Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
4b4e25f2 | 40 | #include "compat.h" |
39279cc3 CM |
41 | #include "ctree.h" |
42 | #include "disk-io.h" | |
43 | #include "transaction.h" | |
44 | #include "btrfs_inode.h" | |
45 | #include "ioctl.h" | |
46 | #include "print-tree.h" | |
0b86a832 | 47 | #include "volumes.h" |
e6dcd2dc | 48 | #include "ordered-data.h" |
95819c05 | 49 | #include "xattr.h" |
e02119d5 | 50 | #include "tree-log.h" |
c8b97818 | 51 | #include "compression.h" |
b4ce94de | 52 | #include "locking.h" |
39279cc3 CM |
53 | |
54 | struct btrfs_iget_args { | |
55 | u64 ino; | |
56 | struct btrfs_root *root; | |
57 | }; | |
58 | ||
6e1d5dcc AD |
59 | static const struct inode_operations btrfs_dir_inode_operations; |
60 | static const struct inode_operations btrfs_symlink_inode_operations; | |
61 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
62 | static const struct inode_operations btrfs_special_inode_operations; | |
63 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
64 | static const struct address_space_operations btrfs_aops; |
65 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 66 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 67 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
68 | |
69 | static struct kmem_cache *btrfs_inode_cachep; | |
70 | struct kmem_cache *btrfs_trans_handle_cachep; | |
71 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
72 | struct kmem_cache *btrfs_path_cachep; |
73 | ||
74 | #define S_SHIFT 12 | |
75 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
76 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
77 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
78 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
79 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
80 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
81 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
82 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
83 | }; | |
84 | ||
7b128766 | 85 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 86 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
87 | static noinline int cow_file_range(struct inode *inode, |
88 | struct page *locked_page, | |
89 | u64 start, u64 end, int *page_started, | |
90 | unsigned long *nr_written, int unlock); | |
7b128766 | 91 | |
f34f57a3 YZ |
92 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
93 | struct inode *inode, struct inode *dir) | |
0279b4cd JO |
94 | { |
95 | int err; | |
96 | ||
f34f57a3 | 97 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 98 | if (!err) |
f34f57a3 | 99 | err = btrfs_xattr_security_init(trans, inode, dir); |
0279b4cd JO |
100 | return err; |
101 | } | |
102 | ||
c8b97818 CM |
103 | /* |
104 | * this does all the hard work for inserting an inline extent into | |
105 | * the btree. The caller should have done a btrfs_drop_extents so that | |
106 | * no overlapping inline items exist in the btree | |
107 | */ | |
d397712b | 108 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
109 | struct btrfs_root *root, struct inode *inode, |
110 | u64 start, size_t size, size_t compressed_size, | |
111 | struct page **compressed_pages) | |
112 | { | |
113 | struct btrfs_key key; | |
114 | struct btrfs_path *path; | |
115 | struct extent_buffer *leaf; | |
116 | struct page *page = NULL; | |
117 | char *kaddr; | |
118 | unsigned long ptr; | |
119 | struct btrfs_file_extent_item *ei; | |
120 | int err = 0; | |
121 | int ret; | |
122 | size_t cur_size = size; | |
123 | size_t datasize; | |
124 | unsigned long offset; | |
125 | int use_compress = 0; | |
126 | ||
127 | if (compressed_size && compressed_pages) { | |
128 | use_compress = 1; | |
129 | cur_size = compressed_size; | |
130 | } | |
131 | ||
d397712b CM |
132 | path = btrfs_alloc_path(); |
133 | if (!path) | |
c8b97818 CM |
134 | return -ENOMEM; |
135 | ||
b9473439 | 136 | path->leave_spinning = 1; |
c8b97818 CM |
137 | btrfs_set_trans_block_group(trans, inode); |
138 | ||
139 | key.objectid = inode->i_ino; | |
140 | key.offset = start; | |
141 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
142 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
143 | ||
144 | inode_add_bytes(inode, size); | |
145 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
146 | datasize); | |
147 | BUG_ON(ret); | |
148 | if (ret) { | |
149 | err = ret; | |
c8b97818 CM |
150 | goto fail; |
151 | } | |
152 | leaf = path->nodes[0]; | |
153 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
154 | struct btrfs_file_extent_item); | |
155 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
156 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
157 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
158 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
159 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
160 | ptr = btrfs_file_extent_inline_start(ei); | |
161 | ||
162 | if (use_compress) { | |
163 | struct page *cpage; | |
164 | int i = 0; | |
d397712b | 165 | while (compressed_size > 0) { |
c8b97818 | 166 | cpage = compressed_pages[i]; |
5b050f04 | 167 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
168 | PAGE_CACHE_SIZE); |
169 | ||
b9473439 | 170 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 171 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 172 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
173 | |
174 | i++; | |
175 | ptr += cur_size; | |
176 | compressed_size -= cur_size; | |
177 | } | |
178 | btrfs_set_file_extent_compression(leaf, ei, | |
179 | BTRFS_COMPRESS_ZLIB); | |
180 | } else { | |
181 | page = find_get_page(inode->i_mapping, | |
182 | start >> PAGE_CACHE_SHIFT); | |
183 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
184 | kaddr = kmap_atomic(page, KM_USER0); | |
185 | offset = start & (PAGE_CACHE_SIZE - 1); | |
186 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
187 | kunmap_atomic(kaddr, KM_USER0); | |
188 | page_cache_release(page); | |
189 | } | |
190 | btrfs_mark_buffer_dirty(leaf); | |
191 | btrfs_free_path(path); | |
192 | ||
c2167754 YZ |
193 | /* |
194 | * we're an inline extent, so nobody can | |
195 | * extend the file past i_size without locking | |
196 | * a page we already have locked. | |
197 | * | |
198 | * We must do any isize and inode updates | |
199 | * before we unlock the pages. Otherwise we | |
200 | * could end up racing with unlink. | |
201 | */ | |
c8b97818 CM |
202 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
203 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 204 | |
c8b97818 CM |
205 | return 0; |
206 | fail: | |
207 | btrfs_free_path(path); | |
208 | return err; | |
209 | } | |
210 | ||
211 | ||
212 | /* | |
213 | * conditionally insert an inline extent into the file. This | |
214 | * does the checks required to make sure the data is small enough | |
215 | * to fit as an inline extent. | |
216 | */ | |
7f366cfe | 217 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
218 | struct btrfs_root *root, |
219 | struct inode *inode, u64 start, u64 end, | |
220 | size_t compressed_size, | |
221 | struct page **compressed_pages) | |
222 | { | |
223 | u64 isize = i_size_read(inode); | |
224 | u64 actual_end = min(end + 1, isize); | |
225 | u64 inline_len = actual_end - start; | |
226 | u64 aligned_end = (end + root->sectorsize - 1) & | |
227 | ~((u64)root->sectorsize - 1); | |
228 | u64 hint_byte; | |
229 | u64 data_len = inline_len; | |
230 | int ret; | |
231 | ||
232 | if (compressed_size) | |
233 | data_len = compressed_size; | |
234 | ||
235 | if (start > 0 || | |
70b99e69 | 236 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
237 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
238 | (!compressed_size && | |
239 | (actual_end & (root->sectorsize - 1)) == 0) || | |
240 | end + 1 < isize || | |
241 | data_len > root->fs_info->max_inline) { | |
242 | return 1; | |
243 | } | |
244 | ||
920bbbfb | 245 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 246 | &hint_byte, 1); |
c8b97818 CM |
247 | BUG_ON(ret); |
248 | ||
249 | if (isize > actual_end) | |
250 | inline_len = min_t(u64, isize, actual_end); | |
251 | ret = insert_inline_extent(trans, root, inode, start, | |
252 | inline_len, compressed_size, | |
253 | compressed_pages); | |
254 | BUG_ON(ret); | |
0ca1f7ce | 255 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 256 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
257 | return 0; |
258 | } | |
259 | ||
771ed689 CM |
260 | struct async_extent { |
261 | u64 start; | |
262 | u64 ram_size; | |
263 | u64 compressed_size; | |
264 | struct page **pages; | |
265 | unsigned long nr_pages; | |
266 | struct list_head list; | |
267 | }; | |
268 | ||
269 | struct async_cow { | |
270 | struct inode *inode; | |
271 | struct btrfs_root *root; | |
272 | struct page *locked_page; | |
273 | u64 start; | |
274 | u64 end; | |
275 | struct list_head extents; | |
276 | struct btrfs_work work; | |
277 | }; | |
278 | ||
279 | static noinline int add_async_extent(struct async_cow *cow, | |
280 | u64 start, u64 ram_size, | |
281 | u64 compressed_size, | |
282 | struct page **pages, | |
283 | unsigned long nr_pages) | |
284 | { | |
285 | struct async_extent *async_extent; | |
286 | ||
287 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
288 | async_extent->start = start; | |
289 | async_extent->ram_size = ram_size; | |
290 | async_extent->compressed_size = compressed_size; | |
291 | async_extent->pages = pages; | |
292 | async_extent->nr_pages = nr_pages; | |
293 | list_add_tail(&async_extent->list, &cow->extents); | |
294 | return 0; | |
295 | } | |
296 | ||
d352ac68 | 297 | /* |
771ed689 CM |
298 | * we create compressed extents in two phases. The first |
299 | * phase compresses a range of pages that have already been | |
300 | * locked (both pages and state bits are locked). | |
c8b97818 | 301 | * |
771ed689 CM |
302 | * This is done inside an ordered work queue, and the compression |
303 | * is spread across many cpus. The actual IO submission is step | |
304 | * two, and the ordered work queue takes care of making sure that | |
305 | * happens in the same order things were put onto the queue by | |
306 | * writepages and friends. | |
c8b97818 | 307 | * |
771ed689 CM |
308 | * If this code finds it can't get good compression, it puts an |
309 | * entry onto the work queue to write the uncompressed bytes. This | |
310 | * makes sure that both compressed inodes and uncompressed inodes | |
311 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 312 | */ |
771ed689 CM |
313 | static noinline int compress_file_range(struct inode *inode, |
314 | struct page *locked_page, | |
315 | u64 start, u64 end, | |
316 | struct async_cow *async_cow, | |
317 | int *num_added) | |
b888db2b CM |
318 | { |
319 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
320 | struct btrfs_trans_handle *trans; | |
db94535d | 321 | u64 num_bytes; |
c8b97818 CM |
322 | u64 orig_start; |
323 | u64 disk_num_bytes; | |
db94535d | 324 | u64 blocksize = root->sectorsize; |
c8b97818 | 325 | u64 actual_end; |
42dc7bab | 326 | u64 isize = i_size_read(inode); |
e6dcd2dc | 327 | int ret = 0; |
c8b97818 CM |
328 | struct page **pages = NULL; |
329 | unsigned long nr_pages; | |
330 | unsigned long nr_pages_ret = 0; | |
331 | unsigned long total_compressed = 0; | |
332 | unsigned long total_in = 0; | |
333 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 334 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
335 | int i; |
336 | int will_compress; | |
b888db2b | 337 | |
c8b97818 CM |
338 | orig_start = start; |
339 | ||
42dc7bab | 340 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
341 | again: |
342 | will_compress = 0; | |
343 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
344 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 345 | |
f03d9301 CM |
346 | /* |
347 | * we don't want to send crud past the end of i_size through | |
348 | * compression, that's just a waste of CPU time. So, if the | |
349 | * end of the file is before the start of our current | |
350 | * requested range of bytes, we bail out to the uncompressed | |
351 | * cleanup code that can deal with all of this. | |
352 | * | |
353 | * It isn't really the fastest way to fix things, but this is a | |
354 | * very uncommon corner. | |
355 | */ | |
356 | if (actual_end <= start) | |
357 | goto cleanup_and_bail_uncompressed; | |
358 | ||
c8b97818 CM |
359 | total_compressed = actual_end - start; |
360 | ||
361 | /* we want to make sure that amount of ram required to uncompress | |
362 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
363 | * of a compressed extent to 128k. This is a crucial number |
364 | * because it also controls how easily we can spread reads across | |
365 | * cpus for decompression. | |
366 | * | |
367 | * We also want to make sure the amount of IO required to do | |
368 | * a random read is reasonably small, so we limit the size of | |
369 | * a compressed extent to 128k. | |
c8b97818 CM |
370 | */ |
371 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 372 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 373 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
374 | disk_num_bytes = num_bytes; |
375 | total_in = 0; | |
376 | ret = 0; | |
db94535d | 377 | |
771ed689 CM |
378 | /* |
379 | * we do compression for mount -o compress and when the | |
380 | * inode has not been flagged as nocompress. This flag can | |
381 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 382 | */ |
6cbff00f | 383 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 CM |
384 | (btrfs_test_opt(root, COMPRESS) || |
385 | (BTRFS_I(inode)->force_compress))) { | |
c8b97818 | 386 | WARN_ON(pages); |
cfbc246e | 387 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 388 | |
c8b97818 CM |
389 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
390 | total_compressed, pages, | |
391 | nr_pages, &nr_pages_ret, | |
392 | &total_in, | |
393 | &total_compressed, | |
394 | max_compressed); | |
395 | ||
396 | if (!ret) { | |
397 | unsigned long offset = total_compressed & | |
398 | (PAGE_CACHE_SIZE - 1); | |
399 | struct page *page = pages[nr_pages_ret - 1]; | |
400 | char *kaddr; | |
401 | ||
402 | /* zero the tail end of the last page, we might be | |
403 | * sending it down to disk | |
404 | */ | |
405 | if (offset) { | |
406 | kaddr = kmap_atomic(page, KM_USER0); | |
407 | memset(kaddr + offset, 0, | |
408 | PAGE_CACHE_SIZE - offset); | |
409 | kunmap_atomic(kaddr, KM_USER0); | |
410 | } | |
411 | will_compress = 1; | |
412 | } | |
413 | } | |
414 | if (start == 0) { | |
771ed689 CM |
415 | trans = btrfs_join_transaction(root, 1); |
416 | BUG_ON(!trans); | |
417 | btrfs_set_trans_block_group(trans, inode); | |
0ca1f7ce | 418 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 419 | |
c8b97818 | 420 | /* lets try to make an inline extent */ |
771ed689 | 421 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 422 | /* we didn't compress the entire range, try |
771ed689 | 423 | * to make an uncompressed inline extent. |
c8b97818 CM |
424 | */ |
425 | ret = cow_file_range_inline(trans, root, inode, | |
426 | start, end, 0, NULL); | |
427 | } else { | |
771ed689 | 428 | /* try making a compressed inline extent */ |
c8b97818 CM |
429 | ret = cow_file_range_inline(trans, root, inode, |
430 | start, end, | |
431 | total_compressed, pages); | |
432 | } | |
433 | if (ret == 0) { | |
771ed689 CM |
434 | /* |
435 | * inline extent creation worked, we don't need | |
436 | * to create any more async work items. Unlock | |
437 | * and free up our temp pages. | |
438 | */ | |
c8b97818 | 439 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
440 | &BTRFS_I(inode)->io_tree, |
441 | start, end, NULL, | |
442 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 443 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 444 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
445 | |
446 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
447 | goto free_pages_out; |
448 | } | |
c2167754 | 449 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
450 | } |
451 | ||
452 | if (will_compress) { | |
453 | /* | |
454 | * we aren't doing an inline extent round the compressed size | |
455 | * up to a block size boundary so the allocator does sane | |
456 | * things | |
457 | */ | |
458 | total_compressed = (total_compressed + blocksize - 1) & | |
459 | ~(blocksize - 1); | |
460 | ||
461 | /* | |
462 | * one last check to make sure the compression is really a | |
463 | * win, compare the page count read with the blocks on disk | |
464 | */ | |
465 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
466 | ~(PAGE_CACHE_SIZE - 1); | |
467 | if (total_compressed >= total_in) { | |
468 | will_compress = 0; | |
469 | } else { | |
470 | disk_num_bytes = total_compressed; | |
471 | num_bytes = total_in; | |
472 | } | |
473 | } | |
474 | if (!will_compress && pages) { | |
475 | /* | |
476 | * the compression code ran but failed to make things smaller, | |
477 | * free any pages it allocated and our page pointer array | |
478 | */ | |
479 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 480 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
481 | page_cache_release(pages[i]); |
482 | } | |
483 | kfree(pages); | |
484 | pages = NULL; | |
485 | total_compressed = 0; | |
486 | nr_pages_ret = 0; | |
487 | ||
488 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
489 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
490 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 491 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 492 | } |
c8b97818 | 493 | } |
771ed689 CM |
494 | if (will_compress) { |
495 | *num_added += 1; | |
c8b97818 | 496 | |
771ed689 CM |
497 | /* the async work queues will take care of doing actual |
498 | * allocation on disk for these compressed pages, | |
499 | * and will submit them to the elevator. | |
500 | */ | |
501 | add_async_extent(async_cow, start, num_bytes, | |
502 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 503 | |
42dc7bab | 504 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
505 | start += num_bytes; |
506 | pages = NULL; | |
507 | cond_resched(); | |
508 | goto again; | |
509 | } | |
510 | } else { | |
f03d9301 | 511 | cleanup_and_bail_uncompressed: |
771ed689 CM |
512 | /* |
513 | * No compression, but we still need to write the pages in | |
514 | * the file we've been given so far. redirty the locked | |
515 | * page if it corresponds to our extent and set things up | |
516 | * for the async work queue to run cow_file_range to do | |
517 | * the normal delalloc dance | |
518 | */ | |
519 | if (page_offset(locked_page) >= start && | |
520 | page_offset(locked_page) <= end) { | |
521 | __set_page_dirty_nobuffers(locked_page); | |
522 | /* unlocked later on in the async handlers */ | |
523 | } | |
524 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
525 | *num_added += 1; | |
526 | } | |
3b951516 | 527 | |
771ed689 CM |
528 | out: |
529 | return 0; | |
530 | ||
531 | free_pages_out: | |
532 | for (i = 0; i < nr_pages_ret; i++) { | |
533 | WARN_ON(pages[i]->mapping); | |
534 | page_cache_release(pages[i]); | |
535 | } | |
d397712b | 536 | kfree(pages); |
771ed689 CM |
537 | |
538 | goto out; | |
539 | } | |
540 | ||
541 | /* | |
542 | * phase two of compressed writeback. This is the ordered portion | |
543 | * of the code, which only gets called in the order the work was | |
544 | * queued. We walk all the async extents created by compress_file_range | |
545 | * and send them down to the disk. | |
546 | */ | |
547 | static noinline int submit_compressed_extents(struct inode *inode, | |
548 | struct async_cow *async_cow) | |
549 | { | |
550 | struct async_extent *async_extent; | |
551 | u64 alloc_hint = 0; | |
552 | struct btrfs_trans_handle *trans; | |
553 | struct btrfs_key ins; | |
554 | struct extent_map *em; | |
555 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
556 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
557 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 558 | int ret = 0; |
771ed689 CM |
559 | |
560 | if (list_empty(&async_cow->extents)) | |
561 | return 0; | |
562 | ||
771ed689 | 563 | |
d397712b | 564 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
565 | async_extent = list_entry(async_cow->extents.next, |
566 | struct async_extent, list); | |
567 | list_del(&async_extent->list); | |
c8b97818 | 568 | |
771ed689 CM |
569 | io_tree = &BTRFS_I(inode)->io_tree; |
570 | ||
f5a84ee3 | 571 | retry: |
771ed689 CM |
572 | /* did the compression code fall back to uncompressed IO? */ |
573 | if (!async_extent->pages) { | |
574 | int page_started = 0; | |
575 | unsigned long nr_written = 0; | |
576 | ||
577 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 JB |
578 | async_extent->start + |
579 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
580 | |
581 | /* allocate blocks */ | |
f5a84ee3 JB |
582 | ret = cow_file_range(inode, async_cow->locked_page, |
583 | async_extent->start, | |
584 | async_extent->start + | |
585 | async_extent->ram_size - 1, | |
586 | &page_started, &nr_written, 0); | |
771ed689 CM |
587 | |
588 | /* | |
589 | * if page_started, cow_file_range inserted an | |
590 | * inline extent and took care of all the unlocking | |
591 | * and IO for us. Otherwise, we need to submit | |
592 | * all those pages down to the drive. | |
593 | */ | |
f5a84ee3 | 594 | if (!page_started && !ret) |
771ed689 CM |
595 | extent_write_locked_range(io_tree, |
596 | inode, async_extent->start, | |
d397712b | 597 | async_extent->start + |
771ed689 CM |
598 | async_extent->ram_size - 1, |
599 | btrfs_get_extent, | |
600 | WB_SYNC_ALL); | |
601 | kfree(async_extent); | |
602 | cond_resched(); | |
603 | continue; | |
604 | } | |
605 | ||
606 | lock_extent(io_tree, async_extent->start, | |
607 | async_extent->start + async_extent->ram_size - 1, | |
608 | GFP_NOFS); | |
771ed689 | 609 | |
c2167754 | 610 | trans = btrfs_join_transaction(root, 1); |
771ed689 CM |
611 | ret = btrfs_reserve_extent(trans, root, |
612 | async_extent->compressed_size, | |
613 | async_extent->compressed_size, | |
614 | 0, alloc_hint, | |
615 | (u64)-1, &ins, 1); | |
c2167754 YZ |
616 | btrfs_end_transaction(trans, root); |
617 | ||
f5a84ee3 JB |
618 | if (ret) { |
619 | int i; | |
620 | for (i = 0; i < async_extent->nr_pages; i++) { | |
621 | WARN_ON(async_extent->pages[i]->mapping); | |
622 | page_cache_release(async_extent->pages[i]); | |
623 | } | |
624 | kfree(async_extent->pages); | |
625 | async_extent->nr_pages = 0; | |
626 | async_extent->pages = NULL; | |
627 | unlock_extent(io_tree, async_extent->start, | |
628 | async_extent->start + | |
629 | async_extent->ram_size - 1, GFP_NOFS); | |
630 | goto retry; | |
631 | } | |
632 | ||
c2167754 YZ |
633 | /* |
634 | * here we're doing allocation and writeback of the | |
635 | * compressed pages | |
636 | */ | |
637 | btrfs_drop_extent_cache(inode, async_extent->start, | |
638 | async_extent->start + | |
639 | async_extent->ram_size - 1, 0); | |
640 | ||
771ed689 CM |
641 | em = alloc_extent_map(GFP_NOFS); |
642 | em->start = async_extent->start; | |
643 | em->len = async_extent->ram_size; | |
445a6944 | 644 | em->orig_start = em->start; |
c8b97818 | 645 | |
771ed689 CM |
646 | em->block_start = ins.objectid; |
647 | em->block_len = ins.offset; | |
648 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
649 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
650 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
651 | ||
d397712b | 652 | while (1) { |
890871be | 653 | write_lock(&em_tree->lock); |
771ed689 | 654 | ret = add_extent_mapping(em_tree, em); |
890871be | 655 | write_unlock(&em_tree->lock); |
771ed689 CM |
656 | if (ret != -EEXIST) { |
657 | free_extent_map(em); | |
658 | break; | |
659 | } | |
660 | btrfs_drop_extent_cache(inode, async_extent->start, | |
661 | async_extent->start + | |
662 | async_extent->ram_size - 1, 0); | |
663 | } | |
664 | ||
665 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
666 | ins.objectid, | |
667 | async_extent->ram_size, | |
668 | ins.offset, | |
669 | BTRFS_ORDERED_COMPRESSED); | |
670 | BUG_ON(ret); | |
671 | ||
771ed689 CM |
672 | /* |
673 | * clear dirty, set writeback and unlock the pages. | |
674 | */ | |
675 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
676 | &BTRFS_I(inode)->io_tree, |
677 | async_extent->start, | |
678 | async_extent->start + | |
679 | async_extent->ram_size - 1, | |
680 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
681 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 682 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 683 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
684 | |
685 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
686 | async_extent->start, |
687 | async_extent->ram_size, | |
688 | ins.objectid, | |
689 | ins.offset, async_extent->pages, | |
690 | async_extent->nr_pages); | |
771ed689 CM |
691 | |
692 | BUG_ON(ret); | |
771ed689 CM |
693 | alloc_hint = ins.objectid + ins.offset; |
694 | kfree(async_extent); | |
695 | cond_resched(); | |
696 | } | |
697 | ||
771ed689 CM |
698 | return 0; |
699 | } | |
700 | ||
701 | /* | |
702 | * when extent_io.c finds a delayed allocation range in the file, | |
703 | * the call backs end up in this code. The basic idea is to | |
704 | * allocate extents on disk for the range, and create ordered data structs | |
705 | * in ram to track those extents. | |
706 | * | |
707 | * locked_page is the page that writepage had locked already. We use | |
708 | * it to make sure we don't do extra locks or unlocks. | |
709 | * | |
710 | * *page_started is set to one if we unlock locked_page and do everything | |
711 | * required to start IO on it. It may be clean and already done with | |
712 | * IO when we return. | |
713 | */ | |
714 | static noinline int cow_file_range(struct inode *inode, | |
715 | struct page *locked_page, | |
716 | u64 start, u64 end, int *page_started, | |
717 | unsigned long *nr_written, | |
718 | int unlock) | |
719 | { | |
720 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
721 | struct btrfs_trans_handle *trans; | |
722 | u64 alloc_hint = 0; | |
723 | u64 num_bytes; | |
724 | unsigned long ram_size; | |
725 | u64 disk_num_bytes; | |
726 | u64 cur_alloc_size; | |
727 | u64 blocksize = root->sectorsize; | |
728 | u64 actual_end; | |
42dc7bab | 729 | u64 isize = i_size_read(inode); |
771ed689 CM |
730 | struct btrfs_key ins; |
731 | struct extent_map *em; | |
732 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
733 | int ret = 0; | |
734 | ||
735 | trans = btrfs_join_transaction(root, 1); | |
736 | BUG_ON(!trans); | |
737 | btrfs_set_trans_block_group(trans, inode); | |
0ca1f7ce | 738 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 739 | |
42dc7bab | 740 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
741 | |
742 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
743 | num_bytes = max(blocksize, num_bytes); | |
744 | disk_num_bytes = num_bytes; | |
745 | ret = 0; | |
746 | ||
747 | if (start == 0) { | |
748 | /* lets try to make an inline extent */ | |
749 | ret = cow_file_range_inline(trans, root, inode, | |
750 | start, end, 0, NULL); | |
751 | if (ret == 0) { | |
752 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
753 | &BTRFS_I(inode)->io_tree, |
754 | start, end, NULL, | |
755 | EXTENT_CLEAR_UNLOCK_PAGE | | |
756 | EXTENT_CLEAR_UNLOCK | | |
757 | EXTENT_CLEAR_DELALLOC | | |
758 | EXTENT_CLEAR_DIRTY | | |
759 | EXTENT_SET_WRITEBACK | | |
760 | EXTENT_END_WRITEBACK); | |
c2167754 | 761 | |
771ed689 CM |
762 | *nr_written = *nr_written + |
763 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
764 | *page_started = 1; | |
765 | ret = 0; | |
766 | goto out; | |
767 | } | |
768 | } | |
769 | ||
770 | BUG_ON(disk_num_bytes > | |
771 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
772 | ||
b917b7c3 CM |
773 | |
774 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
775 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
776 | start, num_bytes); | |
777 | if (em) { | |
6346c939 JB |
778 | /* |
779 | * if block start isn't an actual block number then find the | |
780 | * first block in this inode and use that as a hint. If that | |
781 | * block is also bogus then just don't worry about it. | |
782 | */ | |
783 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
784 | free_extent_map(em); | |
785 | em = search_extent_mapping(em_tree, 0, 0); | |
786 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
787 | alloc_hint = em->block_start; | |
788 | if (em) | |
789 | free_extent_map(em); | |
790 | } else { | |
791 | alloc_hint = em->block_start; | |
792 | free_extent_map(em); | |
793 | } | |
b917b7c3 CM |
794 | } |
795 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
796 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
797 | ||
d397712b | 798 | while (disk_num_bytes > 0) { |
a791e35e CM |
799 | unsigned long op; |
800 | ||
287a0ab9 | 801 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 802 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 803 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 804 | (u64)-1, &ins, 1); |
d397712b CM |
805 | BUG_ON(ret); |
806 | ||
e6dcd2dc CM |
807 | em = alloc_extent_map(GFP_NOFS); |
808 | em->start = start; | |
445a6944 | 809 | em->orig_start = em->start; |
771ed689 CM |
810 | ram_size = ins.offset; |
811 | em->len = ins.offset; | |
c8b97818 | 812 | |
e6dcd2dc | 813 | em->block_start = ins.objectid; |
c8b97818 | 814 | em->block_len = ins.offset; |
e6dcd2dc | 815 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 816 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 817 | |
d397712b | 818 | while (1) { |
890871be | 819 | write_lock(&em_tree->lock); |
e6dcd2dc | 820 | ret = add_extent_mapping(em_tree, em); |
890871be | 821 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
822 | if (ret != -EEXIST) { |
823 | free_extent_map(em); | |
824 | break; | |
825 | } | |
826 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 827 | start + ram_size - 1, 0); |
e6dcd2dc CM |
828 | } |
829 | ||
98d20f67 | 830 | cur_alloc_size = ins.offset; |
e6dcd2dc | 831 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 832 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 833 | BUG_ON(ret); |
c8b97818 | 834 | |
17d217fe YZ |
835 | if (root->root_key.objectid == |
836 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
837 | ret = btrfs_reloc_clone_csums(inode, start, | |
838 | cur_alloc_size); | |
839 | BUG_ON(ret); | |
840 | } | |
841 | ||
d397712b | 842 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 843 | break; |
d397712b | 844 | |
c8b97818 CM |
845 | /* we're not doing compressed IO, don't unlock the first |
846 | * page (which the caller expects to stay locked), don't | |
847 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
848 | * |
849 | * Do set the Private2 bit so we know this page was properly | |
850 | * setup for writepage | |
c8b97818 | 851 | */ |
a791e35e CM |
852 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
853 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
854 | EXTENT_SET_PRIVATE2; | |
855 | ||
c8b97818 CM |
856 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
857 | start, start + ram_size - 1, | |
a791e35e | 858 | locked_page, op); |
c8b97818 | 859 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
860 | num_bytes -= cur_alloc_size; |
861 | alloc_hint = ins.objectid + ins.offset; | |
862 | start += cur_alloc_size; | |
b888db2b | 863 | } |
b888db2b | 864 | out: |
771ed689 | 865 | ret = 0; |
b888db2b | 866 | btrfs_end_transaction(trans, root); |
c8b97818 | 867 | |
be20aa9d | 868 | return ret; |
771ed689 | 869 | } |
c8b97818 | 870 | |
771ed689 CM |
871 | /* |
872 | * work queue call back to started compression on a file and pages | |
873 | */ | |
874 | static noinline void async_cow_start(struct btrfs_work *work) | |
875 | { | |
876 | struct async_cow *async_cow; | |
877 | int num_added = 0; | |
878 | async_cow = container_of(work, struct async_cow, work); | |
879 | ||
880 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
881 | async_cow->start, async_cow->end, async_cow, | |
882 | &num_added); | |
883 | if (num_added == 0) | |
884 | async_cow->inode = NULL; | |
885 | } | |
886 | ||
887 | /* | |
888 | * work queue call back to submit previously compressed pages | |
889 | */ | |
890 | static noinline void async_cow_submit(struct btrfs_work *work) | |
891 | { | |
892 | struct async_cow *async_cow; | |
893 | struct btrfs_root *root; | |
894 | unsigned long nr_pages; | |
895 | ||
896 | async_cow = container_of(work, struct async_cow, work); | |
897 | ||
898 | root = async_cow->root; | |
899 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
900 | PAGE_CACHE_SHIFT; | |
901 | ||
902 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
903 | ||
904 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
905 | 5 * 1042 * 1024 && | |
906 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
907 | wake_up(&root->fs_info->async_submit_wait); | |
908 | ||
d397712b | 909 | if (async_cow->inode) |
771ed689 | 910 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 911 | } |
c8b97818 | 912 | |
771ed689 CM |
913 | static noinline void async_cow_free(struct btrfs_work *work) |
914 | { | |
915 | struct async_cow *async_cow; | |
916 | async_cow = container_of(work, struct async_cow, work); | |
917 | kfree(async_cow); | |
918 | } | |
919 | ||
920 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
921 | u64 start, u64 end, int *page_started, | |
922 | unsigned long *nr_written) | |
923 | { | |
924 | struct async_cow *async_cow; | |
925 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
926 | unsigned long nr_pages; | |
927 | u64 cur_end; | |
928 | int limit = 10 * 1024 * 1042; | |
929 | ||
a3429ab7 CM |
930 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
931 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 932 | while (start < end) { |
771ed689 CM |
933 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
934 | async_cow->inode = inode; | |
935 | async_cow->root = root; | |
936 | async_cow->locked_page = locked_page; | |
937 | async_cow->start = start; | |
938 | ||
6cbff00f | 939 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
940 | cur_end = end; |
941 | else | |
942 | cur_end = min(end, start + 512 * 1024 - 1); | |
943 | ||
944 | async_cow->end = cur_end; | |
945 | INIT_LIST_HEAD(&async_cow->extents); | |
946 | ||
947 | async_cow->work.func = async_cow_start; | |
948 | async_cow->work.ordered_func = async_cow_submit; | |
949 | async_cow->work.ordered_free = async_cow_free; | |
950 | async_cow->work.flags = 0; | |
951 | ||
771ed689 CM |
952 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
953 | PAGE_CACHE_SHIFT; | |
954 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
955 | ||
956 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
957 | &async_cow->work); | |
958 | ||
959 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
960 | wait_event(root->fs_info->async_submit_wait, | |
961 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
962 | limit)); | |
963 | } | |
964 | ||
d397712b | 965 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
966 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
967 | wait_event(root->fs_info->async_submit_wait, | |
968 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
969 | 0)); | |
970 | } | |
971 | ||
972 | *nr_written += nr_pages; | |
973 | start = cur_end + 1; | |
974 | } | |
975 | *page_started = 1; | |
976 | return 0; | |
be20aa9d CM |
977 | } |
978 | ||
d397712b | 979 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
980 | u64 bytenr, u64 num_bytes) |
981 | { | |
982 | int ret; | |
983 | struct btrfs_ordered_sum *sums; | |
984 | LIST_HEAD(list); | |
985 | ||
07d400a6 YZ |
986 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
987 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
988 | if (ret == 0 && list_empty(&list)) |
989 | return 0; | |
990 | ||
991 | while (!list_empty(&list)) { | |
992 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
993 | list_del(&sums->list); | |
994 | kfree(sums); | |
995 | } | |
996 | return 1; | |
997 | } | |
998 | ||
d352ac68 CM |
999 | /* |
1000 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1001 | * of the extents that exist in the file, and COWs the file as required. | |
1002 | * | |
1003 | * If no cow copies or snapshots exist, we write directly to the existing | |
1004 | * blocks on disk | |
1005 | */ | |
7f366cfe CM |
1006 | static noinline int run_delalloc_nocow(struct inode *inode, |
1007 | struct page *locked_page, | |
771ed689 CM |
1008 | u64 start, u64 end, int *page_started, int force, |
1009 | unsigned long *nr_written) | |
be20aa9d | 1010 | { |
be20aa9d | 1011 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1012 | struct btrfs_trans_handle *trans; |
be20aa9d | 1013 | struct extent_buffer *leaf; |
be20aa9d | 1014 | struct btrfs_path *path; |
80ff3856 | 1015 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1016 | struct btrfs_key found_key; |
80ff3856 YZ |
1017 | u64 cow_start; |
1018 | u64 cur_offset; | |
1019 | u64 extent_end; | |
5d4f98a2 | 1020 | u64 extent_offset; |
80ff3856 YZ |
1021 | u64 disk_bytenr; |
1022 | u64 num_bytes; | |
1023 | int extent_type; | |
1024 | int ret; | |
d899e052 | 1025 | int type; |
80ff3856 YZ |
1026 | int nocow; |
1027 | int check_prev = 1; | |
be20aa9d CM |
1028 | |
1029 | path = btrfs_alloc_path(); | |
1030 | BUG_ON(!path); | |
7ea394f1 YZ |
1031 | trans = btrfs_join_transaction(root, 1); |
1032 | BUG_ON(!trans); | |
be20aa9d | 1033 | |
80ff3856 YZ |
1034 | cow_start = (u64)-1; |
1035 | cur_offset = start; | |
1036 | while (1) { | |
1037 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1038 | cur_offset, 0); | |
1039 | BUG_ON(ret < 0); | |
1040 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1041 | leaf = path->nodes[0]; | |
1042 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1043 | path->slots[0] - 1); | |
1044 | if (found_key.objectid == inode->i_ino && | |
1045 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1046 | path->slots[0]--; | |
1047 | } | |
1048 | check_prev = 0; | |
1049 | next_slot: | |
1050 | leaf = path->nodes[0]; | |
1051 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1052 | ret = btrfs_next_leaf(root, path); | |
1053 | if (ret < 0) | |
1054 | BUG_ON(1); | |
1055 | if (ret > 0) | |
1056 | break; | |
1057 | leaf = path->nodes[0]; | |
1058 | } | |
be20aa9d | 1059 | |
80ff3856 YZ |
1060 | nocow = 0; |
1061 | disk_bytenr = 0; | |
17d217fe | 1062 | num_bytes = 0; |
80ff3856 YZ |
1063 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1064 | ||
1065 | if (found_key.objectid > inode->i_ino || | |
1066 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1067 | found_key.offset > end) | |
1068 | break; | |
1069 | ||
1070 | if (found_key.offset > cur_offset) { | |
1071 | extent_end = found_key.offset; | |
e9061e21 | 1072 | extent_type = 0; |
80ff3856 YZ |
1073 | goto out_check; |
1074 | } | |
1075 | ||
1076 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1077 | struct btrfs_file_extent_item); | |
1078 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1079 | ||
d899e052 YZ |
1080 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1081 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1082 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1083 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1084 | extent_end = found_key.offset + |
1085 | btrfs_file_extent_num_bytes(leaf, fi); | |
1086 | if (extent_end <= start) { | |
1087 | path->slots[0]++; | |
1088 | goto next_slot; | |
1089 | } | |
17d217fe YZ |
1090 | if (disk_bytenr == 0) |
1091 | goto out_check; | |
80ff3856 YZ |
1092 | if (btrfs_file_extent_compression(leaf, fi) || |
1093 | btrfs_file_extent_encryption(leaf, fi) || | |
1094 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1095 | goto out_check; | |
d899e052 YZ |
1096 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1097 | goto out_check; | |
d2fb3437 | 1098 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1099 | goto out_check; |
17d217fe | 1100 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1101 | found_key.offset - |
1102 | extent_offset, disk_bytenr)) | |
17d217fe | 1103 | goto out_check; |
5d4f98a2 | 1104 | disk_bytenr += extent_offset; |
17d217fe YZ |
1105 | disk_bytenr += cur_offset - found_key.offset; |
1106 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1107 | /* | |
1108 | * force cow if csum exists in the range. | |
1109 | * this ensure that csum for a given extent are | |
1110 | * either valid or do not exist. | |
1111 | */ | |
1112 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1113 | goto out_check; | |
80ff3856 YZ |
1114 | nocow = 1; |
1115 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1116 | extent_end = found_key.offset + | |
1117 | btrfs_file_extent_inline_len(leaf, fi); | |
1118 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1119 | } else { | |
1120 | BUG_ON(1); | |
1121 | } | |
1122 | out_check: | |
1123 | if (extent_end <= start) { | |
1124 | path->slots[0]++; | |
1125 | goto next_slot; | |
1126 | } | |
1127 | if (!nocow) { | |
1128 | if (cow_start == (u64)-1) | |
1129 | cow_start = cur_offset; | |
1130 | cur_offset = extent_end; | |
1131 | if (cur_offset > end) | |
1132 | break; | |
1133 | path->slots[0]++; | |
1134 | goto next_slot; | |
7ea394f1 YZ |
1135 | } |
1136 | ||
1137 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1138 | if (cow_start != (u64)-1) { |
1139 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1140 | found_key.offset - 1, page_started, |
1141 | nr_written, 1); | |
80ff3856 YZ |
1142 | BUG_ON(ret); |
1143 | cow_start = (u64)-1; | |
7ea394f1 | 1144 | } |
80ff3856 | 1145 | |
d899e052 YZ |
1146 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1147 | struct extent_map *em; | |
1148 | struct extent_map_tree *em_tree; | |
1149 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1150 | em = alloc_extent_map(GFP_NOFS); | |
1151 | em->start = cur_offset; | |
445a6944 | 1152 | em->orig_start = em->start; |
d899e052 YZ |
1153 | em->len = num_bytes; |
1154 | em->block_len = num_bytes; | |
1155 | em->block_start = disk_bytenr; | |
1156 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1157 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1158 | while (1) { | |
890871be | 1159 | write_lock(&em_tree->lock); |
d899e052 | 1160 | ret = add_extent_mapping(em_tree, em); |
890871be | 1161 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1162 | if (ret != -EEXIST) { |
1163 | free_extent_map(em); | |
1164 | break; | |
1165 | } | |
1166 | btrfs_drop_extent_cache(inode, em->start, | |
1167 | em->start + em->len - 1, 0); | |
1168 | } | |
1169 | type = BTRFS_ORDERED_PREALLOC; | |
1170 | } else { | |
1171 | type = BTRFS_ORDERED_NOCOW; | |
1172 | } | |
80ff3856 YZ |
1173 | |
1174 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1175 | num_bytes, num_bytes, type); |
1176 | BUG_ON(ret); | |
771ed689 | 1177 | |
d899e052 | 1178 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1179 | cur_offset, cur_offset + num_bytes - 1, |
1180 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1181 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1182 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1183 | cur_offset = extent_end; |
1184 | if (cur_offset > end) | |
1185 | break; | |
be20aa9d | 1186 | } |
80ff3856 YZ |
1187 | btrfs_release_path(root, path); |
1188 | ||
1189 | if (cur_offset <= end && cow_start == (u64)-1) | |
1190 | cow_start = cur_offset; | |
1191 | if (cow_start != (u64)-1) { | |
1192 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1193 | page_started, nr_written, 1); |
80ff3856 YZ |
1194 | BUG_ON(ret); |
1195 | } | |
1196 | ||
1197 | ret = btrfs_end_transaction(trans, root); | |
1198 | BUG_ON(ret); | |
7ea394f1 | 1199 | btrfs_free_path(path); |
80ff3856 | 1200 | return 0; |
be20aa9d CM |
1201 | } |
1202 | ||
d352ac68 CM |
1203 | /* |
1204 | * extent_io.c call back to do delayed allocation processing | |
1205 | */ | |
c8b97818 | 1206 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1207 | u64 start, u64 end, int *page_started, |
1208 | unsigned long *nr_written) | |
be20aa9d | 1209 | { |
be20aa9d | 1210 | int ret; |
7f366cfe | 1211 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1212 | |
6cbff00f | 1213 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1214 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1215 | page_started, 1, nr_written); |
6cbff00f | 1216 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1217 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1218 | page_started, 0, nr_written); |
1e701a32 CM |
1219 | else if (!btrfs_test_opt(root, COMPRESS) && |
1220 | !(BTRFS_I(inode)->force_compress)) | |
7f366cfe CM |
1221 | ret = cow_file_range(inode, locked_page, start, end, |
1222 | page_started, nr_written, 1); | |
be20aa9d | 1223 | else |
771ed689 | 1224 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1225 | page_started, nr_written); |
b888db2b CM |
1226 | return ret; |
1227 | } | |
1228 | ||
9ed74f2d | 1229 | static int btrfs_split_extent_hook(struct inode *inode, |
0ca1f7ce | 1230 | struct extent_state *orig, u64 split) |
9ed74f2d | 1231 | { |
0ca1f7ce | 1232 | /* not delalloc, ignore it */ |
9ed74f2d JB |
1233 | if (!(orig->state & EXTENT_DELALLOC)) |
1234 | return 0; | |
1235 | ||
0ca1f7ce | 1236 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1237 | return 0; |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1242 | * extents so we can keep track of new extents that are just merged onto old | |
1243 | * extents, such as when we are doing sequential writes, so we can properly | |
1244 | * account for the metadata space we'll need. | |
1245 | */ | |
1246 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1247 | struct extent_state *new, | |
1248 | struct extent_state *other) | |
1249 | { | |
9ed74f2d JB |
1250 | /* not delalloc, ignore it */ |
1251 | if (!(other->state & EXTENT_DELALLOC)) | |
1252 | return 0; | |
1253 | ||
0ca1f7ce | 1254 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1255 | return 0; |
1256 | } | |
1257 | ||
d352ac68 CM |
1258 | /* |
1259 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1260 | * bytes in this file, and to maintain the list of inodes that | |
1261 | * have pending delalloc work to be done. | |
1262 | */ | |
0ca1f7ce YZ |
1263 | static int btrfs_set_bit_hook(struct inode *inode, |
1264 | struct extent_state *state, int *bits) | |
291d673e | 1265 | { |
9ed74f2d | 1266 | |
75eff68e CM |
1267 | /* |
1268 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1269 | * but in this case, we are only testeing for the DELALLOC | |
1270 | * bit, which is only set or cleared with irqs on | |
1271 | */ | |
0ca1f7ce | 1272 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1273 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1274 | u64 len = state->end + 1 - state->start; |
9ed74f2d | 1275 | |
0ca1f7ce YZ |
1276 | if (*bits & EXTENT_FIRST_DELALLOC) |
1277 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1278 | else | |
1279 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); | |
287a0ab9 | 1280 | |
75eff68e | 1281 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1282 | BTRFS_I(inode)->delalloc_bytes += len; |
1283 | root->fs_info->delalloc_bytes += len; | |
ea8c2819 CM |
1284 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1285 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1286 | &root->fs_info->delalloc_inodes); | |
1287 | } | |
75eff68e | 1288 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1289 | } |
1290 | return 0; | |
1291 | } | |
1292 | ||
d352ac68 CM |
1293 | /* |
1294 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1295 | */ | |
9ed74f2d | 1296 | static int btrfs_clear_bit_hook(struct inode *inode, |
0ca1f7ce | 1297 | struct extent_state *state, int *bits) |
291d673e | 1298 | { |
75eff68e CM |
1299 | /* |
1300 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1301 | * but in this case, we are only testeing for the DELALLOC | |
1302 | * bit, which is only set or cleared with irqs on | |
1303 | */ | |
0ca1f7ce | 1304 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1305 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1306 | u64 len = state->end + 1 - state->start; |
bcbfce8a | 1307 | |
0ca1f7ce YZ |
1308 | if (*bits & EXTENT_FIRST_DELALLOC) |
1309 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1310 | else if (!(*bits & EXTENT_DO_ACCOUNTING)) | |
1311 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
1312 | ||
1313 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1314 | btrfs_delalloc_release_metadata(inode, len); | |
1315 | ||
1316 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1317 | btrfs_free_reserved_data_space(inode, len); | |
9ed74f2d | 1318 | |
75eff68e | 1319 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1320 | root->fs_info->delalloc_bytes -= len; |
1321 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1322 | ||
ea8c2819 CM |
1323 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1324 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1325 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1326 | } | |
75eff68e | 1327 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1328 | } |
1329 | return 0; | |
1330 | } | |
1331 | ||
d352ac68 CM |
1332 | /* |
1333 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1334 | * we don't create bios that span stripes or chunks | |
1335 | */ | |
239b14b3 | 1336 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1337 | size_t size, struct bio *bio, |
1338 | unsigned long bio_flags) | |
239b14b3 CM |
1339 | { |
1340 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1341 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1342 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1343 | u64 length = 0; |
1344 | u64 map_length; | |
239b14b3 CM |
1345 | int ret; |
1346 | ||
771ed689 CM |
1347 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1348 | return 0; | |
1349 | ||
f2d8d74d | 1350 | length = bio->bi_size; |
239b14b3 CM |
1351 | map_tree = &root->fs_info->mapping_tree; |
1352 | map_length = length; | |
cea9e445 | 1353 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1354 | &map_length, NULL, 0); |
cea9e445 | 1355 | |
d397712b | 1356 | if (map_length < length + size) |
239b14b3 | 1357 | return 1; |
239b14b3 CM |
1358 | return 0; |
1359 | } | |
1360 | ||
d352ac68 CM |
1361 | /* |
1362 | * in order to insert checksums into the metadata in large chunks, | |
1363 | * we wait until bio submission time. All the pages in the bio are | |
1364 | * checksummed and sums are attached onto the ordered extent record. | |
1365 | * | |
1366 | * At IO completion time the cums attached on the ordered extent record | |
1367 | * are inserted into the btree | |
1368 | */ | |
d397712b CM |
1369 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1370 | struct bio *bio, int mirror_num, | |
1371 | unsigned long bio_flags) | |
065631f6 | 1372 | { |
065631f6 | 1373 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1374 | int ret = 0; |
e015640f | 1375 | |
d20f7043 | 1376 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1377 | BUG_ON(ret); |
4a69a410 CM |
1378 | return 0; |
1379 | } | |
e015640f | 1380 | |
4a69a410 CM |
1381 | /* |
1382 | * in order to insert checksums into the metadata in large chunks, | |
1383 | * we wait until bio submission time. All the pages in the bio are | |
1384 | * checksummed and sums are attached onto the ordered extent record. | |
1385 | * | |
1386 | * At IO completion time the cums attached on the ordered extent record | |
1387 | * are inserted into the btree | |
1388 | */ | |
b2950863 | 1389 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1390 | int mirror_num, unsigned long bio_flags) |
1391 | { | |
1392 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1393 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1394 | } |
1395 | ||
d352ac68 | 1396 | /* |
cad321ad CM |
1397 | * extent_io.c submission hook. This does the right thing for csum calculation |
1398 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1399 | */ |
b2950863 | 1400 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1401 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1402 | { |
1403 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1404 | int ret = 0; | |
19b9bdb0 | 1405 | int skip_sum; |
44b8bd7e | 1406 | |
6cbff00f | 1407 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1408 | |
e6dcd2dc CM |
1409 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1410 | BUG_ON(ret); | |
065631f6 | 1411 | |
4d1b5fb4 | 1412 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1413 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1414 | return btrfs_submit_compressed_read(inode, bio, |
1415 | mirror_num, bio_flags); | |
d20f7043 CM |
1416 | } else if (!skip_sum) |
1417 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1418 | goto mapit; |
19b9bdb0 | 1419 | } else if (!skip_sum) { |
17d217fe YZ |
1420 | /* csum items have already been cloned */ |
1421 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1422 | goto mapit; | |
19b9bdb0 CM |
1423 | /* we're doing a write, do the async checksumming */ |
1424 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1425 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1426 | bio_flags, __btrfs_submit_bio_start, |
1427 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1428 | } |
1429 | ||
0b86a832 | 1430 | mapit: |
8b712842 | 1431 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1432 | } |
6885f308 | 1433 | |
d352ac68 CM |
1434 | /* |
1435 | * given a list of ordered sums record them in the inode. This happens | |
1436 | * at IO completion time based on sums calculated at bio submission time. | |
1437 | */ | |
ba1da2f4 | 1438 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1439 | struct inode *inode, u64 file_offset, |
1440 | struct list_head *list) | |
1441 | { | |
e6dcd2dc CM |
1442 | struct btrfs_ordered_sum *sum; |
1443 | ||
1444 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1445 | |
1446 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1447 | btrfs_csum_file_blocks(trans, |
1448 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1449 | } |
1450 | return 0; | |
1451 | } | |
1452 | ||
2ac55d41 JB |
1453 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1454 | struct extent_state **cached_state) | |
ea8c2819 | 1455 | { |
d397712b | 1456 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1457 | WARN_ON(1); |
ea8c2819 | 1458 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1459 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1460 | } |
1461 | ||
d352ac68 | 1462 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1463 | struct btrfs_writepage_fixup { |
1464 | struct page *page; | |
1465 | struct btrfs_work work; | |
1466 | }; | |
1467 | ||
b2950863 | 1468 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1469 | { |
1470 | struct btrfs_writepage_fixup *fixup; | |
1471 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1472 | struct extent_state *cached_state = NULL; |
247e743c CM |
1473 | struct page *page; |
1474 | struct inode *inode; | |
1475 | u64 page_start; | |
1476 | u64 page_end; | |
1477 | ||
1478 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1479 | page = fixup->page; | |
4a096752 | 1480 | again: |
247e743c CM |
1481 | lock_page(page); |
1482 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1483 | ClearPageChecked(page); | |
1484 | goto out_page; | |
1485 | } | |
1486 | ||
1487 | inode = page->mapping->host; | |
1488 | page_start = page_offset(page); | |
1489 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1490 | ||
2ac55d41 JB |
1491 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1492 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1493 | |
1494 | /* already ordered? We're done */ | |
8b62b72b | 1495 | if (PagePrivate2(page)) |
247e743c | 1496 | goto out; |
4a096752 CM |
1497 | |
1498 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1499 | if (ordered) { | |
2ac55d41 JB |
1500 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1501 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1502 | unlock_page(page); |
1503 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1504 | goto again; | |
1505 | } | |
247e743c | 1506 | |
0ca1f7ce | 1507 | BUG(); |
2ac55d41 | 1508 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c CM |
1509 | ClearPageChecked(page); |
1510 | out: | |
2ac55d41 JB |
1511 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1512 | &cached_state, GFP_NOFS); | |
247e743c CM |
1513 | out_page: |
1514 | unlock_page(page); | |
1515 | page_cache_release(page); | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * There are a few paths in the higher layers of the kernel that directly | |
1520 | * set the page dirty bit without asking the filesystem if it is a | |
1521 | * good idea. This causes problems because we want to make sure COW | |
1522 | * properly happens and the data=ordered rules are followed. | |
1523 | * | |
c8b97818 | 1524 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1525 | * hasn't been properly setup for IO. We kick off an async process |
1526 | * to fix it up. The async helper will wait for ordered extents, set | |
1527 | * the delalloc bit and make it safe to write the page. | |
1528 | */ | |
b2950863 | 1529 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1530 | { |
1531 | struct inode *inode = page->mapping->host; | |
1532 | struct btrfs_writepage_fixup *fixup; | |
1533 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1534 | |
8b62b72b CM |
1535 | /* this page is properly in the ordered list */ |
1536 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1537 | return 0; |
1538 | ||
1539 | if (PageChecked(page)) | |
1540 | return -EAGAIN; | |
1541 | ||
1542 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1543 | if (!fixup) | |
1544 | return -EAGAIN; | |
f421950f | 1545 | |
247e743c CM |
1546 | SetPageChecked(page); |
1547 | page_cache_get(page); | |
1548 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1549 | fixup->page = page; | |
1550 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1551 | return -EAGAIN; | |
1552 | } | |
1553 | ||
d899e052 YZ |
1554 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1555 | struct inode *inode, u64 file_pos, | |
1556 | u64 disk_bytenr, u64 disk_num_bytes, | |
1557 | u64 num_bytes, u64 ram_bytes, | |
1558 | u8 compression, u8 encryption, | |
1559 | u16 other_encoding, int extent_type) | |
1560 | { | |
1561 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1562 | struct btrfs_file_extent_item *fi; | |
1563 | struct btrfs_path *path; | |
1564 | struct extent_buffer *leaf; | |
1565 | struct btrfs_key ins; | |
1566 | u64 hint; | |
1567 | int ret; | |
1568 | ||
1569 | path = btrfs_alloc_path(); | |
1570 | BUG_ON(!path); | |
1571 | ||
b9473439 | 1572 | path->leave_spinning = 1; |
a1ed835e CM |
1573 | |
1574 | /* | |
1575 | * we may be replacing one extent in the tree with another. | |
1576 | * The new extent is pinned in the extent map, and we don't want | |
1577 | * to drop it from the cache until it is completely in the btree. | |
1578 | * | |
1579 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1580 | * the caller is expected to unpin it and allow it to be merged | |
1581 | * with the others. | |
1582 | */ | |
920bbbfb YZ |
1583 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1584 | &hint, 0); | |
d899e052 YZ |
1585 | BUG_ON(ret); |
1586 | ||
1587 | ins.objectid = inode->i_ino; | |
1588 | ins.offset = file_pos; | |
1589 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1590 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1591 | BUG_ON(ret); | |
1592 | leaf = path->nodes[0]; | |
1593 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1594 | struct btrfs_file_extent_item); | |
1595 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1596 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1597 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1598 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1599 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1600 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1601 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1602 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1603 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1604 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1605 | |
1606 | btrfs_unlock_up_safe(path, 1); | |
1607 | btrfs_set_lock_blocking(leaf); | |
1608 | ||
d899e052 YZ |
1609 | btrfs_mark_buffer_dirty(leaf); |
1610 | ||
1611 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1612 | |
1613 | ins.objectid = disk_bytenr; | |
1614 | ins.offset = disk_num_bytes; | |
1615 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1616 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1617 | root->root_key.objectid, | |
1618 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1619 | BUG_ON(ret); |
d899e052 | 1620 | btrfs_free_path(path); |
b9473439 | 1621 | |
d899e052 YZ |
1622 | return 0; |
1623 | } | |
1624 | ||
5d13a98f CM |
1625 | /* |
1626 | * helper function for btrfs_finish_ordered_io, this | |
1627 | * just reads in some of the csum leaves to prime them into ram | |
1628 | * before we start the transaction. It limits the amount of btree | |
1629 | * reads required while inside the transaction. | |
1630 | */ | |
d352ac68 CM |
1631 | /* as ordered data IO finishes, this gets called so we can finish |
1632 | * an ordered extent if the range of bytes in the file it covers are | |
1633 | * fully written. | |
1634 | */ | |
211f90e6 | 1635 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1636 | { |
e6dcd2dc | 1637 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1638 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1639 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1640 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1641 | struct extent_state *cached_state = NULL; |
d899e052 | 1642 | int compressed = 0; |
e6dcd2dc CM |
1643 | int ret; |
1644 | ||
5a1a3df1 JB |
1645 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1646 | end - start + 1); | |
ba1da2f4 | 1647 | if (!ret) |
e6dcd2dc | 1648 | return 0; |
e6dcd2dc | 1649 | BUG_ON(!ordered_extent); |
efd049fb | 1650 | |
c2167754 YZ |
1651 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1652 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1653 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1654 | if (!ret) { | |
1655 | trans = btrfs_join_transaction(root, 1); | |
0ca1f7ce YZ |
1656 | btrfs_set_trans_block_group(trans, inode); |
1657 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 YZ |
1658 | ret = btrfs_update_inode(trans, root, inode); |
1659 | BUG_ON(ret); | |
c2167754 YZ |
1660 | } |
1661 | goto out; | |
1662 | } | |
e6dcd2dc | 1663 | |
2ac55d41 JB |
1664 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1665 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1666 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1667 | |
c2167754 | 1668 | trans = btrfs_join_transaction(root, 1); |
0ca1f7ce YZ |
1669 | btrfs_set_trans_block_group(trans, inode); |
1670 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 | 1671 | |
c8b97818 | 1672 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1673 | compressed = 1; |
1674 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1675 | BUG_ON(compressed); | |
920bbbfb | 1676 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1677 | ordered_extent->file_offset, |
1678 | ordered_extent->file_offset + | |
1679 | ordered_extent->len); | |
1680 | BUG_ON(ret); | |
1681 | } else { | |
1682 | ret = insert_reserved_file_extent(trans, inode, | |
1683 | ordered_extent->file_offset, | |
1684 | ordered_extent->start, | |
1685 | ordered_extent->disk_len, | |
1686 | ordered_extent->len, | |
1687 | ordered_extent->len, | |
1688 | compressed, 0, 0, | |
1689 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1690 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1691 | ordered_extent->file_offset, | |
1692 | ordered_extent->len); | |
d899e052 YZ |
1693 | BUG_ON(ret); |
1694 | } | |
2ac55d41 JB |
1695 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1696 | ordered_extent->file_offset + | |
1697 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1698 | ||
e6dcd2dc CM |
1699 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1700 | &ordered_extent->list); | |
1701 | ||
c2167754 YZ |
1702 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1703 | ret = btrfs_update_inode(trans, root, inode); | |
1704 | BUG_ON(ret); | |
c2167754 | 1705 | out: |
0ca1f7ce YZ |
1706 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
1707 | if (trans) | |
1708 | btrfs_end_transaction(trans, root); | |
e6dcd2dc CM |
1709 | /* once for us */ |
1710 | btrfs_put_ordered_extent(ordered_extent); | |
1711 | /* once for the tree */ | |
1712 | btrfs_put_ordered_extent(ordered_extent); | |
1713 | ||
e6dcd2dc CM |
1714 | return 0; |
1715 | } | |
1716 | ||
b2950863 | 1717 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1718 | struct extent_state *state, int uptodate) |
1719 | { | |
8b62b72b | 1720 | ClearPagePrivate2(page); |
211f90e6 CM |
1721 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1722 | } | |
1723 | ||
d352ac68 CM |
1724 | /* |
1725 | * When IO fails, either with EIO or csum verification fails, we | |
1726 | * try other mirrors that might have a good copy of the data. This | |
1727 | * io_failure_record is used to record state as we go through all the | |
1728 | * mirrors. If another mirror has good data, the page is set up to date | |
1729 | * and things continue. If a good mirror can't be found, the original | |
1730 | * bio end_io callback is called to indicate things have failed. | |
1731 | */ | |
7e38326f CM |
1732 | struct io_failure_record { |
1733 | struct page *page; | |
1734 | u64 start; | |
1735 | u64 len; | |
1736 | u64 logical; | |
d20f7043 | 1737 | unsigned long bio_flags; |
7e38326f CM |
1738 | int last_mirror; |
1739 | }; | |
1740 | ||
b2950863 | 1741 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1742 | struct page *page, u64 start, u64 end, |
1743 | struct extent_state *state) | |
7e38326f CM |
1744 | { |
1745 | struct io_failure_record *failrec = NULL; | |
1746 | u64 private; | |
1747 | struct extent_map *em; | |
1748 | struct inode *inode = page->mapping->host; | |
1749 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1750 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1751 | struct bio *bio; |
1752 | int num_copies; | |
1753 | int ret; | |
1259ab75 | 1754 | int rw; |
7e38326f CM |
1755 | u64 logical; |
1756 | ||
1757 | ret = get_state_private(failure_tree, start, &private); | |
1758 | if (ret) { | |
7e38326f CM |
1759 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1760 | if (!failrec) | |
1761 | return -ENOMEM; | |
1762 | failrec->start = start; | |
1763 | failrec->len = end - start + 1; | |
1764 | failrec->last_mirror = 0; | |
d20f7043 | 1765 | failrec->bio_flags = 0; |
7e38326f | 1766 | |
890871be | 1767 | read_lock(&em_tree->lock); |
3b951516 CM |
1768 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1769 | if (em->start > start || em->start + em->len < start) { | |
1770 | free_extent_map(em); | |
1771 | em = NULL; | |
1772 | } | |
890871be | 1773 | read_unlock(&em_tree->lock); |
7e38326f CM |
1774 | |
1775 | if (!em || IS_ERR(em)) { | |
1776 | kfree(failrec); | |
1777 | return -EIO; | |
1778 | } | |
1779 | logical = start - em->start; | |
1780 | logical = em->block_start + logical; | |
d20f7043 CM |
1781 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1782 | logical = em->block_start; | |
1783 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1784 | } | |
7e38326f CM |
1785 | failrec->logical = logical; |
1786 | free_extent_map(em); | |
1787 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1788 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1789 | set_state_private(failure_tree, start, |
1790 | (u64)(unsigned long)failrec); | |
7e38326f | 1791 | } else { |
587f7704 | 1792 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1793 | } |
1794 | num_copies = btrfs_num_copies( | |
1795 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1796 | failrec->logical, failrec->len); | |
1797 | failrec->last_mirror++; | |
1798 | if (!state) { | |
cad321ad | 1799 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1800 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1801 | failrec->start, | |
1802 | EXTENT_LOCKED); | |
1803 | if (state && state->start != failrec->start) | |
1804 | state = NULL; | |
cad321ad | 1805 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1806 | } |
1807 | if (!state || failrec->last_mirror > num_copies) { | |
1808 | set_state_private(failure_tree, failrec->start, 0); | |
1809 | clear_extent_bits(failure_tree, failrec->start, | |
1810 | failrec->start + failrec->len - 1, | |
1811 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1812 | kfree(failrec); | |
1813 | return -EIO; | |
1814 | } | |
1815 | bio = bio_alloc(GFP_NOFS, 1); | |
1816 | bio->bi_private = state; | |
1817 | bio->bi_end_io = failed_bio->bi_end_io; | |
1818 | bio->bi_sector = failrec->logical >> 9; | |
1819 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1820 | bio->bi_size = 0; |
d20f7043 | 1821 | |
7e38326f | 1822 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1823 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1824 | rw = WRITE; | |
1825 | else | |
1826 | rw = READ; | |
1827 | ||
1828 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1829 | failrec->last_mirror, |
d20f7043 | 1830 | failrec->bio_flags); |
1259ab75 CM |
1831 | return 0; |
1832 | } | |
1833 | ||
d352ac68 CM |
1834 | /* |
1835 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1836 | * to see if we need to process or clean up an io_failure_record | |
1837 | */ | |
b2950863 | 1838 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1839 | { |
1840 | u64 private; | |
1841 | u64 private_failure; | |
1842 | struct io_failure_record *failure; | |
1843 | int ret; | |
1844 | ||
1845 | private = 0; | |
1846 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1847 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1848 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1849 | start, &private_failure); | |
1850 | if (ret == 0) { | |
1851 | failure = (struct io_failure_record *)(unsigned long) | |
1852 | private_failure; | |
1853 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1854 | failure->start, 0); | |
1855 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1856 | failure->start, | |
1857 | failure->start + failure->len - 1, | |
1858 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1859 | GFP_NOFS); | |
1860 | kfree(failure); | |
1861 | } | |
1862 | } | |
7e38326f CM |
1863 | return 0; |
1864 | } | |
1865 | ||
d352ac68 CM |
1866 | /* |
1867 | * when reads are done, we need to check csums to verify the data is correct | |
1868 | * if there's a match, we allow the bio to finish. If not, we go through | |
1869 | * the io_failure_record routines to find good copies | |
1870 | */ | |
b2950863 | 1871 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1872 | struct extent_state *state) |
07157aac | 1873 | { |
35ebb934 | 1874 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1875 | struct inode *inode = page->mapping->host; |
d1310b2e | 1876 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1877 | char *kaddr; |
aadfeb6e | 1878 | u64 private = ~(u32)0; |
07157aac | 1879 | int ret; |
ff79f819 CM |
1880 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1881 | u32 csum = ~(u32)0; | |
d1310b2e | 1882 | |
d20f7043 CM |
1883 | if (PageChecked(page)) { |
1884 | ClearPageChecked(page); | |
1885 | goto good; | |
1886 | } | |
6cbff00f CH |
1887 | |
1888 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1889 | return 0; |
1890 | ||
1891 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1892 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1893 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1894 | GFP_NOFS); | |
b6cda9bc | 1895 | return 0; |
17d217fe | 1896 | } |
d20f7043 | 1897 | |
c2e639f0 | 1898 | if (state && state->start == start) { |
70dec807 CM |
1899 | private = state->private; |
1900 | ret = 0; | |
1901 | } else { | |
1902 | ret = get_state_private(io_tree, start, &private); | |
1903 | } | |
9ab86c8e | 1904 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1905 | if (ret) |
07157aac | 1906 | goto zeroit; |
d397712b | 1907 | |
ff79f819 CM |
1908 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1909 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1910 | if (csum != private) |
07157aac | 1911 | goto zeroit; |
d397712b | 1912 | |
9ab86c8e | 1913 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1914 | good: |
7e38326f CM |
1915 | /* if the io failure tree for this inode is non-empty, |
1916 | * check to see if we've recovered from a failed IO | |
1917 | */ | |
1259ab75 | 1918 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1919 | return 0; |
1920 | ||
1921 | zeroit: | |
193f284d CM |
1922 | if (printk_ratelimit()) { |
1923 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1924 | "private %llu\n", page->mapping->host->i_ino, | |
1925 | (unsigned long long)start, csum, | |
1926 | (unsigned long long)private); | |
1927 | } | |
db94535d CM |
1928 | memset(kaddr + offset, 1, end - start + 1); |
1929 | flush_dcache_page(page); | |
9ab86c8e | 1930 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1931 | if (private == 0) |
1932 | return 0; | |
7e38326f | 1933 | return -EIO; |
07157aac | 1934 | } |
b888db2b | 1935 | |
24bbcf04 YZ |
1936 | struct delayed_iput { |
1937 | struct list_head list; | |
1938 | struct inode *inode; | |
1939 | }; | |
1940 | ||
1941 | void btrfs_add_delayed_iput(struct inode *inode) | |
1942 | { | |
1943 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
1944 | struct delayed_iput *delayed; | |
1945 | ||
1946 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
1947 | return; | |
1948 | ||
1949 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
1950 | delayed->inode = inode; | |
1951 | ||
1952 | spin_lock(&fs_info->delayed_iput_lock); | |
1953 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
1954 | spin_unlock(&fs_info->delayed_iput_lock); | |
1955 | } | |
1956 | ||
1957 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
1958 | { | |
1959 | LIST_HEAD(list); | |
1960 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1961 | struct delayed_iput *delayed; | |
1962 | int empty; | |
1963 | ||
1964 | spin_lock(&fs_info->delayed_iput_lock); | |
1965 | empty = list_empty(&fs_info->delayed_iputs); | |
1966 | spin_unlock(&fs_info->delayed_iput_lock); | |
1967 | if (empty) | |
1968 | return; | |
1969 | ||
1970 | down_read(&root->fs_info->cleanup_work_sem); | |
1971 | spin_lock(&fs_info->delayed_iput_lock); | |
1972 | list_splice_init(&fs_info->delayed_iputs, &list); | |
1973 | spin_unlock(&fs_info->delayed_iput_lock); | |
1974 | ||
1975 | while (!list_empty(&list)) { | |
1976 | delayed = list_entry(list.next, struct delayed_iput, list); | |
1977 | list_del(&delayed->list); | |
1978 | iput(delayed->inode); | |
1979 | kfree(delayed); | |
1980 | } | |
1981 | up_read(&root->fs_info->cleanup_work_sem); | |
1982 | } | |
1983 | ||
d68fc57b YZ |
1984 | /* |
1985 | * calculate extra metadata reservation when snapshotting a subvolume | |
1986 | * contains orphan files. | |
1987 | */ | |
1988 | void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans, | |
1989 | struct btrfs_pending_snapshot *pending, | |
1990 | u64 *bytes_to_reserve) | |
1991 | { | |
1992 | struct btrfs_root *root; | |
1993 | struct btrfs_block_rsv *block_rsv; | |
1994 | u64 num_bytes; | |
1995 | int index; | |
1996 | ||
1997 | root = pending->root; | |
1998 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
1999 | return; | |
2000 | ||
2001 | block_rsv = root->orphan_block_rsv; | |
2002 | ||
2003 | /* orphan block reservation for the snapshot */ | |
2004 | num_bytes = block_rsv->size; | |
2005 | ||
2006 | /* | |
2007 | * after the snapshot is created, COWing tree blocks may use more | |
2008 | * space than it frees. So we should make sure there is enough | |
2009 | * reserved space. | |
2010 | */ | |
2011 | index = trans->transid & 0x1; | |
2012 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2013 | num_bytes += block_rsv->size - | |
2014 | (block_rsv->reserved + block_rsv->freed[index]); | |
2015 | } | |
2016 | ||
2017 | *bytes_to_reserve += num_bytes; | |
2018 | } | |
2019 | ||
2020 | void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans, | |
2021 | struct btrfs_pending_snapshot *pending) | |
2022 | { | |
2023 | struct btrfs_root *root = pending->root; | |
2024 | struct btrfs_root *snap = pending->snap; | |
2025 | struct btrfs_block_rsv *block_rsv; | |
2026 | u64 num_bytes; | |
2027 | int index; | |
2028 | int ret; | |
2029 | ||
2030 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2031 | return; | |
2032 | ||
2033 | /* refill source subvolume's orphan block reservation */ | |
2034 | block_rsv = root->orphan_block_rsv; | |
2035 | index = trans->transid & 0x1; | |
2036 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2037 | num_bytes = block_rsv->size - | |
2038 | (block_rsv->reserved + block_rsv->freed[index]); | |
2039 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2040 | root->orphan_block_rsv, | |
2041 | num_bytes); | |
2042 | BUG_ON(ret); | |
2043 | } | |
2044 | ||
2045 | /* setup orphan block reservation for the snapshot */ | |
2046 | block_rsv = btrfs_alloc_block_rsv(snap); | |
2047 | BUG_ON(!block_rsv); | |
2048 | ||
2049 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
2050 | snap->orphan_block_rsv = block_rsv; | |
2051 | ||
2052 | num_bytes = root->orphan_block_rsv->size; | |
2053 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2054 | block_rsv, num_bytes); | |
2055 | BUG_ON(ret); | |
2056 | ||
2057 | #if 0 | |
2058 | /* insert orphan item for the snapshot */ | |
2059 | WARN_ON(!root->orphan_item_inserted); | |
2060 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2061 | snap->root_key.objectid); | |
2062 | BUG_ON(ret); | |
2063 | snap->orphan_item_inserted = 1; | |
2064 | #endif | |
2065 | } | |
2066 | ||
2067 | enum btrfs_orphan_cleanup_state { | |
2068 | ORPHAN_CLEANUP_STARTED = 1, | |
2069 | ORPHAN_CLEANUP_DONE = 2, | |
2070 | }; | |
2071 | ||
2072 | /* | |
2073 | * This is called in transaction commmit time. If there are no orphan | |
2074 | * files in the subvolume, it removes orphan item and frees block_rsv | |
2075 | * structure. | |
2076 | */ | |
2077 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2078 | struct btrfs_root *root) | |
2079 | { | |
2080 | int ret; | |
2081 | ||
2082 | if (!list_empty(&root->orphan_list) || | |
2083 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2084 | return; | |
2085 | ||
2086 | if (root->orphan_item_inserted && | |
2087 | btrfs_root_refs(&root->root_item) > 0) { | |
2088 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2089 | root->root_key.objectid); | |
2090 | BUG_ON(ret); | |
2091 | root->orphan_item_inserted = 0; | |
2092 | } | |
2093 | ||
2094 | if (root->orphan_block_rsv) { | |
2095 | WARN_ON(root->orphan_block_rsv->size > 0); | |
2096 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
2097 | root->orphan_block_rsv = NULL; | |
2098 | } | |
2099 | } | |
2100 | ||
7b128766 JB |
2101 | /* |
2102 | * This creates an orphan entry for the given inode in case something goes | |
2103 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2104 | * |
2105 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2106 | * this function. | |
7b128766 JB |
2107 | */ |
2108 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2109 | { | |
2110 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2111 | struct btrfs_block_rsv *block_rsv = NULL; |
2112 | int reserve = 0; | |
2113 | int insert = 0; | |
2114 | int ret; | |
7b128766 | 2115 | |
d68fc57b YZ |
2116 | if (!root->orphan_block_rsv) { |
2117 | block_rsv = btrfs_alloc_block_rsv(root); | |
2118 | BUG_ON(!block_rsv); | |
2119 | } | |
7b128766 | 2120 | |
d68fc57b YZ |
2121 | spin_lock(&root->orphan_lock); |
2122 | if (!root->orphan_block_rsv) { | |
2123 | root->orphan_block_rsv = block_rsv; | |
2124 | } else if (block_rsv) { | |
2125 | btrfs_free_block_rsv(root, block_rsv); | |
2126 | block_rsv = NULL; | |
7b128766 JB |
2127 | } |
2128 | ||
d68fc57b YZ |
2129 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2130 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2131 | #if 0 | |
2132 | /* | |
2133 | * For proper ENOSPC handling, we should do orphan | |
2134 | * cleanup when mounting. But this introduces backward | |
2135 | * compatibility issue. | |
2136 | */ | |
2137 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2138 | insert = 2; | |
2139 | else | |
2140 | insert = 1; | |
2141 | #endif | |
2142 | insert = 1; | |
2143 | } else { | |
2144 | WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved); | |
2145 | } | |
7b128766 | 2146 | |
d68fc57b YZ |
2147 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2148 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2149 | reserve = 1; | |
2150 | } | |
2151 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2152 | |
d68fc57b YZ |
2153 | if (block_rsv) |
2154 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
7b128766 | 2155 | |
d68fc57b YZ |
2156 | /* grab metadata reservation from transaction handle */ |
2157 | if (reserve) { | |
2158 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2159 | BUG_ON(ret); | |
2160 | } | |
2161 | ||
2162 | /* insert an orphan item to track this unlinked/truncated file */ | |
2163 | if (insert >= 1) { | |
2164 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2165 | BUG_ON(ret); | |
2166 | } | |
2167 | ||
2168 | /* insert an orphan item to track subvolume contains orphan files */ | |
2169 | if (insert >= 2) { | |
2170 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2171 | root->root_key.objectid); | |
2172 | BUG_ON(ret); | |
2173 | } | |
2174 | return 0; | |
7b128766 JB |
2175 | } |
2176 | ||
2177 | /* | |
2178 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2179 | * item for this particular inode. | |
2180 | */ | |
2181 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2182 | { | |
2183 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2184 | int delete_item = 0; |
2185 | int release_rsv = 0; | |
7b128766 JB |
2186 | int ret = 0; |
2187 | ||
d68fc57b YZ |
2188 | spin_lock(&root->orphan_lock); |
2189 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2190 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2191 | delete_item = 1; | |
7b128766 JB |
2192 | } |
2193 | ||
d68fc57b YZ |
2194 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2195 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2196 | release_rsv = 1; | |
7b128766 | 2197 | } |
d68fc57b | 2198 | spin_unlock(&root->orphan_lock); |
7b128766 | 2199 | |
d68fc57b YZ |
2200 | if (trans && delete_item) { |
2201 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2202 | BUG_ON(ret); | |
2203 | } | |
7b128766 | 2204 | |
d68fc57b YZ |
2205 | if (release_rsv) |
2206 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2207 | |
d68fc57b | 2208 | return 0; |
7b128766 JB |
2209 | } |
2210 | ||
2211 | /* | |
2212 | * this cleans up any orphans that may be left on the list from the last use | |
2213 | * of this root. | |
2214 | */ | |
2215 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2216 | { | |
2217 | struct btrfs_path *path; | |
2218 | struct extent_buffer *leaf; | |
2219 | struct btrfs_item *item; | |
2220 | struct btrfs_key key, found_key; | |
2221 | struct btrfs_trans_handle *trans; | |
2222 | struct inode *inode; | |
2223 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2224 | ||
d68fc57b | 2225 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
7b128766 | 2226 | return; |
c71bf099 YZ |
2227 | |
2228 | path = btrfs_alloc_path(); | |
2229 | BUG_ON(!path); | |
7b128766 JB |
2230 | path->reada = -1; |
2231 | ||
2232 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2233 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2234 | key.offset = (u64)-1; | |
2235 | ||
7b128766 JB |
2236 | while (1) { |
2237 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2238 | if (ret < 0) { | |
2239 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2240 | "\n", ret); | |
2241 | break; | |
2242 | } | |
2243 | ||
2244 | /* | |
2245 | * if ret == 0 means we found what we were searching for, which | |
2246 | * is weird, but possible, so only screw with path if we didnt | |
2247 | * find the key and see if we have stuff that matches | |
2248 | */ | |
2249 | if (ret > 0) { | |
2250 | if (path->slots[0] == 0) | |
2251 | break; | |
2252 | path->slots[0]--; | |
2253 | } | |
2254 | ||
2255 | /* pull out the item */ | |
2256 | leaf = path->nodes[0]; | |
2257 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2258 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2259 | ||
2260 | /* make sure the item matches what we want */ | |
2261 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2262 | break; | |
2263 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2264 | break; | |
2265 | ||
2266 | /* release the path since we're done with it */ | |
2267 | btrfs_release_path(root, path); | |
2268 | ||
2269 | /* | |
2270 | * this is where we are basically btrfs_lookup, without the | |
2271 | * crossing root thing. we store the inode number in the | |
2272 | * offset of the orphan item. | |
2273 | */ | |
5d4f98a2 YZ |
2274 | found_key.objectid = found_key.offset; |
2275 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2276 | found_key.offset = 0; | |
73f73415 | 2277 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
d68fc57b | 2278 | BUG_ON(IS_ERR(inode)); |
7b128766 | 2279 | |
7b128766 JB |
2280 | /* |
2281 | * add this inode to the orphan list so btrfs_orphan_del does | |
2282 | * the proper thing when we hit it | |
2283 | */ | |
d68fc57b | 2284 | spin_lock(&root->orphan_lock); |
7b128766 | 2285 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
d68fc57b | 2286 | spin_unlock(&root->orphan_lock); |
7b128766 JB |
2287 | |
2288 | /* | |
2289 | * if this is a bad inode, means we actually succeeded in | |
2290 | * removing the inode, but not the orphan record, which means | |
2291 | * we need to manually delete the orphan since iput will just | |
2292 | * do a destroy_inode | |
2293 | */ | |
2294 | if (is_bad_inode(inode)) { | |
a22285a6 | 2295 | trans = btrfs_start_transaction(root, 0); |
7b128766 | 2296 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2297 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2298 | iput(inode); |
2299 | continue; | |
2300 | } | |
2301 | ||
2302 | /* if we have links, this was a truncate, lets do that */ | |
2303 | if (inode->i_nlink) { | |
2304 | nr_truncate++; | |
2305 | btrfs_truncate(inode); | |
2306 | } else { | |
2307 | nr_unlink++; | |
2308 | } | |
2309 | ||
2310 | /* this will do delete_inode and everything for us */ | |
2311 | iput(inode); | |
2312 | } | |
d68fc57b YZ |
2313 | btrfs_free_path(path); |
2314 | ||
2315 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; | |
2316 | ||
2317 | if (root->orphan_block_rsv) | |
2318 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2319 | (u64)-1); | |
2320 | ||
2321 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
2322 | trans = btrfs_join_transaction(root, 1); | |
2323 | btrfs_end_transaction(trans, root); | |
2324 | } | |
7b128766 JB |
2325 | |
2326 | if (nr_unlink) | |
2327 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2328 | if (nr_truncate) | |
2329 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
7b128766 JB |
2330 | } |
2331 | ||
46a53cca CM |
2332 | /* |
2333 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2334 | * don't find any xattrs, we know there can't be any acls. | |
2335 | * | |
2336 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2337 | */ | |
2338 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2339 | int slot, u64 objectid) | |
2340 | { | |
2341 | u32 nritems = btrfs_header_nritems(leaf); | |
2342 | struct btrfs_key found_key; | |
2343 | int scanned = 0; | |
2344 | ||
2345 | slot++; | |
2346 | while (slot < nritems) { | |
2347 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2348 | ||
2349 | /* we found a different objectid, there must not be acls */ | |
2350 | if (found_key.objectid != objectid) | |
2351 | return 0; | |
2352 | ||
2353 | /* we found an xattr, assume we've got an acl */ | |
2354 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2355 | return 1; | |
2356 | ||
2357 | /* | |
2358 | * we found a key greater than an xattr key, there can't | |
2359 | * be any acls later on | |
2360 | */ | |
2361 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2362 | return 0; | |
2363 | ||
2364 | slot++; | |
2365 | scanned++; | |
2366 | ||
2367 | /* | |
2368 | * it goes inode, inode backrefs, xattrs, extents, | |
2369 | * so if there are a ton of hard links to an inode there can | |
2370 | * be a lot of backrefs. Don't waste time searching too hard, | |
2371 | * this is just an optimization | |
2372 | */ | |
2373 | if (scanned >= 8) | |
2374 | break; | |
2375 | } | |
2376 | /* we hit the end of the leaf before we found an xattr or | |
2377 | * something larger than an xattr. We have to assume the inode | |
2378 | * has acls | |
2379 | */ | |
2380 | return 1; | |
2381 | } | |
2382 | ||
d352ac68 CM |
2383 | /* |
2384 | * read an inode from the btree into the in-memory inode | |
2385 | */ | |
5d4f98a2 | 2386 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2387 | { |
2388 | struct btrfs_path *path; | |
5f39d397 | 2389 | struct extent_buffer *leaf; |
39279cc3 | 2390 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2391 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2392 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2393 | struct btrfs_key location; | |
46a53cca | 2394 | int maybe_acls; |
39279cc3 | 2395 | u64 alloc_group_block; |
618e21d5 | 2396 | u32 rdev; |
39279cc3 CM |
2397 | int ret; |
2398 | ||
2399 | path = btrfs_alloc_path(); | |
2400 | BUG_ON(!path); | |
39279cc3 | 2401 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2402 | |
39279cc3 | 2403 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2404 | if (ret) |
39279cc3 | 2405 | goto make_bad; |
39279cc3 | 2406 | |
5f39d397 CM |
2407 | leaf = path->nodes[0]; |
2408 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2409 | struct btrfs_inode_item); | |
2410 | ||
2411 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2412 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2413 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2414 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2415 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2416 | |
2417 | tspec = btrfs_inode_atime(inode_item); | |
2418 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2419 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2420 | ||
2421 | tspec = btrfs_inode_mtime(inode_item); | |
2422 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2423 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2424 | ||
2425 | tspec = btrfs_inode_ctime(inode_item); | |
2426 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2427 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2428 | ||
a76a3cd4 | 2429 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2430 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2431 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2432 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2433 | inode->i_rdev = 0; |
5f39d397 CM |
2434 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2435 | ||
aec7477b | 2436 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2437 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2438 | |
5f39d397 | 2439 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2440 | |
46a53cca CM |
2441 | /* |
2442 | * try to precache a NULL acl entry for files that don't have | |
2443 | * any xattrs or acls | |
2444 | */ | |
2445 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2446 | if (!maybe_acls) |
2447 | cache_no_acl(inode); | |
46a53cca | 2448 | |
d2fb3437 YZ |
2449 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2450 | alloc_group_block, 0); | |
39279cc3 CM |
2451 | btrfs_free_path(path); |
2452 | inode_item = NULL; | |
2453 | ||
39279cc3 | 2454 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2455 | case S_IFREG: |
2456 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2457 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2458 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2459 | inode->i_fop = &btrfs_file_operations; |
2460 | inode->i_op = &btrfs_file_inode_operations; | |
2461 | break; | |
2462 | case S_IFDIR: | |
2463 | inode->i_fop = &btrfs_dir_file_operations; | |
2464 | if (root == root->fs_info->tree_root) | |
2465 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2466 | else | |
2467 | inode->i_op = &btrfs_dir_inode_operations; | |
2468 | break; | |
2469 | case S_IFLNK: | |
2470 | inode->i_op = &btrfs_symlink_inode_operations; | |
2471 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2472 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2473 | break; |
618e21d5 | 2474 | default: |
0279b4cd | 2475 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2476 | init_special_inode(inode, inode->i_mode, rdev); |
2477 | break; | |
39279cc3 | 2478 | } |
6cbff00f CH |
2479 | |
2480 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2481 | return; |
2482 | ||
2483 | make_bad: | |
39279cc3 | 2484 | btrfs_free_path(path); |
39279cc3 CM |
2485 | make_bad_inode(inode); |
2486 | } | |
2487 | ||
d352ac68 CM |
2488 | /* |
2489 | * given a leaf and an inode, copy the inode fields into the leaf | |
2490 | */ | |
e02119d5 CM |
2491 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2492 | struct extent_buffer *leaf, | |
5f39d397 | 2493 | struct btrfs_inode_item *item, |
39279cc3 CM |
2494 | struct inode *inode) |
2495 | { | |
5f39d397 CM |
2496 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2497 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2498 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2499 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2500 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2501 | ||
2502 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2503 | inode->i_atime.tv_sec); | |
2504 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2505 | inode->i_atime.tv_nsec); | |
2506 | ||
2507 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2508 | inode->i_mtime.tv_sec); | |
2509 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2510 | inode->i_mtime.tv_nsec); | |
2511 | ||
2512 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2513 | inode->i_ctime.tv_sec); | |
2514 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2515 | inode->i_ctime.tv_nsec); | |
2516 | ||
a76a3cd4 | 2517 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2518 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2519 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2520 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2521 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2522 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2523 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2524 | } |
2525 | ||
d352ac68 CM |
2526 | /* |
2527 | * copy everything in the in-memory inode into the btree. | |
2528 | */ | |
d397712b CM |
2529 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2530 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2531 | { |
2532 | struct btrfs_inode_item *inode_item; | |
2533 | struct btrfs_path *path; | |
5f39d397 | 2534 | struct extent_buffer *leaf; |
39279cc3 CM |
2535 | int ret; |
2536 | ||
2537 | path = btrfs_alloc_path(); | |
2538 | BUG_ON(!path); | |
b9473439 | 2539 | path->leave_spinning = 1; |
39279cc3 CM |
2540 | ret = btrfs_lookup_inode(trans, root, path, |
2541 | &BTRFS_I(inode)->location, 1); | |
2542 | if (ret) { | |
2543 | if (ret > 0) | |
2544 | ret = -ENOENT; | |
2545 | goto failed; | |
2546 | } | |
2547 | ||
b4ce94de | 2548 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2549 | leaf = path->nodes[0]; |
2550 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2551 | struct btrfs_inode_item); |
2552 | ||
e02119d5 | 2553 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2554 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2555 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2556 | ret = 0; |
2557 | failed: | |
39279cc3 CM |
2558 | btrfs_free_path(path); |
2559 | return ret; | |
2560 | } | |
2561 | ||
2562 | ||
d352ac68 CM |
2563 | /* |
2564 | * unlink helper that gets used here in inode.c and in the tree logging | |
2565 | * recovery code. It remove a link in a directory with a given name, and | |
2566 | * also drops the back refs in the inode to the directory | |
2567 | */ | |
e02119d5 CM |
2568 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2569 | struct btrfs_root *root, | |
2570 | struct inode *dir, struct inode *inode, | |
2571 | const char *name, int name_len) | |
39279cc3 CM |
2572 | { |
2573 | struct btrfs_path *path; | |
39279cc3 | 2574 | int ret = 0; |
5f39d397 | 2575 | struct extent_buffer *leaf; |
39279cc3 | 2576 | struct btrfs_dir_item *di; |
5f39d397 | 2577 | struct btrfs_key key; |
aec7477b | 2578 | u64 index; |
39279cc3 CM |
2579 | |
2580 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2581 | if (!path) { |
2582 | ret = -ENOMEM; | |
2583 | goto err; | |
2584 | } | |
2585 | ||
b9473439 | 2586 | path->leave_spinning = 1; |
39279cc3 CM |
2587 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2588 | name, name_len, -1); | |
2589 | if (IS_ERR(di)) { | |
2590 | ret = PTR_ERR(di); | |
2591 | goto err; | |
2592 | } | |
2593 | if (!di) { | |
2594 | ret = -ENOENT; | |
2595 | goto err; | |
2596 | } | |
5f39d397 CM |
2597 | leaf = path->nodes[0]; |
2598 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2599 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2600 | if (ret) |
2601 | goto err; | |
39279cc3 CM |
2602 | btrfs_release_path(root, path); |
2603 | ||
aec7477b | 2604 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2605 | inode->i_ino, |
2606 | dir->i_ino, &index); | |
aec7477b | 2607 | if (ret) { |
d397712b | 2608 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2609 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2610 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2611 | goto err; |
2612 | } | |
2613 | ||
39279cc3 | 2614 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2615 | index, name, name_len, -1); |
39279cc3 CM |
2616 | if (IS_ERR(di)) { |
2617 | ret = PTR_ERR(di); | |
2618 | goto err; | |
2619 | } | |
2620 | if (!di) { | |
2621 | ret = -ENOENT; | |
2622 | goto err; | |
2623 | } | |
2624 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2625 | btrfs_release_path(root, path); |
39279cc3 | 2626 | |
e02119d5 CM |
2627 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2628 | inode, dir->i_ino); | |
49eb7e46 | 2629 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2630 | |
2631 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2632 | dir, index); | |
2633 | BUG_ON(ret); | |
39279cc3 CM |
2634 | err: |
2635 | btrfs_free_path(path); | |
e02119d5 CM |
2636 | if (ret) |
2637 | goto out; | |
2638 | ||
2639 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2640 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2641 | btrfs_update_inode(trans, root, dir); | |
2642 | btrfs_drop_nlink(inode); | |
2643 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2644 | out: |
39279cc3 CM |
2645 | return ret; |
2646 | } | |
2647 | ||
a22285a6 YZ |
2648 | /* helper to check if there is any shared block in the path */ |
2649 | static int check_path_shared(struct btrfs_root *root, | |
2650 | struct btrfs_path *path) | |
2651 | { | |
2652 | struct extent_buffer *eb; | |
2653 | int level; | |
2654 | int ret; | |
2655 | u64 refs; | |
2656 | ||
2657 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
2658 | if (!path->nodes[level]) | |
2659 | break; | |
2660 | eb = path->nodes[level]; | |
2661 | if (!btrfs_block_can_be_shared(root, eb)) | |
2662 | continue; | |
2663 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2664 | &refs, NULL); | |
2665 | if (refs > 1) | |
2666 | return 1; | |
2667 | } | |
2668 | return 0; | |
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * helper to start transaction for unlink and rmdir. | |
2673 | * | |
2674 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2675 | * so in enospc case, we should make sure they will free space before | |
2676 | * allowing them to use the global metadata reservation. | |
2677 | */ | |
2678 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2679 | struct dentry *dentry) | |
39279cc3 | 2680 | { |
39279cc3 | 2681 | struct btrfs_trans_handle *trans; |
a22285a6 YZ |
2682 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2683 | struct btrfs_path *path; | |
2684 | struct btrfs_inode_ref *ref; | |
2685 | struct btrfs_dir_item *di; | |
7b128766 | 2686 | struct inode *inode = dentry->d_inode; |
a22285a6 YZ |
2687 | u64 index; |
2688 | int check_link = 1; | |
2689 | int err = -ENOSPC; | |
39279cc3 CM |
2690 | int ret; |
2691 | ||
a22285a6 YZ |
2692 | trans = btrfs_start_transaction(root, 10); |
2693 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
2694 | return trans; | |
1832a6d5 | 2695 | |
a22285a6 YZ |
2696 | if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
2697 | return ERR_PTR(-ENOSPC); | |
5df6a9f6 | 2698 | |
a22285a6 YZ |
2699 | /* check if there is someone else holds reference */ |
2700 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2701 | return ERR_PTR(-ENOSPC); | |
2702 | ||
2703 | if (atomic_read(&inode->i_count) > 2) | |
2704 | return ERR_PTR(-ENOSPC); | |
2705 | ||
2706 | if (xchg(&root->fs_info->enospc_unlink, 1)) | |
2707 | return ERR_PTR(-ENOSPC); | |
2708 | ||
2709 | path = btrfs_alloc_path(); | |
2710 | if (!path) { | |
2711 | root->fs_info->enospc_unlink = 0; | |
2712 | return ERR_PTR(-ENOMEM); | |
2713 | } | |
2714 | ||
2715 | trans = btrfs_start_transaction(root, 0); | |
5df6a9f6 | 2716 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2717 | btrfs_free_path(path); |
2718 | root->fs_info->enospc_unlink = 0; | |
2719 | return trans; | |
2720 | } | |
2721 | ||
2722 | path->skip_locking = 1; | |
2723 | path->search_commit_root = 1; | |
2724 | ||
2725 | ret = btrfs_lookup_inode(trans, root, path, | |
2726 | &BTRFS_I(dir)->location, 0); | |
2727 | if (ret < 0) { | |
2728 | err = ret; | |
2729 | goto out; | |
2730 | } | |
2731 | if (ret == 0) { | |
2732 | if (check_path_shared(root, path)) | |
2733 | goto out; | |
2734 | } else { | |
2735 | check_link = 0; | |
5df6a9f6 | 2736 | } |
a22285a6 YZ |
2737 | btrfs_release_path(root, path); |
2738 | ||
2739 | ret = btrfs_lookup_inode(trans, root, path, | |
2740 | &BTRFS_I(inode)->location, 0); | |
2741 | if (ret < 0) { | |
2742 | err = ret; | |
2743 | goto out; | |
2744 | } | |
2745 | if (ret == 0) { | |
2746 | if (check_path_shared(root, path)) | |
2747 | goto out; | |
2748 | } else { | |
2749 | check_link = 0; | |
2750 | } | |
2751 | btrfs_release_path(root, path); | |
2752 | ||
2753 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2754 | ret = btrfs_lookup_file_extent(trans, root, path, | |
2755 | inode->i_ino, (u64)-1, 0); | |
2756 | if (ret < 0) { | |
2757 | err = ret; | |
2758 | goto out; | |
2759 | } | |
2760 | BUG_ON(ret == 0); | |
2761 | if (check_path_shared(root, path)) | |
2762 | goto out; | |
2763 | btrfs_release_path(root, path); | |
2764 | } | |
2765 | ||
2766 | if (!check_link) { | |
2767 | err = 0; | |
2768 | goto out; | |
2769 | } | |
2770 | ||
2771 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2772 | dentry->d_name.name, dentry->d_name.len, 0); | |
2773 | if (IS_ERR(di)) { | |
2774 | err = PTR_ERR(di); | |
2775 | goto out; | |
2776 | } | |
2777 | if (di) { | |
2778 | if (check_path_shared(root, path)) | |
2779 | goto out; | |
2780 | } else { | |
2781 | err = 0; | |
2782 | goto out; | |
2783 | } | |
2784 | btrfs_release_path(root, path); | |
2785 | ||
2786 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2787 | dentry->d_name.name, dentry->d_name.len, | |
2788 | inode->i_ino, dir->i_ino, 0); | |
2789 | if (IS_ERR(ref)) { | |
2790 | err = PTR_ERR(ref); | |
2791 | goto out; | |
2792 | } | |
2793 | BUG_ON(!ref); | |
2794 | if (check_path_shared(root, path)) | |
2795 | goto out; | |
2796 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
2797 | btrfs_release_path(root, path); | |
2798 | ||
2799 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index, | |
2800 | dentry->d_name.name, dentry->d_name.len, 0); | |
2801 | if (IS_ERR(di)) { | |
2802 | err = PTR_ERR(di); | |
2803 | goto out; | |
2804 | } | |
2805 | BUG_ON(ret == -ENOENT); | |
2806 | if (check_path_shared(root, path)) | |
2807 | goto out; | |
2808 | ||
2809 | err = 0; | |
2810 | out: | |
2811 | btrfs_free_path(path); | |
2812 | if (err) { | |
2813 | btrfs_end_transaction(trans, root); | |
2814 | root->fs_info->enospc_unlink = 0; | |
2815 | return ERR_PTR(err); | |
2816 | } | |
2817 | ||
2818 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2819 | return trans; | |
2820 | } | |
2821 | ||
2822 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2823 | struct btrfs_root *root) | |
2824 | { | |
2825 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
2826 | BUG_ON(!root->fs_info->enospc_unlink); | |
2827 | root->fs_info->enospc_unlink = 0; | |
2828 | } | |
2829 | btrfs_end_transaction_throttle(trans, root); | |
2830 | } | |
2831 | ||
2832 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2833 | { | |
2834 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2835 | struct btrfs_trans_handle *trans; | |
2836 | struct inode *inode = dentry->d_inode; | |
2837 | int ret; | |
2838 | unsigned long nr = 0; | |
2839 | ||
2840 | trans = __unlink_start_trans(dir, dentry); | |
2841 | if (IS_ERR(trans)) | |
2842 | return PTR_ERR(trans); | |
5f39d397 | 2843 | |
39279cc3 | 2844 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2845 | |
2846 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2847 | ||
e02119d5 CM |
2848 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2849 | dentry->d_name.name, dentry->d_name.len); | |
a22285a6 | 2850 | BUG_ON(ret); |
7b128766 | 2851 | |
a22285a6 | 2852 | if (inode->i_nlink == 0) { |
7b128766 | 2853 | ret = btrfs_orphan_add(trans, inode); |
a22285a6 YZ |
2854 | BUG_ON(ret); |
2855 | } | |
7b128766 | 2856 | |
d3c2fdcf | 2857 | nr = trans->blocks_used; |
a22285a6 | 2858 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2859 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2860 | return ret; |
2861 | } | |
2862 | ||
4df27c4d YZ |
2863 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2864 | struct btrfs_root *root, | |
2865 | struct inode *dir, u64 objectid, | |
2866 | const char *name, int name_len) | |
2867 | { | |
2868 | struct btrfs_path *path; | |
2869 | struct extent_buffer *leaf; | |
2870 | struct btrfs_dir_item *di; | |
2871 | struct btrfs_key key; | |
2872 | u64 index; | |
2873 | int ret; | |
2874 | ||
2875 | path = btrfs_alloc_path(); | |
2876 | if (!path) | |
2877 | return -ENOMEM; | |
2878 | ||
2879 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2880 | name, name_len, -1); | |
2881 | BUG_ON(!di || IS_ERR(di)); | |
2882 | ||
2883 | leaf = path->nodes[0]; | |
2884 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2885 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2886 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2887 | BUG_ON(ret); | |
2888 | btrfs_release_path(root, path); | |
2889 | ||
2890 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2891 | objectid, root->root_key.objectid, | |
2892 | dir->i_ino, &index, name, name_len); | |
2893 | if (ret < 0) { | |
2894 | BUG_ON(ret != -ENOENT); | |
2895 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2896 | name, name_len); | |
2897 | BUG_ON(!di || IS_ERR(di)); | |
2898 | ||
2899 | leaf = path->nodes[0]; | |
2900 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2901 | btrfs_release_path(root, path); | |
2902 | index = key.offset; | |
2903 | } | |
2904 | ||
2905 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2906 | index, name, name_len, -1); | |
2907 | BUG_ON(!di || IS_ERR(di)); | |
2908 | ||
2909 | leaf = path->nodes[0]; | |
2910 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2911 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2912 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2913 | BUG_ON(ret); | |
2914 | btrfs_release_path(root, path); | |
2915 | ||
2916 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2917 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2918 | ret = btrfs_update_inode(trans, root, dir); | |
2919 | BUG_ON(ret); | |
2920 | dir->i_sb->s_dirt = 1; | |
2921 | ||
2922 | btrfs_free_path(path); | |
2923 | return 0; | |
2924 | } | |
2925 | ||
39279cc3 CM |
2926 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2927 | { | |
2928 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2929 | int err = 0; |
39279cc3 | 2930 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 2931 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2932 | unsigned long nr = 0; |
39279cc3 | 2933 | |
3394e160 | 2934 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2935 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2936 | return -ENOTEMPTY; |
2937 | ||
a22285a6 YZ |
2938 | trans = __unlink_start_trans(dir, dentry); |
2939 | if (IS_ERR(trans)) | |
5df6a9f6 | 2940 | return PTR_ERR(trans); |
5df6a9f6 | 2941 | |
39279cc3 | 2942 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 2943 | |
4df27c4d YZ |
2944 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2945 | err = btrfs_unlink_subvol(trans, root, dir, | |
2946 | BTRFS_I(inode)->location.objectid, | |
2947 | dentry->d_name.name, | |
2948 | dentry->d_name.len); | |
2949 | goto out; | |
2950 | } | |
2951 | ||
7b128766 JB |
2952 | err = btrfs_orphan_add(trans, inode); |
2953 | if (err) | |
4df27c4d | 2954 | goto out; |
7b128766 | 2955 | |
39279cc3 | 2956 | /* now the directory is empty */ |
e02119d5 CM |
2957 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2958 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2959 | if (!err) |
dbe674a9 | 2960 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2961 | out: |
d3c2fdcf | 2962 | nr = trans->blocks_used; |
a22285a6 | 2963 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2964 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2965 | |
39279cc3 CM |
2966 | return err; |
2967 | } | |
2968 | ||
d20f7043 | 2969 | #if 0 |
323ac95b CM |
2970 | /* |
2971 | * when truncating bytes in a file, it is possible to avoid reading | |
2972 | * the leaves that contain only checksum items. This can be the | |
2973 | * majority of the IO required to delete a large file, but it must | |
2974 | * be done carefully. | |
2975 | * | |
2976 | * The keys in the level just above the leaves are checked to make sure | |
2977 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2978 | * after the new size. | |
2979 | * | |
2980 | * Then the key for the next leaf is checked to make sure it also has | |
2981 | * a checksum item for the same file. If it does, we know our target leaf | |
2982 | * contains only checksum items, and it can be safely freed without reading | |
2983 | * it. | |
2984 | * | |
2985 | * This is just an optimization targeted at large files. It may do | |
2986 | * nothing. It will return 0 unless things went badly. | |
2987 | */ | |
2988 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2989 | struct btrfs_root *root, | |
2990 | struct btrfs_path *path, | |
2991 | struct inode *inode, u64 new_size) | |
2992 | { | |
2993 | struct btrfs_key key; | |
2994 | int ret; | |
2995 | int nritems; | |
2996 | struct btrfs_key found_key; | |
2997 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2998 | struct btrfs_leaf_ref *ref; |
2999 | u64 leaf_gen; | |
3000 | u64 leaf_start; | |
323ac95b CM |
3001 | |
3002 | path->lowest_level = 1; | |
3003 | key.objectid = inode->i_ino; | |
3004 | key.type = BTRFS_CSUM_ITEM_KEY; | |
3005 | key.offset = new_size; | |
3006 | again: | |
3007 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3008 | if (ret < 0) | |
3009 | goto out; | |
3010 | ||
3011 | if (path->nodes[1] == NULL) { | |
3012 | ret = 0; | |
3013 | goto out; | |
3014 | } | |
3015 | ret = 0; | |
3016 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
3017 | nritems = btrfs_header_nritems(path->nodes[1]); | |
3018 | ||
3019 | if (!nritems) | |
3020 | goto out; | |
3021 | ||
3022 | if (path->slots[1] >= nritems) | |
3023 | goto next_node; | |
3024 | ||
3025 | /* did we find a key greater than anything we want to delete? */ | |
3026 | if (found_key.objectid > inode->i_ino || | |
3027 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
3028 | goto out; | |
3029 | ||
3030 | /* we check the next key in the node to make sure the leave contains | |
3031 | * only checksum items. This comparison doesn't work if our | |
3032 | * leaf is the last one in the node | |
3033 | */ | |
3034 | if (path->slots[1] + 1 >= nritems) { | |
3035 | next_node: | |
3036 | /* search forward from the last key in the node, this | |
3037 | * will bring us into the next node in the tree | |
3038 | */ | |
3039 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
3040 | ||
3041 | /* unlikely, but we inc below, so check to be safe */ | |
3042 | if (found_key.offset == (u64)-1) | |
3043 | goto out; | |
3044 | ||
3045 | /* search_forward needs a path with locks held, do the | |
3046 | * search again for the original key. It is possible | |
3047 | * this will race with a balance and return a path that | |
3048 | * we could modify, but this drop is just an optimization | |
3049 | * and is allowed to miss some leaves. | |
3050 | */ | |
3051 | btrfs_release_path(root, path); | |
3052 | found_key.offset++; | |
3053 | ||
3054 | /* setup a max key for search_forward */ | |
3055 | other_key.offset = (u64)-1; | |
3056 | other_key.type = key.type; | |
3057 | other_key.objectid = key.objectid; | |
3058 | ||
3059 | path->keep_locks = 1; | |
3060 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
3061 | path, 0, 0); | |
3062 | path->keep_locks = 0; | |
3063 | if (ret || found_key.objectid != key.objectid || | |
3064 | found_key.type != key.type) { | |
3065 | ret = 0; | |
3066 | goto out; | |
3067 | } | |
3068 | ||
3069 | key.offset = found_key.offset; | |
3070 | btrfs_release_path(root, path); | |
3071 | cond_resched(); | |
3072 | goto again; | |
3073 | } | |
3074 | ||
3075 | /* we know there's one more slot after us in the tree, | |
3076 | * read that key so we can verify it is also a checksum item | |
3077 | */ | |
3078 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
3079 | ||
3080 | if (found_key.objectid < inode->i_ino) | |
3081 | goto next_key; | |
3082 | ||
3083 | if (found_key.type != key.type || found_key.offset < new_size) | |
3084 | goto next_key; | |
3085 | ||
3086 | /* | |
3087 | * if the key for the next leaf isn't a csum key from this objectid, | |
3088 | * we can't be sure there aren't good items inside this leaf. | |
3089 | * Bail out | |
3090 | */ | |
3091 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
3092 | goto out; | |
3093 | ||
5b84e8d6 YZ |
3094 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
3095 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
3096 | /* |
3097 | * it is safe to delete this leaf, it contains only | |
3098 | * csum items from this inode at an offset >= new_size | |
3099 | */ | |
5b84e8d6 | 3100 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
3101 | BUG_ON(ret); |
3102 | ||
5b84e8d6 YZ |
3103 | if (root->ref_cows && leaf_gen < trans->transid) { |
3104 | ref = btrfs_alloc_leaf_ref(root, 0); | |
3105 | if (ref) { | |
3106 | ref->root_gen = root->root_key.offset; | |
3107 | ref->bytenr = leaf_start; | |
3108 | ref->owner = 0; | |
3109 | ref->generation = leaf_gen; | |
3110 | ref->nritems = 0; | |
3111 | ||
bd56b302 CM |
3112 | btrfs_sort_leaf_ref(ref); |
3113 | ||
5b84e8d6 YZ |
3114 | ret = btrfs_add_leaf_ref(root, ref, 0); |
3115 | WARN_ON(ret); | |
3116 | btrfs_free_leaf_ref(root, ref); | |
3117 | } else { | |
3118 | WARN_ON(1); | |
3119 | } | |
3120 | } | |
323ac95b CM |
3121 | next_key: |
3122 | btrfs_release_path(root, path); | |
3123 | ||
3124 | if (other_key.objectid == inode->i_ino && | |
3125 | other_key.type == key.type && other_key.offset > key.offset) { | |
3126 | key.offset = other_key.offset; | |
3127 | cond_resched(); | |
3128 | goto again; | |
3129 | } | |
3130 | ret = 0; | |
3131 | out: | |
3132 | /* fixup any changes we've made to the path */ | |
3133 | path->lowest_level = 0; | |
3134 | path->keep_locks = 0; | |
3135 | btrfs_release_path(root, path); | |
3136 | return ret; | |
3137 | } | |
3138 | ||
d20f7043 CM |
3139 | #endif |
3140 | ||
39279cc3 CM |
3141 | /* |
3142 | * this can truncate away extent items, csum items and directory items. | |
3143 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3144 | * any higher than new_size |
39279cc3 CM |
3145 | * |
3146 | * csum items that cross the new i_size are truncated to the new size | |
3147 | * as well. | |
7b128766 JB |
3148 | * |
3149 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3150 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3151 | */ |
8082510e YZ |
3152 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3153 | struct btrfs_root *root, | |
3154 | struct inode *inode, | |
3155 | u64 new_size, u32 min_type) | |
39279cc3 | 3156 | { |
39279cc3 | 3157 | struct btrfs_path *path; |
5f39d397 | 3158 | struct extent_buffer *leaf; |
39279cc3 | 3159 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3160 | struct btrfs_key key; |
3161 | struct btrfs_key found_key; | |
39279cc3 | 3162 | u64 extent_start = 0; |
db94535d | 3163 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3164 | u64 extent_offset = 0; |
39279cc3 | 3165 | u64 item_end = 0; |
8082510e YZ |
3166 | u64 mask = root->sectorsize - 1; |
3167 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3168 | int found_extent; |
3169 | int del_item; | |
85e21bac CM |
3170 | int pending_del_nr = 0; |
3171 | int pending_del_slot = 0; | |
179e29e4 | 3172 | int extent_type = -1; |
771ed689 | 3173 | int encoding; |
8082510e YZ |
3174 | int ret; |
3175 | int err = 0; | |
3176 | ||
3177 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3178 | |
e02119d5 | 3179 | if (root->ref_cows) |
5b21f2ed | 3180 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3181 | |
39279cc3 CM |
3182 | path = btrfs_alloc_path(); |
3183 | BUG_ON(!path); | |
33c17ad5 | 3184 | path->reada = -1; |
5f39d397 | 3185 | |
39279cc3 CM |
3186 | key.objectid = inode->i_ino; |
3187 | key.offset = (u64)-1; | |
5f39d397 CM |
3188 | key.type = (u8)-1; |
3189 | ||
85e21bac | 3190 | search_again: |
b9473439 | 3191 | path->leave_spinning = 1; |
85e21bac | 3192 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3193 | if (ret < 0) { |
3194 | err = ret; | |
3195 | goto out; | |
3196 | } | |
d397712b | 3197 | |
85e21bac | 3198 | if (ret > 0) { |
e02119d5 CM |
3199 | /* there are no items in the tree for us to truncate, we're |
3200 | * done | |
3201 | */ | |
8082510e YZ |
3202 | if (path->slots[0] == 0) |
3203 | goto out; | |
85e21bac CM |
3204 | path->slots[0]--; |
3205 | } | |
3206 | ||
d397712b | 3207 | while (1) { |
39279cc3 | 3208 | fi = NULL; |
5f39d397 CM |
3209 | leaf = path->nodes[0]; |
3210 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3211 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 3212 | encoding = 0; |
39279cc3 | 3213 | |
5f39d397 | 3214 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 3215 | break; |
5f39d397 | 3216 | |
85e21bac | 3217 | if (found_type < min_type) |
39279cc3 CM |
3218 | break; |
3219 | ||
5f39d397 | 3220 | item_end = found_key.offset; |
39279cc3 | 3221 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3222 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3223 | struct btrfs_file_extent_item); |
179e29e4 | 3224 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3225 | encoding = btrfs_file_extent_compression(leaf, fi); |
3226 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3227 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3228 | ||
179e29e4 | 3229 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3230 | item_end += |
db94535d | 3231 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3232 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3233 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3234 | fi); |
39279cc3 | 3235 | } |
008630c1 | 3236 | item_end--; |
39279cc3 | 3237 | } |
8082510e YZ |
3238 | if (found_type > min_type) { |
3239 | del_item = 1; | |
3240 | } else { | |
3241 | if (item_end < new_size) | |
b888db2b | 3242 | break; |
8082510e YZ |
3243 | if (found_key.offset >= new_size) |
3244 | del_item = 1; | |
3245 | else | |
3246 | del_item = 0; | |
39279cc3 | 3247 | } |
39279cc3 | 3248 | found_extent = 0; |
39279cc3 | 3249 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3250 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3251 | goto delete; | |
3252 | ||
3253 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3254 | u64 num_dec; |
db94535d | 3255 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3256 | if (!del_item && !encoding) { |
db94535d CM |
3257 | u64 orig_num_bytes = |
3258 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3259 | extent_num_bytes = new_size - |
5f39d397 | 3260 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3261 | extent_num_bytes = extent_num_bytes & |
3262 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3263 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3264 | extent_num_bytes); | |
3265 | num_dec = (orig_num_bytes - | |
9069218d | 3266 | extent_num_bytes); |
e02119d5 | 3267 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3268 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3269 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3270 | } else { |
db94535d CM |
3271 | extent_num_bytes = |
3272 | btrfs_file_extent_disk_num_bytes(leaf, | |
3273 | fi); | |
5d4f98a2 YZ |
3274 | extent_offset = found_key.offset - |
3275 | btrfs_file_extent_offset(leaf, fi); | |
3276 | ||
39279cc3 | 3277 | /* FIXME blocksize != 4096 */ |
9069218d | 3278 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3279 | if (extent_start != 0) { |
3280 | found_extent = 1; | |
e02119d5 | 3281 | if (root->ref_cows) |
a76a3cd4 | 3282 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3283 | } |
39279cc3 | 3284 | } |
9069218d | 3285 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3286 | /* |
3287 | * we can't truncate inline items that have had | |
3288 | * special encodings | |
3289 | */ | |
3290 | if (!del_item && | |
3291 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3292 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3293 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3294 | u32 size = new_size - found_key.offset; |
3295 | ||
3296 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3297 | inode_sub_bytes(inode, item_end + 1 - |
3298 | new_size); | |
e02119d5 CM |
3299 | } |
3300 | size = | |
3301 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3302 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3303 | size, 1); |
9069218d | 3304 | BUG_ON(ret); |
e02119d5 | 3305 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3306 | inode_sub_bytes(inode, item_end + 1 - |
3307 | found_key.offset); | |
9069218d | 3308 | } |
39279cc3 | 3309 | } |
179e29e4 | 3310 | delete: |
39279cc3 | 3311 | if (del_item) { |
85e21bac CM |
3312 | if (!pending_del_nr) { |
3313 | /* no pending yet, add ourselves */ | |
3314 | pending_del_slot = path->slots[0]; | |
3315 | pending_del_nr = 1; | |
3316 | } else if (pending_del_nr && | |
3317 | path->slots[0] + 1 == pending_del_slot) { | |
3318 | /* hop on the pending chunk */ | |
3319 | pending_del_nr++; | |
3320 | pending_del_slot = path->slots[0]; | |
3321 | } else { | |
d397712b | 3322 | BUG(); |
85e21bac | 3323 | } |
39279cc3 CM |
3324 | } else { |
3325 | break; | |
3326 | } | |
5d4f98a2 | 3327 | if (found_extent && root->ref_cows) { |
b9473439 | 3328 | btrfs_set_path_blocking(path); |
39279cc3 | 3329 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3330 | extent_num_bytes, 0, |
3331 | btrfs_header_owner(leaf), | |
3332 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3333 | BUG_ON(ret); |
3334 | } | |
85e21bac | 3335 | |
8082510e YZ |
3336 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3337 | break; | |
3338 | ||
3339 | if (path->slots[0] == 0 || | |
3340 | path->slots[0] != pending_del_slot) { | |
3341 | if (root->ref_cows) { | |
3342 | err = -EAGAIN; | |
3343 | goto out; | |
3344 | } | |
3345 | if (pending_del_nr) { | |
3346 | ret = btrfs_del_items(trans, root, path, | |
3347 | pending_del_slot, | |
3348 | pending_del_nr); | |
3349 | BUG_ON(ret); | |
3350 | pending_del_nr = 0; | |
3351 | } | |
85e21bac CM |
3352 | btrfs_release_path(root, path); |
3353 | goto search_again; | |
8082510e YZ |
3354 | } else { |
3355 | path->slots[0]--; | |
85e21bac | 3356 | } |
39279cc3 | 3357 | } |
8082510e | 3358 | out: |
85e21bac CM |
3359 | if (pending_del_nr) { |
3360 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3361 | pending_del_nr); | |
d68fc57b | 3362 | BUG_ON(ret); |
85e21bac | 3363 | } |
39279cc3 | 3364 | btrfs_free_path(path); |
8082510e | 3365 | return err; |
39279cc3 CM |
3366 | } |
3367 | ||
3368 | /* | |
3369 | * taken from block_truncate_page, but does cow as it zeros out | |
3370 | * any bytes left in the last page in the file. | |
3371 | */ | |
3372 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3373 | { | |
3374 | struct inode *inode = mapping->host; | |
db94535d | 3375 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3376 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3377 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3378 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3379 | char *kaddr; |
db94535d | 3380 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3381 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3382 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3383 | struct page *page; | |
39279cc3 | 3384 | int ret = 0; |
a52d9a80 | 3385 | u64 page_start; |
e6dcd2dc | 3386 | u64 page_end; |
39279cc3 CM |
3387 | |
3388 | if ((offset & (blocksize - 1)) == 0) | |
3389 | goto out; | |
0ca1f7ce | 3390 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3391 | if (ret) |
3392 | goto out; | |
39279cc3 CM |
3393 | |
3394 | ret = -ENOMEM; | |
211c17f5 | 3395 | again: |
39279cc3 | 3396 | page = grab_cache_page(mapping, index); |
5d5e103a | 3397 | if (!page) { |
0ca1f7ce | 3398 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3399 | goto out; |
5d5e103a | 3400 | } |
e6dcd2dc CM |
3401 | |
3402 | page_start = page_offset(page); | |
3403 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3404 | ||
39279cc3 | 3405 | if (!PageUptodate(page)) { |
9ebefb18 | 3406 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3407 | lock_page(page); |
211c17f5 CM |
3408 | if (page->mapping != mapping) { |
3409 | unlock_page(page); | |
3410 | page_cache_release(page); | |
3411 | goto again; | |
3412 | } | |
39279cc3 CM |
3413 | if (!PageUptodate(page)) { |
3414 | ret = -EIO; | |
89642229 | 3415 | goto out_unlock; |
39279cc3 CM |
3416 | } |
3417 | } | |
211c17f5 | 3418 | wait_on_page_writeback(page); |
e6dcd2dc | 3419 | |
2ac55d41 JB |
3420 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3421 | GFP_NOFS); | |
e6dcd2dc CM |
3422 | set_page_extent_mapped(page); |
3423 | ||
3424 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3425 | if (ordered) { | |
2ac55d41 JB |
3426 | unlock_extent_cached(io_tree, page_start, page_end, |
3427 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3428 | unlock_page(page); |
3429 | page_cache_release(page); | |
eb84ae03 | 3430 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3431 | btrfs_put_ordered_extent(ordered); |
3432 | goto again; | |
3433 | } | |
3434 | ||
2ac55d41 | 3435 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3436 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3437 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3438 | |
2ac55d41 JB |
3439 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3440 | &cached_state); | |
9ed74f2d | 3441 | if (ret) { |
2ac55d41 JB |
3442 | unlock_extent_cached(io_tree, page_start, page_end, |
3443 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3444 | goto out_unlock; |
3445 | } | |
3446 | ||
e6dcd2dc CM |
3447 | ret = 0; |
3448 | if (offset != PAGE_CACHE_SIZE) { | |
3449 | kaddr = kmap(page); | |
3450 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3451 | flush_dcache_page(page); | |
3452 | kunmap(page); | |
3453 | } | |
247e743c | 3454 | ClearPageChecked(page); |
e6dcd2dc | 3455 | set_page_dirty(page); |
2ac55d41 JB |
3456 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3457 | GFP_NOFS); | |
39279cc3 | 3458 | |
89642229 | 3459 | out_unlock: |
5d5e103a | 3460 | if (ret) |
0ca1f7ce | 3461 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3462 | unlock_page(page); |
3463 | page_cache_release(page); | |
3464 | out: | |
3465 | return ret; | |
3466 | } | |
3467 | ||
9036c102 | 3468 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3469 | { |
9036c102 YZ |
3470 | struct btrfs_trans_handle *trans; |
3471 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3472 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3473 | struct extent_map *em = NULL; |
2ac55d41 | 3474 | struct extent_state *cached_state = NULL; |
9036c102 YZ |
3475 | u64 mask = root->sectorsize - 1; |
3476 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3477 | u64 block_end = (size + mask) & ~mask; | |
3478 | u64 last_byte; | |
3479 | u64 cur_offset; | |
3480 | u64 hole_size; | |
9ed74f2d | 3481 | int err = 0; |
39279cc3 | 3482 | |
9036c102 YZ |
3483 | if (size <= hole_start) |
3484 | return 0; | |
3485 | ||
9036c102 YZ |
3486 | while (1) { |
3487 | struct btrfs_ordered_extent *ordered; | |
3488 | btrfs_wait_ordered_range(inode, hole_start, | |
3489 | block_end - hole_start); | |
2ac55d41 JB |
3490 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3491 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3492 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3493 | if (!ordered) | |
3494 | break; | |
2ac55d41 JB |
3495 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3496 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3497 | btrfs_put_ordered_extent(ordered); |
3498 | } | |
39279cc3 | 3499 | |
9036c102 YZ |
3500 | cur_offset = hole_start; |
3501 | while (1) { | |
3502 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3503 | block_end - cur_offset, 0); | |
3504 | BUG_ON(IS_ERR(em) || !em); | |
3505 | last_byte = min(extent_map_end(em), block_end); | |
3506 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3507 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3508 | u64 hint_byte = 0; |
9036c102 | 3509 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3510 | |
a22285a6 YZ |
3511 | trans = btrfs_start_transaction(root, 2); |
3512 | if (IS_ERR(trans)) { | |
3513 | err = PTR_ERR(trans); | |
9ed74f2d | 3514 | break; |
a22285a6 | 3515 | } |
8082510e YZ |
3516 | btrfs_set_trans_block_group(trans, inode); |
3517 | ||
3518 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3519 | cur_offset + hole_size, | |
3520 | &hint_byte, 1); | |
3521 | BUG_ON(err); | |
3522 | ||
9036c102 YZ |
3523 | err = btrfs_insert_file_extent(trans, root, |
3524 | inode->i_ino, cur_offset, 0, | |
3525 | 0, hole_size, 0, hole_size, | |
3526 | 0, 0, 0); | |
8082510e YZ |
3527 | BUG_ON(err); |
3528 | ||
9036c102 YZ |
3529 | btrfs_drop_extent_cache(inode, hole_start, |
3530 | last_byte - 1, 0); | |
8082510e YZ |
3531 | |
3532 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3533 | } |
3534 | free_extent_map(em); | |
a22285a6 | 3535 | em = NULL; |
9036c102 | 3536 | cur_offset = last_byte; |
8082510e | 3537 | if (cur_offset >= block_end) |
9036c102 YZ |
3538 | break; |
3539 | } | |
1832a6d5 | 3540 | |
a22285a6 | 3541 | free_extent_map(em); |
2ac55d41 JB |
3542 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3543 | GFP_NOFS); | |
9036c102 YZ |
3544 | return err; |
3545 | } | |
39279cc3 | 3546 | |
8082510e YZ |
3547 | static int btrfs_setattr_size(struct inode *inode, struct iattr *attr) |
3548 | { | |
3549 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3550 | struct btrfs_trans_handle *trans; | |
3551 | unsigned long nr; | |
3552 | int ret; | |
3553 | ||
3554 | if (attr->ia_size == inode->i_size) | |
3555 | return 0; | |
3556 | ||
3557 | if (attr->ia_size > inode->i_size) { | |
3558 | unsigned long limit; | |
3559 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
3560 | if (attr->ia_size > inode->i_sb->s_maxbytes) | |
3561 | return -EFBIG; | |
3562 | if (limit != RLIM_INFINITY && attr->ia_size > limit) { | |
3563 | send_sig(SIGXFSZ, current, 0); | |
3564 | return -EFBIG; | |
3565 | } | |
3566 | } | |
3567 | ||
d68fc57b YZ |
3568 | trans = btrfs_start_transaction(root, 5); |
3569 | if (IS_ERR(trans)) | |
3570 | return PTR_ERR(trans); | |
3571 | ||
8082510e YZ |
3572 | btrfs_set_trans_block_group(trans, inode); |
3573 | ||
3574 | ret = btrfs_orphan_add(trans, inode); | |
3575 | BUG_ON(ret); | |
3576 | ||
3577 | nr = trans->blocks_used; | |
3578 | btrfs_end_transaction(trans, root); | |
8082510e YZ |
3579 | btrfs_btree_balance_dirty(root, nr); |
3580 | ||
3581 | if (attr->ia_size > inode->i_size) { | |
3582 | ret = btrfs_cont_expand(inode, attr->ia_size); | |
3583 | if (ret) { | |
3584 | btrfs_truncate(inode); | |
3585 | return ret; | |
3586 | } | |
3587 | ||
3588 | i_size_write(inode, attr->ia_size); | |
3589 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
3590 | ||
d68fc57b YZ |
3591 | trans = btrfs_start_transaction(root, 0); |
3592 | BUG_ON(IS_ERR(trans)); | |
8082510e | 3593 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b YZ |
3594 | trans->block_rsv = root->orphan_block_rsv; |
3595 | BUG_ON(!trans->block_rsv); | |
8082510e YZ |
3596 | |
3597 | ret = btrfs_update_inode(trans, root, inode); | |
3598 | BUG_ON(ret); | |
3599 | if (inode->i_nlink > 0) { | |
3600 | ret = btrfs_orphan_del(trans, inode); | |
3601 | BUG_ON(ret); | |
3602 | } | |
3603 | nr = trans->blocks_used; | |
3604 | btrfs_end_transaction(trans, root); | |
3605 | btrfs_btree_balance_dirty(root, nr); | |
3606 | return 0; | |
3607 | } | |
3608 | ||
3609 | /* | |
3610 | * We're truncating a file that used to have good data down to | |
3611 | * zero. Make sure it gets into the ordered flush list so that | |
3612 | * any new writes get down to disk quickly. | |
3613 | */ | |
3614 | if (attr->ia_size == 0) | |
3615 | BTRFS_I(inode)->ordered_data_close = 1; | |
3616 | ||
3617 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ | |
3618 | ret = vmtruncate(inode, attr->ia_size); | |
3619 | BUG_ON(ret); | |
3620 | ||
3621 | return 0; | |
3622 | } | |
3623 | ||
9036c102 YZ |
3624 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3625 | { | |
3626 | struct inode *inode = dentry->d_inode; | |
3627 | int err; | |
39279cc3 | 3628 | |
9036c102 YZ |
3629 | err = inode_change_ok(inode, attr); |
3630 | if (err) | |
3631 | return err; | |
2bf5a725 | 3632 | |
5a3f23d5 | 3633 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
8082510e YZ |
3634 | err = btrfs_setattr_size(inode, attr); |
3635 | if (err) | |
3636 | return err; | |
39279cc3 | 3637 | } |
8082510e | 3638 | attr->ia_valid &= ~ATTR_SIZE; |
9036c102 | 3639 | |
8082510e YZ |
3640 | if (attr->ia_valid) |
3641 | err = inode_setattr(inode, attr); | |
33268eaf JB |
3642 | |
3643 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3644 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3645 | return err; |
3646 | } | |
61295eb8 | 3647 | |
39279cc3 CM |
3648 | void btrfs_delete_inode(struct inode *inode) |
3649 | { | |
3650 | struct btrfs_trans_handle *trans; | |
3651 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3652 | unsigned long nr; |
39279cc3 CM |
3653 | int ret; |
3654 | ||
3655 | truncate_inode_pages(&inode->i_data, 0); | |
3656 | if (is_bad_inode(inode)) { | |
7b128766 | 3657 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3658 | goto no_delete; |
3659 | } | |
4a096752 | 3660 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3661 | |
c71bf099 YZ |
3662 | if (root->fs_info->log_root_recovering) { |
3663 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3664 | goto no_delete; | |
3665 | } | |
3666 | ||
76dda93c YZ |
3667 | if (inode->i_nlink > 0) { |
3668 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3669 | goto no_delete; | |
3670 | } | |
3671 | ||
dbe674a9 | 3672 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3673 | |
8082510e | 3674 | while (1) { |
d68fc57b YZ |
3675 | trans = btrfs_start_transaction(root, 0); |
3676 | BUG_ON(IS_ERR(trans)); | |
8082510e | 3677 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b YZ |
3678 | trans->block_rsv = root->orphan_block_rsv; |
3679 | ||
3680 | ret = btrfs_block_rsv_check(trans, root, | |
3681 | root->orphan_block_rsv, 0, 5); | |
3682 | if (ret) { | |
3683 | BUG_ON(ret != -EAGAIN); | |
3684 | ret = btrfs_commit_transaction(trans, root); | |
3685 | BUG_ON(ret); | |
3686 | continue; | |
3687 | } | |
7b128766 | 3688 | |
d68fc57b | 3689 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3690 | if (ret != -EAGAIN) |
3691 | break; | |
85e21bac | 3692 | |
8082510e YZ |
3693 | nr = trans->blocks_used; |
3694 | btrfs_end_transaction(trans, root); | |
3695 | trans = NULL; | |
3696 | btrfs_btree_balance_dirty(root, nr); | |
d68fc57b | 3697 | |
8082510e | 3698 | } |
5f39d397 | 3699 | |
8082510e YZ |
3700 | if (ret == 0) { |
3701 | ret = btrfs_orphan_del(trans, inode); | |
3702 | BUG_ON(ret); | |
3703 | } | |
54aa1f4d | 3704 | |
d3c2fdcf | 3705 | nr = trans->blocks_used; |
54aa1f4d | 3706 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3707 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3708 | no_delete: |
3709 | clear_inode(inode); | |
8082510e | 3710 | return; |
39279cc3 CM |
3711 | } |
3712 | ||
3713 | /* | |
3714 | * this returns the key found in the dir entry in the location pointer. | |
3715 | * If no dir entries were found, location->objectid is 0. | |
3716 | */ | |
3717 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3718 | struct btrfs_key *location) | |
3719 | { | |
3720 | const char *name = dentry->d_name.name; | |
3721 | int namelen = dentry->d_name.len; | |
3722 | struct btrfs_dir_item *di; | |
3723 | struct btrfs_path *path; | |
3724 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3725 | int ret = 0; |
39279cc3 CM |
3726 | |
3727 | path = btrfs_alloc_path(); | |
3728 | BUG_ON(!path); | |
3954401f | 3729 | |
39279cc3 CM |
3730 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3731 | namelen, 0); | |
0d9f7f3e Y |
3732 | if (IS_ERR(di)) |
3733 | ret = PTR_ERR(di); | |
d397712b CM |
3734 | |
3735 | if (!di || IS_ERR(di)) | |
3954401f | 3736 | goto out_err; |
d397712b | 3737 | |
5f39d397 | 3738 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3739 | out: |
39279cc3 CM |
3740 | btrfs_free_path(path); |
3741 | return ret; | |
3954401f CM |
3742 | out_err: |
3743 | location->objectid = 0; | |
3744 | goto out; | |
39279cc3 CM |
3745 | } |
3746 | ||
3747 | /* | |
3748 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3749 | * needs to be changed to reflect the root directory of the tree root. This | |
3750 | * is kind of like crossing a mount point. | |
3751 | */ | |
3752 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3753 | struct inode *dir, |
3754 | struct dentry *dentry, | |
3755 | struct btrfs_key *location, | |
3756 | struct btrfs_root **sub_root) | |
39279cc3 | 3757 | { |
4df27c4d YZ |
3758 | struct btrfs_path *path; |
3759 | struct btrfs_root *new_root; | |
3760 | struct btrfs_root_ref *ref; | |
3761 | struct extent_buffer *leaf; | |
3762 | int ret; | |
3763 | int err = 0; | |
39279cc3 | 3764 | |
4df27c4d YZ |
3765 | path = btrfs_alloc_path(); |
3766 | if (!path) { | |
3767 | err = -ENOMEM; | |
3768 | goto out; | |
3769 | } | |
39279cc3 | 3770 | |
4df27c4d YZ |
3771 | err = -ENOENT; |
3772 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3773 | BTRFS_I(dir)->root->root_key.objectid, | |
3774 | location->objectid); | |
3775 | if (ret) { | |
3776 | if (ret < 0) | |
3777 | err = ret; | |
3778 | goto out; | |
3779 | } | |
39279cc3 | 3780 | |
4df27c4d YZ |
3781 | leaf = path->nodes[0]; |
3782 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3783 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3784 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3785 | goto out; | |
39279cc3 | 3786 | |
4df27c4d YZ |
3787 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3788 | (unsigned long)(ref + 1), | |
3789 | dentry->d_name.len); | |
3790 | if (ret) | |
3791 | goto out; | |
3792 | ||
3793 | btrfs_release_path(root->fs_info->tree_root, path); | |
3794 | ||
3795 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3796 | if (IS_ERR(new_root)) { | |
3797 | err = PTR_ERR(new_root); | |
3798 | goto out; | |
3799 | } | |
3800 | ||
3801 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3802 | err = -ENOENT; | |
3803 | goto out; | |
3804 | } | |
3805 | ||
3806 | *sub_root = new_root; | |
3807 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3808 | location->type = BTRFS_INODE_ITEM_KEY; | |
3809 | location->offset = 0; | |
3810 | err = 0; | |
3811 | out: | |
3812 | btrfs_free_path(path); | |
3813 | return err; | |
39279cc3 CM |
3814 | } |
3815 | ||
5d4f98a2 YZ |
3816 | static void inode_tree_add(struct inode *inode) |
3817 | { | |
3818 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3819 | struct btrfs_inode *entry; | |
03e860bd FNP |
3820 | struct rb_node **p; |
3821 | struct rb_node *parent; | |
03e860bd FNP |
3822 | again: |
3823 | p = &root->inode_tree.rb_node; | |
3824 | parent = NULL; | |
5d4f98a2 | 3825 | |
76dda93c YZ |
3826 | if (hlist_unhashed(&inode->i_hash)) |
3827 | return; | |
3828 | ||
5d4f98a2 YZ |
3829 | spin_lock(&root->inode_lock); |
3830 | while (*p) { | |
3831 | parent = *p; | |
3832 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3833 | ||
3834 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3835 | p = &parent->rb_left; |
5d4f98a2 | 3836 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3837 | p = &parent->rb_right; |
5d4f98a2 YZ |
3838 | else { |
3839 | WARN_ON(!(entry->vfs_inode.i_state & | |
3840 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd FNP |
3841 | rb_erase(parent, &root->inode_tree); |
3842 | RB_CLEAR_NODE(parent); | |
3843 | spin_unlock(&root->inode_lock); | |
3844 | goto again; | |
5d4f98a2 YZ |
3845 | } |
3846 | } | |
3847 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3848 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3849 | spin_unlock(&root->inode_lock); | |
3850 | } | |
3851 | ||
3852 | static void inode_tree_del(struct inode *inode) | |
3853 | { | |
3854 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3855 | int empty = 0; |
5d4f98a2 | 3856 | |
03e860bd | 3857 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3858 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3859 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3860 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3861 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3862 | } |
03e860bd | 3863 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3864 | |
3865 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3866 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3867 | spin_lock(&root->inode_lock); | |
3868 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3869 | spin_unlock(&root->inode_lock); | |
3870 | if (empty) | |
3871 | btrfs_add_dead_root(root); | |
3872 | } | |
3873 | } | |
3874 | ||
3875 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3876 | { | |
3877 | struct rb_node *node; | |
3878 | struct rb_node *prev; | |
3879 | struct btrfs_inode *entry; | |
3880 | struct inode *inode; | |
3881 | u64 objectid = 0; | |
3882 | ||
3883 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3884 | ||
3885 | spin_lock(&root->inode_lock); | |
3886 | again: | |
3887 | node = root->inode_tree.rb_node; | |
3888 | prev = NULL; | |
3889 | while (node) { | |
3890 | prev = node; | |
3891 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3892 | ||
3893 | if (objectid < entry->vfs_inode.i_ino) | |
3894 | node = node->rb_left; | |
3895 | else if (objectid > entry->vfs_inode.i_ino) | |
3896 | node = node->rb_right; | |
3897 | else | |
3898 | break; | |
3899 | } | |
3900 | if (!node) { | |
3901 | while (prev) { | |
3902 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3903 | if (objectid <= entry->vfs_inode.i_ino) { | |
3904 | node = prev; | |
3905 | break; | |
3906 | } | |
3907 | prev = rb_next(prev); | |
3908 | } | |
3909 | } | |
3910 | while (node) { | |
3911 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3912 | objectid = entry->vfs_inode.i_ino + 1; | |
3913 | inode = igrab(&entry->vfs_inode); | |
3914 | if (inode) { | |
3915 | spin_unlock(&root->inode_lock); | |
3916 | if (atomic_read(&inode->i_count) > 1) | |
3917 | d_prune_aliases(inode); | |
3918 | /* | |
3919 | * btrfs_drop_inode will remove it from | |
3920 | * the inode cache when its usage count | |
3921 | * hits zero. | |
3922 | */ | |
3923 | iput(inode); | |
3924 | cond_resched(); | |
3925 | spin_lock(&root->inode_lock); | |
3926 | goto again; | |
3927 | } | |
3928 | ||
3929 | if (cond_resched_lock(&root->inode_lock)) | |
3930 | goto again; | |
3931 | ||
3932 | node = rb_next(node); | |
3933 | } | |
3934 | spin_unlock(&root->inode_lock); | |
3935 | return 0; | |
5d4f98a2 YZ |
3936 | } |
3937 | ||
e02119d5 CM |
3938 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
3939 | { | |
3940 | struct btrfs_iget_args *args = p; | |
3941 | inode->i_ino = args->ino; | |
e02119d5 | 3942 | BTRFS_I(inode)->root = args->root; |
6a63209f | 3943 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3944 | return 0; |
3945 | } | |
3946 | ||
3947 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3948 | { | |
3949 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3950 | return args->ino == inode->i_ino && |
3951 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3952 | } |
3953 | ||
5d4f98a2 YZ |
3954 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3955 | u64 objectid, | |
3956 | struct btrfs_root *root) | |
39279cc3 CM |
3957 | { |
3958 | struct inode *inode; | |
3959 | struct btrfs_iget_args args; | |
3960 | args.ino = objectid; | |
3961 | args.root = root; | |
3962 | ||
3963 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3964 | btrfs_init_locked_inode, | |
3965 | (void *)&args); | |
3966 | return inode; | |
3967 | } | |
3968 | ||
1a54ef8c BR |
3969 | /* Get an inode object given its location and corresponding root. |
3970 | * Returns in *is_new if the inode was read from disk | |
3971 | */ | |
3972 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 3973 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
3974 | { |
3975 | struct inode *inode; | |
3976 | ||
3977 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3978 | if (!inode) | |
5d4f98a2 | 3979 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3980 | |
3981 | if (inode->i_state & I_NEW) { | |
3982 | BTRFS_I(inode)->root = root; | |
3983 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3984 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3985 | |
3986 | inode_tree_add(inode); | |
1a54ef8c | 3987 | unlock_new_inode(inode); |
73f73415 JB |
3988 | if (new) |
3989 | *new = 1; | |
1a54ef8c BR |
3990 | } |
3991 | ||
3992 | return inode; | |
3993 | } | |
3994 | ||
4df27c4d YZ |
3995 | static struct inode *new_simple_dir(struct super_block *s, |
3996 | struct btrfs_key *key, | |
3997 | struct btrfs_root *root) | |
3998 | { | |
3999 | struct inode *inode = new_inode(s); | |
4000 | ||
4001 | if (!inode) | |
4002 | return ERR_PTR(-ENOMEM); | |
4003 | ||
4df27c4d YZ |
4004 | BTRFS_I(inode)->root = root; |
4005 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
4006 | BTRFS_I(inode)->dummy_inode = 1; | |
4007 | ||
4008 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
4009 | inode->i_op = &simple_dir_inode_operations; | |
4010 | inode->i_fop = &simple_dir_operations; | |
4011 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4012 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4013 | ||
4014 | return inode; | |
4015 | } | |
4016 | ||
3de4586c | 4017 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4018 | { |
d397712b | 4019 | struct inode *inode; |
4df27c4d | 4020 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4021 | struct btrfs_root *sub_root = root; |
4022 | struct btrfs_key location; | |
76dda93c | 4023 | int index; |
5d4f98a2 | 4024 | int ret; |
39279cc3 | 4025 | |
76dda93c YZ |
4026 | dentry->d_op = &btrfs_dentry_operations; |
4027 | ||
39279cc3 CM |
4028 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
4029 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4030 | |
39279cc3 | 4031 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 4032 | |
39279cc3 CM |
4033 | if (ret < 0) |
4034 | return ERR_PTR(ret); | |
5f39d397 | 4035 | |
4df27c4d YZ |
4036 | if (location.objectid == 0) |
4037 | return NULL; | |
4038 | ||
4039 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4040 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4041 | return inode; |
4042 | } | |
4043 | ||
4044 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4045 | ||
76dda93c | 4046 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4047 | ret = fixup_tree_root_location(root, dir, dentry, |
4048 | &location, &sub_root); | |
4049 | if (ret < 0) { | |
4050 | if (ret != -ENOENT) | |
4051 | inode = ERR_PTR(ret); | |
4052 | else | |
4053 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4054 | } else { | |
73f73415 | 4055 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4056 | } |
76dda93c YZ |
4057 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4058 | ||
c71bf099 YZ |
4059 | if (root != sub_root) { |
4060 | down_read(&root->fs_info->cleanup_work_sem); | |
4061 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
4062 | btrfs_orphan_cleanup(sub_root); | |
4063 | up_read(&root->fs_info->cleanup_work_sem); | |
4064 | } | |
4065 | ||
3de4586c CM |
4066 | return inode; |
4067 | } | |
4068 | ||
76dda93c YZ |
4069 | static int btrfs_dentry_delete(struct dentry *dentry) |
4070 | { | |
4071 | struct btrfs_root *root; | |
4072 | ||
efefb143 YZ |
4073 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
4074 | dentry = dentry->d_parent; | |
76dda93c | 4075 | |
efefb143 YZ |
4076 | if (dentry->d_inode) { |
4077 | root = BTRFS_I(dentry->d_inode)->root; | |
4078 | if (btrfs_root_refs(&root->root_item) == 0) | |
4079 | return 1; | |
4080 | } | |
76dda93c YZ |
4081 | return 0; |
4082 | } | |
4083 | ||
3de4586c CM |
4084 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4085 | struct nameidata *nd) | |
4086 | { | |
4087 | struct inode *inode; | |
4088 | ||
3de4586c CM |
4089 | inode = btrfs_lookup_dentry(dir, dentry); |
4090 | if (IS_ERR(inode)) | |
4091 | return ERR_CAST(inode); | |
7b128766 | 4092 | |
39279cc3 CM |
4093 | return d_splice_alias(inode, dentry); |
4094 | } | |
4095 | ||
39279cc3 CM |
4096 | static unsigned char btrfs_filetype_table[] = { |
4097 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
4098 | }; | |
4099 | ||
cbdf5a24 DW |
4100 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4101 | filldir_t filldir) | |
39279cc3 | 4102 | { |
6da6abae | 4103 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4104 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4105 | struct btrfs_item *item; | |
4106 | struct btrfs_dir_item *di; | |
4107 | struct btrfs_key key; | |
5f39d397 | 4108 | struct btrfs_key found_key; |
39279cc3 CM |
4109 | struct btrfs_path *path; |
4110 | int ret; | |
4111 | u32 nritems; | |
5f39d397 | 4112 | struct extent_buffer *leaf; |
39279cc3 CM |
4113 | int slot; |
4114 | int advance; | |
4115 | unsigned char d_type; | |
4116 | int over = 0; | |
4117 | u32 di_cur; | |
4118 | u32 di_total; | |
4119 | u32 di_len; | |
4120 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4121 | char tmp_name[32]; |
4122 | char *name_ptr; | |
4123 | int name_len; | |
39279cc3 CM |
4124 | |
4125 | /* FIXME, use a real flag for deciding about the key type */ | |
4126 | if (root->fs_info->tree_root == root) | |
4127 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4128 | |
3954401f CM |
4129 | /* special case for "." */ |
4130 | if (filp->f_pos == 0) { | |
4131 | over = filldir(dirent, ".", 1, | |
4132 | 1, inode->i_ino, | |
4133 | DT_DIR); | |
4134 | if (over) | |
4135 | return 0; | |
4136 | filp->f_pos = 1; | |
4137 | } | |
3954401f CM |
4138 | /* special case for .., just use the back ref */ |
4139 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4140 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4141 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 4142 | 2, pino, DT_DIR); |
3954401f | 4143 | if (over) |
49593bfa | 4144 | return 0; |
3954401f CM |
4145 | filp->f_pos = 2; |
4146 | } | |
49593bfa DW |
4147 | path = btrfs_alloc_path(); |
4148 | path->reada = 2; | |
4149 | ||
39279cc3 CM |
4150 | btrfs_set_key_type(&key, key_type); |
4151 | key.offset = filp->f_pos; | |
49593bfa | 4152 | key.objectid = inode->i_ino; |
5f39d397 | 4153 | |
39279cc3 CM |
4154 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4155 | if (ret < 0) | |
4156 | goto err; | |
4157 | advance = 0; | |
49593bfa DW |
4158 | |
4159 | while (1) { | |
5f39d397 CM |
4160 | leaf = path->nodes[0]; |
4161 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4162 | slot = path->slots[0]; |
4163 | if (advance || slot >= nritems) { | |
49593bfa | 4164 | if (slot >= nritems - 1) { |
39279cc3 CM |
4165 | ret = btrfs_next_leaf(root, path); |
4166 | if (ret) | |
4167 | break; | |
5f39d397 CM |
4168 | leaf = path->nodes[0]; |
4169 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4170 | slot = path->slots[0]; |
4171 | } else { | |
4172 | slot++; | |
4173 | path->slots[0]++; | |
4174 | } | |
4175 | } | |
3de4586c | 4176 | |
39279cc3 | 4177 | advance = 1; |
5f39d397 CM |
4178 | item = btrfs_item_nr(leaf, slot); |
4179 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4180 | ||
4181 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4182 | break; |
5f39d397 | 4183 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4184 | break; |
5f39d397 | 4185 | if (found_key.offset < filp->f_pos) |
39279cc3 | 4186 | continue; |
5f39d397 CM |
4187 | |
4188 | filp->f_pos = found_key.offset; | |
49593bfa | 4189 | |
39279cc3 CM |
4190 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4191 | di_cur = 0; | |
5f39d397 | 4192 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4193 | |
4194 | while (di_cur < di_total) { | |
5f39d397 CM |
4195 | struct btrfs_key location; |
4196 | ||
4197 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 4198 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4199 | name_ptr = tmp_name; |
4200 | } else { | |
4201 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4202 | if (!name_ptr) { |
4203 | ret = -ENOMEM; | |
4204 | goto err; | |
4205 | } | |
5f39d397 CM |
4206 | } |
4207 | read_extent_buffer(leaf, name_ptr, | |
4208 | (unsigned long)(di + 1), name_len); | |
4209 | ||
4210 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4211 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
4212 | |
4213 | /* is this a reference to our own snapshot? If so | |
4214 | * skip it | |
4215 | */ | |
4216 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4217 | location.objectid == root->root_key.objectid) { | |
4218 | over = 0; | |
4219 | goto skip; | |
4220 | } | |
5f39d397 | 4221 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4222 | found_key.offset, location.objectid, |
39279cc3 | 4223 | d_type); |
5f39d397 | 4224 | |
3de4586c | 4225 | skip: |
5f39d397 CM |
4226 | if (name_ptr != tmp_name) |
4227 | kfree(name_ptr); | |
4228 | ||
39279cc3 CM |
4229 | if (over) |
4230 | goto nopos; | |
5103e947 | 4231 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4232 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4233 | di_cur += di_len; |
4234 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4235 | } | |
4236 | } | |
49593bfa DW |
4237 | |
4238 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 4239 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4240 | /* |
4241 | * 32-bit glibc will use getdents64, but then strtol - | |
4242 | * so the last number we can serve is this. | |
4243 | */ | |
4244 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4245 | else |
4246 | filp->f_pos++; | |
39279cc3 CM |
4247 | nopos: |
4248 | ret = 0; | |
4249 | err: | |
39279cc3 | 4250 | btrfs_free_path(path); |
39279cc3 CM |
4251 | return ret; |
4252 | } | |
4253 | ||
a9185b41 | 4254 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4255 | { |
4256 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4257 | struct btrfs_trans_handle *trans; | |
4258 | int ret = 0; | |
4259 | ||
8929ecfa | 4260 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4261 | return 0; |
4262 | ||
a9185b41 | 4263 | if (wbc->sync_mode == WB_SYNC_ALL) { |
f9295749 | 4264 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
4265 | btrfs_set_trans_block_group(trans, inode); |
4266 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4267 | } |
4268 | return ret; | |
4269 | } | |
4270 | ||
4271 | /* | |
54aa1f4d | 4272 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4273 | * inode changes. But, it is most likely to find the inode in cache. |
4274 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4275 | * to keep or drop this code. | |
4276 | */ | |
4277 | void btrfs_dirty_inode(struct inode *inode) | |
4278 | { | |
4279 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4280 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4281 | int ret; |
4282 | ||
4283 | if (BTRFS_I(inode)->dummy_inode) | |
4284 | return; | |
39279cc3 | 4285 | |
f9295749 | 4286 | trans = btrfs_join_transaction(root, 1); |
39279cc3 | 4287 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa YZ |
4288 | |
4289 | ret = btrfs_update_inode(trans, root, inode); | |
4290 | if (ret) | |
4291 | printk(KERN_ERR"btrfs: fail to dirty inode %lu error %d\n", | |
4292 | inode->i_ino, ret); | |
4293 | ||
39279cc3 | 4294 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4295 | } |
4296 | ||
d352ac68 CM |
4297 | /* |
4298 | * find the highest existing sequence number in a directory | |
4299 | * and then set the in-memory index_cnt variable to reflect | |
4300 | * free sequence numbers | |
4301 | */ | |
aec7477b JB |
4302 | static int btrfs_set_inode_index_count(struct inode *inode) |
4303 | { | |
4304 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4305 | struct btrfs_key key, found_key; | |
4306 | struct btrfs_path *path; | |
4307 | struct extent_buffer *leaf; | |
4308 | int ret; | |
4309 | ||
4310 | key.objectid = inode->i_ino; | |
4311 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
4312 | key.offset = (u64)-1; | |
4313 | ||
4314 | path = btrfs_alloc_path(); | |
4315 | if (!path) | |
4316 | return -ENOMEM; | |
4317 | ||
4318 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4319 | if (ret < 0) | |
4320 | goto out; | |
4321 | /* FIXME: we should be able to handle this */ | |
4322 | if (ret == 0) | |
4323 | goto out; | |
4324 | ret = 0; | |
4325 | ||
4326 | /* | |
4327 | * MAGIC NUMBER EXPLANATION: | |
4328 | * since we search a directory based on f_pos we have to start at 2 | |
4329 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4330 | * else has to start at 2 | |
4331 | */ | |
4332 | if (path->slots[0] == 0) { | |
4333 | BTRFS_I(inode)->index_cnt = 2; | |
4334 | goto out; | |
4335 | } | |
4336 | ||
4337 | path->slots[0]--; | |
4338 | ||
4339 | leaf = path->nodes[0]; | |
4340 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4341 | ||
4342 | if (found_key.objectid != inode->i_ino || | |
4343 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
4344 | BTRFS_I(inode)->index_cnt = 2; | |
4345 | goto out; | |
4346 | } | |
4347 | ||
4348 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4349 | out: | |
4350 | btrfs_free_path(path); | |
4351 | return ret; | |
4352 | } | |
4353 | ||
d352ac68 CM |
4354 | /* |
4355 | * helper to find a free sequence number in a given directory. This current | |
4356 | * code is very simple, later versions will do smarter things in the btree | |
4357 | */ | |
3de4586c | 4358 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4359 | { |
4360 | int ret = 0; | |
4361 | ||
4362 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
4363 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 4364 | if (ret) |
aec7477b JB |
4365 | return ret; |
4366 | } | |
4367 | ||
00e4e6b3 | 4368 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4369 | BTRFS_I(dir)->index_cnt++; |
4370 | ||
4371 | return ret; | |
4372 | } | |
4373 | ||
39279cc3 CM |
4374 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4375 | struct btrfs_root *root, | |
aec7477b | 4376 | struct inode *dir, |
9c58309d | 4377 | const char *name, int name_len, |
d2fb3437 YZ |
4378 | u64 ref_objectid, u64 objectid, |
4379 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4380 | { |
4381 | struct inode *inode; | |
5f39d397 | 4382 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4383 | struct btrfs_key *location; |
5f39d397 | 4384 | struct btrfs_path *path; |
9c58309d CM |
4385 | struct btrfs_inode_ref *ref; |
4386 | struct btrfs_key key[2]; | |
4387 | u32 sizes[2]; | |
4388 | unsigned long ptr; | |
39279cc3 CM |
4389 | int ret; |
4390 | int owner; | |
4391 | ||
5f39d397 CM |
4392 | path = btrfs_alloc_path(); |
4393 | BUG_ON(!path); | |
4394 | ||
39279cc3 CM |
4395 | inode = new_inode(root->fs_info->sb); |
4396 | if (!inode) | |
4397 | return ERR_PTR(-ENOMEM); | |
4398 | ||
aec7477b | 4399 | if (dir) { |
3de4586c | 4400 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
4401 | if (ret) { |
4402 | iput(inode); | |
aec7477b | 4403 | return ERR_PTR(ret); |
09771430 | 4404 | } |
aec7477b JB |
4405 | } |
4406 | /* | |
4407 | * index_cnt is ignored for everything but a dir, | |
4408 | * btrfs_get_inode_index_count has an explanation for the magic | |
4409 | * number | |
4410 | */ | |
4411 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4412 | BTRFS_I(inode)->root = root; |
e02119d5 | 4413 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 4414 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4415 | |
39279cc3 CM |
4416 | if (mode & S_IFDIR) |
4417 | owner = 0; | |
4418 | else | |
4419 | owner = 1; | |
d2fb3437 YZ |
4420 | BTRFS_I(inode)->block_group = |
4421 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4422 | |
4423 | key[0].objectid = objectid; | |
4424 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4425 | key[0].offset = 0; | |
4426 | ||
4427 | key[1].objectid = objectid; | |
4428 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4429 | key[1].offset = ref_objectid; | |
4430 | ||
4431 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4432 | sizes[1] = name_len + sizeof(*ref); | |
4433 | ||
b9473439 | 4434 | path->leave_spinning = 1; |
9c58309d CM |
4435 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4436 | if (ret != 0) | |
5f39d397 CM |
4437 | goto fail; |
4438 | ||
79683f2d | 4439 | inode->i_uid = current_fsuid(); |
8c087b51 | 4440 | |
42f15d77 | 4441 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4442 | inode->i_gid = dir->i_gid; |
4443 | if (S_ISDIR(mode)) | |
4444 | mode |= S_ISGID; | |
4445 | } else | |
4446 | inode->i_gid = current_fsgid(); | |
4447 | ||
39279cc3 CM |
4448 | inode->i_mode = mode; |
4449 | inode->i_ino = objectid; | |
a76a3cd4 | 4450 | inode_set_bytes(inode, 0); |
39279cc3 | 4451 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4452 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4453 | struct btrfs_inode_item); | |
e02119d5 | 4454 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4455 | |
4456 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4457 | struct btrfs_inode_ref); | |
4458 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4459 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4460 | ptr = (unsigned long)(ref + 1); |
4461 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4462 | ||
5f39d397 CM |
4463 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4464 | btrfs_free_path(path); | |
4465 | ||
39279cc3 CM |
4466 | location = &BTRFS_I(inode)->location; |
4467 | location->objectid = objectid; | |
39279cc3 CM |
4468 | location->offset = 0; |
4469 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4470 | ||
6cbff00f CH |
4471 | btrfs_inherit_iflags(inode, dir); |
4472 | ||
94272164 CM |
4473 | if ((mode & S_IFREG)) { |
4474 | if (btrfs_test_opt(root, NODATASUM)) | |
4475 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4476 | if (btrfs_test_opt(root, NODATACOW)) | |
4477 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4478 | } | |
4479 | ||
39279cc3 | 4480 | insert_inode_hash(inode); |
5d4f98a2 | 4481 | inode_tree_add(inode); |
39279cc3 | 4482 | return inode; |
5f39d397 | 4483 | fail: |
aec7477b JB |
4484 | if (dir) |
4485 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4486 | btrfs_free_path(path); |
09771430 | 4487 | iput(inode); |
5f39d397 | 4488 | return ERR_PTR(ret); |
39279cc3 CM |
4489 | } |
4490 | ||
4491 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4492 | { | |
4493 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4494 | } | |
4495 | ||
d352ac68 CM |
4496 | /* |
4497 | * utility function to add 'inode' into 'parent_inode' with | |
4498 | * a give name and a given sequence number. | |
4499 | * if 'add_backref' is true, also insert a backref from the | |
4500 | * inode to the parent directory. | |
4501 | */ | |
e02119d5 CM |
4502 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4503 | struct inode *parent_inode, struct inode *inode, | |
4504 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4505 | { |
4df27c4d | 4506 | int ret = 0; |
39279cc3 | 4507 | struct btrfs_key key; |
e02119d5 | 4508 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4509 | |
4df27c4d YZ |
4510 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4511 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4512 | } else { | |
4513 | key.objectid = inode->i_ino; | |
4514 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4515 | key.offset = 0; | |
4516 | } | |
4517 | ||
4518 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4519 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4520 | key.objectid, root->root_key.objectid, | |
4521 | parent_inode->i_ino, | |
4522 | index, name, name_len); | |
4523 | } else if (add_backref) { | |
4524 | ret = btrfs_insert_inode_ref(trans, root, | |
4525 | name, name_len, inode->i_ino, | |
4526 | parent_inode->i_ino, index); | |
4527 | } | |
39279cc3 | 4528 | |
39279cc3 | 4529 | if (ret == 0) { |
4df27c4d YZ |
4530 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4531 | parent_inode->i_ino, &key, | |
4532 | btrfs_inode_type(inode), index); | |
4533 | BUG_ON(ret); | |
4534 | ||
dbe674a9 | 4535 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4536 | name_len * 2); |
79c44584 | 4537 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4538 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4539 | } |
4540 | return ret; | |
4541 | } | |
4542 | ||
4543 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4544 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4545 | int backref, u64 index) |
39279cc3 | 4546 | { |
e02119d5 CM |
4547 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4548 | inode, dentry->d_name.name, | |
4549 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4550 | if (!err) { |
4551 | d_instantiate(dentry, inode); | |
4552 | return 0; | |
4553 | } | |
4554 | if (err > 0) | |
4555 | err = -EEXIST; | |
4556 | return err; | |
4557 | } | |
4558 | ||
618e21d5 JB |
4559 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4560 | int mode, dev_t rdev) | |
4561 | { | |
4562 | struct btrfs_trans_handle *trans; | |
4563 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4564 | struct inode *inode = NULL; |
618e21d5 JB |
4565 | int err; |
4566 | int drop_inode = 0; | |
4567 | u64 objectid; | |
1832a6d5 | 4568 | unsigned long nr = 0; |
00e4e6b3 | 4569 | u64 index = 0; |
618e21d5 JB |
4570 | |
4571 | if (!new_valid_dev(rdev)) | |
4572 | return -EINVAL; | |
4573 | ||
a22285a6 YZ |
4574 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4575 | if (err) | |
4576 | return err; | |
4577 | ||
9ed74f2d JB |
4578 | /* |
4579 | * 2 for inode item and ref | |
4580 | * 2 for dir items | |
4581 | * 1 for xattr if selinux is on | |
4582 | */ | |
a22285a6 YZ |
4583 | trans = btrfs_start_transaction(root, 5); |
4584 | if (IS_ERR(trans)) | |
4585 | return PTR_ERR(trans); | |
1832a6d5 | 4586 | |
618e21d5 JB |
4587 | btrfs_set_trans_block_group(trans, dir); |
4588 | ||
aec7477b | 4589 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4590 | dentry->d_name.len, |
4591 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4592 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4593 | err = PTR_ERR(inode); |
4594 | if (IS_ERR(inode)) | |
4595 | goto out_unlock; | |
4596 | ||
f34f57a3 | 4597 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4598 | if (err) { |
4599 | drop_inode = 1; | |
4600 | goto out_unlock; | |
4601 | } | |
4602 | ||
618e21d5 | 4603 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4604 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4605 | if (err) |
4606 | drop_inode = 1; | |
4607 | else { | |
4608 | inode->i_op = &btrfs_special_inode_operations; | |
4609 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4610 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4611 | } |
618e21d5 JB |
4612 | btrfs_update_inode_block_group(trans, inode); |
4613 | btrfs_update_inode_block_group(trans, dir); | |
4614 | out_unlock: | |
d3c2fdcf | 4615 | nr = trans->blocks_used; |
89ce8a63 | 4616 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4617 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4618 | if (drop_inode) { |
4619 | inode_dec_link_count(inode); | |
4620 | iput(inode); | |
4621 | } | |
618e21d5 JB |
4622 | return err; |
4623 | } | |
4624 | ||
39279cc3 CM |
4625 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4626 | int mode, struct nameidata *nd) | |
4627 | { | |
4628 | struct btrfs_trans_handle *trans; | |
4629 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4630 | struct inode *inode = NULL; |
39279cc3 | 4631 | int drop_inode = 0; |
a22285a6 | 4632 | int err; |
1832a6d5 | 4633 | unsigned long nr = 0; |
39279cc3 | 4634 | u64 objectid; |
00e4e6b3 | 4635 | u64 index = 0; |
39279cc3 | 4636 | |
a22285a6 YZ |
4637 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4638 | if (err) | |
4639 | return err; | |
9ed74f2d JB |
4640 | /* |
4641 | * 2 for inode item and ref | |
4642 | * 2 for dir items | |
4643 | * 1 for xattr if selinux is on | |
4644 | */ | |
a22285a6 YZ |
4645 | trans = btrfs_start_transaction(root, 5); |
4646 | if (IS_ERR(trans)) | |
4647 | return PTR_ERR(trans); | |
9ed74f2d | 4648 | |
39279cc3 CM |
4649 | btrfs_set_trans_block_group(trans, dir); |
4650 | ||
aec7477b | 4651 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4652 | dentry->d_name.len, |
4653 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4654 | objectid, BTRFS_I(dir)->block_group, mode, |
4655 | &index); | |
39279cc3 CM |
4656 | err = PTR_ERR(inode); |
4657 | if (IS_ERR(inode)) | |
4658 | goto out_unlock; | |
4659 | ||
f34f57a3 | 4660 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4661 | if (err) { |
4662 | drop_inode = 1; | |
4663 | goto out_unlock; | |
4664 | } | |
4665 | ||
39279cc3 | 4666 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4667 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4668 | if (err) |
4669 | drop_inode = 1; | |
4670 | else { | |
4671 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4672 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4673 | inode->i_fop = &btrfs_file_operations; |
4674 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4675 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4676 | } |
39279cc3 CM |
4677 | btrfs_update_inode_block_group(trans, inode); |
4678 | btrfs_update_inode_block_group(trans, dir); | |
4679 | out_unlock: | |
d3c2fdcf | 4680 | nr = trans->blocks_used; |
ab78c84d | 4681 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4682 | if (drop_inode) { |
4683 | inode_dec_link_count(inode); | |
4684 | iput(inode); | |
4685 | } | |
d3c2fdcf | 4686 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4687 | return err; |
4688 | } | |
4689 | ||
4690 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4691 | struct dentry *dentry) | |
4692 | { | |
4693 | struct btrfs_trans_handle *trans; | |
4694 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4695 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4696 | u64 index; |
1832a6d5 | 4697 | unsigned long nr = 0; |
39279cc3 CM |
4698 | int err; |
4699 | int drop_inode = 0; | |
4700 | ||
4701 | if (inode->i_nlink == 0) | |
4702 | return -ENOENT; | |
4703 | ||
4a8be425 TH |
4704 | /* do not allow sys_link's with other subvols of the same device */ |
4705 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
4706 | return -EPERM; | |
4707 | ||
9ed74f2d JB |
4708 | btrfs_inc_nlink(inode); |
4709 | ||
3de4586c | 4710 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4711 | if (err) |
4712 | goto fail; | |
4713 | ||
a22285a6 YZ |
4714 | /* |
4715 | * 1 item for inode ref | |
4716 | * 2 items for dir items | |
4717 | */ | |
4718 | trans = btrfs_start_transaction(root, 3); | |
4719 | if (IS_ERR(trans)) { | |
4720 | err = PTR_ERR(trans); | |
4721 | goto fail; | |
4722 | } | |
5f39d397 | 4723 | |
39279cc3 CM |
4724 | btrfs_set_trans_block_group(trans, dir); |
4725 | atomic_inc(&inode->i_count); | |
aec7477b | 4726 | |
00e4e6b3 | 4727 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4728 | |
a5719521 | 4729 | if (err) { |
54aa1f4d | 4730 | drop_inode = 1; |
a5719521 YZ |
4731 | } else { |
4732 | btrfs_update_inode_block_group(trans, dir); | |
4733 | err = btrfs_update_inode(trans, root, inode); | |
4734 | BUG_ON(err); | |
4735 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4736 | } | |
39279cc3 | 4737 | |
d3c2fdcf | 4738 | nr = trans->blocks_used; |
ab78c84d | 4739 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4740 | fail: |
39279cc3 CM |
4741 | if (drop_inode) { |
4742 | inode_dec_link_count(inode); | |
4743 | iput(inode); | |
4744 | } | |
d3c2fdcf | 4745 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4746 | return err; |
4747 | } | |
4748 | ||
39279cc3 CM |
4749 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4750 | { | |
b9d86667 | 4751 | struct inode *inode = NULL; |
39279cc3 CM |
4752 | struct btrfs_trans_handle *trans; |
4753 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4754 | int err = 0; | |
4755 | int drop_on_err = 0; | |
b9d86667 | 4756 | u64 objectid = 0; |
00e4e6b3 | 4757 | u64 index = 0; |
d3c2fdcf | 4758 | unsigned long nr = 1; |
39279cc3 | 4759 | |
a22285a6 YZ |
4760 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4761 | if (err) | |
4762 | return err; | |
4763 | ||
9ed74f2d JB |
4764 | /* |
4765 | * 2 items for inode and ref | |
4766 | * 2 items for dir items | |
4767 | * 1 for xattr if selinux is on | |
4768 | */ | |
a22285a6 YZ |
4769 | trans = btrfs_start_transaction(root, 5); |
4770 | if (IS_ERR(trans)) | |
4771 | return PTR_ERR(trans); | |
9ed74f2d | 4772 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 4773 | |
aec7477b | 4774 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4775 | dentry->d_name.len, |
4776 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4777 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4778 | &index); | |
39279cc3 CM |
4779 | if (IS_ERR(inode)) { |
4780 | err = PTR_ERR(inode); | |
4781 | goto out_fail; | |
4782 | } | |
5f39d397 | 4783 | |
39279cc3 | 4784 | drop_on_err = 1; |
33268eaf | 4785 | |
f34f57a3 | 4786 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4787 | if (err) |
4788 | goto out_fail; | |
4789 | ||
39279cc3 CM |
4790 | inode->i_op = &btrfs_dir_inode_operations; |
4791 | inode->i_fop = &btrfs_dir_file_operations; | |
4792 | btrfs_set_trans_block_group(trans, inode); | |
4793 | ||
dbe674a9 | 4794 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4795 | err = btrfs_update_inode(trans, root, inode); |
4796 | if (err) | |
4797 | goto out_fail; | |
5f39d397 | 4798 | |
e02119d5 CM |
4799 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4800 | inode, dentry->d_name.name, | |
4801 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4802 | if (err) |
4803 | goto out_fail; | |
5f39d397 | 4804 | |
39279cc3 CM |
4805 | d_instantiate(dentry, inode); |
4806 | drop_on_err = 0; | |
39279cc3 CM |
4807 | btrfs_update_inode_block_group(trans, inode); |
4808 | btrfs_update_inode_block_group(trans, dir); | |
4809 | ||
4810 | out_fail: | |
d3c2fdcf | 4811 | nr = trans->blocks_used; |
ab78c84d | 4812 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4813 | if (drop_on_err) |
4814 | iput(inode); | |
d3c2fdcf | 4815 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4816 | return err; |
4817 | } | |
4818 | ||
d352ac68 CM |
4819 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4820 | * and an extent that you want to insert, deal with overlap and insert | |
4821 | * the new extent into the tree. | |
4822 | */ | |
3b951516 CM |
4823 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4824 | struct extent_map *existing, | |
e6dcd2dc CM |
4825 | struct extent_map *em, |
4826 | u64 map_start, u64 map_len) | |
3b951516 CM |
4827 | { |
4828 | u64 start_diff; | |
3b951516 | 4829 | |
e6dcd2dc CM |
4830 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4831 | start_diff = map_start - em->start; | |
4832 | em->start = map_start; | |
4833 | em->len = map_len; | |
c8b97818 CM |
4834 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4835 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4836 | em->block_start += start_diff; |
c8b97818 CM |
4837 | em->block_len -= start_diff; |
4838 | } | |
e6dcd2dc | 4839 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4840 | } |
4841 | ||
c8b97818 CM |
4842 | static noinline int uncompress_inline(struct btrfs_path *path, |
4843 | struct inode *inode, struct page *page, | |
4844 | size_t pg_offset, u64 extent_offset, | |
4845 | struct btrfs_file_extent_item *item) | |
4846 | { | |
4847 | int ret; | |
4848 | struct extent_buffer *leaf = path->nodes[0]; | |
4849 | char *tmp; | |
4850 | size_t max_size; | |
4851 | unsigned long inline_size; | |
4852 | unsigned long ptr; | |
4853 | ||
4854 | WARN_ON(pg_offset != 0); | |
4855 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4856 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4857 | btrfs_item_nr(leaf, path->slots[0])); | |
4858 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4859 | ptr = btrfs_file_extent_inline_start(item); | |
4860 | ||
4861 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4862 | ||
5b050f04 | 4863 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4864 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4865 | inline_size, max_size); | |
4866 | if (ret) { | |
4867 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4868 | unsigned long copy_size = min_t(u64, | |
4869 | PAGE_CACHE_SIZE - pg_offset, | |
4870 | max_size - extent_offset); | |
4871 | memset(kaddr + pg_offset, 0, copy_size); | |
4872 | kunmap_atomic(kaddr, KM_USER0); | |
4873 | } | |
4874 | kfree(tmp); | |
4875 | return 0; | |
4876 | } | |
4877 | ||
d352ac68 CM |
4878 | /* |
4879 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4880 | * the ugly parts come from merging extents from the disk with the in-ram |
4881 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4882 | * where the in-ram extents might be locked pending data=ordered completion. |
4883 | * | |
4884 | * This also copies inline extents directly into the page. | |
4885 | */ | |
d397712b | 4886 | |
a52d9a80 | 4887 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4888 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4889 | int create) |
4890 | { | |
4891 | int ret; | |
4892 | int err = 0; | |
db94535d | 4893 | u64 bytenr; |
a52d9a80 CM |
4894 | u64 extent_start = 0; |
4895 | u64 extent_end = 0; | |
4896 | u64 objectid = inode->i_ino; | |
4897 | u32 found_type; | |
f421950f | 4898 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4899 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4900 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4901 | struct extent_buffer *leaf; |
4902 | struct btrfs_key found_key; | |
a52d9a80 CM |
4903 | struct extent_map *em = NULL; |
4904 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4905 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4906 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4907 | int compressed; |
a52d9a80 | 4908 | |
a52d9a80 | 4909 | again: |
890871be | 4910 | read_lock(&em_tree->lock); |
d1310b2e | 4911 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4912 | if (em) |
4913 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4914 | read_unlock(&em_tree->lock); |
d1310b2e | 4915 | |
a52d9a80 | 4916 | if (em) { |
e1c4b745 CM |
4917 | if (em->start > start || em->start + em->len <= start) |
4918 | free_extent_map(em); | |
4919 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4920 | free_extent_map(em); |
4921 | else | |
4922 | goto out; | |
a52d9a80 | 4923 | } |
d1310b2e | 4924 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4925 | if (!em) { |
d1310b2e CM |
4926 | err = -ENOMEM; |
4927 | goto out; | |
a52d9a80 | 4928 | } |
e6dcd2dc | 4929 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4930 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4931 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4932 | em->len = (u64)-1; |
c8b97818 | 4933 | em->block_len = (u64)-1; |
f421950f CM |
4934 | |
4935 | if (!path) { | |
4936 | path = btrfs_alloc_path(); | |
4937 | BUG_ON(!path); | |
4938 | } | |
4939 | ||
179e29e4 CM |
4940 | ret = btrfs_lookup_file_extent(trans, root, path, |
4941 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4942 | if (ret < 0) { |
4943 | err = ret; | |
4944 | goto out; | |
4945 | } | |
4946 | ||
4947 | if (ret != 0) { | |
4948 | if (path->slots[0] == 0) | |
4949 | goto not_found; | |
4950 | path->slots[0]--; | |
4951 | } | |
4952 | ||
5f39d397 CM |
4953 | leaf = path->nodes[0]; |
4954 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4955 | struct btrfs_file_extent_item); |
a52d9a80 | 4956 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4957 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4958 | found_type = btrfs_key_type(&found_key); | |
4959 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4960 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4961 | goto not_found; | |
4962 | } | |
4963 | ||
5f39d397 CM |
4964 | found_type = btrfs_file_extent_type(leaf, item); |
4965 | extent_start = found_key.offset; | |
c8b97818 | 4966 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4967 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4968 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4969 | extent_end = extent_start + |
db94535d | 4970 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4971 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4972 | size_t size; | |
4973 | size = btrfs_file_extent_inline_len(leaf, item); | |
4974 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4975 | ~((u64)root->sectorsize - 1); | |
4976 | } | |
4977 | ||
4978 | if (start >= extent_end) { | |
4979 | path->slots[0]++; | |
4980 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4981 | ret = btrfs_next_leaf(root, path); | |
4982 | if (ret < 0) { | |
4983 | err = ret; | |
4984 | goto out; | |
a52d9a80 | 4985 | } |
9036c102 YZ |
4986 | if (ret > 0) |
4987 | goto not_found; | |
4988 | leaf = path->nodes[0]; | |
a52d9a80 | 4989 | } |
9036c102 YZ |
4990 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4991 | if (found_key.objectid != objectid || | |
4992 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4993 | goto not_found; | |
4994 | if (start + len <= found_key.offset) | |
4995 | goto not_found; | |
4996 | em->start = start; | |
4997 | em->len = found_key.offset - start; | |
4998 | goto not_found_em; | |
4999 | } | |
5000 | ||
d899e052 YZ |
5001 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5002 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5003 | em->start = extent_start; |
5004 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5005 | em->orig_start = extent_start - |
5006 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5007 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5008 | if (bytenr == 0) { | |
5f39d397 | 5009 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5010 | goto insert; |
5011 | } | |
c8b97818 CM |
5012 | if (compressed) { |
5013 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
5014 | em->block_start = bytenr; | |
5015 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5016 | item); | |
5017 | } else { | |
5018 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5019 | em->block_start = bytenr; | |
5020 | em->block_len = em->len; | |
d899e052 YZ |
5021 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5022 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5023 | } |
a52d9a80 CM |
5024 | goto insert; |
5025 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5026 | unsigned long ptr; |
a52d9a80 | 5027 | char *map; |
3326d1b0 CM |
5028 | size_t size; |
5029 | size_t extent_offset; | |
5030 | size_t copy_size; | |
a52d9a80 | 5031 | |
689f9346 | 5032 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5033 | if (!page || create) { |
689f9346 | 5034 | em->start = extent_start; |
9036c102 | 5035 | em->len = extent_end - extent_start; |
689f9346 Y |
5036 | goto out; |
5037 | } | |
5f39d397 | 5038 | |
9036c102 YZ |
5039 | size = btrfs_file_extent_inline_len(leaf, item); |
5040 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5041 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5042 | size - extent_offset); |
3326d1b0 | 5043 | em->start = extent_start + extent_offset; |
70dec807 CM |
5044 | em->len = (copy_size + root->sectorsize - 1) & |
5045 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5046 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
5047 | if (compressed) |
5048 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 5049 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5050 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
5051 | if (btrfs_file_extent_compression(leaf, item) == |
5052 | BTRFS_COMPRESS_ZLIB) { | |
5053 | ret = uncompress_inline(path, inode, page, | |
5054 | pg_offset, | |
5055 | extent_offset, item); | |
5056 | BUG_ON(ret); | |
5057 | } else { | |
5058 | map = kmap(page); | |
5059 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5060 | copy_size); | |
93c82d57 CM |
5061 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5062 | memset(map + pg_offset + copy_size, 0, | |
5063 | PAGE_CACHE_SIZE - pg_offset - | |
5064 | copy_size); | |
5065 | } | |
c8b97818 CM |
5066 | kunmap(page); |
5067 | } | |
179e29e4 CM |
5068 | flush_dcache_page(page); |
5069 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 5070 | WARN_ON(1); |
179e29e4 CM |
5071 | if (!trans) { |
5072 | kunmap(page); | |
5073 | free_extent_map(em); | |
5074 | em = NULL; | |
5075 | btrfs_release_path(root, path); | |
f9295749 | 5076 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
5077 | goto again; |
5078 | } | |
c8b97818 | 5079 | map = kmap(page); |
70dec807 | 5080 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5081 | copy_size); |
c8b97818 | 5082 | kunmap(page); |
179e29e4 | 5083 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5084 | } |
d1310b2e CM |
5085 | set_extent_uptodate(io_tree, em->start, |
5086 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
5087 | goto insert; |
5088 | } else { | |
d397712b | 5089 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5090 | WARN_ON(1); |
5091 | } | |
5092 | not_found: | |
5093 | em->start = start; | |
d1310b2e | 5094 | em->len = len; |
a52d9a80 | 5095 | not_found_em: |
5f39d397 | 5096 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5097 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
5098 | insert: |
5099 | btrfs_release_path(root, path); | |
d1310b2e | 5100 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5101 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5102 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5103 | (unsigned long long)em->len, | |
5104 | (unsigned long long)start, | |
5105 | (unsigned long long)len); | |
a52d9a80 CM |
5106 | err = -EIO; |
5107 | goto out; | |
5108 | } | |
d1310b2e CM |
5109 | |
5110 | err = 0; | |
890871be | 5111 | write_lock(&em_tree->lock); |
a52d9a80 | 5112 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5113 | /* it is possible that someone inserted the extent into the tree |
5114 | * while we had the lock dropped. It is also possible that | |
5115 | * an overlapping map exists in the tree | |
5116 | */ | |
a52d9a80 | 5117 | if (ret == -EEXIST) { |
3b951516 | 5118 | struct extent_map *existing; |
e6dcd2dc CM |
5119 | |
5120 | ret = 0; | |
5121 | ||
3b951516 | 5122 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5123 | if (existing && (existing->start > start || |
5124 | existing->start + existing->len <= start)) { | |
5125 | free_extent_map(existing); | |
5126 | existing = NULL; | |
5127 | } | |
3b951516 CM |
5128 | if (!existing) { |
5129 | existing = lookup_extent_mapping(em_tree, em->start, | |
5130 | em->len); | |
5131 | if (existing) { | |
5132 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5133 | em, start, |
5134 | root->sectorsize); | |
3b951516 CM |
5135 | free_extent_map(existing); |
5136 | if (err) { | |
5137 | free_extent_map(em); | |
5138 | em = NULL; | |
5139 | } | |
5140 | } else { | |
5141 | err = -EIO; | |
3b951516 CM |
5142 | free_extent_map(em); |
5143 | em = NULL; | |
5144 | } | |
5145 | } else { | |
5146 | free_extent_map(em); | |
5147 | em = existing; | |
e6dcd2dc | 5148 | err = 0; |
a52d9a80 | 5149 | } |
a52d9a80 | 5150 | } |
890871be | 5151 | write_unlock(&em_tree->lock); |
a52d9a80 | 5152 | out: |
f421950f CM |
5153 | if (path) |
5154 | btrfs_free_path(path); | |
a52d9a80 CM |
5155 | if (trans) { |
5156 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5157 | if (!err) |
a52d9a80 CM |
5158 | err = ret; |
5159 | } | |
a52d9a80 CM |
5160 | if (err) { |
5161 | free_extent_map(em); | |
a52d9a80 CM |
5162 | return ERR_PTR(err); |
5163 | } | |
5164 | return em; | |
5165 | } | |
5166 | ||
16432985 CM |
5167 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
5168 | const struct iovec *iov, loff_t offset, | |
5169 | unsigned long nr_segs) | |
5170 | { | |
e1c4b745 | 5171 | return -EINVAL; |
16432985 CM |
5172 | } |
5173 | ||
1506fcc8 YS |
5174 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
5175 | __u64 start, __u64 len) | |
5176 | { | |
5177 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
5178 | } | |
5179 | ||
a52d9a80 | 5180 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 5181 | { |
d1310b2e CM |
5182 | struct extent_io_tree *tree; |
5183 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 5184 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 5185 | } |
1832a6d5 | 5186 | |
a52d9a80 | 5187 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 5188 | { |
d1310b2e | 5189 | struct extent_io_tree *tree; |
b888db2b CM |
5190 | |
5191 | ||
5192 | if (current->flags & PF_MEMALLOC) { | |
5193 | redirty_page_for_writepage(wbc, page); | |
5194 | unlock_page(page); | |
5195 | return 0; | |
5196 | } | |
d1310b2e | 5197 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 5198 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
5199 | } |
5200 | ||
f421950f CM |
5201 | int btrfs_writepages(struct address_space *mapping, |
5202 | struct writeback_control *wbc) | |
b293f02e | 5203 | { |
d1310b2e | 5204 | struct extent_io_tree *tree; |
771ed689 | 5205 | |
d1310b2e | 5206 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
5207 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
5208 | } | |
5209 | ||
3ab2fb5a CM |
5210 | static int |
5211 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
5212 | struct list_head *pages, unsigned nr_pages) | |
5213 | { | |
d1310b2e CM |
5214 | struct extent_io_tree *tree; |
5215 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
5216 | return extent_readpages(tree, mapping, pages, nr_pages, |
5217 | btrfs_get_extent); | |
5218 | } | |
e6dcd2dc | 5219 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 5220 | { |
d1310b2e CM |
5221 | struct extent_io_tree *tree; |
5222 | struct extent_map_tree *map; | |
a52d9a80 | 5223 | int ret; |
8c2383c3 | 5224 | |
d1310b2e CM |
5225 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
5226 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 5227 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
5228 | if (ret == 1) { |
5229 | ClearPagePrivate(page); | |
5230 | set_page_private(page, 0); | |
5231 | page_cache_release(page); | |
39279cc3 | 5232 | } |
a52d9a80 | 5233 | return ret; |
39279cc3 CM |
5234 | } |
5235 | ||
e6dcd2dc CM |
5236 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
5237 | { | |
98509cfc CM |
5238 | if (PageWriteback(page) || PageDirty(page)) |
5239 | return 0; | |
b335b003 | 5240 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
5241 | } |
5242 | ||
a52d9a80 | 5243 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 5244 | { |
d1310b2e | 5245 | struct extent_io_tree *tree; |
e6dcd2dc | 5246 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 5247 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
5248 | u64 page_start = page_offset(page); |
5249 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 5250 | |
8b62b72b CM |
5251 | |
5252 | /* | |
5253 | * we have the page locked, so new writeback can't start, | |
5254 | * and the dirty bit won't be cleared while we are here. | |
5255 | * | |
5256 | * Wait for IO on this page so that we can safely clear | |
5257 | * the PagePrivate2 bit and do ordered accounting | |
5258 | */ | |
e6dcd2dc | 5259 | wait_on_page_writeback(page); |
8b62b72b | 5260 | |
d1310b2e | 5261 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
5262 | if (offset) { |
5263 | btrfs_releasepage(page, GFP_NOFS); | |
5264 | return; | |
5265 | } | |
2ac55d41 JB |
5266 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
5267 | GFP_NOFS); | |
e6dcd2dc CM |
5268 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
5269 | page_offset(page)); | |
5270 | if (ordered) { | |
eb84ae03 CM |
5271 | /* |
5272 | * IO on this page will never be started, so we need | |
5273 | * to account for any ordered extents now | |
5274 | */ | |
e6dcd2dc CM |
5275 | clear_extent_bit(tree, page_start, page_end, |
5276 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 5277 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 5278 | &cached_state, GFP_NOFS); |
8b62b72b CM |
5279 | /* |
5280 | * whoever cleared the private bit is responsible | |
5281 | * for the finish_ordered_io | |
5282 | */ | |
5283 | if (TestClearPagePrivate2(page)) { | |
5284 | btrfs_finish_ordered_io(page->mapping->host, | |
5285 | page_start, page_end); | |
5286 | } | |
e6dcd2dc | 5287 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
5288 | cached_state = NULL; |
5289 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
5290 | GFP_NOFS); | |
e6dcd2dc CM |
5291 | } |
5292 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 5293 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 5294 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
5295 | __btrfs_releasepage(page, GFP_NOFS); |
5296 | ||
4a096752 | 5297 | ClearPageChecked(page); |
9ad6b7bc | 5298 | if (PagePrivate(page)) { |
9ad6b7bc CM |
5299 | ClearPagePrivate(page); |
5300 | set_page_private(page, 0); | |
5301 | page_cache_release(page); | |
5302 | } | |
39279cc3 CM |
5303 | } |
5304 | ||
9ebefb18 CM |
5305 | /* |
5306 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
5307 | * called from a page fault handler when a page is first dirtied. Hence we must | |
5308 | * be careful to check for EOF conditions here. We set the page up correctly | |
5309 | * for a written page which means we get ENOSPC checking when writing into | |
5310 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
5311 | * support these features. | |
5312 | * | |
5313 | * We are not allowed to take the i_mutex here so we have to play games to | |
5314 | * protect against truncate races as the page could now be beyond EOF. Because | |
5315 | * vmtruncate() writes the inode size before removing pages, once we have the | |
5316 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
5317 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
5318 | * unlock the page. | |
5319 | */ | |
c2ec175c | 5320 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 5321 | { |
c2ec175c | 5322 | struct page *page = vmf->page; |
6da6abae | 5323 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 5324 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
5325 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
5326 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 5327 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
5328 | char *kaddr; |
5329 | unsigned long zero_start; | |
9ebefb18 | 5330 | loff_t size; |
1832a6d5 | 5331 | int ret; |
a52d9a80 | 5332 | u64 page_start; |
e6dcd2dc | 5333 | u64 page_end; |
9ebefb18 | 5334 | |
0ca1f7ce | 5335 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
5336 | if (ret) { |
5337 | if (ret == -ENOMEM) | |
5338 | ret = VM_FAULT_OOM; | |
5339 | else /* -ENOSPC, -EIO, etc */ | |
5340 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 5341 | goto out; |
56a76f82 | 5342 | } |
1832a6d5 | 5343 | |
56a76f82 | 5344 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 5345 | again: |
9ebefb18 | 5346 | lock_page(page); |
9ebefb18 | 5347 | size = i_size_read(inode); |
e6dcd2dc CM |
5348 | page_start = page_offset(page); |
5349 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 5350 | |
9ebefb18 | 5351 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 5352 | (page_start >= size)) { |
9ebefb18 CM |
5353 | /* page got truncated out from underneath us */ |
5354 | goto out_unlock; | |
5355 | } | |
e6dcd2dc CM |
5356 | wait_on_page_writeback(page); |
5357 | ||
2ac55d41 JB |
5358 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
5359 | GFP_NOFS); | |
e6dcd2dc CM |
5360 | set_page_extent_mapped(page); |
5361 | ||
eb84ae03 CM |
5362 | /* |
5363 | * we can't set the delalloc bits if there are pending ordered | |
5364 | * extents. Drop our locks and wait for them to finish | |
5365 | */ | |
e6dcd2dc CM |
5366 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5367 | if (ordered) { | |
2ac55d41 JB |
5368 | unlock_extent_cached(io_tree, page_start, page_end, |
5369 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 5370 | unlock_page(page); |
eb84ae03 | 5371 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5372 | btrfs_put_ordered_extent(ordered); |
5373 | goto again; | |
5374 | } | |
5375 | ||
fbf19087 JB |
5376 | /* |
5377 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
5378 | * if it was already dirty, so for space accounting reasons we need to | |
5379 | * clear any delalloc bits for the range we are fixing to save. There | |
5380 | * is probably a better way to do this, but for now keep consistent with | |
5381 | * prepare_pages in the normal write path. | |
5382 | */ | |
2ac55d41 | 5383 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 5384 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 5385 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 5386 | |
2ac55d41 JB |
5387 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
5388 | &cached_state); | |
9ed74f2d | 5389 | if (ret) { |
2ac55d41 JB |
5390 | unlock_extent_cached(io_tree, page_start, page_end, |
5391 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
5392 | ret = VM_FAULT_SIGBUS; |
5393 | goto out_unlock; | |
5394 | } | |
e6dcd2dc | 5395 | ret = 0; |
9ebefb18 CM |
5396 | |
5397 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 5398 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 5399 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 5400 | else |
e6dcd2dc | 5401 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 5402 | |
e6dcd2dc CM |
5403 | if (zero_start != PAGE_CACHE_SIZE) { |
5404 | kaddr = kmap(page); | |
5405 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
5406 | flush_dcache_page(page); | |
5407 | kunmap(page); | |
5408 | } | |
247e743c | 5409 | ClearPageChecked(page); |
e6dcd2dc | 5410 | set_page_dirty(page); |
50a9b214 | 5411 | SetPageUptodate(page); |
5a3f23d5 | 5412 | |
257c62e1 CM |
5413 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5414 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5415 | ||
2ac55d41 | 5416 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
5417 | |
5418 | out_unlock: | |
50a9b214 CM |
5419 | if (!ret) |
5420 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5421 | unlock_page(page); |
0ca1f7ce | 5422 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 5423 | out: |
9ebefb18 CM |
5424 | return ret; |
5425 | } | |
5426 | ||
39279cc3 CM |
5427 | static void btrfs_truncate(struct inode *inode) |
5428 | { | |
5429 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5430 | int ret; | |
5431 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5432 | unsigned long nr; |
dbe674a9 | 5433 | u64 mask = root->sectorsize - 1; |
39279cc3 | 5434 | |
8082510e YZ |
5435 | if (!S_ISREG(inode->i_mode)) { |
5436 | WARN_ON(1); | |
39279cc3 | 5437 | return; |
8082510e | 5438 | } |
39279cc3 | 5439 | |
5d5e103a JB |
5440 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5441 | if (ret) | |
5442 | return; | |
8082510e | 5443 | |
4a096752 | 5444 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 5445 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 5446 | |
d68fc57b YZ |
5447 | trans = btrfs_start_transaction(root, 0); |
5448 | BUG_ON(IS_ERR(trans)); | |
8082510e | 5449 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b | 5450 | trans->block_rsv = root->orphan_block_rsv; |
5a3f23d5 CM |
5451 | |
5452 | /* | |
5453 | * setattr is responsible for setting the ordered_data_close flag, | |
5454 | * but that is only tested during the last file release. That | |
5455 | * could happen well after the next commit, leaving a great big | |
5456 | * window where new writes may get lost if someone chooses to write | |
5457 | * to this file after truncating to zero | |
5458 | * | |
5459 | * The inode doesn't have any dirty data here, and so if we commit | |
5460 | * this is a noop. If someone immediately starts writing to the inode | |
5461 | * it is very likely we'll catch some of their writes in this | |
5462 | * transaction, and the commit will find this file on the ordered | |
5463 | * data list with good things to send down. | |
5464 | * | |
5465 | * This is a best effort solution, there is still a window where | |
5466 | * using truncate to replace the contents of the file will | |
5467 | * end up with a zero length file after a crash. | |
5468 | */ | |
5469 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5470 | btrfs_add_ordered_operation(trans, root, inode); | |
5471 | ||
8082510e | 5472 | while (1) { |
d68fc57b YZ |
5473 | if (!trans) { |
5474 | trans = btrfs_start_transaction(root, 0); | |
5475 | BUG_ON(IS_ERR(trans)); | |
5476 | btrfs_set_trans_block_group(trans, inode); | |
5477 | trans->block_rsv = root->orphan_block_rsv; | |
5478 | } | |
5479 | ||
5480 | ret = btrfs_block_rsv_check(trans, root, | |
5481 | root->orphan_block_rsv, 0, 5); | |
5482 | if (ret) { | |
5483 | BUG_ON(ret != -EAGAIN); | |
5484 | ret = btrfs_commit_transaction(trans, root); | |
5485 | BUG_ON(ret); | |
5486 | trans = NULL; | |
5487 | continue; | |
5488 | } | |
5489 | ||
8082510e YZ |
5490 | ret = btrfs_truncate_inode_items(trans, root, inode, |
5491 | inode->i_size, | |
5492 | BTRFS_EXTENT_DATA_KEY); | |
5493 | if (ret != -EAGAIN) | |
5494 | break; | |
39279cc3 | 5495 | |
8082510e YZ |
5496 | ret = btrfs_update_inode(trans, root, inode); |
5497 | BUG_ON(ret); | |
5f39d397 | 5498 | |
8082510e YZ |
5499 | nr = trans->blocks_used; |
5500 | btrfs_end_transaction(trans, root); | |
d68fc57b | 5501 | trans = NULL; |
8082510e | 5502 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
5503 | } |
5504 | ||
5505 | if (ret == 0 && inode->i_nlink > 0) { | |
5506 | ret = btrfs_orphan_del(trans, inode); | |
5507 | BUG_ON(ret); | |
5508 | } | |
5509 | ||
5510 | ret = btrfs_update_inode(trans, root, inode); | |
7b128766 JB |
5511 | BUG_ON(ret); |
5512 | ||
7b128766 | 5513 | nr = trans->blocks_used; |
89ce8a63 | 5514 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5515 | BUG_ON(ret); |
d3c2fdcf | 5516 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5517 | } |
5518 | ||
d352ac68 CM |
5519 | /* |
5520 | * create a new subvolume directory/inode (helper for the ioctl). | |
5521 | */ | |
d2fb3437 | 5522 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5523 | struct btrfs_root *new_root, |
d2fb3437 | 5524 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5525 | { |
39279cc3 | 5526 | struct inode *inode; |
76dda93c | 5527 | int err; |
00e4e6b3 | 5528 | u64 index = 0; |
39279cc3 | 5529 | |
aec7477b | 5530 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5531 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5532 | if (IS_ERR(inode)) |
f46b5a66 | 5533 | return PTR_ERR(inode); |
39279cc3 CM |
5534 | inode->i_op = &btrfs_dir_inode_operations; |
5535 | inode->i_fop = &btrfs_dir_file_operations; | |
5536 | ||
39279cc3 | 5537 | inode->i_nlink = 1; |
dbe674a9 | 5538 | btrfs_i_size_write(inode, 0); |
3b96362c | 5539 | |
76dda93c YZ |
5540 | err = btrfs_update_inode(trans, new_root, inode); |
5541 | BUG_ON(err); | |
cb8e7090 | 5542 | |
76dda93c | 5543 | iput(inode); |
cb8e7090 | 5544 | return 0; |
39279cc3 CM |
5545 | } |
5546 | ||
d352ac68 CM |
5547 | /* helper function for file defrag and space balancing. This |
5548 | * forces readahead on a given range of bytes in an inode | |
5549 | */ | |
edbd8d4e | 5550 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5551 | struct file_ra_state *ra, struct file *file, |
5552 | pgoff_t offset, pgoff_t last_index) | |
5553 | { | |
8e7bf94f | 5554 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5555 | |
86479a04 CM |
5556 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5557 | return offset + req_size; | |
86479a04 CM |
5558 | } |
5559 | ||
39279cc3 CM |
5560 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5561 | { | |
5562 | struct btrfs_inode *ei; | |
2ead6ae7 | 5563 | struct inode *inode; |
39279cc3 CM |
5564 | |
5565 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5566 | if (!ei) | |
5567 | return NULL; | |
2ead6ae7 YZ |
5568 | |
5569 | ei->root = NULL; | |
5570 | ei->space_info = NULL; | |
5571 | ei->generation = 0; | |
5572 | ei->sequence = 0; | |
15ee9bc7 | 5573 | ei->last_trans = 0; |
257c62e1 | 5574 | ei->last_sub_trans = 0; |
e02119d5 | 5575 | ei->logged_trans = 0; |
2ead6ae7 YZ |
5576 | ei->delalloc_bytes = 0; |
5577 | ei->reserved_bytes = 0; | |
5578 | ei->disk_i_size = 0; | |
5579 | ei->flags = 0; | |
5580 | ei->index_cnt = (u64)-1; | |
5581 | ei->last_unlink_trans = 0; | |
5582 | ||
5583 | spin_lock_init(&ei->accounting_lock); | |
0ca1f7ce | 5584 | atomic_set(&ei->outstanding_extents, 0); |
32c00aff | 5585 | ei->reserved_extents = 0; |
2ead6ae7 YZ |
5586 | |
5587 | ei->ordered_data_close = 0; | |
d68fc57b | 5588 | ei->orphan_meta_reserved = 0; |
2ead6ae7 YZ |
5589 | ei->dummy_inode = 0; |
5590 | ei->force_compress = 0; | |
5591 | ||
5592 | inode = &ei->vfs_inode; | |
5593 | extent_map_tree_init(&ei->extent_tree, GFP_NOFS); | |
5594 | extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS); | |
5595 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS); | |
5596 | mutex_init(&ei->log_mutex); | |
e6dcd2dc | 5597 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5598 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 5599 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 5600 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
5601 | RB_CLEAR_NODE(&ei->rb_node); |
5602 | ||
5603 | return inode; | |
39279cc3 CM |
5604 | } |
5605 | ||
5606 | void btrfs_destroy_inode(struct inode *inode) | |
5607 | { | |
e6dcd2dc | 5608 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5609 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5610 | ||
39279cc3 CM |
5611 | WARN_ON(!list_empty(&inode->i_dentry)); |
5612 | WARN_ON(inode->i_data.nrpages); | |
0ca1f7ce YZ |
5613 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents)); |
5614 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
39279cc3 | 5615 | |
a6dbd429 JB |
5616 | /* |
5617 | * This can happen where we create an inode, but somebody else also | |
5618 | * created the same inode and we need to destroy the one we already | |
5619 | * created. | |
5620 | */ | |
5621 | if (!root) | |
5622 | goto free; | |
5623 | ||
5a3f23d5 CM |
5624 | /* |
5625 | * Make sure we're properly removed from the ordered operation | |
5626 | * lists. | |
5627 | */ | |
5628 | smp_mb(); | |
5629 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5630 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5631 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5632 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5633 | } | |
5634 | ||
d68fc57b | 5635 | spin_lock(&root->orphan_lock); |
7b128766 | 5636 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
8082510e YZ |
5637 | printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", |
5638 | inode->i_ino); | |
5639 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
7b128766 | 5640 | } |
d68fc57b | 5641 | spin_unlock(&root->orphan_lock); |
7b128766 | 5642 | |
d397712b | 5643 | while (1) { |
e6dcd2dc CM |
5644 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5645 | if (!ordered) | |
5646 | break; | |
5647 | else { | |
d397712b CM |
5648 | printk(KERN_ERR "btrfs found ordered " |
5649 | "extent %llu %llu on inode cleanup\n", | |
5650 | (unsigned long long)ordered->file_offset, | |
5651 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5652 | btrfs_remove_ordered_extent(inode, ordered); |
5653 | btrfs_put_ordered_extent(ordered); | |
5654 | btrfs_put_ordered_extent(ordered); | |
5655 | } | |
5656 | } | |
5d4f98a2 | 5657 | inode_tree_del(inode); |
5b21f2ed | 5658 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 5659 | free: |
39279cc3 CM |
5660 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5661 | } | |
5662 | ||
76dda93c YZ |
5663 | void btrfs_drop_inode(struct inode *inode) |
5664 | { | |
5665 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c YZ |
5666 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) |
5667 | generic_delete_inode(inode); | |
5668 | else | |
5669 | generic_drop_inode(inode); | |
5670 | } | |
5671 | ||
0ee0fda0 | 5672 | static void init_once(void *foo) |
39279cc3 CM |
5673 | { |
5674 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5675 | ||
5676 | inode_init_once(&ei->vfs_inode); | |
5677 | } | |
5678 | ||
5679 | void btrfs_destroy_cachep(void) | |
5680 | { | |
5681 | if (btrfs_inode_cachep) | |
5682 | kmem_cache_destroy(btrfs_inode_cachep); | |
5683 | if (btrfs_trans_handle_cachep) | |
5684 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5685 | if (btrfs_transaction_cachep) | |
5686 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5687 | if (btrfs_path_cachep) |
5688 | kmem_cache_destroy(btrfs_path_cachep); | |
5689 | } | |
5690 | ||
5691 | int btrfs_init_cachep(void) | |
5692 | { | |
9601e3f6 CH |
5693 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5694 | sizeof(struct btrfs_inode), 0, | |
5695 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5696 | if (!btrfs_inode_cachep) |
5697 | goto fail; | |
9601e3f6 CH |
5698 | |
5699 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5700 | sizeof(struct btrfs_trans_handle), 0, | |
5701 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5702 | if (!btrfs_trans_handle_cachep) |
5703 | goto fail; | |
9601e3f6 CH |
5704 | |
5705 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5706 | sizeof(struct btrfs_transaction), 0, | |
5707 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5708 | if (!btrfs_transaction_cachep) |
5709 | goto fail; | |
9601e3f6 CH |
5710 | |
5711 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5712 | sizeof(struct btrfs_path), 0, | |
5713 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5714 | if (!btrfs_path_cachep) |
5715 | goto fail; | |
9601e3f6 | 5716 | |
39279cc3 CM |
5717 | return 0; |
5718 | fail: | |
5719 | btrfs_destroy_cachep(); | |
5720 | return -ENOMEM; | |
5721 | } | |
5722 | ||
5723 | static int btrfs_getattr(struct vfsmount *mnt, | |
5724 | struct dentry *dentry, struct kstat *stat) | |
5725 | { | |
5726 | struct inode *inode = dentry->d_inode; | |
5727 | generic_fillattr(inode, stat); | |
3394e160 | 5728 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5729 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5730 | stat->blocks = (inode_get_bytes(inode) + |
5731 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5732 | return 0; |
5733 | } | |
5734 | ||
d397712b CM |
5735 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5736 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5737 | { |
5738 | struct btrfs_trans_handle *trans; | |
5739 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5740 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5741 | struct inode *new_inode = new_dentry->d_inode; |
5742 | struct inode *old_inode = old_dentry->d_inode; | |
5743 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5744 | u64 index = 0; |
4df27c4d | 5745 | u64 root_objectid; |
39279cc3 CM |
5746 | int ret; |
5747 | ||
f679a840 YZ |
5748 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5749 | return -EPERM; | |
5750 | ||
4df27c4d YZ |
5751 | /* we only allow rename subvolume link between subvolumes */ |
5752 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5753 | return -EXDEV; |
5754 | ||
4df27c4d YZ |
5755 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5756 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5757 | return -ENOTEMPTY; |
5f39d397 | 5758 | |
4df27c4d YZ |
5759 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5760 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5761 | return -ENOTEMPTY; | |
5a3f23d5 CM |
5762 | /* |
5763 | * we're using rename to replace one file with another. | |
5764 | * and the replacement file is large. Start IO on it now so | |
5765 | * we don't add too much work to the end of the transaction | |
5766 | */ | |
4baf8c92 | 5767 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5768 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5769 | filemap_flush(old_inode->i_mapping); | |
5770 | ||
76dda93c YZ |
5771 | /* close the racy window with snapshot create/destroy ioctl */ |
5772 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5773 | down_read(&root->fs_info->subvol_sem); | |
a22285a6 YZ |
5774 | /* |
5775 | * We want to reserve the absolute worst case amount of items. So if | |
5776 | * both inodes are subvols and we need to unlink them then that would | |
5777 | * require 4 item modifications, but if they are both normal inodes it | |
5778 | * would require 5 item modifications, so we'll assume their normal | |
5779 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
5780 | * should cover the worst case number of items we'll modify. | |
5781 | */ | |
5782 | trans = btrfs_start_transaction(root, 20); | |
5783 | if (IS_ERR(trans)) | |
5784 | return PTR_ERR(trans); | |
76dda93c | 5785 | |
a5719521 | 5786 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5787 | |
4df27c4d YZ |
5788 | if (dest != root) |
5789 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 5790 | |
a5719521 YZ |
5791 | ret = btrfs_set_inode_index(new_dir, &index); |
5792 | if (ret) | |
5793 | goto out_fail; | |
5a3f23d5 | 5794 | |
a5719521 | 5795 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5796 | /* force full log commit if subvolume involved. */ |
5797 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5798 | } else { | |
a5719521 YZ |
5799 | ret = btrfs_insert_inode_ref(trans, dest, |
5800 | new_dentry->d_name.name, | |
5801 | new_dentry->d_name.len, | |
5802 | old_inode->i_ino, | |
5803 | new_dir->i_ino, index); | |
5804 | if (ret) | |
5805 | goto out_fail; | |
4df27c4d YZ |
5806 | /* |
5807 | * this is an ugly little race, but the rename is required | |
5808 | * to make sure that if we crash, the inode is either at the | |
5809 | * old name or the new one. pinning the log transaction lets | |
5810 | * us make sure we don't allow a log commit to come in after | |
5811 | * we unlink the name but before we add the new name back in. | |
5812 | */ | |
5813 | btrfs_pin_log_trans(root); | |
5814 | } | |
5a3f23d5 CM |
5815 | /* |
5816 | * make sure the inode gets flushed if it is replacing | |
5817 | * something. | |
5818 | */ | |
5819 | if (new_inode && new_inode->i_size && | |
5820 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5821 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5822 | } | |
5823 | ||
39279cc3 CM |
5824 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5825 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5826 | old_inode->i_ctime = ctime; | |
5f39d397 | 5827 | |
12fcfd22 CM |
5828 | if (old_dentry->d_parent != new_dentry->d_parent) |
5829 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5830 | ||
4df27c4d YZ |
5831 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5832 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5833 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5834 | old_dentry->d_name.name, | |
5835 | old_dentry->d_name.len); | |
5836 | } else { | |
5837 | btrfs_inc_nlink(old_dentry->d_inode); | |
5838 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5839 | old_dentry->d_inode, | |
5840 | old_dentry->d_name.name, | |
5841 | old_dentry->d_name.len); | |
5842 | } | |
5843 | BUG_ON(ret); | |
39279cc3 CM |
5844 | |
5845 | if (new_inode) { | |
5846 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5847 | if (unlikely(new_inode->i_ino == |
5848 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5849 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5850 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5851 | root_objectid, | |
5852 | new_dentry->d_name.name, | |
5853 | new_dentry->d_name.len); | |
5854 | BUG_ON(new_inode->i_nlink == 0); | |
5855 | } else { | |
5856 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5857 | new_dentry->d_inode, | |
5858 | new_dentry->d_name.name, | |
5859 | new_dentry->d_name.len); | |
5860 | } | |
5861 | BUG_ON(ret); | |
7b128766 | 5862 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5863 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5864 | BUG_ON(ret); |
7b128766 | 5865 | } |
39279cc3 | 5866 | } |
aec7477b | 5867 | |
4df27c4d YZ |
5868 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5869 | new_dentry->d_name.name, | |
a5719521 | 5870 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5871 | BUG_ON(ret); |
39279cc3 | 5872 | |
4df27c4d YZ |
5873 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5874 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5875 | new_dentry->d_parent); | |
5876 | btrfs_end_log_trans(root); | |
5877 | } | |
39279cc3 | 5878 | out_fail: |
ab78c84d | 5879 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5880 | |
76dda93c YZ |
5881 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5882 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 5883 | |
39279cc3 CM |
5884 | return ret; |
5885 | } | |
5886 | ||
d352ac68 CM |
5887 | /* |
5888 | * some fairly slow code that needs optimization. This walks the list | |
5889 | * of all the inodes with pending delalloc and forces them to disk. | |
5890 | */ | |
24bbcf04 | 5891 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
5892 | { |
5893 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5894 | struct btrfs_inode *binode; | |
5b21f2ed | 5895 | struct inode *inode; |
ea8c2819 | 5896 | |
c146afad YZ |
5897 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5898 | return -EROFS; | |
5899 | ||
75eff68e | 5900 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5901 | while (!list_empty(head)) { |
ea8c2819 CM |
5902 | binode = list_entry(head->next, struct btrfs_inode, |
5903 | delalloc_inodes); | |
5b21f2ed ZY |
5904 | inode = igrab(&binode->vfs_inode); |
5905 | if (!inode) | |
5906 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5907 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5908 | if (inode) { |
8c8bee1d | 5909 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
5910 | if (delay_iput) |
5911 | btrfs_add_delayed_iput(inode); | |
5912 | else | |
5913 | iput(inode); | |
5b21f2ed ZY |
5914 | } |
5915 | cond_resched(); | |
75eff68e | 5916 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5917 | } |
75eff68e | 5918 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5919 | |
5920 | /* the filemap_flush will queue IO into the worker threads, but | |
5921 | * we have to make sure the IO is actually started and that | |
5922 | * ordered extents get created before we return | |
5923 | */ | |
5924 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5925 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5926 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5927 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5928 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5929 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5930 | } |
5931 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5932 | return 0; |
5933 | } | |
5934 | ||
5da9d01b YZ |
5935 | int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput) |
5936 | { | |
5937 | struct btrfs_inode *binode; | |
5938 | struct inode *inode = NULL; | |
5939 | ||
5940 | spin_lock(&root->fs_info->delalloc_lock); | |
5941 | while (!list_empty(&root->fs_info->delalloc_inodes)) { | |
5942 | binode = list_entry(root->fs_info->delalloc_inodes.next, | |
5943 | struct btrfs_inode, delalloc_inodes); | |
5944 | inode = igrab(&binode->vfs_inode); | |
5945 | if (inode) { | |
5946 | list_move_tail(&binode->delalloc_inodes, | |
5947 | &root->fs_info->delalloc_inodes); | |
5948 | break; | |
5949 | } | |
5950 | ||
5951 | list_del_init(&binode->delalloc_inodes); | |
5952 | cond_resched_lock(&root->fs_info->delalloc_lock); | |
5953 | } | |
5954 | spin_unlock(&root->fs_info->delalloc_lock); | |
5955 | ||
5956 | if (inode) { | |
5957 | write_inode_now(inode, 0); | |
5958 | if (delay_iput) | |
5959 | btrfs_add_delayed_iput(inode); | |
5960 | else | |
5961 | iput(inode); | |
5962 | return 1; | |
5963 | } | |
5964 | return 0; | |
5965 | } | |
5966 | ||
39279cc3 CM |
5967 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5968 | const char *symname) | |
5969 | { | |
5970 | struct btrfs_trans_handle *trans; | |
5971 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5972 | struct btrfs_path *path; | |
5973 | struct btrfs_key key; | |
1832a6d5 | 5974 | struct inode *inode = NULL; |
39279cc3 CM |
5975 | int err; |
5976 | int drop_inode = 0; | |
5977 | u64 objectid; | |
00e4e6b3 | 5978 | u64 index = 0 ; |
39279cc3 CM |
5979 | int name_len; |
5980 | int datasize; | |
5f39d397 | 5981 | unsigned long ptr; |
39279cc3 | 5982 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5983 | struct extent_buffer *leaf; |
1832a6d5 | 5984 | unsigned long nr = 0; |
39279cc3 CM |
5985 | |
5986 | name_len = strlen(symname) + 1; | |
5987 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5988 | return -ENAMETOOLONG; | |
1832a6d5 | 5989 | |
a22285a6 YZ |
5990 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
5991 | if (err) | |
5992 | return err; | |
9ed74f2d JB |
5993 | /* |
5994 | * 2 items for inode item and ref | |
5995 | * 2 items for dir items | |
5996 | * 1 item for xattr if selinux is on | |
5997 | */ | |
a22285a6 YZ |
5998 | trans = btrfs_start_transaction(root, 5); |
5999 | if (IS_ERR(trans)) | |
6000 | return PTR_ERR(trans); | |
1832a6d5 | 6001 | |
39279cc3 CM |
6002 | btrfs_set_trans_block_group(trans, dir); |
6003 | ||
aec7477b | 6004 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
6005 | dentry->d_name.len, |
6006 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
6007 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
6008 | &index); | |
39279cc3 CM |
6009 | err = PTR_ERR(inode); |
6010 | if (IS_ERR(inode)) | |
6011 | goto out_unlock; | |
6012 | ||
f34f57a3 | 6013 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
6014 | if (err) { |
6015 | drop_inode = 1; | |
6016 | goto out_unlock; | |
6017 | } | |
6018 | ||
39279cc3 | 6019 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 6020 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
6021 | if (err) |
6022 | drop_inode = 1; | |
6023 | else { | |
6024 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 6025 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
6026 | inode->i_fop = &btrfs_file_operations; |
6027 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 6028 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 6029 | } |
39279cc3 CM |
6030 | btrfs_update_inode_block_group(trans, inode); |
6031 | btrfs_update_inode_block_group(trans, dir); | |
6032 | if (drop_inode) | |
6033 | goto out_unlock; | |
6034 | ||
6035 | path = btrfs_alloc_path(); | |
6036 | BUG_ON(!path); | |
6037 | key.objectid = inode->i_ino; | |
6038 | key.offset = 0; | |
39279cc3 CM |
6039 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
6040 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
6041 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
6042 | datasize); | |
54aa1f4d CM |
6043 | if (err) { |
6044 | drop_inode = 1; | |
6045 | goto out_unlock; | |
6046 | } | |
5f39d397 CM |
6047 | leaf = path->nodes[0]; |
6048 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
6049 | struct btrfs_file_extent_item); | |
6050 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
6051 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 6052 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
6053 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
6054 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
6055 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
6056 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
6057 | ||
39279cc3 | 6058 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
6059 | write_extent_buffer(leaf, symname, ptr, name_len); |
6060 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 6061 | btrfs_free_path(path); |
5f39d397 | 6062 | |
39279cc3 CM |
6063 | inode->i_op = &btrfs_symlink_inode_operations; |
6064 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 6065 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 6066 | inode_set_bytes(inode, name_len); |
dbe674a9 | 6067 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
6068 | err = btrfs_update_inode(trans, root, inode); |
6069 | if (err) | |
6070 | drop_inode = 1; | |
39279cc3 CM |
6071 | |
6072 | out_unlock: | |
d3c2fdcf | 6073 | nr = trans->blocks_used; |
ab78c84d | 6074 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
6075 | if (drop_inode) { |
6076 | inode_dec_link_count(inode); | |
6077 | iput(inode); | |
6078 | } | |
d3c2fdcf | 6079 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
6080 | return err; |
6081 | } | |
16432985 | 6082 | |
5a303d5d | 6083 | static int prealloc_file_range(struct inode *inode, u64 start, u64 end, |
d1ea6a61 | 6084 | u64 alloc_hint, int mode, loff_t actual_len) |
d899e052 | 6085 | { |
5a303d5d | 6086 | struct btrfs_trans_handle *trans; |
d899e052 YZ |
6087 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6088 | struct btrfs_key ins; | |
d899e052 YZ |
6089 | u64 cur_offset = start; |
6090 | u64 num_bytes = end - start; | |
6091 | int ret = 0; | |
d1ea6a61 | 6092 | u64 i_size; |
d899e052 | 6093 | |
d899e052 | 6094 | while (num_bytes > 0) { |
a22285a6 YZ |
6095 | trans = btrfs_start_transaction(root, 3); |
6096 | if (IS_ERR(trans)) { | |
6097 | ret = PTR_ERR(trans); | |
6098 | break; | |
6099 | } | |
3a1abec9 | 6100 | |
287a0ab9 | 6101 | ret = btrfs_reserve_extent(trans, root, num_bytes, |
d899e052 YZ |
6102 | root->sectorsize, 0, alloc_hint, |
6103 | (u64)-1, &ins, 1); | |
6104 | if (ret) { | |
a22285a6 YZ |
6105 | btrfs_end_transaction(trans, root); |
6106 | break; | |
d899e052 | 6107 | } |
5a303d5d | 6108 | |
d899e052 YZ |
6109 | ret = insert_reserved_file_extent(trans, inode, |
6110 | cur_offset, ins.objectid, | |
6111 | ins.offset, ins.offset, | |
920bbbfb | 6112 | ins.offset, 0, 0, 0, |
d899e052 YZ |
6113 | BTRFS_FILE_EXTENT_PREALLOC); |
6114 | BUG_ON(ret); | |
a1ed835e CM |
6115 | btrfs_drop_extent_cache(inode, cur_offset, |
6116 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 6117 | |
d899e052 YZ |
6118 | num_bytes -= ins.offset; |
6119 | cur_offset += ins.offset; | |
6120 | alloc_hint = ins.objectid + ins.offset; | |
5a303d5d | 6121 | |
d899e052 | 6122 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 6123 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 6124 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
23b5c509 AK |
6125 | (actual_len > inode->i_size) && |
6126 | (cur_offset > inode->i_size)) { | |
6127 | ||
d1ea6a61 AK |
6128 | if (cur_offset > actual_len) |
6129 | i_size = actual_len; | |
6130 | else | |
6131 | i_size = cur_offset; | |
6132 | i_size_write(inode, i_size); | |
6133 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
6134 | } |
6135 | ||
d899e052 YZ |
6136 | ret = btrfs_update_inode(trans, root, inode); |
6137 | BUG_ON(ret); | |
d899e052 | 6138 | |
5a303d5d | 6139 | btrfs_end_transaction(trans, root); |
5a303d5d | 6140 | } |
d899e052 YZ |
6141 | return ret; |
6142 | } | |
6143 | ||
6144 | static long btrfs_fallocate(struct inode *inode, int mode, | |
6145 | loff_t offset, loff_t len) | |
6146 | { | |
2ac55d41 | 6147 | struct extent_state *cached_state = NULL; |
d899e052 YZ |
6148 | u64 cur_offset; |
6149 | u64 last_byte; | |
6150 | u64 alloc_start; | |
6151 | u64 alloc_end; | |
6152 | u64 alloc_hint = 0; | |
e980b50c | 6153 | u64 locked_end; |
d899e052 YZ |
6154 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
6155 | struct extent_map *em; | |
6156 | int ret; | |
6157 | ||
6158 | alloc_start = offset & ~mask; | |
6159 | alloc_end = (offset + len + mask) & ~mask; | |
6160 | ||
546888da CM |
6161 | /* |
6162 | * wait for ordered IO before we have any locks. We'll loop again | |
6163 | * below with the locks held. | |
6164 | */ | |
6165 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
6166 | ||
d899e052 YZ |
6167 | mutex_lock(&inode->i_mutex); |
6168 | if (alloc_start > inode->i_size) { | |
6169 | ret = btrfs_cont_expand(inode, alloc_start); | |
6170 | if (ret) | |
6171 | goto out; | |
6172 | } | |
6173 | ||
0ca1f7ce | 6174 | ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start); |
a970b0a1 JB |
6175 | if (ret) |
6176 | goto out; | |
6177 | ||
e980b50c | 6178 | locked_end = alloc_end - 1; |
d899e052 YZ |
6179 | while (1) { |
6180 | struct btrfs_ordered_extent *ordered; | |
546888da | 6181 | |
546888da CM |
6182 | /* the extent lock is ordered inside the running |
6183 | * transaction | |
6184 | */ | |
2ac55d41 JB |
6185 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, |
6186 | locked_end, 0, &cached_state, GFP_NOFS); | |
d899e052 YZ |
6187 | ordered = btrfs_lookup_first_ordered_extent(inode, |
6188 | alloc_end - 1); | |
6189 | if (ordered && | |
6190 | ordered->file_offset + ordered->len > alloc_start && | |
6191 | ordered->file_offset < alloc_end) { | |
6192 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
6193 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
6194 | alloc_start, locked_end, | |
6195 | &cached_state, GFP_NOFS); | |
546888da CM |
6196 | /* |
6197 | * we can't wait on the range with the transaction | |
6198 | * running or with the extent lock held | |
6199 | */ | |
d899e052 YZ |
6200 | btrfs_wait_ordered_range(inode, alloc_start, |
6201 | alloc_end - alloc_start); | |
6202 | } else { | |
6203 | if (ordered) | |
6204 | btrfs_put_ordered_extent(ordered); | |
6205 | break; | |
6206 | } | |
6207 | } | |
6208 | ||
6209 | cur_offset = alloc_start; | |
6210 | while (1) { | |
6211 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
6212 | alloc_end - cur_offset, 0); | |
6213 | BUG_ON(IS_ERR(em) || !em); | |
6214 | last_byte = min(extent_map_end(em), alloc_end); | |
6215 | last_byte = (last_byte + mask) & ~mask; | |
5a303d5d YZ |
6216 | if (em->block_start == EXTENT_MAP_HOLE || |
6217 | (cur_offset >= inode->i_size && | |
6218 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6219 | ret = prealloc_file_range(inode, | |
6220 | cur_offset, last_byte, | |
d1ea6a61 | 6221 | alloc_hint, mode, offset+len); |
d899e052 YZ |
6222 | if (ret < 0) { |
6223 | free_extent_map(em); | |
6224 | break; | |
6225 | } | |
6226 | } | |
6227 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
6228 | alloc_hint = em->block_start; | |
6229 | free_extent_map(em); | |
6230 | ||
6231 | cur_offset = last_byte; | |
6232 | if (cur_offset >= alloc_end) { | |
6233 | ret = 0; | |
6234 | break; | |
6235 | } | |
6236 | } | |
2ac55d41 JB |
6237 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
6238 | &cached_state, GFP_NOFS); | |
546888da | 6239 | |
0ca1f7ce | 6240 | btrfs_free_reserved_data_space(inode, alloc_end - alloc_start); |
d899e052 YZ |
6241 | out: |
6242 | mutex_unlock(&inode->i_mutex); | |
6243 | return ret; | |
6244 | } | |
6245 | ||
e6dcd2dc CM |
6246 | static int btrfs_set_page_dirty(struct page *page) |
6247 | { | |
e6dcd2dc CM |
6248 | return __set_page_dirty_nobuffers(page); |
6249 | } | |
6250 | ||
0ee0fda0 | 6251 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 6252 | { |
6cbff00f | 6253 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 6254 | return -EACCES; |
33268eaf | 6255 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 6256 | } |
39279cc3 | 6257 | |
6e1d5dcc | 6258 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 6259 | .getattr = btrfs_getattr, |
39279cc3 CM |
6260 | .lookup = btrfs_lookup, |
6261 | .create = btrfs_create, | |
6262 | .unlink = btrfs_unlink, | |
6263 | .link = btrfs_link, | |
6264 | .mkdir = btrfs_mkdir, | |
6265 | .rmdir = btrfs_rmdir, | |
6266 | .rename = btrfs_rename, | |
6267 | .symlink = btrfs_symlink, | |
6268 | .setattr = btrfs_setattr, | |
618e21d5 | 6269 | .mknod = btrfs_mknod, |
95819c05 CH |
6270 | .setxattr = btrfs_setxattr, |
6271 | .getxattr = btrfs_getxattr, | |
5103e947 | 6272 | .listxattr = btrfs_listxattr, |
95819c05 | 6273 | .removexattr = btrfs_removexattr, |
fdebe2bd | 6274 | .permission = btrfs_permission, |
39279cc3 | 6275 | }; |
6e1d5dcc | 6276 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 6277 | .lookup = btrfs_lookup, |
fdebe2bd | 6278 | .permission = btrfs_permission, |
39279cc3 | 6279 | }; |
76dda93c | 6280 | |
828c0950 | 6281 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
6282 | .llseek = generic_file_llseek, |
6283 | .read = generic_read_dir, | |
cbdf5a24 | 6284 | .readdir = btrfs_real_readdir, |
34287aa3 | 6285 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 6286 | #ifdef CONFIG_COMPAT |
34287aa3 | 6287 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 6288 | #endif |
6bf13c0c | 6289 | .release = btrfs_release_file, |
e02119d5 | 6290 | .fsync = btrfs_sync_file, |
39279cc3 CM |
6291 | }; |
6292 | ||
d1310b2e | 6293 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 6294 | .fill_delalloc = run_delalloc_range, |
065631f6 | 6295 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 6296 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 6297 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 6298 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 6299 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 6300 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
6301 | .set_bit_hook = btrfs_set_bit_hook, |
6302 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
6303 | .merge_extent_hook = btrfs_merge_extent_hook, |
6304 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
6305 | }; |
6306 | ||
35054394 CM |
6307 | /* |
6308 | * btrfs doesn't support the bmap operation because swapfiles | |
6309 | * use bmap to make a mapping of extents in the file. They assume | |
6310 | * these extents won't change over the life of the file and they | |
6311 | * use the bmap result to do IO directly to the drive. | |
6312 | * | |
6313 | * the btrfs bmap call would return logical addresses that aren't | |
6314 | * suitable for IO and they also will change frequently as COW | |
6315 | * operations happen. So, swapfile + btrfs == corruption. | |
6316 | * | |
6317 | * For now we're avoiding this by dropping bmap. | |
6318 | */ | |
7f09410b | 6319 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
6320 | .readpage = btrfs_readpage, |
6321 | .writepage = btrfs_writepage, | |
b293f02e | 6322 | .writepages = btrfs_writepages, |
3ab2fb5a | 6323 | .readpages = btrfs_readpages, |
39279cc3 | 6324 | .sync_page = block_sync_page, |
16432985 | 6325 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
6326 | .invalidatepage = btrfs_invalidatepage, |
6327 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 6328 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 6329 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
6330 | }; |
6331 | ||
7f09410b | 6332 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
6333 | .readpage = btrfs_readpage, |
6334 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
6335 | .invalidatepage = btrfs_invalidatepage, |
6336 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
6337 | }; |
6338 | ||
6e1d5dcc | 6339 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
6340 | .truncate = btrfs_truncate, |
6341 | .getattr = btrfs_getattr, | |
6342 | .setattr = btrfs_setattr, | |
95819c05 CH |
6343 | .setxattr = btrfs_setxattr, |
6344 | .getxattr = btrfs_getxattr, | |
5103e947 | 6345 | .listxattr = btrfs_listxattr, |
95819c05 | 6346 | .removexattr = btrfs_removexattr, |
fdebe2bd | 6347 | .permission = btrfs_permission, |
d899e052 | 6348 | .fallocate = btrfs_fallocate, |
1506fcc8 | 6349 | .fiemap = btrfs_fiemap, |
39279cc3 | 6350 | }; |
6e1d5dcc | 6351 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
6352 | .getattr = btrfs_getattr, |
6353 | .setattr = btrfs_setattr, | |
fdebe2bd | 6354 | .permission = btrfs_permission, |
95819c05 CH |
6355 | .setxattr = btrfs_setxattr, |
6356 | .getxattr = btrfs_getxattr, | |
33268eaf | 6357 | .listxattr = btrfs_listxattr, |
95819c05 | 6358 | .removexattr = btrfs_removexattr, |
618e21d5 | 6359 | }; |
6e1d5dcc | 6360 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
6361 | .readlink = generic_readlink, |
6362 | .follow_link = page_follow_link_light, | |
6363 | .put_link = page_put_link, | |
fdebe2bd | 6364 | .permission = btrfs_permission, |
0279b4cd JO |
6365 | .setxattr = btrfs_setxattr, |
6366 | .getxattr = btrfs_getxattr, | |
6367 | .listxattr = btrfs_listxattr, | |
6368 | .removexattr = btrfs_removexattr, | |
39279cc3 | 6369 | }; |
76dda93c | 6370 | |
82d339d9 | 6371 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
6372 | .d_delete = btrfs_dentry_delete, |
6373 | }; |