we need | |
4254 | * to account for any ordered extents now | |
4255 | */ | |
e6dcd2dc CM |
4256 | clear_extent_bit(tree, page_start, page_end, |
4257 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4258 | EXTENT_LOCKED, 1, 0, GFP_NOFS); | |
211f90e6 CM |
4259 | btrfs_finish_ordered_io(page->mapping->host, |
4260 | page_start, page_end); | |
e6dcd2dc CM |
4261 | btrfs_put_ordered_extent(ordered); |
4262 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4263 | } | |
4264 | clear_extent_bit(tree, page_start, page_end, | |
4265 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4266 | EXTENT_ORDERED, | |
4267 | 1, 1, GFP_NOFS); | |
4268 | __btrfs_releasepage(page, GFP_NOFS); | |
4269 | ||
4a096752 | 4270 | ClearPageChecked(page); |
9ad6b7bc | 4271 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4272 | ClearPagePrivate(page); |
4273 | set_page_private(page, 0); | |
4274 | page_cache_release(page); | |
4275 | } | |
39279cc3 CM |
4276 | } |
4277 | ||
9ebefb18 CM |
4278 | /* |
4279 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4280 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4281 | * be careful to check for EOF conditions here. We set the page up correctly | |
4282 | * for a written page which means we get ENOSPC checking when writing into | |
4283 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4284 | * support these features. | |
4285 | * | |
4286 | * We are not allowed to take the i_mutex here so we have to play games to | |
4287 | * protect against truncate races as the page could now be beyond EOF. Because | |
4288 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4289 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4290 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4291 | * unlock the page. | |
4292 | */ | |
4293 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) | |
4294 | { | |
6da6abae | 4295 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4296 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4297 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4298 | struct btrfs_ordered_extent *ordered; | |
4299 | char *kaddr; | |
4300 | unsigned long zero_start; | |
9ebefb18 | 4301 | loff_t size; |
1832a6d5 | 4302 | int ret; |
a52d9a80 | 4303 | u64 page_start; |
e6dcd2dc | 4304 | u64 page_end; |
9ebefb18 | 4305 | |
1832a6d5 | 4306 | ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0); |
1832a6d5 CM |
4307 | if (ret) |
4308 | goto out; | |
4309 | ||
4310 | ret = -EINVAL; | |
e6dcd2dc | 4311 | again: |
9ebefb18 | 4312 | lock_page(page); |
9ebefb18 | 4313 | size = i_size_read(inode); |
e6dcd2dc CM |
4314 | page_start = page_offset(page); |
4315 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4316 | |
9ebefb18 | 4317 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4318 | (page_start >= size)) { |
9ebefb18 CM |
4319 | /* page got truncated out from underneath us */ |
4320 | goto out_unlock; | |
4321 | } | |
e6dcd2dc CM |
4322 | wait_on_page_writeback(page); |
4323 | ||
4324 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4325 | set_page_extent_mapped(page); | |
4326 | ||
eb84ae03 CM |
4327 | /* |
4328 | * we can't set the delalloc bits if there are pending ordered | |
4329 | * extents. Drop our locks and wait for them to finish | |
4330 | */ | |
e6dcd2dc CM |
4331 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4332 | if (ordered) { | |
4333 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4334 | unlock_page(page); | |
eb84ae03 | 4335 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4336 | btrfs_put_ordered_extent(ordered); |
4337 | goto again; | |
4338 | } | |
4339 | ||
ea8c2819 | 4340 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc | 4341 | ret = 0; |
9ebefb18 CM |
4342 | |
4343 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4344 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4345 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4346 | else |
e6dcd2dc | 4347 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4348 | |
e6dcd2dc CM |
4349 | if (zero_start != PAGE_CACHE_SIZE) { |
4350 | kaddr = kmap(page); | |
4351 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4352 | flush_dcache_page(page); | |
4353 | kunmap(page); | |
4354 | } | |
247e743c | 4355 | ClearPageChecked(page); |
e6dcd2dc CM |
4356 | set_page_dirty(page); |
4357 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
9ebefb18 CM |
4358 | |
4359 | out_unlock: | |
4360 | unlock_page(page); | |
1832a6d5 | 4361 | out: |
9ebefb18 CM |
4362 | return ret; |
4363 | } | |
4364 | ||
39279cc3 CM |
4365 | static void btrfs_truncate(struct inode *inode) |
4366 | { | |
4367 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4368 | int ret; | |
4369 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 4370 | unsigned long nr; |
dbe674a9 | 4371 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
4372 | |
4373 | if (!S_ISREG(inode->i_mode)) | |
4374 | return; | |
4375 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4376 | return; | |
4377 | ||
4378 | btrfs_truncate_page(inode->i_mapping, inode->i_size); | |
4a096752 | 4379 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 4380 | |
39279cc3 CM |
4381 | trans = btrfs_start_transaction(root, 1); |
4382 | btrfs_set_trans_block_group(trans, inode); | |
dbe674a9 | 4383 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 4384 | |
7b128766 JB |
4385 | ret = btrfs_orphan_add(trans, inode); |
4386 | if (ret) | |
4387 | goto out; | |
39279cc3 | 4388 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 4389 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 4390 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 4391 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 4392 | |
7b128766 JB |
4393 | ret = btrfs_orphan_del(trans, inode); |
4394 | BUG_ON(ret); | |
4395 | ||
4396 | out: | |
4397 | nr = trans->blocks_used; | |
89ce8a63 | 4398 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 4399 | BUG_ON(ret); |
d3c2fdcf | 4400 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4401 | } |
4402 | ||
d352ac68 CM |
4403 | /* |
4404 | * create a new subvolume directory/inode (helper for the ioctl). | |
4405 | */ | |
d2fb3437 YZ |
4406 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
4407 | struct btrfs_root *new_root, struct dentry *dentry, | |
4408 | u64 new_dirid, u64 alloc_hint) | |
39279cc3 | 4409 | { |
39279cc3 | 4410 | struct inode *inode; |
cb8e7090 | 4411 | int error; |
00e4e6b3 | 4412 | u64 index = 0; |
39279cc3 | 4413 | |
aec7477b | 4414 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 4415 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 4416 | if (IS_ERR(inode)) |
f46b5a66 | 4417 | return PTR_ERR(inode); |
39279cc3 CM |
4418 | inode->i_op = &btrfs_dir_inode_operations; |
4419 | inode->i_fop = &btrfs_dir_file_operations; | |
4420 | ||
39279cc3 | 4421 | inode->i_nlink = 1; |
dbe674a9 | 4422 | btrfs_i_size_write(inode, 0); |
3b96362c | 4423 | |
cb8e7090 CH |
4424 | error = btrfs_update_inode(trans, new_root, inode); |
4425 | if (error) | |
4426 | return error; | |
4427 | ||
4428 | d_instantiate(dentry, inode); | |
4429 | return 0; | |
39279cc3 CM |
4430 | } |
4431 | ||
d352ac68 CM |
4432 | /* helper function for file defrag and space balancing. This |
4433 | * forces readahead on a given range of bytes in an inode | |
4434 | */ | |
edbd8d4e | 4435 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
4436 | struct file_ra_state *ra, struct file *file, |
4437 | pgoff_t offset, pgoff_t last_index) | |
4438 | { | |
8e7bf94f | 4439 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 4440 | |
86479a04 CM |
4441 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
4442 | return offset + req_size; | |
86479a04 CM |
4443 | } |
4444 | ||
39279cc3 CM |
4445 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
4446 | { | |
4447 | struct btrfs_inode *ei; | |
4448 | ||
4449 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
4450 | if (!ei) | |
4451 | return NULL; | |
15ee9bc7 | 4452 | ei->last_trans = 0; |
e02119d5 | 4453 | ei->logged_trans = 0; |
e6dcd2dc | 4454 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
33268eaf JB |
4455 | ei->i_acl = BTRFS_ACL_NOT_CACHED; |
4456 | ei->i_default_acl = BTRFS_ACL_NOT_CACHED; | |
7b128766 | 4457 | INIT_LIST_HEAD(&ei->i_orphan); |
39279cc3 CM |
4458 | return &ei->vfs_inode; |
4459 | } | |
4460 | ||
4461 | void btrfs_destroy_inode(struct inode *inode) | |
4462 | { | |
e6dcd2dc | 4463 | struct btrfs_ordered_extent *ordered; |
39279cc3 CM |
4464 | WARN_ON(!list_empty(&inode->i_dentry)); |
4465 | WARN_ON(inode->i_data.nrpages); | |
4466 | ||
33268eaf JB |
4467 | if (BTRFS_I(inode)->i_acl && |
4468 | BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED) | |
4469 | posix_acl_release(BTRFS_I(inode)->i_acl); | |
4470 | if (BTRFS_I(inode)->i_default_acl && | |
4471 | BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED) | |
4472 | posix_acl_release(BTRFS_I(inode)->i_default_acl); | |
4473 | ||
bcc63abb | 4474 | spin_lock(&BTRFS_I(inode)->root->list_lock); |
7b128766 JB |
4475 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
4476 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
4477 | " list\n", inode->i_ino); | |
4478 | dump_stack(); | |
4479 | } | |
bcc63abb | 4480 | spin_unlock(&BTRFS_I(inode)->root->list_lock); |
7b128766 | 4481 | |
d397712b | 4482 | while (1) { |
e6dcd2dc CM |
4483 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
4484 | if (!ordered) | |
4485 | break; | |
4486 | else { | |
d397712b CM |
4487 | printk(KERN_ERR "btrfs found ordered " |
4488 | "extent %llu %llu on inode cleanup\n", | |
4489 | (unsigned long long)ordered->file_offset, | |
4490 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
4491 | btrfs_remove_ordered_extent(inode, ordered); |
4492 | btrfs_put_ordered_extent(ordered); | |
4493 | btrfs_put_ordered_extent(ordered); | |
4494 | } | |
4495 | } | |
5b21f2ed | 4496 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
4497 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
4498 | } | |
4499 | ||
0ee0fda0 | 4500 | static void init_once(void *foo) |
39279cc3 CM |
4501 | { |
4502 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
4503 | ||
4504 | inode_init_once(&ei->vfs_inode); | |
4505 | } | |
4506 | ||
4507 | void btrfs_destroy_cachep(void) | |
4508 | { | |
4509 | if (btrfs_inode_cachep) | |
4510 | kmem_cache_destroy(btrfs_inode_cachep); | |
4511 | if (btrfs_trans_handle_cachep) | |
4512 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
4513 | if (btrfs_transaction_cachep) | |
4514 | kmem_cache_destroy(btrfs_transaction_cachep); | |
4515 | if (btrfs_bit_radix_cachep) | |
4516 | kmem_cache_destroy(btrfs_bit_radix_cachep); | |
4517 | if (btrfs_path_cachep) | |
4518 | kmem_cache_destroy(btrfs_path_cachep); | |
4519 | } | |
4520 | ||
86479a04 | 4521 | struct kmem_cache *btrfs_cache_create(const char *name, size_t size, |
92fee66d | 4522 | unsigned long extra_flags, |
2b1f55b0 | 4523 | void (*ctor)(void *)) |
92fee66d CM |
4524 | { |
4525 | return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT | | |
2b1f55b0 | 4526 | SLAB_MEM_SPREAD | extra_flags), ctor); |
92fee66d CM |
4527 | } |
4528 | ||
39279cc3 CM |
4529 | int btrfs_init_cachep(void) |
4530 | { | |
86479a04 | 4531 | btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache", |
92fee66d CM |
4532 | sizeof(struct btrfs_inode), |
4533 | 0, init_once); | |
39279cc3 CM |
4534 | if (!btrfs_inode_cachep) |
4535 | goto fail; | |
86479a04 CM |
4536 | btrfs_trans_handle_cachep = |
4537 | btrfs_cache_create("btrfs_trans_handle_cache", | |
4538 | sizeof(struct btrfs_trans_handle), | |
4539 | 0, NULL); | |
39279cc3 CM |
4540 | if (!btrfs_trans_handle_cachep) |
4541 | goto fail; | |
86479a04 | 4542 | btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache", |
39279cc3 | 4543 | sizeof(struct btrfs_transaction), |
92fee66d | 4544 | 0, NULL); |
39279cc3 CM |
4545 | if (!btrfs_transaction_cachep) |
4546 | goto fail; | |
86479a04 | 4547 | btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache", |
23223584 | 4548 | sizeof(struct btrfs_path), |
92fee66d | 4549 | 0, NULL); |
39279cc3 CM |
4550 | if (!btrfs_path_cachep) |
4551 | goto fail; | |
86479a04 | 4552 | btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256, |
92fee66d | 4553 | SLAB_DESTROY_BY_RCU, NULL); |
39279cc3 CM |
4554 | if (!btrfs_bit_radix_cachep) |
4555 | goto fail; | |
4556 | return 0; | |
4557 | fail: | |
4558 | btrfs_destroy_cachep(); | |
4559 | return -ENOMEM; | |
4560 | } | |
4561 | ||
4562 | static int btrfs_getattr(struct vfsmount *mnt, | |
4563 | struct dentry *dentry, struct kstat *stat) | |
4564 | { | |
4565 | struct inode *inode = dentry->d_inode; | |
4566 | generic_fillattr(inode, stat); | |
3394e160 | 4567 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 4568 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
4569 | stat->blocks = (inode_get_bytes(inode) + |
4570 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
4571 | return 0; |
4572 | } | |
4573 | ||
d397712b CM |
4574 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4575 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
4576 | { |
4577 | struct btrfs_trans_handle *trans; | |
4578 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4579 | struct inode *new_inode = new_dentry->d_inode; | |
4580 | struct inode *old_inode = old_dentry->d_inode; | |
4581 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 4582 | u64 index = 0; |
39279cc3 CM |
4583 | int ret; |
4584 | ||
3394e160 CM |
4585 | /* we're not allowed to rename between subvolumes */ |
4586 | if (BTRFS_I(old_inode)->root->root_key.objectid != | |
4587 | BTRFS_I(new_dir)->root->root_key.objectid) | |
4588 | return -EXDEV; | |
4589 | ||
39279cc3 CM |
4590 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
4591 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) { | |
4592 | return -ENOTEMPTY; | |
4593 | } | |
5f39d397 | 4594 | |
0660b5af CM |
4595 | /* to rename a snapshot or subvolume, we need to juggle the |
4596 | * backrefs. This isn't coded yet | |
4597 | */ | |
4598 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
4599 | return -EXDEV; | |
4600 | ||
1832a6d5 CM |
4601 | ret = btrfs_check_free_space(root, 1, 0); |
4602 | if (ret) | |
4603 | goto out_unlock; | |
4604 | ||
39279cc3 | 4605 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4606 | |
39279cc3 | 4607 | btrfs_set_trans_block_group(trans, new_dir); |
39279cc3 | 4608 | |
e02119d5 | 4609 | btrfs_inc_nlink(old_dentry->d_inode); |
39279cc3 CM |
4610 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
4611 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
4612 | old_inode->i_ctime = ctime; | |
5f39d397 | 4613 | |
e02119d5 CM |
4614 | ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, |
4615 | old_dentry->d_name.name, | |
4616 | old_dentry->d_name.len); | |
39279cc3 CM |
4617 | if (ret) |
4618 | goto out_fail; | |
4619 | ||
4620 | if (new_inode) { | |
4621 | new_inode->i_ctime = CURRENT_TIME; | |
e02119d5 CM |
4622 | ret = btrfs_unlink_inode(trans, root, new_dir, |
4623 | new_dentry->d_inode, | |
4624 | new_dentry->d_name.name, | |
4625 | new_dentry->d_name.len); | |
39279cc3 CM |
4626 | if (ret) |
4627 | goto out_fail; | |
7b128766 | 4628 | if (new_inode->i_nlink == 0) { |
e02119d5 | 4629 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
7b128766 JB |
4630 | if (ret) |
4631 | goto out_fail; | |
4632 | } | |
e02119d5 | 4633 | |
39279cc3 | 4634 | } |
3de4586c | 4635 | ret = btrfs_set_inode_index(new_dir, &index); |
aec7477b JB |
4636 | if (ret) |
4637 | goto out_fail; | |
4638 | ||
e02119d5 CM |
4639 | ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode, |
4640 | old_inode, new_dentry->d_name.name, | |
4641 | new_dentry->d_name.len, 1, index); | |
39279cc3 CM |
4642 | if (ret) |
4643 | goto out_fail; | |
4644 | ||
4645 | out_fail: | |
ab78c84d | 4646 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4647 | out_unlock: |
39279cc3 CM |
4648 | return ret; |
4649 | } | |
4650 | ||
d352ac68 CM |
4651 | /* |
4652 | * some fairly slow code that needs optimization. This walks the list | |
4653 | * of all the inodes with pending delalloc and forces them to disk. | |
4654 | */ | |
ea8c2819 CM |
4655 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
4656 | { | |
4657 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
4658 | struct btrfs_inode *binode; | |
5b21f2ed | 4659 | struct inode *inode; |
ea8c2819 | 4660 | |
c146afad YZ |
4661 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
4662 | return -EROFS; | |
4663 | ||
75eff68e | 4664 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 4665 | while (!list_empty(head)) { |
ea8c2819 CM |
4666 | binode = list_entry(head->next, struct btrfs_inode, |
4667 | delalloc_inodes); | |
5b21f2ed ZY |
4668 | inode = igrab(&binode->vfs_inode); |
4669 | if (!inode) | |
4670 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 4671 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 4672 | if (inode) { |
8c8bee1d | 4673 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
4674 | iput(inode); |
4675 | } | |
4676 | cond_resched(); | |
75eff68e | 4677 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 4678 | } |
75eff68e | 4679 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
4680 | |
4681 | /* the filemap_flush will queue IO into the worker threads, but | |
4682 | * we have to make sure the IO is actually started and that | |
4683 | * ordered extents get created before we return | |
4684 | */ | |
4685 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 4686 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 4687 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 4688 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
4689 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
4690 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
4691 | } |
4692 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
4693 | return 0; |
4694 | } | |
4695 | ||
39279cc3 CM |
4696 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
4697 | const char *symname) | |
4698 | { | |
4699 | struct btrfs_trans_handle *trans; | |
4700 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4701 | struct btrfs_path *path; | |
4702 | struct btrfs_key key; | |
1832a6d5 | 4703 | struct inode *inode = NULL; |
39279cc3 CM |
4704 | int err; |
4705 | int drop_inode = 0; | |
4706 | u64 objectid; | |
00e4e6b3 | 4707 | u64 index = 0 ; |
39279cc3 CM |
4708 | int name_len; |
4709 | int datasize; | |
5f39d397 | 4710 | unsigned long ptr; |
39279cc3 | 4711 | struct btrfs_file_extent_item *ei; |
5f39d397 | 4712 | struct extent_buffer *leaf; |
1832a6d5 | 4713 | unsigned long nr = 0; |
39279cc3 CM |
4714 | |
4715 | name_len = strlen(symname) + 1; | |
4716 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
4717 | return -ENAMETOOLONG; | |
1832a6d5 | 4718 | |
1832a6d5 CM |
4719 | err = btrfs_check_free_space(root, 1, 0); |
4720 | if (err) | |
4721 | goto out_fail; | |
4722 | ||
39279cc3 CM |
4723 | trans = btrfs_start_transaction(root, 1); |
4724 | btrfs_set_trans_block_group(trans, dir); | |
4725 | ||
4726 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4727 | if (err) { | |
4728 | err = -ENOSPC; | |
4729 | goto out_unlock; | |
4730 | } | |
4731 | ||
aec7477b | 4732 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4733 | dentry->d_name.len, |
4734 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4735 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
4736 | &index); | |
39279cc3 CM |
4737 | err = PTR_ERR(inode); |
4738 | if (IS_ERR(inode)) | |
4739 | goto out_unlock; | |
4740 | ||
33268eaf JB |
4741 | err = btrfs_init_acl(inode, dir); |
4742 | if (err) { | |
4743 | drop_inode = 1; | |
4744 | goto out_unlock; | |
4745 | } | |
4746 | ||
39279cc3 | 4747 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4748 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4749 | if (err) |
4750 | drop_inode = 1; | |
4751 | else { | |
4752 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4753 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4754 | inode->i_fop = &btrfs_file_operations; |
4755 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4756 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
4757 | } |
4758 | dir->i_sb->s_dirt = 1; | |
4759 | btrfs_update_inode_block_group(trans, inode); | |
4760 | btrfs_update_inode_block_group(trans, dir); | |
4761 | if (drop_inode) | |
4762 | goto out_unlock; | |
4763 | ||
4764 | path = btrfs_alloc_path(); | |
4765 | BUG_ON(!path); | |
4766 | key.objectid = inode->i_ino; | |
4767 | key.offset = 0; | |
39279cc3 CM |
4768 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
4769 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
4770 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
4771 | datasize); | |
54aa1f4d CM |
4772 | if (err) { |
4773 | drop_inode = 1; | |
4774 | goto out_unlock; | |
4775 | } | |
5f39d397 CM |
4776 | leaf = path->nodes[0]; |
4777 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
4778 | struct btrfs_file_extent_item); | |
4779 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
4780 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 4781 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
4782 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
4783 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
4784 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
4785 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
4786 | ||
39279cc3 | 4787 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
4788 | write_extent_buffer(leaf, symname, ptr, name_len); |
4789 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 4790 | btrfs_free_path(path); |
5f39d397 | 4791 | |
39279cc3 CM |
4792 | inode->i_op = &btrfs_symlink_inode_operations; |
4793 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 4794 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 4795 | inode_set_bytes(inode, name_len); |
dbe674a9 | 4796 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
4797 | err = btrfs_update_inode(trans, root, inode); |
4798 | if (err) | |
4799 | drop_inode = 1; | |
39279cc3 CM |
4800 | |
4801 | out_unlock: | |
d3c2fdcf | 4802 | nr = trans->blocks_used; |
ab78c84d | 4803 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4804 | out_fail: |
39279cc3 CM |
4805 | if (drop_inode) { |
4806 | inode_dec_link_count(inode); | |
4807 | iput(inode); | |
4808 | } | |
d3c2fdcf | 4809 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4810 | return err; |
4811 | } | |
16432985 | 4812 | |
d899e052 YZ |
4813 | static int prealloc_file_range(struct inode *inode, u64 start, u64 end, |
4814 | u64 alloc_hint, int mode) | |
4815 | { | |
4816 | struct btrfs_trans_handle *trans; | |
4817 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4818 | struct btrfs_key ins; | |
4819 | u64 alloc_size; | |
4820 | u64 cur_offset = start; | |
4821 | u64 num_bytes = end - start; | |
4822 | int ret = 0; | |
4823 | ||
4824 | trans = btrfs_join_transaction(root, 1); | |
4825 | BUG_ON(!trans); | |
4826 | btrfs_set_trans_block_group(trans, inode); | |
4827 | ||
4828 | while (num_bytes > 0) { | |
4829 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
4830 | ret = btrfs_reserve_extent(trans, root, alloc_size, | |
4831 | root->sectorsize, 0, alloc_hint, | |
4832 | (u64)-1, &ins, 1); | |
4833 | if (ret) { | |
4834 | WARN_ON(1); | |
4835 | goto out; | |
4836 | } | |
4837 | ret = insert_reserved_file_extent(trans, inode, | |
4838 | cur_offset, ins.objectid, | |
4839 | ins.offset, ins.offset, | |
4840 | ins.offset, 0, 0, 0, | |
4841 | BTRFS_FILE_EXTENT_PREALLOC); | |
4842 | BUG_ON(ret); | |
4843 | num_bytes -= ins.offset; | |
4844 | cur_offset += ins.offset; | |
4845 | alloc_hint = ins.objectid + ins.offset; | |
4846 | } | |
4847 | out: | |
4848 | if (cur_offset > start) { | |
4849 | inode->i_ctime = CURRENT_TIME; | |
4850 | btrfs_set_flag(inode, PREALLOC); | |
4851 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4852 | cur_offset > i_size_read(inode)) | |
4853 | btrfs_i_size_write(inode, cur_offset); | |
4854 | ret = btrfs_update_inode(trans, root, inode); | |
4855 | BUG_ON(ret); | |
4856 | } | |
4857 | ||
4858 | btrfs_end_transaction(trans, root); | |
4859 | return ret; | |
4860 | } | |
4861 | ||
4862 | static long btrfs_fallocate(struct inode *inode, int mode, | |
4863 | loff_t offset, loff_t len) | |
4864 | { | |
4865 | u64 cur_offset; | |
4866 | u64 last_byte; | |
4867 | u64 alloc_start; | |
4868 | u64 alloc_end; | |
4869 | u64 alloc_hint = 0; | |
4870 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
4871 | struct extent_map *em; | |
4872 | int ret; | |
4873 | ||
4874 | alloc_start = offset & ~mask; | |
4875 | alloc_end = (offset + len + mask) & ~mask; | |
4876 | ||
4877 | mutex_lock(&inode->i_mutex); | |
4878 | if (alloc_start > inode->i_size) { | |
4879 | ret = btrfs_cont_expand(inode, alloc_start); | |
4880 | if (ret) | |
4881 | goto out; | |
4882 | } | |
4883 | ||
4884 | while (1) { | |
4885 | struct btrfs_ordered_extent *ordered; | |
4886 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, | |
4887 | alloc_end - 1, GFP_NOFS); | |
4888 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
4889 | alloc_end - 1); | |
4890 | if (ordered && | |
4891 | ordered->file_offset + ordered->len > alloc_start && | |
4892 | ordered->file_offset < alloc_end) { | |
4893 | btrfs_put_ordered_extent(ordered); | |
4894 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4895 | alloc_start, alloc_end - 1, GFP_NOFS); | |
4896 | btrfs_wait_ordered_range(inode, alloc_start, | |
4897 | alloc_end - alloc_start); | |
4898 | } else { | |
4899 | if (ordered) | |
4900 | btrfs_put_ordered_extent(ordered); | |
4901 | break; | |
4902 | } | |
4903 | } | |
4904 | ||
4905 | cur_offset = alloc_start; | |
4906 | while (1) { | |
4907 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4908 | alloc_end - cur_offset, 0); | |
4909 | BUG_ON(IS_ERR(em) || !em); | |
4910 | last_byte = min(extent_map_end(em), alloc_end); | |
4911 | last_byte = (last_byte + mask) & ~mask; | |
4912 | if (em->block_start == EXTENT_MAP_HOLE) { | |
4913 | ret = prealloc_file_range(inode, cur_offset, | |
4914 | last_byte, alloc_hint, mode); | |
4915 | if (ret < 0) { | |
4916 | free_extent_map(em); | |
4917 | break; | |
4918 | } | |
4919 | } | |
4920 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
4921 | alloc_hint = em->block_start; | |
4922 | free_extent_map(em); | |
4923 | ||
4924 | cur_offset = last_byte; | |
4925 | if (cur_offset >= alloc_end) { | |
4926 | ret = 0; | |
4927 | break; | |
4928 | } | |
4929 | } | |
4930 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1, | |
4931 | GFP_NOFS); | |
4932 | out: | |
4933 | mutex_unlock(&inode->i_mutex); | |
4934 | return ret; | |
4935 | } | |
4936 | ||
e6dcd2dc CM |
4937 | static int btrfs_set_page_dirty(struct page *page) |
4938 | { | |
e6dcd2dc CM |
4939 | return __set_page_dirty_nobuffers(page); |
4940 | } | |
4941 | ||
0ee0fda0 | 4942 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd Y |
4943 | { |
4944 | if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE)) | |
4945 | return -EACCES; | |
33268eaf | 4946 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 4947 | } |
39279cc3 CM |
4948 | |
4949 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 4950 | .getattr = btrfs_getattr, |
39279cc3 CM |
4951 | .lookup = btrfs_lookup, |
4952 | .create = btrfs_create, | |
4953 | .unlink = btrfs_unlink, | |
4954 | .link = btrfs_link, | |
4955 | .mkdir = btrfs_mkdir, | |
4956 | .rmdir = btrfs_rmdir, | |
4957 | .rename = btrfs_rename, | |
4958 | .symlink = btrfs_symlink, | |
4959 | .setattr = btrfs_setattr, | |
618e21d5 | 4960 | .mknod = btrfs_mknod, |
95819c05 CH |
4961 | .setxattr = btrfs_setxattr, |
4962 | .getxattr = btrfs_getxattr, | |
5103e947 | 4963 | .listxattr = btrfs_listxattr, |
95819c05 | 4964 | .removexattr = btrfs_removexattr, |
fdebe2bd | 4965 | .permission = btrfs_permission, |
39279cc3 | 4966 | }; |
39279cc3 CM |
4967 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
4968 | .lookup = btrfs_lookup, | |
fdebe2bd | 4969 | .permission = btrfs_permission, |
39279cc3 | 4970 | }; |
39279cc3 CM |
4971 | static struct file_operations btrfs_dir_file_operations = { |
4972 | .llseek = generic_file_llseek, | |
4973 | .read = generic_read_dir, | |
cbdf5a24 | 4974 | .readdir = btrfs_real_readdir, |
34287aa3 | 4975 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 4976 | #ifdef CONFIG_COMPAT |
34287aa3 | 4977 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 4978 | #endif |
6bf13c0c | 4979 | .release = btrfs_release_file, |
e02119d5 | 4980 | .fsync = btrfs_sync_file, |
39279cc3 CM |
4981 | }; |
4982 | ||
d1310b2e | 4983 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 4984 | .fill_delalloc = run_delalloc_range, |
065631f6 | 4985 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 4986 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 4987 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 4988 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 4989 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 4990 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
4991 | .set_bit_hook = btrfs_set_bit_hook, |
4992 | .clear_bit_hook = btrfs_clear_bit_hook, | |
07157aac CM |
4993 | }; |
4994 | ||
39279cc3 CM |
4995 | static struct address_space_operations btrfs_aops = { |
4996 | .readpage = btrfs_readpage, | |
4997 | .writepage = btrfs_writepage, | |
b293f02e | 4998 | .writepages = btrfs_writepages, |
3ab2fb5a | 4999 | .readpages = btrfs_readpages, |
39279cc3 | 5000 | .sync_page = block_sync_page, |
39279cc3 | 5001 | .bmap = btrfs_bmap, |
16432985 | 5002 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5003 | .invalidatepage = btrfs_invalidatepage, |
5004 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5005 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5006 | }; |
5007 | ||
5008 | static struct address_space_operations btrfs_symlink_aops = { | |
5009 | .readpage = btrfs_readpage, | |
5010 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5011 | .invalidatepage = btrfs_invalidatepage, |
5012 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5013 | }; |
5014 | ||
5015 | static struct inode_operations btrfs_file_inode_operations = { | |
5016 | .truncate = btrfs_truncate, | |
5017 | .getattr = btrfs_getattr, | |
5018 | .setattr = btrfs_setattr, | |
95819c05 CH |
5019 | .setxattr = btrfs_setxattr, |
5020 | .getxattr = btrfs_getxattr, | |
5103e947 | 5021 | .listxattr = btrfs_listxattr, |
95819c05 | 5022 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5023 | .permission = btrfs_permission, |
d899e052 | 5024 | .fallocate = btrfs_fallocate, |
39279cc3 | 5025 | }; |
618e21d5 JB |
5026 | static struct inode_operations btrfs_special_inode_operations = { |
5027 | .getattr = btrfs_getattr, | |
5028 | .setattr = btrfs_setattr, | |
fdebe2bd | 5029 | .permission = btrfs_permission, |
95819c05 CH |
5030 | .setxattr = btrfs_setxattr, |
5031 | .getxattr = btrfs_getxattr, | |
33268eaf | 5032 | .listxattr = btrfs_listxattr, |
95819c05 | 5033 | .removexattr = btrfs_removexattr, |
618e21d5 | 5034 | }; |
39279cc3 CM |
5035 | static struct inode_operations btrfs_symlink_inode_operations = { |
5036 | .readlink = generic_readlink, | |
5037 | .follow_link = page_follow_link_light, | |
5038 | .put_link = page_put_link, | |
fdebe2bd | 5039 | .permission = btrfs_permission, |
39279cc3 | 5040 | }; |