]>
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> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
4b4e25f2 | 41 | #include "compat.h" |
39279cc3 CM |
42 | #include "ctree.h" |
43 | #include "disk-io.h" | |
44 | #include "transaction.h" | |
45 | #include "btrfs_inode.h" | |
46 | #include "ioctl.h" | |
47 | #include "print-tree.h" | |
0b86a832 | 48 | #include "volumes.h" |
e6dcd2dc | 49 | #include "ordered-data.h" |
95819c05 | 50 | #include "xattr.h" |
e02119d5 | 51 | #include "tree-log.h" |
c8b97818 | 52 | #include "compression.h" |
b4ce94de | 53 | #include "locking.h" |
dc89e982 | 54 | #include "free-space-cache.h" |
581bb050 | 55 | #include "inode-map.h" |
39279cc3 CM |
56 | |
57 | struct btrfs_iget_args { | |
58 | u64 ino; | |
59 | struct btrfs_root *root; | |
60 | }; | |
61 | ||
6e1d5dcc AD |
62 | static const struct inode_operations btrfs_dir_inode_operations; |
63 | static const struct inode_operations btrfs_symlink_inode_operations; | |
64 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
65 | static const struct inode_operations btrfs_special_inode_operations; | |
66 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
67 | static const struct address_space_operations btrfs_aops; |
68 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 69 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 70 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
71 | |
72 | static struct kmem_cache *btrfs_inode_cachep; | |
73 | struct kmem_cache *btrfs_trans_handle_cachep; | |
74 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 75 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 76 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
77 | |
78 | #define S_SHIFT 12 | |
79 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
80 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
81 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
82 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
83 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
84 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
85 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
86 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
87 | }; | |
88 | ||
a41ad394 JB |
89 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
90 | static int btrfs_truncate(struct inode *inode); | |
c8b97818 | 91 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
92 | static noinline int cow_file_range(struct inode *inode, |
93 | struct page *locked_page, | |
94 | u64 start, u64 end, int *page_started, | |
95 | unsigned long *nr_written, int unlock); | |
7b128766 | 96 | |
f34f57a3 | 97 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
98 | struct inode *inode, struct inode *dir, |
99 | const struct qstr *qstr) | |
0279b4cd JO |
100 | { |
101 | int err; | |
102 | ||
f34f57a3 | 103 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 104 | if (!err) |
2a7dba39 | 105 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
106 | return err; |
107 | } | |
108 | ||
c8b97818 CM |
109 | /* |
110 | * this does all the hard work for inserting an inline extent into | |
111 | * the btree. The caller should have done a btrfs_drop_extents so that | |
112 | * no overlapping inline items exist in the btree | |
113 | */ | |
d397712b | 114 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
115 | struct btrfs_root *root, struct inode *inode, |
116 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 117 | int compress_type, |
c8b97818 CM |
118 | struct page **compressed_pages) |
119 | { | |
120 | struct btrfs_key key; | |
121 | struct btrfs_path *path; | |
122 | struct extent_buffer *leaf; | |
123 | struct page *page = NULL; | |
124 | char *kaddr; | |
125 | unsigned long ptr; | |
126 | struct btrfs_file_extent_item *ei; | |
127 | int err = 0; | |
128 | int ret; | |
129 | size_t cur_size = size; | |
130 | size_t datasize; | |
131 | unsigned long offset; | |
c8b97818 | 132 | |
fe3f566c | 133 | if (compressed_size && compressed_pages) |
c8b97818 | 134 | cur_size = compressed_size; |
c8b97818 | 135 | |
d397712b CM |
136 | path = btrfs_alloc_path(); |
137 | if (!path) | |
c8b97818 CM |
138 | return -ENOMEM; |
139 | ||
b9473439 | 140 | path->leave_spinning = 1; |
c8b97818 | 141 | |
33345d01 | 142 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
143 | key.offset = start; |
144 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
145 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
146 | ||
147 | inode_add_bytes(inode, size); | |
148 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
149 | datasize); | |
150 | BUG_ON(ret); | |
151 | if (ret) { | |
152 | err = ret; | |
c8b97818 CM |
153 | goto fail; |
154 | } | |
155 | leaf = path->nodes[0]; | |
156 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
157 | struct btrfs_file_extent_item); | |
158 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
159 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
160 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
161 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
162 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
163 | ptr = btrfs_file_extent_inline_start(ei); | |
164 | ||
261507a0 | 165 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
166 | struct page *cpage; |
167 | int i = 0; | |
d397712b | 168 | while (compressed_size > 0) { |
c8b97818 | 169 | cpage = compressed_pages[i]; |
5b050f04 | 170 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
171 | PAGE_CACHE_SIZE); |
172 | ||
b9473439 | 173 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 174 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 175 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
176 | |
177 | i++; | |
178 | ptr += cur_size; | |
179 | compressed_size -= cur_size; | |
180 | } | |
181 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 182 | compress_type); |
c8b97818 CM |
183 | } else { |
184 | page = find_get_page(inode->i_mapping, | |
185 | start >> PAGE_CACHE_SHIFT); | |
186 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
187 | kaddr = kmap_atomic(page, KM_USER0); | |
188 | offset = start & (PAGE_CACHE_SIZE - 1); | |
189 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
190 | kunmap_atomic(kaddr, KM_USER0); | |
191 | page_cache_release(page); | |
192 | } | |
193 | btrfs_mark_buffer_dirty(leaf); | |
194 | btrfs_free_path(path); | |
195 | ||
c2167754 YZ |
196 | /* |
197 | * we're an inline extent, so nobody can | |
198 | * extend the file past i_size without locking | |
199 | * a page we already have locked. | |
200 | * | |
201 | * We must do any isize and inode updates | |
202 | * before we unlock the pages. Otherwise we | |
203 | * could end up racing with unlink. | |
204 | */ | |
c8b97818 CM |
205 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
206 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 207 | |
c8b97818 CM |
208 | return 0; |
209 | fail: | |
210 | btrfs_free_path(path); | |
211 | return err; | |
212 | } | |
213 | ||
214 | ||
215 | /* | |
216 | * conditionally insert an inline extent into the file. This | |
217 | * does the checks required to make sure the data is small enough | |
218 | * to fit as an inline extent. | |
219 | */ | |
7f366cfe | 220 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
221 | struct btrfs_root *root, |
222 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 223 | size_t compressed_size, int compress_type, |
c8b97818 CM |
224 | struct page **compressed_pages) |
225 | { | |
226 | u64 isize = i_size_read(inode); | |
227 | u64 actual_end = min(end + 1, isize); | |
228 | u64 inline_len = actual_end - start; | |
229 | u64 aligned_end = (end + root->sectorsize - 1) & | |
230 | ~((u64)root->sectorsize - 1); | |
231 | u64 hint_byte; | |
232 | u64 data_len = inline_len; | |
233 | int ret; | |
234 | ||
235 | if (compressed_size) | |
236 | data_len = compressed_size; | |
237 | ||
238 | if (start > 0 || | |
70b99e69 | 239 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
240 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
241 | (!compressed_size && | |
242 | (actual_end & (root->sectorsize - 1)) == 0) || | |
243 | end + 1 < isize || | |
244 | data_len > root->fs_info->max_inline) { | |
245 | return 1; | |
246 | } | |
247 | ||
920bbbfb | 248 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 249 | &hint_byte, 1); |
c8b97818 CM |
250 | BUG_ON(ret); |
251 | ||
252 | if (isize > actual_end) | |
253 | inline_len = min_t(u64, isize, actual_end); | |
254 | ret = insert_inline_extent(trans, root, inode, start, | |
255 | inline_len, compressed_size, | |
fe3f566c | 256 | compress_type, compressed_pages); |
c8b97818 | 257 | BUG_ON(ret); |
0ca1f7ce | 258 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 259 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
260 | return 0; |
261 | } | |
262 | ||
771ed689 CM |
263 | struct async_extent { |
264 | u64 start; | |
265 | u64 ram_size; | |
266 | u64 compressed_size; | |
267 | struct page **pages; | |
268 | unsigned long nr_pages; | |
261507a0 | 269 | int compress_type; |
771ed689 CM |
270 | struct list_head list; |
271 | }; | |
272 | ||
273 | struct async_cow { | |
274 | struct inode *inode; | |
275 | struct btrfs_root *root; | |
276 | struct page *locked_page; | |
277 | u64 start; | |
278 | u64 end; | |
279 | struct list_head extents; | |
280 | struct btrfs_work work; | |
281 | }; | |
282 | ||
283 | static noinline int add_async_extent(struct async_cow *cow, | |
284 | u64 start, u64 ram_size, | |
285 | u64 compressed_size, | |
286 | struct page **pages, | |
261507a0 LZ |
287 | unsigned long nr_pages, |
288 | int compress_type) | |
771ed689 CM |
289 | { |
290 | struct async_extent *async_extent; | |
291 | ||
292 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
dac97e51 | 293 | BUG_ON(!async_extent); |
771ed689 CM |
294 | async_extent->start = start; |
295 | async_extent->ram_size = ram_size; | |
296 | async_extent->compressed_size = compressed_size; | |
297 | async_extent->pages = pages; | |
298 | async_extent->nr_pages = nr_pages; | |
261507a0 | 299 | async_extent->compress_type = compress_type; |
771ed689 CM |
300 | list_add_tail(&async_extent->list, &cow->extents); |
301 | return 0; | |
302 | } | |
303 | ||
d352ac68 | 304 | /* |
771ed689 CM |
305 | * we create compressed extents in two phases. The first |
306 | * phase compresses a range of pages that have already been | |
307 | * locked (both pages and state bits are locked). | |
c8b97818 | 308 | * |
771ed689 CM |
309 | * This is done inside an ordered work queue, and the compression |
310 | * is spread across many cpus. The actual IO submission is step | |
311 | * two, and the ordered work queue takes care of making sure that | |
312 | * happens in the same order things were put onto the queue by | |
313 | * writepages and friends. | |
c8b97818 | 314 | * |
771ed689 CM |
315 | * If this code finds it can't get good compression, it puts an |
316 | * entry onto the work queue to write the uncompressed bytes. This | |
317 | * makes sure that both compressed inodes and uncompressed inodes | |
318 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 319 | */ |
771ed689 CM |
320 | static noinline int compress_file_range(struct inode *inode, |
321 | struct page *locked_page, | |
322 | u64 start, u64 end, | |
323 | struct async_cow *async_cow, | |
324 | int *num_added) | |
b888db2b CM |
325 | { |
326 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
327 | struct btrfs_trans_handle *trans; | |
db94535d | 328 | u64 num_bytes; |
db94535d | 329 | u64 blocksize = root->sectorsize; |
c8b97818 | 330 | u64 actual_end; |
42dc7bab | 331 | u64 isize = i_size_read(inode); |
e6dcd2dc | 332 | int ret = 0; |
c8b97818 CM |
333 | struct page **pages = NULL; |
334 | unsigned long nr_pages; | |
335 | unsigned long nr_pages_ret = 0; | |
336 | unsigned long total_compressed = 0; | |
337 | unsigned long total_in = 0; | |
338 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 339 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
340 | int i; |
341 | int will_compress; | |
261507a0 | 342 | int compress_type = root->fs_info->compress_type; |
b888db2b | 343 | |
4cb5300b CM |
344 | /* if this is a small write inside eof, kick off a defragbot */ |
345 | if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024) | |
346 | btrfs_add_inode_defrag(NULL, inode); | |
347 | ||
42dc7bab | 348 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
349 | again: |
350 | will_compress = 0; | |
351 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
352 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 353 | |
f03d9301 CM |
354 | /* |
355 | * we don't want to send crud past the end of i_size through | |
356 | * compression, that's just a waste of CPU time. So, if the | |
357 | * end of the file is before the start of our current | |
358 | * requested range of bytes, we bail out to the uncompressed | |
359 | * cleanup code that can deal with all of this. | |
360 | * | |
361 | * It isn't really the fastest way to fix things, but this is a | |
362 | * very uncommon corner. | |
363 | */ | |
364 | if (actual_end <= start) | |
365 | goto cleanup_and_bail_uncompressed; | |
366 | ||
c8b97818 CM |
367 | total_compressed = actual_end - start; |
368 | ||
369 | /* we want to make sure that amount of ram required to uncompress | |
370 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
371 | * of a compressed extent to 128k. This is a crucial number |
372 | * because it also controls how easily we can spread reads across | |
373 | * cpus for decompression. | |
374 | * | |
375 | * We also want to make sure the amount of IO required to do | |
376 | * a random read is reasonably small, so we limit the size of | |
377 | * a compressed extent to 128k. | |
c8b97818 CM |
378 | */ |
379 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 380 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 381 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
382 | total_in = 0; |
383 | ret = 0; | |
db94535d | 384 | |
771ed689 CM |
385 | /* |
386 | * we do compression for mount -o compress and when the | |
387 | * inode has not been flagged as nocompress. This flag can | |
388 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 389 | */ |
6cbff00f | 390 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 391 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
392 | (BTRFS_I(inode)->force_compress) || |
393 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 394 | WARN_ON(pages); |
cfbc246e | 395 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
dac97e51 | 396 | BUG_ON(!pages); |
c8b97818 | 397 | |
261507a0 LZ |
398 | if (BTRFS_I(inode)->force_compress) |
399 | compress_type = BTRFS_I(inode)->force_compress; | |
400 | ||
401 | ret = btrfs_compress_pages(compress_type, | |
402 | inode->i_mapping, start, | |
403 | total_compressed, pages, | |
404 | nr_pages, &nr_pages_ret, | |
405 | &total_in, | |
406 | &total_compressed, | |
407 | max_compressed); | |
c8b97818 CM |
408 | |
409 | if (!ret) { | |
410 | unsigned long offset = total_compressed & | |
411 | (PAGE_CACHE_SIZE - 1); | |
412 | struct page *page = pages[nr_pages_ret - 1]; | |
413 | char *kaddr; | |
414 | ||
415 | /* zero the tail end of the last page, we might be | |
416 | * sending it down to disk | |
417 | */ | |
418 | if (offset) { | |
419 | kaddr = kmap_atomic(page, KM_USER0); | |
420 | memset(kaddr + offset, 0, | |
421 | PAGE_CACHE_SIZE - offset); | |
422 | kunmap_atomic(kaddr, KM_USER0); | |
423 | } | |
424 | will_compress = 1; | |
425 | } | |
426 | } | |
427 | if (start == 0) { | |
7a7eaa40 | 428 | trans = btrfs_join_transaction(root); |
3612b495 | 429 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 430 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 431 | |
c8b97818 | 432 | /* lets try to make an inline extent */ |
771ed689 | 433 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 434 | /* we didn't compress the entire range, try |
771ed689 | 435 | * to make an uncompressed inline extent. |
c8b97818 CM |
436 | */ |
437 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 438 | start, end, 0, 0, NULL); |
c8b97818 | 439 | } else { |
771ed689 | 440 | /* try making a compressed inline extent */ |
c8b97818 CM |
441 | ret = cow_file_range_inline(trans, root, inode, |
442 | start, end, | |
fe3f566c LZ |
443 | total_compressed, |
444 | compress_type, pages); | |
c8b97818 CM |
445 | } |
446 | if (ret == 0) { | |
771ed689 CM |
447 | /* |
448 | * inline extent creation worked, we don't need | |
449 | * to create any more async work items. Unlock | |
450 | * and free up our temp pages. | |
451 | */ | |
c8b97818 | 452 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
453 | &BTRFS_I(inode)->io_tree, |
454 | start, end, NULL, | |
455 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 456 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 457 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
458 | |
459 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
460 | goto free_pages_out; |
461 | } | |
c2167754 | 462 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
463 | } |
464 | ||
465 | if (will_compress) { | |
466 | /* | |
467 | * we aren't doing an inline extent round the compressed size | |
468 | * up to a block size boundary so the allocator does sane | |
469 | * things | |
470 | */ | |
471 | total_compressed = (total_compressed + blocksize - 1) & | |
472 | ~(blocksize - 1); | |
473 | ||
474 | /* | |
475 | * one last check to make sure the compression is really a | |
476 | * win, compare the page count read with the blocks on disk | |
477 | */ | |
478 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
479 | ~(PAGE_CACHE_SIZE - 1); | |
480 | if (total_compressed >= total_in) { | |
481 | will_compress = 0; | |
482 | } else { | |
c8b97818 CM |
483 | num_bytes = total_in; |
484 | } | |
485 | } | |
486 | if (!will_compress && pages) { | |
487 | /* | |
488 | * the compression code ran but failed to make things smaller, | |
489 | * free any pages it allocated and our page pointer array | |
490 | */ | |
491 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 492 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
493 | page_cache_release(pages[i]); |
494 | } | |
495 | kfree(pages); | |
496 | pages = NULL; | |
497 | total_compressed = 0; | |
498 | nr_pages_ret = 0; | |
499 | ||
500 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
501 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
502 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 503 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 504 | } |
c8b97818 | 505 | } |
771ed689 CM |
506 | if (will_compress) { |
507 | *num_added += 1; | |
c8b97818 | 508 | |
771ed689 CM |
509 | /* the async work queues will take care of doing actual |
510 | * allocation on disk for these compressed pages, | |
511 | * and will submit them to the elevator. | |
512 | */ | |
513 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
514 | total_compressed, pages, nr_pages_ret, |
515 | compress_type); | |
179e29e4 | 516 | |
24ae6365 | 517 | if (start + num_bytes < end) { |
771ed689 CM |
518 | start += num_bytes; |
519 | pages = NULL; | |
520 | cond_resched(); | |
521 | goto again; | |
522 | } | |
523 | } else { | |
f03d9301 | 524 | cleanup_and_bail_uncompressed: |
771ed689 CM |
525 | /* |
526 | * No compression, but we still need to write the pages in | |
527 | * the file we've been given so far. redirty the locked | |
528 | * page if it corresponds to our extent and set things up | |
529 | * for the async work queue to run cow_file_range to do | |
530 | * the normal delalloc dance | |
531 | */ | |
532 | if (page_offset(locked_page) >= start && | |
533 | page_offset(locked_page) <= end) { | |
534 | __set_page_dirty_nobuffers(locked_page); | |
535 | /* unlocked later on in the async handlers */ | |
536 | } | |
261507a0 LZ |
537 | add_async_extent(async_cow, start, end - start + 1, |
538 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
539 | *num_added += 1; |
540 | } | |
3b951516 | 541 | |
771ed689 CM |
542 | out: |
543 | return 0; | |
544 | ||
545 | free_pages_out: | |
546 | for (i = 0; i < nr_pages_ret; i++) { | |
547 | WARN_ON(pages[i]->mapping); | |
548 | page_cache_release(pages[i]); | |
549 | } | |
d397712b | 550 | kfree(pages); |
771ed689 CM |
551 | |
552 | goto out; | |
553 | } | |
554 | ||
555 | /* | |
556 | * phase two of compressed writeback. This is the ordered portion | |
557 | * of the code, which only gets called in the order the work was | |
558 | * queued. We walk all the async extents created by compress_file_range | |
559 | * and send them down to the disk. | |
560 | */ | |
561 | static noinline int submit_compressed_extents(struct inode *inode, | |
562 | struct async_cow *async_cow) | |
563 | { | |
564 | struct async_extent *async_extent; | |
565 | u64 alloc_hint = 0; | |
566 | struct btrfs_trans_handle *trans; | |
567 | struct btrfs_key ins; | |
568 | struct extent_map *em; | |
569 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
570 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
571 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 572 | int ret = 0; |
771ed689 CM |
573 | |
574 | if (list_empty(&async_cow->extents)) | |
575 | return 0; | |
576 | ||
771ed689 | 577 | |
d397712b | 578 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
579 | async_extent = list_entry(async_cow->extents.next, |
580 | struct async_extent, list); | |
581 | list_del(&async_extent->list); | |
c8b97818 | 582 | |
771ed689 CM |
583 | io_tree = &BTRFS_I(inode)->io_tree; |
584 | ||
f5a84ee3 | 585 | retry: |
771ed689 CM |
586 | /* did the compression code fall back to uncompressed IO? */ |
587 | if (!async_extent->pages) { | |
588 | int page_started = 0; | |
589 | unsigned long nr_written = 0; | |
590 | ||
591 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 JB |
592 | async_extent->start + |
593 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
594 | |
595 | /* allocate blocks */ | |
f5a84ee3 JB |
596 | ret = cow_file_range(inode, async_cow->locked_page, |
597 | async_extent->start, | |
598 | async_extent->start + | |
599 | async_extent->ram_size - 1, | |
600 | &page_started, &nr_written, 0); | |
771ed689 CM |
601 | |
602 | /* | |
603 | * if page_started, cow_file_range inserted an | |
604 | * inline extent and took care of all the unlocking | |
605 | * and IO for us. Otherwise, we need to submit | |
606 | * all those pages down to the drive. | |
607 | */ | |
f5a84ee3 | 608 | if (!page_started && !ret) |
771ed689 CM |
609 | extent_write_locked_range(io_tree, |
610 | inode, async_extent->start, | |
d397712b | 611 | async_extent->start + |
771ed689 CM |
612 | async_extent->ram_size - 1, |
613 | btrfs_get_extent, | |
614 | WB_SYNC_ALL); | |
615 | kfree(async_extent); | |
616 | cond_resched(); | |
617 | continue; | |
618 | } | |
619 | ||
620 | lock_extent(io_tree, async_extent->start, | |
621 | async_extent->start + async_extent->ram_size - 1, | |
622 | GFP_NOFS); | |
771ed689 | 623 | |
7a7eaa40 | 624 | trans = btrfs_join_transaction(root); |
3612b495 | 625 | BUG_ON(IS_ERR(trans)); |
74b21075 | 626 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 CM |
627 | ret = btrfs_reserve_extent(trans, root, |
628 | async_extent->compressed_size, | |
629 | async_extent->compressed_size, | |
630 | 0, alloc_hint, | |
631 | (u64)-1, &ins, 1); | |
c2167754 YZ |
632 | btrfs_end_transaction(trans, root); |
633 | ||
f5a84ee3 JB |
634 | if (ret) { |
635 | int i; | |
636 | for (i = 0; i < async_extent->nr_pages; i++) { | |
637 | WARN_ON(async_extent->pages[i]->mapping); | |
638 | page_cache_release(async_extent->pages[i]); | |
639 | } | |
640 | kfree(async_extent->pages); | |
641 | async_extent->nr_pages = 0; | |
642 | async_extent->pages = NULL; | |
643 | unlock_extent(io_tree, async_extent->start, | |
644 | async_extent->start + | |
645 | async_extent->ram_size - 1, GFP_NOFS); | |
646 | goto retry; | |
647 | } | |
648 | ||
c2167754 YZ |
649 | /* |
650 | * here we're doing allocation and writeback of the | |
651 | * compressed pages | |
652 | */ | |
653 | btrfs_drop_extent_cache(inode, async_extent->start, | |
654 | async_extent->start + | |
655 | async_extent->ram_size - 1, 0); | |
656 | ||
172ddd60 | 657 | em = alloc_extent_map(); |
c26a9203 | 658 | BUG_ON(!em); |
771ed689 CM |
659 | em->start = async_extent->start; |
660 | em->len = async_extent->ram_size; | |
445a6944 | 661 | em->orig_start = em->start; |
c8b97818 | 662 | |
771ed689 CM |
663 | em->block_start = ins.objectid; |
664 | em->block_len = ins.offset; | |
665 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 666 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
667 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
668 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
669 | ||
d397712b | 670 | while (1) { |
890871be | 671 | write_lock(&em_tree->lock); |
771ed689 | 672 | ret = add_extent_mapping(em_tree, em); |
890871be | 673 | write_unlock(&em_tree->lock); |
771ed689 CM |
674 | if (ret != -EEXIST) { |
675 | free_extent_map(em); | |
676 | break; | |
677 | } | |
678 | btrfs_drop_extent_cache(inode, async_extent->start, | |
679 | async_extent->start + | |
680 | async_extent->ram_size - 1, 0); | |
681 | } | |
682 | ||
261507a0 LZ |
683 | ret = btrfs_add_ordered_extent_compress(inode, |
684 | async_extent->start, | |
685 | ins.objectid, | |
686 | async_extent->ram_size, | |
687 | ins.offset, | |
688 | BTRFS_ORDERED_COMPRESSED, | |
689 | async_extent->compress_type); | |
771ed689 CM |
690 | BUG_ON(ret); |
691 | ||
771ed689 CM |
692 | /* |
693 | * clear dirty, set writeback and unlock the pages. | |
694 | */ | |
695 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
696 | &BTRFS_I(inode)->io_tree, |
697 | async_extent->start, | |
698 | async_extent->start + | |
699 | async_extent->ram_size - 1, | |
700 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
701 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 702 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 703 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
704 | |
705 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
706 | async_extent->start, |
707 | async_extent->ram_size, | |
708 | ins.objectid, | |
709 | ins.offset, async_extent->pages, | |
710 | async_extent->nr_pages); | |
771ed689 CM |
711 | |
712 | BUG_ON(ret); | |
771ed689 CM |
713 | alloc_hint = ins.objectid + ins.offset; |
714 | kfree(async_extent); | |
715 | cond_resched(); | |
716 | } | |
717 | ||
771ed689 CM |
718 | return 0; |
719 | } | |
720 | ||
4b46fce2 JB |
721 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
722 | u64 num_bytes) | |
723 | { | |
724 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
725 | struct extent_map *em; | |
726 | u64 alloc_hint = 0; | |
727 | ||
728 | read_lock(&em_tree->lock); | |
729 | em = search_extent_mapping(em_tree, start, num_bytes); | |
730 | if (em) { | |
731 | /* | |
732 | * if block start isn't an actual block number then find the | |
733 | * first block in this inode and use that as a hint. If that | |
734 | * block is also bogus then just don't worry about it. | |
735 | */ | |
736 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
737 | free_extent_map(em); | |
738 | em = search_extent_mapping(em_tree, 0, 0); | |
739 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
740 | alloc_hint = em->block_start; | |
741 | if (em) | |
742 | free_extent_map(em); | |
743 | } else { | |
744 | alloc_hint = em->block_start; | |
745 | free_extent_map(em); | |
746 | } | |
747 | } | |
748 | read_unlock(&em_tree->lock); | |
749 | ||
750 | return alloc_hint; | |
751 | } | |
752 | ||
82d5902d LZ |
753 | static inline bool is_free_space_inode(struct btrfs_root *root, |
754 | struct inode *inode) | |
755 | { | |
756 | if (root == root->fs_info->tree_root || | |
757 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) | |
758 | return true; | |
759 | return false; | |
760 | } | |
761 | ||
771ed689 CM |
762 | /* |
763 | * when extent_io.c finds a delayed allocation range in the file, | |
764 | * the call backs end up in this code. The basic idea is to | |
765 | * allocate extents on disk for the range, and create ordered data structs | |
766 | * in ram to track those extents. | |
767 | * | |
768 | * locked_page is the page that writepage had locked already. We use | |
769 | * it to make sure we don't do extra locks or unlocks. | |
770 | * | |
771 | * *page_started is set to one if we unlock locked_page and do everything | |
772 | * required to start IO on it. It may be clean and already done with | |
773 | * IO when we return. | |
774 | */ | |
775 | static noinline int cow_file_range(struct inode *inode, | |
776 | struct page *locked_page, | |
777 | u64 start, u64 end, int *page_started, | |
778 | unsigned long *nr_written, | |
779 | int unlock) | |
780 | { | |
781 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
782 | struct btrfs_trans_handle *trans; | |
783 | u64 alloc_hint = 0; | |
784 | u64 num_bytes; | |
785 | unsigned long ram_size; | |
786 | u64 disk_num_bytes; | |
787 | u64 cur_alloc_size; | |
788 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
789 | struct btrfs_key ins; |
790 | struct extent_map *em; | |
791 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
792 | int ret = 0; | |
793 | ||
82d5902d | 794 | BUG_ON(is_free_space_inode(root, inode)); |
7a7eaa40 | 795 | trans = btrfs_join_transaction(root); |
3612b495 | 796 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 797 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 798 | |
771ed689 CM |
799 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
800 | num_bytes = max(blocksize, num_bytes); | |
801 | disk_num_bytes = num_bytes; | |
802 | ret = 0; | |
803 | ||
4cb5300b CM |
804 | /* if this is a small write inside eof, kick off defrag */ |
805 | if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024) | |
806 | btrfs_add_inode_defrag(trans, inode); | |
807 | ||
771ed689 CM |
808 | if (start == 0) { |
809 | /* lets try to make an inline extent */ | |
810 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 811 | start, end, 0, 0, NULL); |
771ed689 CM |
812 | if (ret == 0) { |
813 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
814 | &BTRFS_I(inode)->io_tree, |
815 | start, end, NULL, | |
816 | EXTENT_CLEAR_UNLOCK_PAGE | | |
817 | EXTENT_CLEAR_UNLOCK | | |
818 | EXTENT_CLEAR_DELALLOC | | |
819 | EXTENT_CLEAR_DIRTY | | |
820 | EXTENT_SET_WRITEBACK | | |
821 | EXTENT_END_WRITEBACK); | |
c2167754 | 822 | |
771ed689 CM |
823 | *nr_written = *nr_written + |
824 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
825 | *page_started = 1; | |
826 | ret = 0; | |
827 | goto out; | |
828 | } | |
829 | } | |
830 | ||
831 | BUG_ON(disk_num_bytes > | |
832 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
833 | ||
4b46fce2 | 834 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
835 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
836 | ||
d397712b | 837 | while (disk_num_bytes > 0) { |
a791e35e CM |
838 | unsigned long op; |
839 | ||
287a0ab9 | 840 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 841 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 842 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 843 | (u64)-1, &ins, 1); |
d397712b CM |
844 | BUG_ON(ret); |
845 | ||
172ddd60 | 846 | em = alloc_extent_map(); |
c26a9203 | 847 | BUG_ON(!em); |
e6dcd2dc | 848 | em->start = start; |
445a6944 | 849 | em->orig_start = em->start; |
771ed689 CM |
850 | ram_size = ins.offset; |
851 | em->len = ins.offset; | |
c8b97818 | 852 | |
e6dcd2dc | 853 | em->block_start = ins.objectid; |
c8b97818 | 854 | em->block_len = ins.offset; |
e6dcd2dc | 855 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 856 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 857 | |
d397712b | 858 | while (1) { |
890871be | 859 | write_lock(&em_tree->lock); |
e6dcd2dc | 860 | ret = add_extent_mapping(em_tree, em); |
890871be | 861 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
862 | if (ret != -EEXIST) { |
863 | free_extent_map(em); | |
864 | break; | |
865 | } | |
866 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 867 | start + ram_size - 1, 0); |
e6dcd2dc CM |
868 | } |
869 | ||
98d20f67 | 870 | cur_alloc_size = ins.offset; |
e6dcd2dc | 871 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 872 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 873 | BUG_ON(ret); |
c8b97818 | 874 | |
17d217fe YZ |
875 | if (root->root_key.objectid == |
876 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
877 | ret = btrfs_reloc_clone_csums(inode, start, | |
878 | cur_alloc_size); | |
879 | BUG_ON(ret); | |
880 | } | |
881 | ||
d397712b | 882 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 883 | break; |
d397712b | 884 | |
c8b97818 CM |
885 | /* we're not doing compressed IO, don't unlock the first |
886 | * page (which the caller expects to stay locked), don't | |
887 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
888 | * |
889 | * Do set the Private2 bit so we know this page was properly | |
890 | * setup for writepage | |
c8b97818 | 891 | */ |
a791e35e CM |
892 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
893 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
894 | EXTENT_SET_PRIVATE2; | |
895 | ||
c8b97818 CM |
896 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
897 | start, start + ram_size - 1, | |
a791e35e | 898 | locked_page, op); |
c8b97818 | 899 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
900 | num_bytes -= cur_alloc_size; |
901 | alloc_hint = ins.objectid + ins.offset; | |
902 | start += cur_alloc_size; | |
b888db2b | 903 | } |
b888db2b | 904 | out: |
771ed689 | 905 | ret = 0; |
b888db2b | 906 | btrfs_end_transaction(trans, root); |
c8b97818 | 907 | |
be20aa9d | 908 | return ret; |
771ed689 | 909 | } |
c8b97818 | 910 | |
771ed689 CM |
911 | /* |
912 | * work queue call back to started compression on a file and pages | |
913 | */ | |
914 | static noinline void async_cow_start(struct btrfs_work *work) | |
915 | { | |
916 | struct async_cow *async_cow; | |
917 | int num_added = 0; | |
918 | async_cow = container_of(work, struct async_cow, work); | |
919 | ||
920 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
921 | async_cow->start, async_cow->end, async_cow, | |
922 | &num_added); | |
923 | if (num_added == 0) | |
924 | async_cow->inode = NULL; | |
925 | } | |
926 | ||
927 | /* | |
928 | * work queue call back to submit previously compressed pages | |
929 | */ | |
930 | static noinline void async_cow_submit(struct btrfs_work *work) | |
931 | { | |
932 | struct async_cow *async_cow; | |
933 | struct btrfs_root *root; | |
934 | unsigned long nr_pages; | |
935 | ||
936 | async_cow = container_of(work, struct async_cow, work); | |
937 | ||
938 | root = async_cow->root; | |
939 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
940 | PAGE_CACHE_SHIFT; | |
941 | ||
942 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
943 | ||
944 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
945 | 5 * 1042 * 1024 && | |
946 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
947 | wake_up(&root->fs_info->async_submit_wait); | |
948 | ||
d397712b | 949 | if (async_cow->inode) |
771ed689 | 950 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 951 | } |
c8b97818 | 952 | |
771ed689 CM |
953 | static noinline void async_cow_free(struct btrfs_work *work) |
954 | { | |
955 | struct async_cow *async_cow; | |
956 | async_cow = container_of(work, struct async_cow, work); | |
957 | kfree(async_cow); | |
958 | } | |
959 | ||
960 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
961 | u64 start, u64 end, int *page_started, | |
962 | unsigned long *nr_written) | |
963 | { | |
964 | struct async_cow *async_cow; | |
965 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
966 | unsigned long nr_pages; | |
967 | u64 cur_end; | |
968 | int limit = 10 * 1024 * 1042; | |
969 | ||
a3429ab7 CM |
970 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
971 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 972 | while (start < end) { |
771ed689 | 973 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
8d413713 | 974 | BUG_ON(!async_cow); |
771ed689 CM |
975 | async_cow->inode = inode; |
976 | async_cow->root = root; | |
977 | async_cow->locked_page = locked_page; | |
978 | async_cow->start = start; | |
979 | ||
6cbff00f | 980 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
981 | cur_end = end; |
982 | else | |
983 | cur_end = min(end, start + 512 * 1024 - 1); | |
984 | ||
985 | async_cow->end = cur_end; | |
986 | INIT_LIST_HEAD(&async_cow->extents); | |
987 | ||
988 | async_cow->work.func = async_cow_start; | |
989 | async_cow->work.ordered_func = async_cow_submit; | |
990 | async_cow->work.ordered_free = async_cow_free; | |
991 | async_cow->work.flags = 0; | |
992 | ||
771ed689 CM |
993 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
994 | PAGE_CACHE_SHIFT; | |
995 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
996 | ||
997 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
998 | &async_cow->work); | |
999 | ||
1000 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1001 | wait_event(root->fs_info->async_submit_wait, | |
1002 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1003 | limit)); | |
1004 | } | |
1005 | ||
d397712b | 1006 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1007 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1008 | wait_event(root->fs_info->async_submit_wait, | |
1009 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1010 | 0)); | |
1011 | } | |
1012 | ||
1013 | *nr_written += nr_pages; | |
1014 | start = cur_end + 1; | |
1015 | } | |
1016 | *page_started = 1; | |
1017 | return 0; | |
be20aa9d CM |
1018 | } |
1019 | ||
d397712b | 1020 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1021 | u64 bytenr, u64 num_bytes) |
1022 | { | |
1023 | int ret; | |
1024 | struct btrfs_ordered_sum *sums; | |
1025 | LIST_HEAD(list); | |
1026 | ||
07d400a6 | 1027 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1028 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1029 | if (ret == 0 && list_empty(&list)) |
1030 | return 0; | |
1031 | ||
1032 | while (!list_empty(&list)) { | |
1033 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1034 | list_del(&sums->list); | |
1035 | kfree(sums); | |
1036 | } | |
1037 | return 1; | |
1038 | } | |
1039 | ||
d352ac68 CM |
1040 | /* |
1041 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1042 | * of the extents that exist in the file, and COWs the file as required. | |
1043 | * | |
1044 | * If no cow copies or snapshots exist, we write directly to the existing | |
1045 | * blocks on disk | |
1046 | */ | |
7f366cfe CM |
1047 | static noinline int run_delalloc_nocow(struct inode *inode, |
1048 | struct page *locked_page, | |
771ed689 CM |
1049 | u64 start, u64 end, int *page_started, int force, |
1050 | unsigned long *nr_written) | |
be20aa9d | 1051 | { |
be20aa9d | 1052 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1053 | struct btrfs_trans_handle *trans; |
be20aa9d | 1054 | struct extent_buffer *leaf; |
be20aa9d | 1055 | struct btrfs_path *path; |
80ff3856 | 1056 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1057 | struct btrfs_key found_key; |
80ff3856 YZ |
1058 | u64 cow_start; |
1059 | u64 cur_offset; | |
1060 | u64 extent_end; | |
5d4f98a2 | 1061 | u64 extent_offset; |
80ff3856 YZ |
1062 | u64 disk_bytenr; |
1063 | u64 num_bytes; | |
1064 | int extent_type; | |
1065 | int ret; | |
d899e052 | 1066 | int type; |
80ff3856 YZ |
1067 | int nocow; |
1068 | int check_prev = 1; | |
82d5902d | 1069 | bool nolock; |
33345d01 | 1070 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1071 | |
1072 | path = btrfs_alloc_path(); | |
1073 | BUG_ON(!path); | |
82d5902d LZ |
1074 | |
1075 | nolock = is_free_space_inode(root, inode); | |
1076 | ||
1077 | if (nolock) | |
7a7eaa40 | 1078 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1079 | else |
7a7eaa40 | 1080 | trans = btrfs_join_transaction(root); |
ff5714cc | 1081 | |
3612b495 | 1082 | BUG_ON(IS_ERR(trans)); |
74b21075 | 1083 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1084 | |
80ff3856 YZ |
1085 | cow_start = (u64)-1; |
1086 | cur_offset = start; | |
1087 | while (1) { | |
33345d01 | 1088 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 YZ |
1089 | cur_offset, 0); |
1090 | BUG_ON(ret < 0); | |
1091 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1092 | leaf = path->nodes[0]; | |
1093 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1094 | path->slots[0] - 1); | |
33345d01 | 1095 | if (found_key.objectid == ino && |
80ff3856 YZ |
1096 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1097 | path->slots[0]--; | |
1098 | } | |
1099 | check_prev = 0; | |
1100 | next_slot: | |
1101 | leaf = path->nodes[0]; | |
1102 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1103 | ret = btrfs_next_leaf(root, path); | |
1104 | if (ret < 0) | |
1105 | BUG_ON(1); | |
1106 | if (ret > 0) | |
1107 | break; | |
1108 | leaf = path->nodes[0]; | |
1109 | } | |
be20aa9d | 1110 | |
80ff3856 YZ |
1111 | nocow = 0; |
1112 | disk_bytenr = 0; | |
17d217fe | 1113 | num_bytes = 0; |
80ff3856 YZ |
1114 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1115 | ||
33345d01 | 1116 | if (found_key.objectid > ino || |
80ff3856 YZ |
1117 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1118 | found_key.offset > end) | |
1119 | break; | |
1120 | ||
1121 | if (found_key.offset > cur_offset) { | |
1122 | extent_end = found_key.offset; | |
e9061e21 | 1123 | extent_type = 0; |
80ff3856 YZ |
1124 | goto out_check; |
1125 | } | |
1126 | ||
1127 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1128 | struct btrfs_file_extent_item); | |
1129 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1130 | ||
d899e052 YZ |
1131 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1132 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1133 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1134 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1135 | extent_end = found_key.offset + |
1136 | btrfs_file_extent_num_bytes(leaf, fi); | |
1137 | if (extent_end <= start) { | |
1138 | path->slots[0]++; | |
1139 | goto next_slot; | |
1140 | } | |
17d217fe YZ |
1141 | if (disk_bytenr == 0) |
1142 | goto out_check; | |
80ff3856 YZ |
1143 | if (btrfs_file_extent_compression(leaf, fi) || |
1144 | btrfs_file_extent_encryption(leaf, fi) || | |
1145 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1146 | goto out_check; | |
d899e052 YZ |
1147 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1148 | goto out_check; | |
d2fb3437 | 1149 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1150 | goto out_check; |
33345d01 | 1151 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1152 | found_key.offset - |
1153 | extent_offset, disk_bytenr)) | |
17d217fe | 1154 | goto out_check; |
5d4f98a2 | 1155 | disk_bytenr += extent_offset; |
17d217fe YZ |
1156 | disk_bytenr += cur_offset - found_key.offset; |
1157 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1158 | /* | |
1159 | * force cow if csum exists in the range. | |
1160 | * this ensure that csum for a given extent are | |
1161 | * either valid or do not exist. | |
1162 | */ | |
1163 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1164 | goto out_check; | |
80ff3856 YZ |
1165 | nocow = 1; |
1166 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1167 | extent_end = found_key.offset + | |
1168 | btrfs_file_extent_inline_len(leaf, fi); | |
1169 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1170 | } else { | |
1171 | BUG_ON(1); | |
1172 | } | |
1173 | out_check: | |
1174 | if (extent_end <= start) { | |
1175 | path->slots[0]++; | |
1176 | goto next_slot; | |
1177 | } | |
1178 | if (!nocow) { | |
1179 | if (cow_start == (u64)-1) | |
1180 | cow_start = cur_offset; | |
1181 | cur_offset = extent_end; | |
1182 | if (cur_offset > end) | |
1183 | break; | |
1184 | path->slots[0]++; | |
1185 | goto next_slot; | |
7ea394f1 YZ |
1186 | } |
1187 | ||
b3b4aa74 | 1188 | btrfs_release_path(path); |
80ff3856 YZ |
1189 | if (cow_start != (u64)-1) { |
1190 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1191 | found_key.offset - 1, page_started, |
1192 | nr_written, 1); | |
80ff3856 YZ |
1193 | BUG_ON(ret); |
1194 | cow_start = (u64)-1; | |
7ea394f1 | 1195 | } |
80ff3856 | 1196 | |
d899e052 YZ |
1197 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1198 | struct extent_map *em; | |
1199 | struct extent_map_tree *em_tree; | |
1200 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1201 | em = alloc_extent_map(); |
c26a9203 | 1202 | BUG_ON(!em); |
d899e052 | 1203 | em->start = cur_offset; |
445a6944 | 1204 | em->orig_start = em->start; |
d899e052 YZ |
1205 | em->len = num_bytes; |
1206 | em->block_len = num_bytes; | |
1207 | em->block_start = disk_bytenr; | |
1208 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1209 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1210 | while (1) { | |
890871be | 1211 | write_lock(&em_tree->lock); |
d899e052 | 1212 | ret = add_extent_mapping(em_tree, em); |
890871be | 1213 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1214 | if (ret != -EEXIST) { |
1215 | free_extent_map(em); | |
1216 | break; | |
1217 | } | |
1218 | btrfs_drop_extent_cache(inode, em->start, | |
1219 | em->start + em->len - 1, 0); | |
1220 | } | |
1221 | type = BTRFS_ORDERED_PREALLOC; | |
1222 | } else { | |
1223 | type = BTRFS_ORDERED_NOCOW; | |
1224 | } | |
80ff3856 YZ |
1225 | |
1226 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1227 | num_bytes, num_bytes, type); |
1228 | BUG_ON(ret); | |
771ed689 | 1229 | |
efa56464 YZ |
1230 | if (root->root_key.objectid == |
1231 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1232 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1233 | num_bytes); | |
1234 | BUG_ON(ret); | |
1235 | } | |
1236 | ||
d899e052 | 1237 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1238 | cur_offset, cur_offset + num_bytes - 1, |
1239 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1240 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1241 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1242 | cur_offset = extent_end; |
1243 | if (cur_offset > end) | |
1244 | break; | |
be20aa9d | 1245 | } |
b3b4aa74 | 1246 | btrfs_release_path(path); |
80ff3856 YZ |
1247 | |
1248 | if (cur_offset <= end && cow_start == (u64)-1) | |
1249 | cow_start = cur_offset; | |
1250 | if (cow_start != (u64)-1) { | |
1251 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1252 | page_started, nr_written, 1); |
80ff3856 YZ |
1253 | BUG_ON(ret); |
1254 | } | |
1255 | ||
0cb59c99 JB |
1256 | if (nolock) { |
1257 | ret = btrfs_end_transaction_nolock(trans, root); | |
1258 | BUG_ON(ret); | |
1259 | } else { | |
1260 | ret = btrfs_end_transaction(trans, root); | |
1261 | BUG_ON(ret); | |
1262 | } | |
7ea394f1 | 1263 | btrfs_free_path(path); |
80ff3856 | 1264 | return 0; |
be20aa9d CM |
1265 | } |
1266 | ||
d352ac68 CM |
1267 | /* |
1268 | * extent_io.c call back to do delayed allocation processing | |
1269 | */ | |
c8b97818 | 1270 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1271 | u64 start, u64 end, int *page_started, |
1272 | unsigned long *nr_written) | |
be20aa9d | 1273 | { |
be20aa9d | 1274 | int ret; |
7f366cfe | 1275 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1276 | |
6cbff00f | 1277 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1278 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1279 | page_started, 1, nr_written); |
6cbff00f | 1280 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1281 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1282 | page_started, 0, nr_written); |
1e701a32 | 1283 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1284 | !(BTRFS_I(inode)->force_compress) && |
1285 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1286 | ret = cow_file_range(inode, locked_page, start, end, |
1287 | page_started, nr_written, 1); | |
be20aa9d | 1288 | else |
771ed689 | 1289 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1290 | page_started, nr_written); |
b888db2b CM |
1291 | return ret; |
1292 | } | |
1293 | ||
9ed74f2d | 1294 | static int btrfs_split_extent_hook(struct inode *inode, |
0ca1f7ce | 1295 | struct extent_state *orig, u64 split) |
9ed74f2d | 1296 | { |
0ca1f7ce | 1297 | /* not delalloc, ignore it */ |
9ed74f2d JB |
1298 | if (!(orig->state & EXTENT_DELALLOC)) |
1299 | return 0; | |
1300 | ||
0ca1f7ce | 1301 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1302 | return 0; |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1307 | * extents so we can keep track of new extents that are just merged onto old | |
1308 | * extents, such as when we are doing sequential writes, so we can properly | |
1309 | * account for the metadata space we'll need. | |
1310 | */ | |
1311 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1312 | struct extent_state *new, | |
1313 | struct extent_state *other) | |
1314 | { | |
9ed74f2d JB |
1315 | /* not delalloc, ignore it */ |
1316 | if (!(other->state & EXTENT_DELALLOC)) | |
1317 | return 0; | |
1318 | ||
0ca1f7ce | 1319 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1320 | return 0; |
1321 | } | |
1322 | ||
d352ac68 CM |
1323 | /* |
1324 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1325 | * bytes in this file, and to maintain the list of inodes that | |
1326 | * have pending delalloc work to be done. | |
1327 | */ | |
0ca1f7ce YZ |
1328 | static int btrfs_set_bit_hook(struct inode *inode, |
1329 | struct extent_state *state, int *bits) | |
291d673e | 1330 | { |
9ed74f2d | 1331 | |
75eff68e CM |
1332 | /* |
1333 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1334 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1335 | * bit, which is only set or cleared with irqs on |
1336 | */ | |
0ca1f7ce | 1337 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1338 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1339 | u64 len = state->end + 1 - state->start; |
82d5902d | 1340 | bool do_list = !is_free_space_inode(root, inode); |
9ed74f2d | 1341 | |
0ca1f7ce YZ |
1342 | if (*bits & EXTENT_FIRST_DELALLOC) |
1343 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1344 | else | |
1345 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); | |
287a0ab9 | 1346 | |
75eff68e | 1347 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1348 | BTRFS_I(inode)->delalloc_bytes += len; |
1349 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1350 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1351 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1352 | &root->fs_info->delalloc_inodes); | |
1353 | } | |
75eff68e | 1354 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1355 | } |
1356 | return 0; | |
1357 | } | |
1358 | ||
d352ac68 CM |
1359 | /* |
1360 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1361 | */ | |
9ed74f2d | 1362 | static int btrfs_clear_bit_hook(struct inode *inode, |
0ca1f7ce | 1363 | struct extent_state *state, int *bits) |
291d673e | 1364 | { |
75eff68e CM |
1365 | /* |
1366 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1367 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1368 | * bit, which is only set or cleared with irqs on |
1369 | */ | |
0ca1f7ce | 1370 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1371 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1372 | u64 len = state->end + 1 - state->start; |
82d5902d | 1373 | bool do_list = !is_free_space_inode(root, inode); |
bcbfce8a | 1374 | |
0ca1f7ce YZ |
1375 | if (*bits & EXTENT_FIRST_DELALLOC) |
1376 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1377 | else if (!(*bits & EXTENT_DO_ACCOUNTING)) | |
1378 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
1379 | ||
1380 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1381 | btrfs_delalloc_release_metadata(inode, len); | |
1382 | ||
0cb59c99 JB |
1383 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1384 | && do_list) | |
0ca1f7ce | 1385 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1386 | |
75eff68e | 1387 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1388 | root->fs_info->delalloc_bytes -= len; |
1389 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1390 | ||
0cb59c99 | 1391 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1392 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1393 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1394 | } | |
75eff68e | 1395 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1396 | } |
1397 | return 0; | |
1398 | } | |
1399 | ||
d352ac68 CM |
1400 | /* |
1401 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1402 | * we don't create bios that span stripes or chunks | |
1403 | */ | |
239b14b3 | 1404 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1405 | size_t size, struct bio *bio, |
1406 | unsigned long bio_flags) | |
239b14b3 CM |
1407 | { |
1408 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1409 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1410 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1411 | u64 length = 0; |
1412 | u64 map_length; | |
239b14b3 CM |
1413 | int ret; |
1414 | ||
771ed689 CM |
1415 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1416 | return 0; | |
1417 | ||
f2d8d74d | 1418 | length = bio->bi_size; |
239b14b3 CM |
1419 | map_tree = &root->fs_info->mapping_tree; |
1420 | map_length = length; | |
cea9e445 | 1421 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1422 | &map_length, NULL, 0); |
cea9e445 | 1423 | |
d397712b | 1424 | if (map_length < length + size) |
239b14b3 | 1425 | return 1; |
411fc6bc | 1426 | return ret; |
239b14b3 CM |
1427 | } |
1428 | ||
d352ac68 CM |
1429 | /* |
1430 | * in order to insert checksums into the metadata in large chunks, | |
1431 | * we wait until bio submission time. All the pages in the bio are | |
1432 | * checksummed and sums are attached onto the ordered extent record. | |
1433 | * | |
1434 | * At IO completion time the cums attached on the ordered extent record | |
1435 | * are inserted into the btree | |
1436 | */ | |
d397712b CM |
1437 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1438 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1439 | unsigned long bio_flags, |
1440 | u64 bio_offset) | |
065631f6 | 1441 | { |
065631f6 | 1442 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1443 | int ret = 0; |
e015640f | 1444 | |
d20f7043 | 1445 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1446 | BUG_ON(ret); |
4a69a410 CM |
1447 | return 0; |
1448 | } | |
e015640f | 1449 | |
4a69a410 CM |
1450 | /* |
1451 | * in order to insert checksums into the metadata in large chunks, | |
1452 | * we wait until bio submission time. All the pages in the bio are | |
1453 | * checksummed and sums are attached onto the ordered extent record. | |
1454 | * | |
1455 | * At IO completion time the cums attached on the ordered extent record | |
1456 | * are inserted into the btree | |
1457 | */ | |
b2950863 | 1458 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1459 | int mirror_num, unsigned long bio_flags, |
1460 | u64 bio_offset) | |
4a69a410 CM |
1461 | { |
1462 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1463 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1464 | } |
1465 | ||
d352ac68 | 1466 | /* |
cad321ad CM |
1467 | * extent_io.c submission hook. This does the right thing for csum calculation |
1468 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1469 | */ |
b2950863 | 1470 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1471 | int mirror_num, unsigned long bio_flags, |
1472 | u64 bio_offset) | |
44b8bd7e CM |
1473 | { |
1474 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1475 | int ret = 0; | |
19b9bdb0 | 1476 | int skip_sum; |
44b8bd7e | 1477 | |
6cbff00f | 1478 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1479 | |
82d5902d | 1480 | if (is_free_space_inode(root, inode)) |
0cb59c99 JB |
1481 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2); |
1482 | else | |
1483 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
e6dcd2dc | 1484 | BUG_ON(ret); |
065631f6 | 1485 | |
7b6d91da | 1486 | if (!(rw & REQ_WRITE)) { |
d20f7043 | 1487 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1488 | return btrfs_submit_compressed_read(inode, bio, |
1489 | mirror_num, bio_flags); | |
c2db1073 TI |
1490 | } else if (!skip_sum) { |
1491 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1492 | if (ret) | |
1493 | return ret; | |
1494 | } | |
4d1b5fb4 | 1495 | goto mapit; |
19b9bdb0 | 1496 | } else if (!skip_sum) { |
17d217fe YZ |
1497 | /* csum items have already been cloned */ |
1498 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1499 | goto mapit; | |
19b9bdb0 CM |
1500 | /* we're doing a write, do the async checksumming */ |
1501 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1502 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1503 | bio_flags, bio_offset, |
1504 | __btrfs_submit_bio_start, | |
4a69a410 | 1505 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1506 | } |
1507 | ||
0b86a832 | 1508 | mapit: |
8b712842 | 1509 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1510 | } |
6885f308 | 1511 | |
d352ac68 CM |
1512 | /* |
1513 | * given a list of ordered sums record them in the inode. This happens | |
1514 | * at IO completion time based on sums calculated at bio submission time. | |
1515 | */ | |
ba1da2f4 | 1516 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1517 | struct inode *inode, u64 file_offset, |
1518 | struct list_head *list) | |
1519 | { | |
e6dcd2dc CM |
1520 | struct btrfs_ordered_sum *sum; |
1521 | ||
c6e30871 | 1522 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1523 | btrfs_csum_file_blocks(trans, |
1524 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1525 | } |
1526 | return 0; | |
1527 | } | |
1528 | ||
2ac55d41 JB |
1529 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1530 | struct extent_state **cached_state) | |
ea8c2819 | 1531 | { |
d397712b | 1532 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1533 | WARN_ON(1); |
ea8c2819 | 1534 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1535 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1536 | } |
1537 | ||
d352ac68 | 1538 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1539 | struct btrfs_writepage_fixup { |
1540 | struct page *page; | |
1541 | struct btrfs_work work; | |
1542 | }; | |
1543 | ||
b2950863 | 1544 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1545 | { |
1546 | struct btrfs_writepage_fixup *fixup; | |
1547 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1548 | struct extent_state *cached_state = NULL; |
247e743c CM |
1549 | struct page *page; |
1550 | struct inode *inode; | |
1551 | u64 page_start; | |
1552 | u64 page_end; | |
1553 | ||
1554 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1555 | page = fixup->page; | |
4a096752 | 1556 | again: |
247e743c CM |
1557 | lock_page(page); |
1558 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1559 | ClearPageChecked(page); | |
1560 | goto out_page; | |
1561 | } | |
1562 | ||
1563 | inode = page->mapping->host; | |
1564 | page_start = page_offset(page); | |
1565 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1566 | ||
2ac55d41 JB |
1567 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1568 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1569 | |
1570 | /* already ordered? We're done */ | |
8b62b72b | 1571 | if (PagePrivate2(page)) |
247e743c | 1572 | goto out; |
4a096752 CM |
1573 | |
1574 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1575 | if (ordered) { | |
2ac55d41 JB |
1576 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1577 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1578 | unlock_page(page); |
1579 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1580 | goto again; | |
1581 | } | |
247e743c | 1582 | |
0ca1f7ce | 1583 | BUG(); |
2ac55d41 | 1584 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c CM |
1585 | ClearPageChecked(page); |
1586 | out: | |
2ac55d41 JB |
1587 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1588 | &cached_state, GFP_NOFS); | |
247e743c CM |
1589 | out_page: |
1590 | unlock_page(page); | |
1591 | page_cache_release(page); | |
b897abec | 1592 | kfree(fixup); |
247e743c CM |
1593 | } |
1594 | ||
1595 | /* | |
1596 | * There are a few paths in the higher layers of the kernel that directly | |
1597 | * set the page dirty bit without asking the filesystem if it is a | |
1598 | * good idea. This causes problems because we want to make sure COW | |
1599 | * properly happens and the data=ordered rules are followed. | |
1600 | * | |
c8b97818 | 1601 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1602 | * hasn't been properly setup for IO. We kick off an async process |
1603 | * to fix it up. The async helper will wait for ordered extents, set | |
1604 | * the delalloc bit and make it safe to write the page. | |
1605 | */ | |
b2950863 | 1606 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1607 | { |
1608 | struct inode *inode = page->mapping->host; | |
1609 | struct btrfs_writepage_fixup *fixup; | |
1610 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1611 | |
8b62b72b CM |
1612 | /* this page is properly in the ordered list */ |
1613 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1614 | return 0; |
1615 | ||
1616 | if (PageChecked(page)) | |
1617 | return -EAGAIN; | |
1618 | ||
1619 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1620 | if (!fixup) | |
1621 | return -EAGAIN; | |
f421950f | 1622 | |
247e743c CM |
1623 | SetPageChecked(page); |
1624 | page_cache_get(page); | |
1625 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1626 | fixup->page = page; | |
1627 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1628 | return -EAGAIN; | |
1629 | } | |
1630 | ||
d899e052 YZ |
1631 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1632 | struct inode *inode, u64 file_pos, | |
1633 | u64 disk_bytenr, u64 disk_num_bytes, | |
1634 | u64 num_bytes, u64 ram_bytes, | |
1635 | u8 compression, u8 encryption, | |
1636 | u16 other_encoding, int extent_type) | |
1637 | { | |
1638 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1639 | struct btrfs_file_extent_item *fi; | |
1640 | struct btrfs_path *path; | |
1641 | struct extent_buffer *leaf; | |
1642 | struct btrfs_key ins; | |
1643 | u64 hint; | |
1644 | int ret; | |
1645 | ||
1646 | path = btrfs_alloc_path(); | |
1647 | BUG_ON(!path); | |
1648 | ||
b9473439 | 1649 | path->leave_spinning = 1; |
a1ed835e CM |
1650 | |
1651 | /* | |
1652 | * we may be replacing one extent in the tree with another. | |
1653 | * The new extent is pinned in the extent map, and we don't want | |
1654 | * to drop it from the cache until it is completely in the btree. | |
1655 | * | |
1656 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1657 | * the caller is expected to unpin it and allow it to be merged | |
1658 | * with the others. | |
1659 | */ | |
920bbbfb YZ |
1660 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1661 | &hint, 0); | |
d899e052 YZ |
1662 | BUG_ON(ret); |
1663 | ||
33345d01 | 1664 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1665 | ins.offset = file_pos; |
1666 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1667 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1668 | BUG_ON(ret); | |
1669 | leaf = path->nodes[0]; | |
1670 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1671 | struct btrfs_file_extent_item); | |
1672 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1673 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1674 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1675 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1676 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1677 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1678 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1679 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1680 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1681 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1682 | |
1683 | btrfs_unlock_up_safe(path, 1); | |
1684 | btrfs_set_lock_blocking(leaf); | |
1685 | ||
d899e052 YZ |
1686 | btrfs_mark_buffer_dirty(leaf); |
1687 | ||
1688 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1689 | |
1690 | ins.objectid = disk_bytenr; | |
1691 | ins.offset = disk_num_bytes; | |
1692 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1693 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1694 | root->root_key.objectid, | |
33345d01 | 1695 | btrfs_ino(inode), file_pos, &ins); |
d899e052 | 1696 | BUG_ON(ret); |
d899e052 | 1697 | btrfs_free_path(path); |
b9473439 | 1698 | |
d899e052 YZ |
1699 | return 0; |
1700 | } | |
1701 | ||
5d13a98f CM |
1702 | /* |
1703 | * helper function for btrfs_finish_ordered_io, this | |
1704 | * just reads in some of the csum leaves to prime them into ram | |
1705 | * before we start the transaction. It limits the amount of btree | |
1706 | * reads required while inside the transaction. | |
1707 | */ | |
d352ac68 CM |
1708 | /* as ordered data IO finishes, this gets called so we can finish |
1709 | * an ordered extent if the range of bytes in the file it covers are | |
1710 | * fully written. | |
1711 | */ | |
211f90e6 | 1712 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1713 | { |
e6dcd2dc | 1714 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1715 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1716 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1717 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1718 | struct extent_state *cached_state = NULL; |
261507a0 | 1719 | int compress_type = 0; |
e6dcd2dc | 1720 | int ret; |
82d5902d | 1721 | bool nolock; |
e6dcd2dc | 1722 | |
5a1a3df1 JB |
1723 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1724 | end - start + 1); | |
ba1da2f4 | 1725 | if (!ret) |
e6dcd2dc | 1726 | return 0; |
e6dcd2dc | 1727 | BUG_ON(!ordered_extent); |
efd049fb | 1728 | |
82d5902d | 1729 | nolock = is_free_space_inode(root, inode); |
0cb59c99 | 1730 | |
c2167754 YZ |
1731 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1732 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1733 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1734 | if (!ret) { | |
0cb59c99 | 1735 | if (nolock) |
7a7eaa40 | 1736 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1737 | else |
7a7eaa40 | 1738 | trans = btrfs_join_transaction(root); |
3612b495 | 1739 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1740 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 YZ |
1741 | ret = btrfs_update_inode(trans, root, inode); |
1742 | BUG_ON(ret); | |
c2167754 YZ |
1743 | } |
1744 | goto out; | |
1745 | } | |
e6dcd2dc | 1746 | |
2ac55d41 JB |
1747 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1748 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1749 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1750 | |
0cb59c99 | 1751 | if (nolock) |
7a7eaa40 | 1752 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1753 | else |
7a7eaa40 | 1754 | trans = btrfs_join_transaction(root); |
3612b495 | 1755 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1756 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1757 | |
c8b97818 | 1758 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1759 | compress_type = ordered_extent->compress_type; |
d899e052 | 1760 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1761 | BUG_ON(compress_type); |
920bbbfb | 1762 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1763 | ordered_extent->file_offset, |
1764 | ordered_extent->file_offset + | |
1765 | ordered_extent->len); | |
1766 | BUG_ON(ret); | |
1767 | } else { | |
0af3d00b | 1768 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1769 | ret = insert_reserved_file_extent(trans, inode, |
1770 | ordered_extent->file_offset, | |
1771 | ordered_extent->start, | |
1772 | ordered_extent->disk_len, | |
1773 | ordered_extent->len, | |
1774 | ordered_extent->len, | |
261507a0 | 1775 | compress_type, 0, 0, |
d899e052 | 1776 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1777 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1778 | ordered_extent->file_offset, | |
1779 | ordered_extent->len); | |
d899e052 YZ |
1780 | BUG_ON(ret); |
1781 | } | |
2ac55d41 JB |
1782 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1783 | ordered_extent->file_offset + | |
1784 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1785 | ||
e6dcd2dc CM |
1786 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1787 | &ordered_extent->list); | |
1788 | ||
1ef30be1 JB |
1789 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1790 | if (!ret) { | |
1791 | ret = btrfs_update_inode(trans, root, inode); | |
1792 | BUG_ON(ret); | |
1793 | } | |
1794 | ret = 0; | |
c2167754 | 1795 | out: |
0cb59c99 JB |
1796 | if (nolock) { |
1797 | if (trans) | |
1798 | btrfs_end_transaction_nolock(trans, root); | |
1799 | } else { | |
1800 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); | |
1801 | if (trans) | |
1802 | btrfs_end_transaction(trans, root); | |
1803 | } | |
1804 | ||
e6dcd2dc CM |
1805 | /* once for us */ |
1806 | btrfs_put_ordered_extent(ordered_extent); | |
1807 | /* once for the tree */ | |
1808 | btrfs_put_ordered_extent(ordered_extent); | |
1809 | ||
e6dcd2dc CM |
1810 | return 0; |
1811 | } | |
1812 | ||
b2950863 | 1813 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1814 | struct extent_state *state, int uptodate) |
1815 | { | |
1abe9b8a | 1816 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1817 | ||
8b62b72b | 1818 | ClearPagePrivate2(page); |
211f90e6 CM |
1819 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1820 | } | |
1821 | ||
d352ac68 CM |
1822 | /* |
1823 | * When IO fails, either with EIO or csum verification fails, we | |
1824 | * try other mirrors that might have a good copy of the data. This | |
1825 | * io_failure_record is used to record state as we go through all the | |
1826 | * mirrors. If another mirror has good data, the page is set up to date | |
1827 | * and things continue. If a good mirror can't be found, the original | |
1828 | * bio end_io callback is called to indicate things have failed. | |
1829 | */ | |
7e38326f CM |
1830 | struct io_failure_record { |
1831 | struct page *page; | |
1832 | u64 start; | |
1833 | u64 len; | |
1834 | u64 logical; | |
d20f7043 | 1835 | unsigned long bio_flags; |
7e38326f CM |
1836 | int last_mirror; |
1837 | }; | |
1838 | ||
b2950863 | 1839 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1840 | struct page *page, u64 start, u64 end, |
1841 | struct extent_state *state) | |
7e38326f CM |
1842 | { |
1843 | struct io_failure_record *failrec = NULL; | |
1844 | u64 private; | |
1845 | struct extent_map *em; | |
1846 | struct inode *inode = page->mapping->host; | |
1847 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1848 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1849 | struct bio *bio; |
1850 | int num_copies; | |
1851 | int ret; | |
1259ab75 | 1852 | int rw; |
7e38326f CM |
1853 | u64 logical; |
1854 | ||
1855 | ret = get_state_private(failure_tree, start, &private); | |
1856 | if (ret) { | |
7e38326f CM |
1857 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1858 | if (!failrec) | |
1859 | return -ENOMEM; | |
1860 | failrec->start = start; | |
1861 | failrec->len = end - start + 1; | |
1862 | failrec->last_mirror = 0; | |
d20f7043 | 1863 | failrec->bio_flags = 0; |
7e38326f | 1864 | |
890871be | 1865 | read_lock(&em_tree->lock); |
3b951516 CM |
1866 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1867 | if (em->start > start || em->start + em->len < start) { | |
1868 | free_extent_map(em); | |
1869 | em = NULL; | |
1870 | } | |
890871be | 1871 | read_unlock(&em_tree->lock); |
7e38326f | 1872 | |
c704005d | 1873 | if (IS_ERR_OR_NULL(em)) { |
7e38326f CM |
1874 | kfree(failrec); |
1875 | return -EIO; | |
1876 | } | |
1877 | logical = start - em->start; | |
1878 | logical = em->block_start + logical; | |
d20f7043 CM |
1879 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1880 | logical = em->block_start; | |
1881 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
261507a0 LZ |
1882 | extent_set_compress_type(&failrec->bio_flags, |
1883 | em->compress_type); | |
d20f7043 | 1884 | } |
7e38326f CM |
1885 | failrec->logical = logical; |
1886 | free_extent_map(em); | |
1887 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1888 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1889 | set_state_private(failure_tree, start, |
1890 | (u64)(unsigned long)failrec); | |
7e38326f | 1891 | } else { |
587f7704 | 1892 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1893 | } |
1894 | num_copies = btrfs_num_copies( | |
1895 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1896 | failrec->logical, failrec->len); | |
1897 | failrec->last_mirror++; | |
1898 | if (!state) { | |
cad321ad | 1899 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1900 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1901 | failrec->start, | |
1902 | EXTENT_LOCKED); | |
1903 | if (state && state->start != failrec->start) | |
1904 | state = NULL; | |
cad321ad | 1905 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1906 | } |
1907 | if (!state || failrec->last_mirror > num_copies) { | |
1908 | set_state_private(failure_tree, failrec->start, 0); | |
1909 | clear_extent_bits(failure_tree, failrec->start, | |
1910 | failrec->start + failrec->len - 1, | |
1911 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1912 | kfree(failrec); | |
1913 | return -EIO; | |
1914 | } | |
1915 | bio = bio_alloc(GFP_NOFS, 1); | |
1916 | bio->bi_private = state; | |
1917 | bio->bi_end_io = failed_bio->bi_end_io; | |
1918 | bio->bi_sector = failrec->logical >> 9; | |
1919 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1920 | bio->bi_size = 0; |
d20f7043 | 1921 | |
7e38326f | 1922 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
7b6d91da | 1923 | if (failed_bio->bi_rw & REQ_WRITE) |
1259ab75 CM |
1924 | rw = WRITE; |
1925 | else | |
1926 | rw = READ; | |
1927 | ||
c2db1073 | 1928 | ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, |
c8b97818 | 1929 | failrec->last_mirror, |
eaf25d93 | 1930 | failrec->bio_flags, 0); |
c2db1073 | 1931 | return ret; |
1259ab75 CM |
1932 | } |
1933 | ||
d352ac68 CM |
1934 | /* |
1935 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1936 | * to see if we need to process or clean up an io_failure_record | |
1937 | */ | |
b2950863 | 1938 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1939 | { |
1940 | u64 private; | |
1941 | u64 private_failure; | |
1942 | struct io_failure_record *failure; | |
1943 | int ret; | |
1944 | ||
1945 | private = 0; | |
1946 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
ec29ed5b | 1947 | (u64)-1, 1, EXTENT_DIRTY, 0)) { |
1259ab75 CM |
1948 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, |
1949 | start, &private_failure); | |
1950 | if (ret == 0) { | |
1951 | failure = (struct io_failure_record *)(unsigned long) | |
1952 | private_failure; | |
1953 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1954 | failure->start, 0); | |
1955 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1956 | failure->start, | |
1957 | failure->start + failure->len - 1, | |
1958 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1959 | GFP_NOFS); | |
1960 | kfree(failure); | |
1961 | } | |
1962 | } | |
7e38326f CM |
1963 | return 0; |
1964 | } | |
1965 | ||
d352ac68 CM |
1966 | /* |
1967 | * when reads are done, we need to check csums to verify the data is correct | |
1968 | * if there's a match, we allow the bio to finish. If not, we go through | |
1969 | * the io_failure_record routines to find good copies | |
1970 | */ | |
b2950863 | 1971 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1972 | struct extent_state *state) |
07157aac | 1973 | { |
35ebb934 | 1974 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1975 | struct inode *inode = page->mapping->host; |
d1310b2e | 1976 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1977 | char *kaddr; |
aadfeb6e | 1978 | u64 private = ~(u32)0; |
07157aac | 1979 | int ret; |
ff79f819 CM |
1980 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1981 | u32 csum = ~(u32)0; | |
d1310b2e | 1982 | |
d20f7043 CM |
1983 | if (PageChecked(page)) { |
1984 | ClearPageChecked(page); | |
1985 | goto good; | |
1986 | } | |
6cbff00f CH |
1987 | |
1988 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 1989 | goto good; |
17d217fe YZ |
1990 | |
1991 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1992 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1993 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1994 | GFP_NOFS); | |
b6cda9bc | 1995 | return 0; |
17d217fe | 1996 | } |
d20f7043 | 1997 | |
c2e639f0 | 1998 | if (state && state->start == start) { |
70dec807 CM |
1999 | private = state->private; |
2000 | ret = 0; | |
2001 | } else { | |
2002 | ret = get_state_private(io_tree, start, &private); | |
2003 | } | |
9ab86c8e | 2004 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 2005 | if (ret) |
07157aac | 2006 | goto zeroit; |
d397712b | 2007 | |
ff79f819 CM |
2008 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2009 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2010 | if (csum != private) |
07157aac | 2011 | goto zeroit; |
d397712b | 2012 | |
9ab86c8e | 2013 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 2014 | good: |
7e38326f CM |
2015 | /* if the io failure tree for this inode is non-empty, |
2016 | * check to see if we've recovered from a failed IO | |
2017 | */ | |
1259ab75 | 2018 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
2019 | return 0; |
2020 | ||
2021 | zeroit: | |
945d8962 | 2022 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2023 | "private %llu\n", |
2024 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2025 | (unsigned long long)start, csum, |
2026 | (unsigned long long)private); | |
db94535d CM |
2027 | memset(kaddr + offset, 1, end - start + 1); |
2028 | flush_dcache_page(page); | |
9ab86c8e | 2029 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2030 | if (private == 0) |
2031 | return 0; | |
7e38326f | 2032 | return -EIO; |
07157aac | 2033 | } |
b888db2b | 2034 | |
24bbcf04 YZ |
2035 | struct delayed_iput { |
2036 | struct list_head list; | |
2037 | struct inode *inode; | |
2038 | }; | |
2039 | ||
2040 | void btrfs_add_delayed_iput(struct inode *inode) | |
2041 | { | |
2042 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2043 | struct delayed_iput *delayed; | |
2044 | ||
2045 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2046 | return; | |
2047 | ||
2048 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2049 | delayed->inode = inode; | |
2050 | ||
2051 | spin_lock(&fs_info->delayed_iput_lock); | |
2052 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2053 | spin_unlock(&fs_info->delayed_iput_lock); | |
2054 | } | |
2055 | ||
2056 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2057 | { | |
2058 | LIST_HEAD(list); | |
2059 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2060 | struct delayed_iput *delayed; | |
2061 | int empty; | |
2062 | ||
2063 | spin_lock(&fs_info->delayed_iput_lock); | |
2064 | empty = list_empty(&fs_info->delayed_iputs); | |
2065 | spin_unlock(&fs_info->delayed_iput_lock); | |
2066 | if (empty) | |
2067 | return; | |
2068 | ||
2069 | down_read(&root->fs_info->cleanup_work_sem); | |
2070 | spin_lock(&fs_info->delayed_iput_lock); | |
2071 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2072 | spin_unlock(&fs_info->delayed_iput_lock); | |
2073 | ||
2074 | while (!list_empty(&list)) { | |
2075 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2076 | list_del(&delayed->list); | |
2077 | iput(delayed->inode); | |
2078 | kfree(delayed); | |
2079 | } | |
2080 | up_read(&root->fs_info->cleanup_work_sem); | |
2081 | } | |
2082 | ||
d68fc57b YZ |
2083 | /* |
2084 | * calculate extra metadata reservation when snapshotting a subvolume | |
2085 | * contains orphan files. | |
2086 | */ | |
2087 | void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans, | |
2088 | struct btrfs_pending_snapshot *pending, | |
2089 | u64 *bytes_to_reserve) | |
2090 | { | |
2091 | struct btrfs_root *root; | |
2092 | struct btrfs_block_rsv *block_rsv; | |
2093 | u64 num_bytes; | |
2094 | int index; | |
2095 | ||
2096 | root = pending->root; | |
2097 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2098 | return; | |
2099 | ||
2100 | block_rsv = root->orphan_block_rsv; | |
2101 | ||
2102 | /* orphan block reservation for the snapshot */ | |
2103 | num_bytes = block_rsv->size; | |
2104 | ||
2105 | /* | |
2106 | * after the snapshot is created, COWing tree blocks may use more | |
2107 | * space than it frees. So we should make sure there is enough | |
2108 | * reserved space. | |
2109 | */ | |
2110 | index = trans->transid & 0x1; | |
2111 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2112 | num_bytes += block_rsv->size - | |
2113 | (block_rsv->reserved + block_rsv->freed[index]); | |
2114 | } | |
2115 | ||
2116 | *bytes_to_reserve += num_bytes; | |
2117 | } | |
2118 | ||
2119 | void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans, | |
2120 | struct btrfs_pending_snapshot *pending) | |
2121 | { | |
2122 | struct btrfs_root *root = pending->root; | |
2123 | struct btrfs_root *snap = pending->snap; | |
2124 | struct btrfs_block_rsv *block_rsv; | |
2125 | u64 num_bytes; | |
2126 | int index; | |
2127 | int ret; | |
2128 | ||
2129 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2130 | return; | |
2131 | ||
2132 | /* refill source subvolume's orphan block reservation */ | |
2133 | block_rsv = root->orphan_block_rsv; | |
2134 | index = trans->transid & 0x1; | |
2135 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2136 | num_bytes = block_rsv->size - | |
2137 | (block_rsv->reserved + block_rsv->freed[index]); | |
2138 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2139 | root->orphan_block_rsv, | |
2140 | num_bytes); | |
2141 | BUG_ON(ret); | |
2142 | } | |
2143 | ||
2144 | /* setup orphan block reservation for the snapshot */ | |
2145 | block_rsv = btrfs_alloc_block_rsv(snap); | |
2146 | BUG_ON(!block_rsv); | |
2147 | ||
2148 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
2149 | snap->orphan_block_rsv = block_rsv; | |
2150 | ||
2151 | num_bytes = root->orphan_block_rsv->size; | |
2152 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2153 | block_rsv, num_bytes); | |
2154 | BUG_ON(ret); | |
2155 | ||
2156 | #if 0 | |
2157 | /* insert orphan item for the snapshot */ | |
2158 | WARN_ON(!root->orphan_item_inserted); | |
2159 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2160 | snap->root_key.objectid); | |
2161 | BUG_ON(ret); | |
2162 | snap->orphan_item_inserted = 1; | |
2163 | #endif | |
2164 | } | |
2165 | ||
2166 | enum btrfs_orphan_cleanup_state { | |
2167 | ORPHAN_CLEANUP_STARTED = 1, | |
2168 | ORPHAN_CLEANUP_DONE = 2, | |
2169 | }; | |
2170 | ||
2171 | /* | |
2172 | * This is called in transaction commmit time. If there are no orphan | |
2173 | * files in the subvolume, it removes orphan item and frees block_rsv | |
2174 | * structure. | |
2175 | */ | |
2176 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2177 | struct btrfs_root *root) | |
2178 | { | |
2179 | int ret; | |
2180 | ||
2181 | if (!list_empty(&root->orphan_list) || | |
2182 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2183 | return; | |
2184 | ||
2185 | if (root->orphan_item_inserted && | |
2186 | btrfs_root_refs(&root->root_item) > 0) { | |
2187 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2188 | root->root_key.objectid); | |
2189 | BUG_ON(ret); | |
2190 | root->orphan_item_inserted = 0; | |
2191 | } | |
2192 | ||
2193 | if (root->orphan_block_rsv) { | |
2194 | WARN_ON(root->orphan_block_rsv->size > 0); | |
2195 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
2196 | root->orphan_block_rsv = NULL; | |
2197 | } | |
2198 | } | |
2199 | ||
7b128766 JB |
2200 | /* |
2201 | * This creates an orphan entry for the given inode in case something goes | |
2202 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2203 | * |
2204 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2205 | * this function. | |
7b128766 JB |
2206 | */ |
2207 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2208 | { | |
2209 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2210 | struct btrfs_block_rsv *block_rsv = NULL; |
2211 | int reserve = 0; | |
2212 | int insert = 0; | |
2213 | int ret; | |
7b128766 | 2214 | |
d68fc57b YZ |
2215 | if (!root->orphan_block_rsv) { |
2216 | block_rsv = btrfs_alloc_block_rsv(root); | |
2217 | BUG_ON(!block_rsv); | |
2218 | } | |
7b128766 | 2219 | |
d68fc57b YZ |
2220 | spin_lock(&root->orphan_lock); |
2221 | if (!root->orphan_block_rsv) { | |
2222 | root->orphan_block_rsv = block_rsv; | |
2223 | } else if (block_rsv) { | |
2224 | btrfs_free_block_rsv(root, block_rsv); | |
2225 | block_rsv = NULL; | |
7b128766 | 2226 | } |
7b128766 | 2227 | |
d68fc57b YZ |
2228 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2229 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2230 | #if 0 | |
2231 | /* | |
2232 | * For proper ENOSPC handling, we should do orphan | |
2233 | * cleanup when mounting. But this introduces backward | |
2234 | * compatibility issue. | |
2235 | */ | |
2236 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2237 | insert = 2; | |
2238 | else | |
2239 | insert = 1; | |
2240 | #endif | |
2241 | insert = 1; | |
7b128766 JB |
2242 | } |
2243 | ||
d68fc57b YZ |
2244 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2245 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2246 | reserve = 1; | |
2247 | } | |
2248 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2249 | |
d68fc57b YZ |
2250 | if (block_rsv) |
2251 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
7b128766 | 2252 | |
d68fc57b YZ |
2253 | /* grab metadata reservation from transaction handle */ |
2254 | if (reserve) { | |
2255 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2256 | BUG_ON(ret); | |
2257 | } | |
7b128766 | 2258 | |
d68fc57b YZ |
2259 | /* insert an orphan item to track this unlinked/truncated file */ |
2260 | if (insert >= 1) { | |
33345d01 | 2261 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
d68fc57b YZ |
2262 | BUG_ON(ret); |
2263 | } | |
2264 | ||
2265 | /* insert an orphan item to track subvolume contains orphan files */ | |
2266 | if (insert >= 2) { | |
2267 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2268 | root->root_key.objectid); | |
2269 | BUG_ON(ret); | |
2270 | } | |
2271 | return 0; | |
7b128766 JB |
2272 | } |
2273 | ||
2274 | /* | |
2275 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2276 | * item for this particular inode. | |
2277 | */ | |
2278 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2279 | { | |
2280 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2281 | int delete_item = 0; |
2282 | int release_rsv = 0; | |
7b128766 JB |
2283 | int ret = 0; |
2284 | ||
d68fc57b YZ |
2285 | spin_lock(&root->orphan_lock); |
2286 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2287 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2288 | delete_item = 1; | |
7b128766 JB |
2289 | } |
2290 | ||
d68fc57b YZ |
2291 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2292 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2293 | release_rsv = 1; | |
7b128766 | 2294 | } |
d68fc57b | 2295 | spin_unlock(&root->orphan_lock); |
7b128766 | 2296 | |
d68fc57b | 2297 | if (trans && delete_item) { |
33345d01 | 2298 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
d68fc57b YZ |
2299 | BUG_ON(ret); |
2300 | } | |
7b128766 | 2301 | |
d68fc57b YZ |
2302 | if (release_rsv) |
2303 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2304 | |
d68fc57b | 2305 | return 0; |
7b128766 JB |
2306 | } |
2307 | ||
2308 | /* | |
2309 | * this cleans up any orphans that may be left on the list from the last use | |
2310 | * of this root. | |
2311 | */ | |
66b4ffd1 | 2312 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2313 | { |
2314 | struct btrfs_path *path; | |
2315 | struct extent_buffer *leaf; | |
7b128766 JB |
2316 | struct btrfs_key key, found_key; |
2317 | struct btrfs_trans_handle *trans; | |
2318 | struct inode *inode; | |
2319 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2320 | ||
d68fc57b | 2321 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2322 | return 0; |
c71bf099 YZ |
2323 | |
2324 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2325 | if (!path) { |
2326 | ret = -ENOMEM; | |
2327 | goto out; | |
2328 | } | |
7b128766 JB |
2329 | path->reada = -1; |
2330 | ||
2331 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2332 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2333 | key.offset = (u64)-1; | |
2334 | ||
7b128766 JB |
2335 | while (1) { |
2336 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2337 | if (ret < 0) |
2338 | goto out; | |
7b128766 JB |
2339 | |
2340 | /* | |
2341 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2342 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2343 | * find the key and see if we have stuff that matches |
2344 | */ | |
2345 | if (ret > 0) { | |
66b4ffd1 | 2346 | ret = 0; |
7b128766 JB |
2347 | if (path->slots[0] == 0) |
2348 | break; | |
2349 | path->slots[0]--; | |
2350 | } | |
2351 | ||
2352 | /* pull out the item */ | |
2353 | leaf = path->nodes[0]; | |
7b128766 JB |
2354 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2355 | ||
2356 | /* make sure the item matches what we want */ | |
2357 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2358 | break; | |
2359 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2360 | break; | |
2361 | ||
2362 | /* release the path since we're done with it */ | |
b3b4aa74 | 2363 | btrfs_release_path(path); |
7b128766 JB |
2364 | |
2365 | /* | |
2366 | * this is where we are basically btrfs_lookup, without the | |
2367 | * crossing root thing. we store the inode number in the | |
2368 | * offset of the orphan item. | |
2369 | */ | |
5d4f98a2 YZ |
2370 | found_key.objectid = found_key.offset; |
2371 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2372 | found_key.offset = 0; | |
73f73415 | 2373 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
66b4ffd1 JB |
2374 | if (IS_ERR(inode)) { |
2375 | ret = PTR_ERR(inode); | |
2376 | goto out; | |
2377 | } | |
7b128766 | 2378 | |
7b128766 JB |
2379 | /* |
2380 | * add this inode to the orphan list so btrfs_orphan_del does | |
2381 | * the proper thing when we hit it | |
2382 | */ | |
d68fc57b | 2383 | spin_lock(&root->orphan_lock); |
7b128766 | 2384 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
d68fc57b | 2385 | spin_unlock(&root->orphan_lock); |
7b128766 JB |
2386 | |
2387 | /* | |
2388 | * if this is a bad inode, means we actually succeeded in | |
2389 | * removing the inode, but not the orphan record, which means | |
2390 | * we need to manually delete the orphan since iput will just | |
2391 | * do a destroy_inode | |
2392 | */ | |
2393 | if (is_bad_inode(inode)) { | |
a22285a6 | 2394 | trans = btrfs_start_transaction(root, 0); |
66b4ffd1 JB |
2395 | if (IS_ERR(trans)) { |
2396 | ret = PTR_ERR(trans); | |
2397 | goto out; | |
2398 | } | |
7b128766 | 2399 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2400 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2401 | iput(inode); |
2402 | continue; | |
2403 | } | |
2404 | ||
2405 | /* if we have links, this was a truncate, lets do that */ | |
2406 | if (inode->i_nlink) { | |
a41ad394 JB |
2407 | if (!S_ISREG(inode->i_mode)) { |
2408 | WARN_ON(1); | |
2409 | iput(inode); | |
2410 | continue; | |
2411 | } | |
7b128766 | 2412 | nr_truncate++; |
66b4ffd1 | 2413 | ret = btrfs_truncate(inode); |
7b128766 JB |
2414 | } else { |
2415 | nr_unlink++; | |
2416 | } | |
2417 | ||
2418 | /* this will do delete_inode and everything for us */ | |
2419 | iput(inode); | |
66b4ffd1 JB |
2420 | if (ret) |
2421 | goto out; | |
7b128766 | 2422 | } |
d68fc57b YZ |
2423 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2424 | ||
2425 | if (root->orphan_block_rsv) | |
2426 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2427 | (u64)-1); | |
2428 | ||
2429 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2430 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2431 | if (!IS_ERR(trans)) |
2432 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2433 | } |
7b128766 JB |
2434 | |
2435 | if (nr_unlink) | |
2436 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2437 | if (nr_truncate) | |
2438 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2439 | |
2440 | out: | |
2441 | if (ret) | |
2442 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2443 | btrfs_free_path(path); | |
2444 | return ret; | |
7b128766 JB |
2445 | } |
2446 | ||
46a53cca CM |
2447 | /* |
2448 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2449 | * don't find any xattrs, we know there can't be any acls. | |
2450 | * | |
2451 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2452 | */ | |
2453 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2454 | int slot, u64 objectid) | |
2455 | { | |
2456 | u32 nritems = btrfs_header_nritems(leaf); | |
2457 | struct btrfs_key found_key; | |
2458 | int scanned = 0; | |
2459 | ||
2460 | slot++; | |
2461 | while (slot < nritems) { | |
2462 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2463 | ||
2464 | /* we found a different objectid, there must not be acls */ | |
2465 | if (found_key.objectid != objectid) | |
2466 | return 0; | |
2467 | ||
2468 | /* we found an xattr, assume we've got an acl */ | |
2469 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2470 | return 1; | |
2471 | ||
2472 | /* | |
2473 | * we found a key greater than an xattr key, there can't | |
2474 | * be any acls later on | |
2475 | */ | |
2476 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2477 | return 0; | |
2478 | ||
2479 | slot++; | |
2480 | scanned++; | |
2481 | ||
2482 | /* | |
2483 | * it goes inode, inode backrefs, xattrs, extents, | |
2484 | * so if there are a ton of hard links to an inode there can | |
2485 | * be a lot of backrefs. Don't waste time searching too hard, | |
2486 | * this is just an optimization | |
2487 | */ | |
2488 | if (scanned >= 8) | |
2489 | break; | |
2490 | } | |
2491 | /* we hit the end of the leaf before we found an xattr or | |
2492 | * something larger than an xattr. We have to assume the inode | |
2493 | * has acls | |
2494 | */ | |
2495 | return 1; | |
2496 | } | |
2497 | ||
d352ac68 CM |
2498 | /* |
2499 | * read an inode from the btree into the in-memory inode | |
2500 | */ | |
5d4f98a2 | 2501 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2502 | { |
2503 | struct btrfs_path *path; | |
5f39d397 | 2504 | struct extent_buffer *leaf; |
39279cc3 | 2505 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2506 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2507 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2508 | struct btrfs_key location; | |
46a53cca | 2509 | int maybe_acls; |
618e21d5 | 2510 | u32 rdev; |
39279cc3 CM |
2511 | int ret; |
2512 | ||
2513 | path = btrfs_alloc_path(); | |
2514 | BUG_ON(!path); | |
d90c7321 | 2515 | path->leave_spinning = 1; |
39279cc3 | 2516 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2517 | |
39279cc3 | 2518 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2519 | if (ret) |
39279cc3 | 2520 | goto make_bad; |
39279cc3 | 2521 | |
5f39d397 CM |
2522 | leaf = path->nodes[0]; |
2523 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2524 | struct btrfs_inode_item); | |
d90c7321 JB |
2525 | if (!leaf->map_token) |
2526 | map_private_extent_buffer(leaf, (unsigned long)inode_item, | |
2527 | sizeof(struct btrfs_inode_item), | |
2528 | &leaf->map_token, &leaf->kaddr, | |
2529 | &leaf->map_start, &leaf->map_len, | |
2530 | KM_USER1); | |
5f39d397 CM |
2531 | |
2532 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2533 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2534 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2535 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2536 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2537 | |
2538 | tspec = btrfs_inode_atime(inode_item); | |
2539 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2540 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2541 | ||
2542 | tspec = btrfs_inode_mtime(inode_item); | |
2543 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2544 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2545 | ||
2546 | tspec = btrfs_inode_ctime(inode_item); | |
2547 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2548 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2549 | ||
a76a3cd4 | 2550 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2551 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2552 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2553 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2554 | inode->i_rdev = 0; |
5f39d397 CM |
2555 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2556 | ||
aec7477b | 2557 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2558 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2559 | |
46a53cca CM |
2560 | /* |
2561 | * try to precache a NULL acl entry for files that don't have | |
2562 | * any xattrs or acls | |
2563 | */ | |
33345d01 LZ |
2564 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2565 | btrfs_ino(inode)); | |
72c04902 AV |
2566 | if (!maybe_acls) |
2567 | cache_no_acl(inode); | |
46a53cca | 2568 | |
d90c7321 JB |
2569 | if (leaf->map_token) { |
2570 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2571 | leaf->map_token = NULL; | |
2572 | } | |
2573 | ||
39279cc3 CM |
2574 | btrfs_free_path(path); |
2575 | inode_item = NULL; | |
2576 | ||
39279cc3 | 2577 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2578 | case S_IFREG: |
2579 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2580 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2581 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2582 | inode->i_fop = &btrfs_file_operations; |
2583 | inode->i_op = &btrfs_file_inode_operations; | |
2584 | break; | |
2585 | case S_IFDIR: | |
2586 | inode->i_fop = &btrfs_dir_file_operations; | |
2587 | if (root == root->fs_info->tree_root) | |
2588 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2589 | else | |
2590 | inode->i_op = &btrfs_dir_inode_operations; | |
2591 | break; | |
2592 | case S_IFLNK: | |
2593 | inode->i_op = &btrfs_symlink_inode_operations; | |
2594 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2595 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2596 | break; |
618e21d5 | 2597 | default: |
0279b4cd | 2598 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2599 | init_special_inode(inode, inode->i_mode, rdev); |
2600 | break; | |
39279cc3 | 2601 | } |
6cbff00f CH |
2602 | |
2603 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2604 | return; |
2605 | ||
2606 | make_bad: | |
39279cc3 | 2607 | btrfs_free_path(path); |
39279cc3 CM |
2608 | make_bad_inode(inode); |
2609 | } | |
2610 | ||
d352ac68 CM |
2611 | /* |
2612 | * given a leaf and an inode, copy the inode fields into the leaf | |
2613 | */ | |
e02119d5 CM |
2614 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2615 | struct extent_buffer *leaf, | |
5f39d397 | 2616 | struct btrfs_inode_item *item, |
39279cc3 CM |
2617 | struct inode *inode) |
2618 | { | |
12ddb96c JB |
2619 | if (!leaf->map_token) |
2620 | map_private_extent_buffer(leaf, (unsigned long)item, | |
2621 | sizeof(struct btrfs_inode_item), | |
2622 | &leaf->map_token, &leaf->kaddr, | |
2623 | &leaf->map_start, &leaf->map_len, | |
2624 | KM_USER1); | |
2625 | ||
5f39d397 CM |
2626 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2627 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2628 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2629 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2630 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2631 | ||
2632 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2633 | inode->i_atime.tv_sec); | |
2634 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2635 | inode->i_atime.tv_nsec); | |
2636 | ||
2637 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2638 | inode->i_mtime.tv_sec); | |
2639 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2640 | inode->i_mtime.tv_nsec); | |
2641 | ||
2642 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2643 | inode->i_ctime.tv_sec); | |
2644 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2645 | inode->i_ctime.tv_nsec); | |
2646 | ||
a76a3cd4 | 2647 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2648 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2649 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2650 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2651 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2652 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2653 | btrfs_set_inode_block_group(leaf, item, 0); |
12ddb96c JB |
2654 | |
2655 | if (leaf->map_token) { | |
2656 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2657 | leaf->map_token = NULL; | |
2658 | } | |
39279cc3 CM |
2659 | } |
2660 | ||
d352ac68 CM |
2661 | /* |
2662 | * copy everything in the in-memory inode into the btree. | |
2663 | */ | |
d397712b CM |
2664 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2665 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2666 | { |
2667 | struct btrfs_inode_item *inode_item; | |
2668 | struct btrfs_path *path; | |
5f39d397 | 2669 | struct extent_buffer *leaf; |
39279cc3 CM |
2670 | int ret; |
2671 | ||
16cdcec7 MX |
2672 | /* |
2673 | * If root is tree root, it means this inode is used to | |
2674 | * store free space information. And these inodes are updated | |
2675 | * when committing the transaction, so they needn't delaye to | |
2676 | * be updated, or deadlock will occured. | |
2677 | */ | |
dcc6d073 | 2678 | if (!is_free_space_inode(root, inode)) { |
16cdcec7 MX |
2679 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2680 | if (!ret) | |
2681 | btrfs_set_inode_last_trans(trans, inode); | |
2682 | return ret; | |
2683 | } | |
2684 | ||
39279cc3 | 2685 | path = btrfs_alloc_path(); |
16cdcec7 MX |
2686 | if (!path) |
2687 | return -ENOMEM; | |
2688 | ||
b9473439 | 2689 | path->leave_spinning = 1; |
16cdcec7 MX |
2690 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2691 | 1); | |
39279cc3 CM |
2692 | if (ret) { |
2693 | if (ret > 0) | |
2694 | ret = -ENOENT; | |
2695 | goto failed; | |
2696 | } | |
2697 | ||
b4ce94de | 2698 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2699 | leaf = path->nodes[0]; |
2700 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2701 | struct btrfs_inode_item); |
39279cc3 | 2702 | |
e02119d5 | 2703 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2704 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2705 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2706 | ret = 0; |
2707 | failed: | |
39279cc3 CM |
2708 | btrfs_free_path(path); |
2709 | return ret; | |
2710 | } | |
2711 | ||
d352ac68 CM |
2712 | /* |
2713 | * unlink helper that gets used here in inode.c and in the tree logging | |
2714 | * recovery code. It remove a link in a directory with a given name, and | |
2715 | * also drops the back refs in the inode to the directory | |
2716 | */ | |
92986796 AV |
2717 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2718 | struct btrfs_root *root, | |
2719 | struct inode *dir, struct inode *inode, | |
2720 | const char *name, int name_len) | |
39279cc3 CM |
2721 | { |
2722 | struct btrfs_path *path; | |
39279cc3 | 2723 | int ret = 0; |
5f39d397 | 2724 | struct extent_buffer *leaf; |
39279cc3 | 2725 | struct btrfs_dir_item *di; |
5f39d397 | 2726 | struct btrfs_key key; |
aec7477b | 2727 | u64 index; |
33345d01 LZ |
2728 | u64 ino = btrfs_ino(inode); |
2729 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2730 | |
2731 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2732 | if (!path) { |
2733 | ret = -ENOMEM; | |
554233a6 | 2734 | goto out; |
54aa1f4d CM |
2735 | } |
2736 | ||
b9473439 | 2737 | path->leave_spinning = 1; |
33345d01 | 2738 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2739 | name, name_len, -1); |
2740 | if (IS_ERR(di)) { | |
2741 | ret = PTR_ERR(di); | |
2742 | goto err; | |
2743 | } | |
2744 | if (!di) { | |
2745 | ret = -ENOENT; | |
2746 | goto err; | |
2747 | } | |
5f39d397 CM |
2748 | leaf = path->nodes[0]; |
2749 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2750 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2751 | if (ret) |
2752 | goto err; | |
b3b4aa74 | 2753 | btrfs_release_path(path); |
39279cc3 | 2754 | |
33345d01 LZ |
2755 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2756 | dir_ino, &index); | |
aec7477b | 2757 | if (ret) { |
d397712b | 2758 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2759 | "inode %llu parent %llu\n", name_len, name, |
2760 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
aec7477b JB |
2761 | goto err; |
2762 | } | |
2763 | ||
16cdcec7 MX |
2764 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
2765 | if (ret) | |
39279cc3 | 2766 | goto err; |
39279cc3 | 2767 | |
e02119d5 | 2768 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2769 | inode, dir_ino); |
49eb7e46 | 2770 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2771 | |
2772 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2773 | dir, index); | |
6418c961 CM |
2774 | if (ret == -ENOENT) |
2775 | ret = 0; | |
39279cc3 CM |
2776 | err: |
2777 | btrfs_free_path(path); | |
e02119d5 CM |
2778 | if (ret) |
2779 | goto out; | |
2780 | ||
2781 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2782 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2783 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2784 | out: |
39279cc3 CM |
2785 | return ret; |
2786 | } | |
2787 | ||
92986796 AV |
2788 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2789 | struct btrfs_root *root, | |
2790 | struct inode *dir, struct inode *inode, | |
2791 | const char *name, int name_len) | |
2792 | { | |
2793 | int ret; | |
2794 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2795 | if (!ret) { | |
2796 | btrfs_drop_nlink(inode); | |
2797 | ret = btrfs_update_inode(trans, root, inode); | |
2798 | } | |
2799 | return ret; | |
2800 | } | |
2801 | ||
2802 | ||
a22285a6 YZ |
2803 | /* helper to check if there is any shared block in the path */ |
2804 | static int check_path_shared(struct btrfs_root *root, | |
2805 | struct btrfs_path *path) | |
39279cc3 | 2806 | { |
a22285a6 YZ |
2807 | struct extent_buffer *eb; |
2808 | int level; | |
0e4dcbef | 2809 | u64 refs = 1; |
5df6a9f6 | 2810 | |
a22285a6 | 2811 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2812 | int ret; |
2813 | ||
a22285a6 YZ |
2814 | if (!path->nodes[level]) |
2815 | break; | |
2816 | eb = path->nodes[level]; | |
2817 | if (!btrfs_block_can_be_shared(root, eb)) | |
2818 | continue; | |
2819 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2820 | &refs, NULL); | |
2821 | if (refs > 1) | |
2822 | return 1; | |
5df6a9f6 | 2823 | } |
dedefd72 | 2824 | return 0; |
39279cc3 CM |
2825 | } |
2826 | ||
a22285a6 YZ |
2827 | /* |
2828 | * helper to start transaction for unlink and rmdir. | |
2829 | * | |
2830 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2831 | * so in enospc case, we should make sure they will free space before | |
2832 | * allowing them to use the global metadata reservation. | |
2833 | */ | |
2834 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2835 | struct dentry *dentry) | |
4df27c4d | 2836 | { |
39279cc3 | 2837 | struct btrfs_trans_handle *trans; |
a22285a6 | 2838 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2839 | struct btrfs_path *path; |
a22285a6 | 2840 | struct btrfs_inode_ref *ref; |
4df27c4d | 2841 | struct btrfs_dir_item *di; |
7b128766 | 2842 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2843 | u64 index; |
a22285a6 YZ |
2844 | int check_link = 1; |
2845 | int err = -ENOSPC; | |
4df27c4d | 2846 | int ret; |
33345d01 LZ |
2847 | u64 ino = btrfs_ino(inode); |
2848 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2849 | |
a22285a6 YZ |
2850 | trans = btrfs_start_transaction(root, 10); |
2851 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
2852 | return trans; | |
4df27c4d | 2853 | |
33345d01 | 2854 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2855 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2856 | |
a22285a6 YZ |
2857 | /* check if there is someone else holds reference */ |
2858 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2859 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2860 | |
a22285a6 YZ |
2861 | if (atomic_read(&inode->i_count) > 2) |
2862 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2863 | |
a22285a6 YZ |
2864 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2865 | return ERR_PTR(-ENOSPC); | |
2866 | ||
2867 | path = btrfs_alloc_path(); | |
2868 | if (!path) { | |
2869 | root->fs_info->enospc_unlink = 0; | |
2870 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2871 | } |
2872 | ||
a22285a6 | 2873 | trans = btrfs_start_transaction(root, 0); |
5df6a9f6 | 2874 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2875 | btrfs_free_path(path); |
2876 | root->fs_info->enospc_unlink = 0; | |
2877 | return trans; | |
2878 | } | |
4df27c4d | 2879 | |
a22285a6 YZ |
2880 | path->skip_locking = 1; |
2881 | path->search_commit_root = 1; | |
4df27c4d | 2882 | |
a22285a6 YZ |
2883 | ret = btrfs_lookup_inode(trans, root, path, |
2884 | &BTRFS_I(dir)->location, 0); | |
2885 | if (ret < 0) { | |
2886 | err = ret; | |
2887 | goto out; | |
2888 | } | |
2889 | if (ret == 0) { | |
2890 | if (check_path_shared(root, path)) | |
2891 | goto out; | |
2892 | } else { | |
2893 | check_link = 0; | |
5df6a9f6 | 2894 | } |
b3b4aa74 | 2895 | btrfs_release_path(path); |
a22285a6 YZ |
2896 | |
2897 | ret = btrfs_lookup_inode(trans, root, path, | |
2898 | &BTRFS_I(inode)->location, 0); | |
2899 | if (ret < 0) { | |
2900 | err = ret; | |
2901 | goto out; | |
2902 | } | |
2903 | if (ret == 0) { | |
2904 | if (check_path_shared(root, path)) | |
2905 | goto out; | |
2906 | } else { | |
2907 | check_link = 0; | |
2908 | } | |
b3b4aa74 | 2909 | btrfs_release_path(path); |
a22285a6 YZ |
2910 | |
2911 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2912 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2913 | ino, (u64)-1, 0); |
a22285a6 YZ |
2914 | if (ret < 0) { |
2915 | err = ret; | |
2916 | goto out; | |
2917 | } | |
2918 | BUG_ON(ret == 0); | |
2919 | if (check_path_shared(root, path)) | |
2920 | goto out; | |
b3b4aa74 | 2921 | btrfs_release_path(path); |
a22285a6 YZ |
2922 | } |
2923 | ||
2924 | if (!check_link) { | |
2925 | err = 0; | |
2926 | goto out; | |
2927 | } | |
2928 | ||
33345d01 | 2929 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2930 | dentry->d_name.name, dentry->d_name.len, 0); |
2931 | if (IS_ERR(di)) { | |
2932 | err = PTR_ERR(di); | |
2933 | goto out; | |
2934 | } | |
2935 | if (di) { | |
2936 | if (check_path_shared(root, path)) | |
2937 | goto out; | |
2938 | } else { | |
2939 | err = 0; | |
2940 | goto out; | |
2941 | } | |
b3b4aa74 | 2942 | btrfs_release_path(path); |
a22285a6 YZ |
2943 | |
2944 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2945 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 2946 | ino, dir_ino, 0); |
a22285a6 YZ |
2947 | if (IS_ERR(ref)) { |
2948 | err = PTR_ERR(ref); | |
2949 | goto out; | |
2950 | } | |
2951 | BUG_ON(!ref); | |
2952 | if (check_path_shared(root, path)) | |
2953 | goto out; | |
2954 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 2955 | btrfs_release_path(path); |
a22285a6 | 2956 | |
16cdcec7 MX |
2957 | /* |
2958 | * This is a commit root search, if we can lookup inode item and other | |
2959 | * relative items in the commit root, it means the transaction of | |
2960 | * dir/file creation has been committed, and the dir index item that we | |
2961 | * delay to insert has also been inserted into the commit root. So | |
2962 | * we needn't worry about the delayed insertion of the dir index item | |
2963 | * here. | |
2964 | */ | |
33345d01 | 2965 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
2966 | dentry->d_name.name, dentry->d_name.len, 0); |
2967 | if (IS_ERR(di)) { | |
2968 | err = PTR_ERR(di); | |
2969 | goto out; | |
2970 | } | |
2971 | BUG_ON(ret == -ENOENT); | |
2972 | if (check_path_shared(root, path)) | |
2973 | goto out; | |
2974 | ||
2975 | err = 0; | |
2976 | out: | |
2977 | btrfs_free_path(path); | |
2978 | if (err) { | |
2979 | btrfs_end_transaction(trans, root); | |
2980 | root->fs_info->enospc_unlink = 0; | |
2981 | return ERR_PTR(err); | |
2982 | } | |
2983 | ||
2984 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2985 | return trans; | |
2986 | } | |
2987 | ||
2988 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2989 | struct btrfs_root *root) | |
2990 | { | |
2991 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
2992 | BUG_ON(!root->fs_info->enospc_unlink); | |
2993 | root->fs_info->enospc_unlink = 0; | |
2994 | } | |
2995 | btrfs_end_transaction_throttle(trans, root); | |
2996 | } | |
2997 | ||
2998 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2999 | { | |
3000 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3001 | struct btrfs_trans_handle *trans; | |
3002 | struct inode *inode = dentry->d_inode; | |
3003 | int ret; | |
3004 | unsigned long nr = 0; | |
3005 | ||
3006 | trans = __unlink_start_trans(dir, dentry); | |
3007 | if (IS_ERR(trans)) | |
3008 | return PTR_ERR(trans); | |
5f39d397 | 3009 | |
12fcfd22 CM |
3010 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3011 | ||
e02119d5 CM |
3012 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3013 | dentry->d_name.name, dentry->d_name.len); | |
a22285a6 | 3014 | BUG_ON(ret); |
7b128766 | 3015 | |
a22285a6 | 3016 | if (inode->i_nlink == 0) { |
7b128766 | 3017 | ret = btrfs_orphan_add(trans, inode); |
a22285a6 YZ |
3018 | BUG_ON(ret); |
3019 | } | |
7b128766 | 3020 | |
d3c2fdcf | 3021 | nr = trans->blocks_used; |
a22285a6 | 3022 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3023 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3024 | return ret; |
3025 | } | |
3026 | ||
4df27c4d YZ |
3027 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3028 | struct btrfs_root *root, | |
3029 | struct inode *dir, u64 objectid, | |
3030 | const char *name, int name_len) | |
3031 | { | |
3032 | struct btrfs_path *path; | |
3033 | struct extent_buffer *leaf; | |
3034 | struct btrfs_dir_item *di; | |
3035 | struct btrfs_key key; | |
3036 | u64 index; | |
3037 | int ret; | |
33345d01 | 3038 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3039 | |
3040 | path = btrfs_alloc_path(); | |
3041 | if (!path) | |
3042 | return -ENOMEM; | |
3043 | ||
33345d01 | 3044 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3045 | name, name_len, -1); |
c704005d | 3046 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
3047 | |
3048 | leaf = path->nodes[0]; | |
3049 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3050 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3051 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3052 | BUG_ON(ret); | |
b3b4aa74 | 3053 | btrfs_release_path(path); |
4df27c4d YZ |
3054 | |
3055 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3056 | objectid, root->root_key.objectid, | |
33345d01 | 3057 | dir_ino, &index, name, name_len); |
4df27c4d YZ |
3058 | if (ret < 0) { |
3059 | BUG_ON(ret != -ENOENT); | |
33345d01 | 3060 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3061 | name, name_len); |
c704005d | 3062 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
3063 | |
3064 | leaf = path->nodes[0]; | |
3065 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3066 | btrfs_release_path(path); |
4df27c4d YZ |
3067 | index = key.offset; |
3068 | } | |
945d8962 | 3069 | btrfs_release_path(path); |
4df27c4d | 3070 | |
16cdcec7 | 3071 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
4df27c4d | 3072 | BUG_ON(ret); |
4df27c4d YZ |
3073 | |
3074 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
3075 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
3076 | ret = btrfs_update_inode(trans, root, dir); | |
3077 | BUG_ON(ret); | |
4df27c4d | 3078 | |
71d7aed0 | 3079 | btrfs_free_path(path); |
4df27c4d YZ |
3080 | return 0; |
3081 | } | |
3082 | ||
39279cc3 CM |
3083 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3084 | { | |
3085 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3086 | int err = 0; |
39279cc3 | 3087 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3088 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3089 | unsigned long nr = 0; |
39279cc3 | 3090 | |
3394e160 | 3091 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 3092 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3093 | return -ENOTEMPTY; |
3094 | ||
a22285a6 YZ |
3095 | trans = __unlink_start_trans(dir, dentry); |
3096 | if (IS_ERR(trans)) | |
5df6a9f6 | 3097 | return PTR_ERR(trans); |
5df6a9f6 | 3098 | |
33345d01 | 3099 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3100 | err = btrfs_unlink_subvol(trans, root, dir, |
3101 | BTRFS_I(inode)->location.objectid, | |
3102 | dentry->d_name.name, | |
3103 | dentry->d_name.len); | |
3104 | goto out; | |
3105 | } | |
3106 | ||
7b128766 JB |
3107 | err = btrfs_orphan_add(trans, inode); |
3108 | if (err) | |
4df27c4d | 3109 | goto out; |
7b128766 | 3110 | |
39279cc3 | 3111 | /* now the directory is empty */ |
e02119d5 CM |
3112 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3113 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3114 | if (!err) |
dbe674a9 | 3115 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3116 | out: |
d3c2fdcf | 3117 | nr = trans->blocks_used; |
a22285a6 | 3118 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3119 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3120 | |
39279cc3 CM |
3121 | return err; |
3122 | } | |
3123 | ||
39279cc3 CM |
3124 | /* |
3125 | * this can truncate away extent items, csum items and directory items. | |
3126 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3127 | * any higher than new_size |
39279cc3 CM |
3128 | * |
3129 | * csum items that cross the new i_size are truncated to the new size | |
3130 | * as well. | |
7b128766 JB |
3131 | * |
3132 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3133 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3134 | */ |
8082510e YZ |
3135 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3136 | struct btrfs_root *root, | |
3137 | struct inode *inode, | |
3138 | u64 new_size, u32 min_type) | |
39279cc3 | 3139 | { |
39279cc3 | 3140 | struct btrfs_path *path; |
5f39d397 | 3141 | struct extent_buffer *leaf; |
39279cc3 | 3142 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3143 | struct btrfs_key key; |
3144 | struct btrfs_key found_key; | |
39279cc3 | 3145 | u64 extent_start = 0; |
db94535d | 3146 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3147 | u64 extent_offset = 0; |
39279cc3 | 3148 | u64 item_end = 0; |
8082510e YZ |
3149 | u64 mask = root->sectorsize - 1; |
3150 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3151 | int found_extent; |
3152 | int del_item; | |
85e21bac CM |
3153 | int pending_del_nr = 0; |
3154 | int pending_del_slot = 0; | |
179e29e4 | 3155 | int extent_type = -1; |
771ed689 | 3156 | int encoding; |
8082510e YZ |
3157 | int ret; |
3158 | int err = 0; | |
33345d01 | 3159 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3160 | |
3161 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3162 | |
0af3d00b | 3163 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3164 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3165 | |
16cdcec7 MX |
3166 | /* |
3167 | * This function is also used to drop the items in the log tree before | |
3168 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3169 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3170 | * items. | |
3171 | */ | |
3172 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3173 | btrfs_kill_delayed_inode_items(inode); | |
3174 | ||
39279cc3 CM |
3175 | path = btrfs_alloc_path(); |
3176 | BUG_ON(!path); | |
33c17ad5 | 3177 | path->reada = -1; |
5f39d397 | 3178 | |
33345d01 | 3179 | key.objectid = ino; |
39279cc3 | 3180 | key.offset = (u64)-1; |
5f39d397 CM |
3181 | key.type = (u8)-1; |
3182 | ||
85e21bac | 3183 | search_again: |
b9473439 | 3184 | path->leave_spinning = 1; |
85e21bac | 3185 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3186 | if (ret < 0) { |
3187 | err = ret; | |
3188 | goto out; | |
3189 | } | |
d397712b | 3190 | |
85e21bac | 3191 | if (ret > 0) { |
e02119d5 CM |
3192 | /* there are no items in the tree for us to truncate, we're |
3193 | * done | |
3194 | */ | |
8082510e YZ |
3195 | if (path->slots[0] == 0) |
3196 | goto out; | |
85e21bac CM |
3197 | path->slots[0]--; |
3198 | } | |
3199 | ||
d397712b | 3200 | while (1) { |
39279cc3 | 3201 | fi = NULL; |
5f39d397 CM |
3202 | leaf = path->nodes[0]; |
3203 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3204 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 3205 | encoding = 0; |
39279cc3 | 3206 | |
33345d01 | 3207 | if (found_key.objectid != ino) |
39279cc3 | 3208 | break; |
5f39d397 | 3209 | |
85e21bac | 3210 | if (found_type < min_type) |
39279cc3 CM |
3211 | break; |
3212 | ||
5f39d397 | 3213 | item_end = found_key.offset; |
39279cc3 | 3214 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3215 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3216 | struct btrfs_file_extent_item); |
179e29e4 | 3217 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3218 | encoding = btrfs_file_extent_compression(leaf, fi); |
3219 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3220 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3221 | ||
179e29e4 | 3222 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3223 | item_end += |
db94535d | 3224 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3225 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3226 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3227 | fi); |
39279cc3 | 3228 | } |
008630c1 | 3229 | item_end--; |
39279cc3 | 3230 | } |
8082510e YZ |
3231 | if (found_type > min_type) { |
3232 | del_item = 1; | |
3233 | } else { | |
3234 | if (item_end < new_size) | |
b888db2b | 3235 | break; |
8082510e YZ |
3236 | if (found_key.offset >= new_size) |
3237 | del_item = 1; | |
3238 | else | |
3239 | del_item = 0; | |
39279cc3 | 3240 | } |
39279cc3 | 3241 | found_extent = 0; |
39279cc3 | 3242 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3243 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3244 | goto delete; | |
3245 | ||
3246 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3247 | u64 num_dec; |
db94535d | 3248 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3249 | if (!del_item && !encoding) { |
db94535d CM |
3250 | u64 orig_num_bytes = |
3251 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3252 | extent_num_bytes = new_size - |
5f39d397 | 3253 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3254 | extent_num_bytes = extent_num_bytes & |
3255 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3256 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3257 | extent_num_bytes); | |
3258 | num_dec = (orig_num_bytes - | |
9069218d | 3259 | extent_num_bytes); |
e02119d5 | 3260 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3261 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3262 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3263 | } else { |
db94535d CM |
3264 | extent_num_bytes = |
3265 | btrfs_file_extent_disk_num_bytes(leaf, | |
3266 | fi); | |
5d4f98a2 YZ |
3267 | extent_offset = found_key.offset - |
3268 | btrfs_file_extent_offset(leaf, fi); | |
3269 | ||
39279cc3 | 3270 | /* FIXME blocksize != 4096 */ |
9069218d | 3271 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3272 | if (extent_start != 0) { |
3273 | found_extent = 1; | |
e02119d5 | 3274 | if (root->ref_cows) |
a76a3cd4 | 3275 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3276 | } |
39279cc3 | 3277 | } |
9069218d | 3278 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3279 | /* |
3280 | * we can't truncate inline items that have had | |
3281 | * special encodings | |
3282 | */ | |
3283 | if (!del_item && | |
3284 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3285 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3286 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3287 | u32 size = new_size - found_key.offset; |
3288 | ||
3289 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3290 | inode_sub_bytes(inode, item_end + 1 - |
3291 | new_size); | |
e02119d5 CM |
3292 | } |
3293 | size = | |
3294 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3295 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3296 | size, 1); |
e02119d5 | 3297 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3298 | inode_sub_bytes(inode, item_end + 1 - |
3299 | found_key.offset); | |
9069218d | 3300 | } |
39279cc3 | 3301 | } |
179e29e4 | 3302 | delete: |
39279cc3 | 3303 | if (del_item) { |
85e21bac CM |
3304 | if (!pending_del_nr) { |
3305 | /* no pending yet, add ourselves */ | |
3306 | pending_del_slot = path->slots[0]; | |
3307 | pending_del_nr = 1; | |
3308 | } else if (pending_del_nr && | |
3309 | path->slots[0] + 1 == pending_del_slot) { | |
3310 | /* hop on the pending chunk */ | |
3311 | pending_del_nr++; | |
3312 | pending_del_slot = path->slots[0]; | |
3313 | } else { | |
d397712b | 3314 | BUG(); |
85e21bac | 3315 | } |
39279cc3 CM |
3316 | } else { |
3317 | break; | |
3318 | } | |
0af3d00b JB |
3319 | if (found_extent && (root->ref_cows || |
3320 | root == root->fs_info->tree_root)) { | |
b9473439 | 3321 | btrfs_set_path_blocking(path); |
39279cc3 | 3322 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3323 | extent_num_bytes, 0, |
3324 | btrfs_header_owner(leaf), | |
33345d01 | 3325 | ino, extent_offset); |
39279cc3 CM |
3326 | BUG_ON(ret); |
3327 | } | |
85e21bac | 3328 | |
8082510e YZ |
3329 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3330 | break; | |
3331 | ||
3332 | if (path->slots[0] == 0 || | |
3333 | path->slots[0] != pending_del_slot) { | |
82d5902d LZ |
3334 | if (root->ref_cows && |
3335 | BTRFS_I(inode)->location.objectid != | |
3336 | BTRFS_FREE_INO_OBJECTID) { | |
8082510e YZ |
3337 | err = -EAGAIN; |
3338 | goto out; | |
3339 | } | |
3340 | if (pending_del_nr) { | |
3341 | ret = btrfs_del_items(trans, root, path, | |
3342 | pending_del_slot, | |
3343 | pending_del_nr); | |
3344 | BUG_ON(ret); | |
3345 | pending_del_nr = 0; | |
3346 | } | |
b3b4aa74 | 3347 | btrfs_release_path(path); |
85e21bac | 3348 | goto search_again; |
8082510e YZ |
3349 | } else { |
3350 | path->slots[0]--; | |
85e21bac | 3351 | } |
39279cc3 | 3352 | } |
8082510e | 3353 | out: |
85e21bac CM |
3354 | if (pending_del_nr) { |
3355 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3356 | pending_del_nr); | |
d68fc57b | 3357 | BUG_ON(ret); |
85e21bac | 3358 | } |
39279cc3 | 3359 | btrfs_free_path(path); |
8082510e | 3360 | return err; |
39279cc3 CM |
3361 | } |
3362 | ||
3363 | /* | |
3364 | * taken from block_truncate_page, but does cow as it zeros out | |
3365 | * any bytes left in the last page in the file. | |
3366 | */ | |
3367 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3368 | { | |
3369 | struct inode *inode = mapping->host; | |
db94535d | 3370 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3371 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3372 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3373 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3374 | char *kaddr; |
db94535d | 3375 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3376 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3377 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3378 | struct page *page; | |
39279cc3 | 3379 | int ret = 0; |
a52d9a80 | 3380 | u64 page_start; |
e6dcd2dc | 3381 | u64 page_end; |
39279cc3 CM |
3382 | |
3383 | if ((offset & (blocksize - 1)) == 0) | |
3384 | goto out; | |
0ca1f7ce | 3385 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3386 | if (ret) |
3387 | goto out; | |
39279cc3 CM |
3388 | |
3389 | ret = -ENOMEM; | |
211c17f5 | 3390 | again: |
39279cc3 | 3391 | page = grab_cache_page(mapping, index); |
5d5e103a | 3392 | if (!page) { |
0ca1f7ce | 3393 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3394 | goto out; |
5d5e103a | 3395 | } |
e6dcd2dc CM |
3396 | |
3397 | page_start = page_offset(page); | |
3398 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3399 | ||
39279cc3 | 3400 | if (!PageUptodate(page)) { |
9ebefb18 | 3401 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3402 | lock_page(page); |
211c17f5 CM |
3403 | if (page->mapping != mapping) { |
3404 | unlock_page(page); | |
3405 | page_cache_release(page); | |
3406 | goto again; | |
3407 | } | |
39279cc3 CM |
3408 | if (!PageUptodate(page)) { |
3409 | ret = -EIO; | |
89642229 | 3410 | goto out_unlock; |
39279cc3 CM |
3411 | } |
3412 | } | |
211c17f5 | 3413 | wait_on_page_writeback(page); |
e6dcd2dc | 3414 | |
2ac55d41 JB |
3415 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3416 | GFP_NOFS); | |
e6dcd2dc CM |
3417 | set_page_extent_mapped(page); |
3418 | ||
3419 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3420 | if (ordered) { | |
2ac55d41 JB |
3421 | unlock_extent_cached(io_tree, page_start, page_end, |
3422 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3423 | unlock_page(page); |
3424 | page_cache_release(page); | |
eb84ae03 | 3425 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3426 | btrfs_put_ordered_extent(ordered); |
3427 | goto again; | |
3428 | } | |
3429 | ||
2ac55d41 | 3430 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3431 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3432 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3433 | |
2ac55d41 JB |
3434 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3435 | &cached_state); | |
9ed74f2d | 3436 | if (ret) { |
2ac55d41 JB |
3437 | unlock_extent_cached(io_tree, page_start, page_end, |
3438 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3439 | goto out_unlock; |
3440 | } | |
3441 | ||
e6dcd2dc CM |
3442 | ret = 0; |
3443 | if (offset != PAGE_CACHE_SIZE) { | |
3444 | kaddr = kmap(page); | |
3445 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3446 | flush_dcache_page(page); | |
3447 | kunmap(page); | |
3448 | } | |
247e743c | 3449 | ClearPageChecked(page); |
e6dcd2dc | 3450 | set_page_dirty(page); |
2ac55d41 JB |
3451 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3452 | GFP_NOFS); | |
39279cc3 | 3453 | |
89642229 | 3454 | out_unlock: |
5d5e103a | 3455 | if (ret) |
0ca1f7ce | 3456 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3457 | unlock_page(page); |
3458 | page_cache_release(page); | |
3459 | out: | |
3460 | return ret; | |
3461 | } | |
3462 | ||
695a0d0d JB |
3463 | /* |
3464 | * This function puts in dummy file extents for the area we're creating a hole | |
3465 | * for. So if we are truncating this file to a larger size we need to insert | |
3466 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3467 | * the range between oldsize and size | |
3468 | */ | |
a41ad394 | 3469 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3470 | { |
9036c102 YZ |
3471 | struct btrfs_trans_handle *trans; |
3472 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3473 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3474 | struct extent_map *em = NULL; |
2ac55d41 | 3475 | struct extent_state *cached_state = NULL; |
9036c102 | 3476 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3477 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3478 | u64 block_end = (size + mask) & ~mask; |
3479 | u64 last_byte; | |
3480 | u64 cur_offset; | |
3481 | u64 hole_size; | |
9ed74f2d | 3482 | int err = 0; |
39279cc3 | 3483 | |
9036c102 YZ |
3484 | if (size <= hole_start) |
3485 | return 0; | |
3486 | ||
9036c102 YZ |
3487 | while (1) { |
3488 | struct btrfs_ordered_extent *ordered; | |
3489 | btrfs_wait_ordered_range(inode, hole_start, | |
3490 | block_end - hole_start); | |
2ac55d41 JB |
3491 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3492 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3493 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3494 | if (!ordered) | |
3495 | break; | |
2ac55d41 JB |
3496 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3497 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3498 | btrfs_put_ordered_extent(ordered); |
3499 | } | |
39279cc3 | 3500 | |
9036c102 YZ |
3501 | cur_offset = hole_start; |
3502 | while (1) { | |
3503 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3504 | block_end - cur_offset, 0); | |
c704005d | 3505 | BUG_ON(IS_ERR_OR_NULL(em)); |
9036c102 YZ |
3506 | last_byte = min(extent_map_end(em), block_end); |
3507 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3508 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3509 | u64 hint_byte = 0; |
9036c102 | 3510 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3511 | |
a22285a6 YZ |
3512 | trans = btrfs_start_transaction(root, 2); |
3513 | if (IS_ERR(trans)) { | |
3514 | err = PTR_ERR(trans); | |
9ed74f2d | 3515 | break; |
a22285a6 | 3516 | } |
8082510e YZ |
3517 | |
3518 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3519 | cur_offset + hole_size, | |
3520 | &hint_byte, 1); | |
3893e33b JB |
3521 | if (err) |
3522 | break; | |
8082510e | 3523 | |
9036c102 | 3524 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3525 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3526 | 0, hole_size, 0, hole_size, |
3527 | 0, 0, 0); | |
3893e33b JB |
3528 | if (err) |
3529 | break; | |
8082510e | 3530 | |
9036c102 YZ |
3531 | btrfs_drop_extent_cache(inode, hole_start, |
3532 | last_byte - 1, 0); | |
8082510e YZ |
3533 | |
3534 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3535 | } |
3536 | free_extent_map(em); | |
a22285a6 | 3537 | em = NULL; |
9036c102 | 3538 | cur_offset = last_byte; |
8082510e | 3539 | if (cur_offset >= block_end) |
9036c102 YZ |
3540 | break; |
3541 | } | |
1832a6d5 | 3542 | |
a22285a6 | 3543 | free_extent_map(em); |
2ac55d41 JB |
3544 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3545 | GFP_NOFS); | |
9036c102 YZ |
3546 | return err; |
3547 | } | |
39279cc3 | 3548 | |
a41ad394 | 3549 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3550 | { |
a41ad394 | 3551 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3552 | int ret; |
3553 | ||
a41ad394 | 3554 | if (newsize == oldsize) |
8082510e YZ |
3555 | return 0; |
3556 | ||
a41ad394 JB |
3557 | if (newsize > oldsize) { |
3558 | i_size_write(inode, newsize); | |
3559 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3560 | truncate_pagecache(inode, oldsize, newsize); | |
3561 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
8082510e | 3562 | if (ret) { |
a41ad394 | 3563 | btrfs_setsize(inode, oldsize); |
8082510e YZ |
3564 | return ret; |
3565 | } | |
3566 | ||
930f028a | 3567 | mark_inode_dirty(inode); |
a41ad394 | 3568 | } else { |
8082510e | 3569 | |
a41ad394 JB |
3570 | /* |
3571 | * We're truncating a file that used to have good data down to | |
3572 | * zero. Make sure it gets into the ordered flush list so that | |
3573 | * any new writes get down to disk quickly. | |
3574 | */ | |
3575 | if (newsize == 0) | |
3576 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3577 | |
a41ad394 JB |
3578 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3579 | truncate_setsize(inode, newsize); | |
3580 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3581 | } |
3582 | ||
a41ad394 | 3583 | return ret; |
8082510e YZ |
3584 | } |
3585 | ||
9036c102 YZ |
3586 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3587 | { | |
3588 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3589 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3590 | int err; |
39279cc3 | 3591 | |
b83cc969 LZ |
3592 | if (btrfs_root_readonly(root)) |
3593 | return -EROFS; | |
3594 | ||
9036c102 YZ |
3595 | err = inode_change_ok(inode, attr); |
3596 | if (err) | |
3597 | return err; | |
2bf5a725 | 3598 | |
5a3f23d5 | 3599 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3600 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3601 | if (err) |
3602 | return err; | |
39279cc3 | 3603 | } |
9036c102 | 3604 | |
1025774c CH |
3605 | if (attr->ia_valid) { |
3606 | setattr_copy(inode, attr); | |
3607 | mark_inode_dirty(inode); | |
3608 | ||
3609 | if (attr->ia_valid & ATTR_MODE) | |
3610 | err = btrfs_acl_chmod(inode); | |
3611 | } | |
33268eaf | 3612 | |
39279cc3 CM |
3613 | return err; |
3614 | } | |
61295eb8 | 3615 | |
bd555975 | 3616 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3617 | { |
3618 | struct btrfs_trans_handle *trans; | |
3619 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3620 | unsigned long nr; |
39279cc3 CM |
3621 | int ret; |
3622 | ||
1abe9b8a | 3623 | trace_btrfs_inode_evict(inode); |
3624 | ||
39279cc3 | 3625 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3626 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
82d5902d | 3627 | is_free_space_inode(root, inode))) |
bd555975 AV |
3628 | goto no_delete; |
3629 | ||
39279cc3 | 3630 | if (is_bad_inode(inode)) { |
7b128766 | 3631 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3632 | goto no_delete; |
3633 | } | |
bd555975 | 3634 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3635 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3636 | |
c71bf099 YZ |
3637 | if (root->fs_info->log_root_recovering) { |
3638 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3639 | goto no_delete; | |
3640 | } | |
3641 | ||
76dda93c YZ |
3642 | if (inode->i_nlink > 0) { |
3643 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3644 | goto no_delete; | |
3645 | } | |
3646 | ||
dbe674a9 | 3647 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3648 | |
8082510e | 3649 | while (1) { |
30b4caf5 | 3650 | trans = btrfs_join_transaction(root); |
d68fc57b | 3651 | BUG_ON(IS_ERR(trans)); |
d68fc57b YZ |
3652 | trans->block_rsv = root->orphan_block_rsv; |
3653 | ||
3654 | ret = btrfs_block_rsv_check(trans, root, | |
3655 | root->orphan_block_rsv, 0, 5); | |
3656 | if (ret) { | |
3657 | BUG_ON(ret != -EAGAIN); | |
3658 | ret = btrfs_commit_transaction(trans, root); | |
3659 | BUG_ON(ret); | |
3660 | continue; | |
3661 | } | |
7b128766 | 3662 | |
d68fc57b | 3663 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3664 | if (ret != -EAGAIN) |
3665 | break; | |
85e21bac | 3666 | |
8082510e YZ |
3667 | nr = trans->blocks_used; |
3668 | btrfs_end_transaction(trans, root); | |
3669 | trans = NULL; | |
3670 | btrfs_btree_balance_dirty(root, nr); | |
d68fc57b | 3671 | |
8082510e | 3672 | } |
5f39d397 | 3673 | |
8082510e YZ |
3674 | if (ret == 0) { |
3675 | ret = btrfs_orphan_del(trans, inode); | |
3676 | BUG_ON(ret); | |
3677 | } | |
54aa1f4d | 3678 | |
581bb050 LZ |
3679 | if (!(root == root->fs_info->tree_root || |
3680 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3681 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3682 | |
d3c2fdcf | 3683 | nr = trans->blocks_used; |
54aa1f4d | 3684 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3685 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3686 | no_delete: |
bd555975 | 3687 | end_writeback(inode); |
8082510e | 3688 | return; |
39279cc3 CM |
3689 | } |
3690 | ||
3691 | /* | |
3692 | * this returns the key found in the dir entry in the location pointer. | |
3693 | * If no dir entries were found, location->objectid is 0. | |
3694 | */ | |
3695 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3696 | struct btrfs_key *location) | |
3697 | { | |
3698 | const char *name = dentry->d_name.name; | |
3699 | int namelen = dentry->d_name.len; | |
3700 | struct btrfs_dir_item *di; | |
3701 | struct btrfs_path *path; | |
3702 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3703 | int ret = 0; |
39279cc3 CM |
3704 | |
3705 | path = btrfs_alloc_path(); | |
3706 | BUG_ON(!path); | |
3954401f | 3707 | |
33345d01 | 3708 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3709 | namelen, 0); |
0d9f7f3e Y |
3710 | if (IS_ERR(di)) |
3711 | ret = PTR_ERR(di); | |
d397712b | 3712 | |
c704005d | 3713 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3714 | goto out_err; |
d397712b | 3715 | |
5f39d397 | 3716 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3717 | out: |
39279cc3 CM |
3718 | btrfs_free_path(path); |
3719 | return ret; | |
3954401f CM |
3720 | out_err: |
3721 | location->objectid = 0; | |
3722 | goto out; | |
39279cc3 CM |
3723 | } |
3724 | ||
3725 | /* | |
3726 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3727 | * needs to be changed to reflect the root directory of the tree root. This | |
3728 | * is kind of like crossing a mount point. | |
3729 | */ | |
3730 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3731 | struct inode *dir, |
3732 | struct dentry *dentry, | |
3733 | struct btrfs_key *location, | |
3734 | struct btrfs_root **sub_root) | |
39279cc3 | 3735 | { |
4df27c4d YZ |
3736 | struct btrfs_path *path; |
3737 | struct btrfs_root *new_root; | |
3738 | struct btrfs_root_ref *ref; | |
3739 | struct extent_buffer *leaf; | |
3740 | int ret; | |
3741 | int err = 0; | |
39279cc3 | 3742 | |
4df27c4d YZ |
3743 | path = btrfs_alloc_path(); |
3744 | if (!path) { | |
3745 | err = -ENOMEM; | |
3746 | goto out; | |
3747 | } | |
39279cc3 | 3748 | |
4df27c4d YZ |
3749 | err = -ENOENT; |
3750 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3751 | BTRFS_I(dir)->root->root_key.objectid, | |
3752 | location->objectid); | |
3753 | if (ret) { | |
3754 | if (ret < 0) | |
3755 | err = ret; | |
3756 | goto out; | |
3757 | } | |
39279cc3 | 3758 | |
4df27c4d YZ |
3759 | leaf = path->nodes[0]; |
3760 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3761 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3762 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3763 | goto out; | |
39279cc3 | 3764 | |
4df27c4d YZ |
3765 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3766 | (unsigned long)(ref + 1), | |
3767 | dentry->d_name.len); | |
3768 | if (ret) | |
3769 | goto out; | |
3770 | ||
b3b4aa74 | 3771 | btrfs_release_path(path); |
4df27c4d YZ |
3772 | |
3773 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3774 | if (IS_ERR(new_root)) { | |
3775 | err = PTR_ERR(new_root); | |
3776 | goto out; | |
3777 | } | |
3778 | ||
3779 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3780 | err = -ENOENT; | |
3781 | goto out; | |
3782 | } | |
3783 | ||
3784 | *sub_root = new_root; | |
3785 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3786 | location->type = BTRFS_INODE_ITEM_KEY; | |
3787 | location->offset = 0; | |
3788 | err = 0; | |
3789 | out: | |
3790 | btrfs_free_path(path); | |
3791 | return err; | |
39279cc3 CM |
3792 | } |
3793 | ||
5d4f98a2 YZ |
3794 | static void inode_tree_add(struct inode *inode) |
3795 | { | |
3796 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3797 | struct btrfs_inode *entry; | |
03e860bd NP |
3798 | struct rb_node **p; |
3799 | struct rb_node *parent; | |
33345d01 | 3800 | u64 ino = btrfs_ino(inode); |
03e860bd NP |
3801 | again: |
3802 | p = &root->inode_tree.rb_node; | |
3803 | parent = NULL; | |
5d4f98a2 | 3804 | |
1d3382cb | 3805 | if (inode_unhashed(inode)) |
76dda93c YZ |
3806 | return; |
3807 | ||
5d4f98a2 YZ |
3808 | spin_lock(&root->inode_lock); |
3809 | while (*p) { | |
3810 | parent = *p; | |
3811 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3812 | ||
33345d01 | 3813 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3814 | p = &parent->rb_left; |
33345d01 | 3815 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3816 | p = &parent->rb_right; |
5d4f98a2 YZ |
3817 | else { |
3818 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3819 | (I_WILL_FREE | I_FREEING))); |
03e860bd NP |
3820 | rb_erase(parent, &root->inode_tree); |
3821 | RB_CLEAR_NODE(parent); | |
3822 | spin_unlock(&root->inode_lock); | |
3823 | goto again; | |
5d4f98a2 YZ |
3824 | } |
3825 | } | |
3826 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3827 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3828 | spin_unlock(&root->inode_lock); | |
3829 | } | |
3830 | ||
3831 | static void inode_tree_del(struct inode *inode) | |
3832 | { | |
3833 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3834 | int empty = 0; |
5d4f98a2 | 3835 | |
03e860bd | 3836 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3837 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3838 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3839 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3840 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3841 | } |
03e860bd | 3842 | spin_unlock(&root->inode_lock); |
76dda93c | 3843 | |
0af3d00b JB |
3844 | /* |
3845 | * Free space cache has inodes in the tree root, but the tree root has a | |
3846 | * root_refs of 0, so this could end up dropping the tree root as a | |
3847 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3848 | * make sure we don't drop it. | |
3849 | */ | |
3850 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3851 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3852 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3853 | spin_lock(&root->inode_lock); | |
3854 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3855 | spin_unlock(&root->inode_lock); | |
3856 | if (empty) | |
3857 | btrfs_add_dead_root(root); | |
3858 | } | |
3859 | } | |
3860 | ||
3861 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3862 | { | |
3863 | struct rb_node *node; | |
3864 | struct rb_node *prev; | |
3865 | struct btrfs_inode *entry; | |
3866 | struct inode *inode; | |
3867 | u64 objectid = 0; | |
3868 | ||
3869 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3870 | ||
3871 | spin_lock(&root->inode_lock); | |
3872 | again: | |
3873 | node = root->inode_tree.rb_node; | |
3874 | prev = NULL; | |
3875 | while (node) { | |
3876 | prev = node; | |
3877 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3878 | ||
33345d01 | 3879 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 3880 | node = node->rb_left; |
33345d01 | 3881 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
3882 | node = node->rb_right; |
3883 | else | |
3884 | break; | |
3885 | } | |
3886 | if (!node) { | |
3887 | while (prev) { | |
3888 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 3889 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
3890 | node = prev; |
3891 | break; | |
3892 | } | |
3893 | prev = rb_next(prev); | |
3894 | } | |
3895 | } | |
3896 | while (node) { | |
3897 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 3898 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
3899 | inode = igrab(&entry->vfs_inode); |
3900 | if (inode) { | |
3901 | spin_unlock(&root->inode_lock); | |
3902 | if (atomic_read(&inode->i_count) > 1) | |
3903 | d_prune_aliases(inode); | |
3904 | /* | |
45321ac5 | 3905 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
3906 | * the inode cache when its usage count |
3907 | * hits zero. | |
3908 | */ | |
3909 | iput(inode); | |
3910 | cond_resched(); | |
3911 | spin_lock(&root->inode_lock); | |
3912 | goto again; | |
3913 | } | |
3914 | ||
3915 | if (cond_resched_lock(&root->inode_lock)) | |
3916 | goto again; | |
3917 | ||
3918 | node = rb_next(node); | |
3919 | } | |
3920 | spin_unlock(&root->inode_lock); | |
3921 | return 0; | |
5d4f98a2 YZ |
3922 | } |
3923 | ||
e02119d5 CM |
3924 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
3925 | { | |
3926 | struct btrfs_iget_args *args = p; | |
3927 | inode->i_ino = args->ino; | |
e02119d5 | 3928 | BTRFS_I(inode)->root = args->root; |
6a63209f | 3929 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3930 | return 0; |
3931 | } | |
3932 | ||
3933 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3934 | { | |
3935 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 3936 | return args->ino == btrfs_ino(inode) && |
d397712b | 3937 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
3938 | } |
3939 | ||
5d4f98a2 YZ |
3940 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3941 | u64 objectid, | |
3942 | struct btrfs_root *root) | |
39279cc3 CM |
3943 | { |
3944 | struct inode *inode; | |
3945 | struct btrfs_iget_args args; | |
3946 | args.ino = objectid; | |
3947 | args.root = root; | |
3948 | ||
3949 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3950 | btrfs_init_locked_inode, | |
3951 | (void *)&args); | |
3952 | return inode; | |
3953 | } | |
3954 | ||
1a54ef8c BR |
3955 | /* Get an inode object given its location and corresponding root. |
3956 | * Returns in *is_new if the inode was read from disk | |
3957 | */ | |
3958 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 3959 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
3960 | { |
3961 | struct inode *inode; | |
3962 | ||
3963 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3964 | if (!inode) | |
5d4f98a2 | 3965 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3966 | |
3967 | if (inode->i_state & I_NEW) { | |
3968 | BTRFS_I(inode)->root = root; | |
3969 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3970 | btrfs_read_locked_inode(inode); | |
5d4f98a2 | 3971 | inode_tree_add(inode); |
1a54ef8c | 3972 | unlock_new_inode(inode); |
73f73415 JB |
3973 | if (new) |
3974 | *new = 1; | |
1a54ef8c BR |
3975 | } |
3976 | ||
3977 | return inode; | |
3978 | } | |
3979 | ||
4df27c4d YZ |
3980 | static struct inode *new_simple_dir(struct super_block *s, |
3981 | struct btrfs_key *key, | |
3982 | struct btrfs_root *root) | |
3983 | { | |
3984 | struct inode *inode = new_inode(s); | |
3985 | ||
3986 | if (!inode) | |
3987 | return ERR_PTR(-ENOMEM); | |
3988 | ||
4df27c4d YZ |
3989 | BTRFS_I(inode)->root = root; |
3990 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3991 | BTRFS_I(inode)->dummy_inode = 1; | |
3992 | ||
3993 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3994 | inode->i_op = &simple_dir_inode_operations; | |
3995 | inode->i_fop = &simple_dir_operations; | |
3996 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3997 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3998 | ||
3999 | return inode; | |
4000 | } | |
4001 | ||
3de4586c | 4002 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4003 | { |
d397712b | 4004 | struct inode *inode; |
4df27c4d | 4005 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4006 | struct btrfs_root *sub_root = root; |
4007 | struct btrfs_key location; | |
76dda93c | 4008 | int index; |
5d4f98a2 | 4009 | int ret; |
39279cc3 CM |
4010 | |
4011 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4012 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4013 | |
39279cc3 | 4014 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 4015 | |
39279cc3 CM |
4016 | if (ret < 0) |
4017 | return ERR_PTR(ret); | |
5f39d397 | 4018 | |
4df27c4d YZ |
4019 | if (location.objectid == 0) |
4020 | return NULL; | |
4021 | ||
4022 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4023 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4024 | return inode; |
4025 | } | |
4026 | ||
4027 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4028 | ||
76dda93c | 4029 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4030 | ret = fixup_tree_root_location(root, dir, dentry, |
4031 | &location, &sub_root); | |
4032 | if (ret < 0) { | |
4033 | if (ret != -ENOENT) | |
4034 | inode = ERR_PTR(ret); | |
4035 | else | |
4036 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4037 | } else { | |
73f73415 | 4038 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4039 | } |
76dda93c YZ |
4040 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4041 | ||
34d19bad | 4042 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4043 | down_read(&root->fs_info->cleanup_work_sem); |
4044 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4045 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4046 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4047 | if (ret) |
4048 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4049 | } |
4050 | ||
3de4586c CM |
4051 | return inode; |
4052 | } | |
4053 | ||
fe15ce44 | 4054 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4055 | { |
4056 | struct btrfs_root *root; | |
4057 | ||
efefb143 YZ |
4058 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
4059 | dentry = dentry->d_parent; | |
76dda93c | 4060 | |
efefb143 YZ |
4061 | if (dentry->d_inode) { |
4062 | root = BTRFS_I(dentry->d_inode)->root; | |
4063 | if (btrfs_root_refs(&root->root_item) == 0) | |
4064 | return 1; | |
4065 | } | |
76dda93c YZ |
4066 | return 0; |
4067 | } | |
4068 | ||
3de4586c CM |
4069 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4070 | struct nameidata *nd) | |
4071 | { | |
4072 | struct inode *inode; | |
4073 | ||
3de4586c CM |
4074 | inode = btrfs_lookup_dentry(dir, dentry); |
4075 | if (IS_ERR(inode)) | |
4076 | return ERR_CAST(inode); | |
7b128766 | 4077 | |
39279cc3 CM |
4078 | return d_splice_alias(inode, dentry); |
4079 | } | |
4080 | ||
16cdcec7 | 4081 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4082 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4083 | }; | |
4084 | ||
cbdf5a24 DW |
4085 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4086 | filldir_t filldir) | |
39279cc3 | 4087 | { |
6da6abae | 4088 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4089 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4090 | struct btrfs_item *item; | |
4091 | struct btrfs_dir_item *di; | |
4092 | struct btrfs_key key; | |
5f39d397 | 4093 | struct btrfs_key found_key; |
39279cc3 | 4094 | struct btrfs_path *path; |
16cdcec7 MX |
4095 | struct list_head ins_list; |
4096 | struct list_head del_list; | |
39279cc3 | 4097 | int ret; |
5f39d397 | 4098 | struct extent_buffer *leaf; |
39279cc3 | 4099 | int slot; |
39279cc3 CM |
4100 | unsigned char d_type; |
4101 | int over = 0; | |
4102 | u32 di_cur; | |
4103 | u32 di_total; | |
4104 | u32 di_len; | |
4105 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4106 | char tmp_name[32]; |
4107 | char *name_ptr; | |
4108 | int name_len; | |
16cdcec7 | 4109 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4110 | |
4111 | /* FIXME, use a real flag for deciding about the key type */ | |
4112 | if (root->fs_info->tree_root == root) | |
4113 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4114 | |
3954401f CM |
4115 | /* special case for "." */ |
4116 | if (filp->f_pos == 0) { | |
33345d01 | 4117 | over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR); |
3954401f CM |
4118 | if (over) |
4119 | return 0; | |
4120 | filp->f_pos = 1; | |
4121 | } | |
3954401f CM |
4122 | /* special case for .., just use the back ref */ |
4123 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4124 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4125 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 4126 | 2, pino, DT_DIR); |
3954401f | 4127 | if (over) |
49593bfa | 4128 | return 0; |
3954401f CM |
4129 | filp->f_pos = 2; |
4130 | } | |
49593bfa | 4131 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4132 | if (!path) |
4133 | return -ENOMEM; | |
ff5714cc | 4134 | |
026fd317 | 4135 | path->reada = 1; |
49593bfa | 4136 | |
16cdcec7 MX |
4137 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4138 | INIT_LIST_HEAD(&ins_list); | |
4139 | INIT_LIST_HEAD(&del_list); | |
4140 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4141 | } | |
4142 | ||
39279cc3 CM |
4143 | btrfs_set_key_type(&key, key_type); |
4144 | key.offset = filp->f_pos; | |
33345d01 | 4145 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4146 | |
39279cc3 CM |
4147 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4148 | if (ret < 0) | |
4149 | goto err; | |
49593bfa DW |
4150 | |
4151 | while (1) { | |
5f39d397 | 4152 | leaf = path->nodes[0]; |
39279cc3 | 4153 | slot = path->slots[0]; |
b9e03af0 LZ |
4154 | if (slot >= btrfs_header_nritems(leaf)) { |
4155 | ret = btrfs_next_leaf(root, path); | |
4156 | if (ret < 0) | |
4157 | goto err; | |
4158 | else if (ret > 0) | |
4159 | break; | |
4160 | continue; | |
39279cc3 | 4161 | } |
3de4586c | 4162 | |
5f39d397 CM |
4163 | item = btrfs_item_nr(leaf, slot); |
4164 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4165 | ||
4166 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4167 | break; |
5f39d397 | 4168 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4169 | break; |
5f39d397 | 4170 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4171 | goto next; |
16cdcec7 MX |
4172 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4173 | btrfs_should_delete_dir_index(&del_list, | |
4174 | found_key.offset)) | |
4175 | goto next; | |
5f39d397 CM |
4176 | |
4177 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4178 | is_curr = 1; |
49593bfa | 4179 | |
39279cc3 CM |
4180 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4181 | di_cur = 0; | |
5f39d397 | 4182 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4183 | |
4184 | while (di_cur < di_total) { | |
5f39d397 CM |
4185 | struct btrfs_key location; |
4186 | ||
22a94d44 JB |
4187 | if (verify_dir_item(root, leaf, di)) |
4188 | break; | |
4189 | ||
5f39d397 | 4190 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4191 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4192 | name_ptr = tmp_name; |
4193 | } else { | |
4194 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4195 | if (!name_ptr) { |
4196 | ret = -ENOMEM; | |
4197 | goto err; | |
4198 | } | |
5f39d397 CM |
4199 | } |
4200 | read_extent_buffer(leaf, name_ptr, | |
4201 | (unsigned long)(di + 1), name_len); | |
4202 | ||
4203 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4204 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
4205 | |
4206 | /* is this a reference to our own snapshot? If so | |
4207 | * skip it | |
4208 | */ | |
4209 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4210 | location.objectid == root->root_key.objectid) { | |
4211 | over = 0; | |
4212 | goto skip; | |
4213 | } | |
5f39d397 | 4214 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4215 | found_key.offset, location.objectid, |
39279cc3 | 4216 | d_type); |
5f39d397 | 4217 | |
3de4586c | 4218 | skip: |
5f39d397 CM |
4219 | if (name_ptr != tmp_name) |
4220 | kfree(name_ptr); | |
4221 | ||
39279cc3 CM |
4222 | if (over) |
4223 | goto nopos; | |
5103e947 | 4224 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4225 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4226 | di_cur += di_len; |
4227 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4228 | } | |
b9e03af0 LZ |
4229 | next: |
4230 | path->slots[0]++; | |
39279cc3 | 4231 | } |
49593bfa | 4232 | |
16cdcec7 MX |
4233 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4234 | if (is_curr) | |
4235 | filp->f_pos++; | |
4236 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4237 | &ins_list); | |
4238 | if (ret) | |
4239 | goto nopos; | |
4240 | } | |
4241 | ||
49593bfa | 4242 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4243 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4244 | /* |
4245 | * 32-bit glibc will use getdents64, but then strtol - | |
4246 | * so the last number we can serve is this. | |
4247 | */ | |
4248 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4249 | else |
4250 | filp->f_pos++; | |
39279cc3 CM |
4251 | nopos: |
4252 | ret = 0; | |
4253 | err: | |
16cdcec7 MX |
4254 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4255 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4256 | btrfs_free_path(path); |
39279cc3 CM |
4257 | return ret; |
4258 | } | |
4259 | ||
a9185b41 | 4260 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4261 | { |
4262 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4263 | struct btrfs_trans_handle *trans; | |
4264 | int ret = 0; | |
0af3d00b | 4265 | bool nolock = false; |
39279cc3 | 4266 | |
8929ecfa | 4267 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4268 | return 0; |
4269 | ||
7841cb28 | 4270 | if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode)) |
82d5902d | 4271 | nolock = true; |
0af3d00b | 4272 | |
a9185b41 | 4273 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4274 | if (nolock) |
7a7eaa40 | 4275 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4276 | else |
7a7eaa40 | 4277 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4278 | if (IS_ERR(trans)) |
4279 | return PTR_ERR(trans); | |
0af3d00b JB |
4280 | if (nolock) |
4281 | ret = btrfs_end_transaction_nolock(trans, root); | |
4282 | else | |
4283 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4284 | } |
4285 | return ret; | |
4286 | } | |
4287 | ||
4288 | /* | |
54aa1f4d | 4289 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4290 | * inode changes. But, it is most likely to find the inode in cache. |
4291 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4292 | * to keep or drop this code. | |
4293 | */ | |
4294 | void btrfs_dirty_inode(struct inode *inode) | |
4295 | { | |
4296 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4297 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4298 | int ret; |
4299 | ||
4300 | if (BTRFS_I(inode)->dummy_inode) | |
4301 | return; | |
39279cc3 | 4302 | |
7a7eaa40 | 4303 | trans = btrfs_join_transaction(root); |
3612b495 | 4304 | BUG_ON(IS_ERR(trans)); |
8929ecfa YZ |
4305 | |
4306 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4307 | if (ret && ret == -ENOSPC) { |
4308 | /* whoops, lets try again with the full transaction */ | |
4309 | btrfs_end_transaction(trans, root); | |
4310 | trans = btrfs_start_transaction(root, 1); | |
9aeead73 | 4311 | if (IS_ERR(trans)) { |
7a36ddec | 4312 | printk_ratelimited(KERN_ERR "btrfs: fail to " |
33345d01 LZ |
4313 | "dirty inode %llu error %ld\n", |
4314 | (unsigned long long)btrfs_ino(inode), | |
4315 | PTR_ERR(trans)); | |
9aeead73 CM |
4316 | return; |
4317 | } | |
8929ecfa | 4318 | |
94b60442 CM |
4319 | ret = btrfs_update_inode(trans, root, inode); |
4320 | if (ret) { | |
7a36ddec | 4321 | printk_ratelimited(KERN_ERR "btrfs: fail to " |
33345d01 LZ |
4322 | "dirty inode %llu error %d\n", |
4323 | (unsigned long long)btrfs_ino(inode), | |
4324 | ret); | |
94b60442 CM |
4325 | } |
4326 | } | |
39279cc3 | 4327 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4328 | if (BTRFS_I(inode)->delayed_node) |
4329 | btrfs_balance_delayed_items(root); | |
39279cc3 CM |
4330 | } |
4331 | ||
d352ac68 CM |
4332 | /* |
4333 | * find the highest existing sequence number in a directory | |
4334 | * and then set the in-memory index_cnt variable to reflect | |
4335 | * free sequence numbers | |
4336 | */ | |
aec7477b JB |
4337 | static int btrfs_set_inode_index_count(struct inode *inode) |
4338 | { | |
4339 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4340 | struct btrfs_key key, found_key; | |
4341 | struct btrfs_path *path; | |
4342 | struct extent_buffer *leaf; | |
4343 | int ret; | |
4344 | ||
33345d01 | 4345 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4346 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4347 | key.offset = (u64)-1; | |
4348 | ||
4349 | path = btrfs_alloc_path(); | |
4350 | if (!path) | |
4351 | return -ENOMEM; | |
4352 | ||
4353 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4354 | if (ret < 0) | |
4355 | goto out; | |
4356 | /* FIXME: we should be able to handle this */ | |
4357 | if (ret == 0) | |
4358 | goto out; | |
4359 | ret = 0; | |
4360 | ||
4361 | /* | |
4362 | * MAGIC NUMBER EXPLANATION: | |
4363 | * since we search a directory based on f_pos we have to start at 2 | |
4364 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4365 | * else has to start at 2 | |
4366 | */ | |
4367 | if (path->slots[0] == 0) { | |
4368 | BTRFS_I(inode)->index_cnt = 2; | |
4369 | goto out; | |
4370 | } | |
4371 | ||
4372 | path->slots[0]--; | |
4373 | ||
4374 | leaf = path->nodes[0]; | |
4375 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4376 | ||
33345d01 | 4377 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4378 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4379 | BTRFS_I(inode)->index_cnt = 2; | |
4380 | goto out; | |
4381 | } | |
4382 | ||
4383 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4384 | out: | |
4385 | btrfs_free_path(path); | |
4386 | return ret; | |
4387 | } | |
4388 | ||
d352ac68 CM |
4389 | /* |
4390 | * helper to find a free sequence number in a given directory. This current | |
4391 | * code is very simple, later versions will do smarter things in the btree | |
4392 | */ | |
3de4586c | 4393 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4394 | { |
4395 | int ret = 0; | |
4396 | ||
4397 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4398 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4399 | if (ret) { | |
4400 | ret = btrfs_set_inode_index_count(dir); | |
4401 | if (ret) | |
4402 | return ret; | |
4403 | } | |
aec7477b JB |
4404 | } |
4405 | ||
00e4e6b3 | 4406 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4407 | BTRFS_I(dir)->index_cnt++; |
4408 | ||
4409 | return ret; | |
4410 | } | |
4411 | ||
39279cc3 CM |
4412 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4413 | struct btrfs_root *root, | |
aec7477b | 4414 | struct inode *dir, |
9c58309d | 4415 | const char *name, int name_len, |
d82a6f1d JB |
4416 | u64 ref_objectid, u64 objectid, int mode, |
4417 | u64 *index) | |
39279cc3 CM |
4418 | { |
4419 | struct inode *inode; | |
5f39d397 | 4420 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4421 | struct btrfs_key *location; |
5f39d397 | 4422 | struct btrfs_path *path; |
9c58309d CM |
4423 | struct btrfs_inode_ref *ref; |
4424 | struct btrfs_key key[2]; | |
4425 | u32 sizes[2]; | |
4426 | unsigned long ptr; | |
39279cc3 CM |
4427 | int ret; |
4428 | int owner; | |
4429 | ||
5f39d397 CM |
4430 | path = btrfs_alloc_path(); |
4431 | BUG_ON(!path); | |
4432 | ||
39279cc3 | 4433 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4434 | if (!inode) { |
4435 | btrfs_free_path(path); | |
39279cc3 | 4436 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4437 | } |
39279cc3 | 4438 | |
581bb050 LZ |
4439 | /* |
4440 | * we have to initialize this early, so we can reclaim the inode | |
4441 | * number if we fail afterwards in this function. | |
4442 | */ | |
4443 | inode->i_ino = objectid; | |
4444 | ||
aec7477b | 4445 | if (dir) { |
1abe9b8a | 4446 | trace_btrfs_inode_request(dir); |
4447 | ||
3de4586c | 4448 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4449 | if (ret) { |
8fb27640 | 4450 | btrfs_free_path(path); |
09771430 | 4451 | iput(inode); |
aec7477b | 4452 | return ERR_PTR(ret); |
09771430 | 4453 | } |
aec7477b JB |
4454 | } |
4455 | /* | |
4456 | * index_cnt is ignored for everything but a dir, | |
4457 | * btrfs_get_inode_index_count has an explanation for the magic | |
4458 | * number | |
4459 | */ | |
4460 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4461 | BTRFS_I(inode)->root = root; |
e02119d5 | 4462 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4463 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4464 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4465 | |
39279cc3 CM |
4466 | if (mode & S_IFDIR) |
4467 | owner = 0; | |
4468 | else | |
4469 | owner = 1; | |
9c58309d CM |
4470 | |
4471 | key[0].objectid = objectid; | |
4472 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4473 | key[0].offset = 0; | |
4474 | ||
4475 | key[1].objectid = objectid; | |
4476 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4477 | key[1].offset = ref_objectid; | |
4478 | ||
4479 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4480 | sizes[1] = name_len + sizeof(*ref); | |
4481 | ||
b9473439 | 4482 | path->leave_spinning = 1; |
9c58309d CM |
4483 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4484 | if (ret != 0) | |
5f39d397 CM |
4485 | goto fail; |
4486 | ||
ecc11fab | 4487 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4488 | inode_set_bytes(inode, 0); |
39279cc3 | 4489 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4490 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4491 | struct btrfs_inode_item); | |
e02119d5 | 4492 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4493 | |
4494 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4495 | struct btrfs_inode_ref); | |
4496 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4497 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4498 | ptr = (unsigned long)(ref + 1); |
4499 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4500 | ||
5f39d397 CM |
4501 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4502 | btrfs_free_path(path); | |
4503 | ||
39279cc3 CM |
4504 | location = &BTRFS_I(inode)->location; |
4505 | location->objectid = objectid; | |
39279cc3 CM |
4506 | location->offset = 0; |
4507 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4508 | ||
6cbff00f CH |
4509 | btrfs_inherit_iflags(inode, dir); |
4510 | ||
94272164 CM |
4511 | if ((mode & S_IFREG)) { |
4512 | if (btrfs_test_opt(root, NODATASUM)) | |
4513 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4514 | if (btrfs_test_opt(root, NODATACOW) || |
4515 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4516 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4517 | } | |
4518 | ||
39279cc3 | 4519 | insert_inode_hash(inode); |
5d4f98a2 | 4520 | inode_tree_add(inode); |
1abe9b8a | 4521 | |
4522 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4523 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4524 | |
39279cc3 | 4525 | return inode; |
5f39d397 | 4526 | fail: |
aec7477b JB |
4527 | if (dir) |
4528 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4529 | btrfs_free_path(path); |
09771430 | 4530 | iput(inode); |
5f39d397 | 4531 | return ERR_PTR(ret); |
39279cc3 CM |
4532 | } |
4533 | ||
4534 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4535 | { | |
4536 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4537 | } | |
4538 | ||
d352ac68 CM |
4539 | /* |
4540 | * utility function to add 'inode' into 'parent_inode' with | |
4541 | * a give name and a given sequence number. | |
4542 | * if 'add_backref' is true, also insert a backref from the | |
4543 | * inode to the parent directory. | |
4544 | */ | |
e02119d5 CM |
4545 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4546 | struct inode *parent_inode, struct inode *inode, | |
4547 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4548 | { |
4df27c4d | 4549 | int ret = 0; |
39279cc3 | 4550 | struct btrfs_key key; |
e02119d5 | 4551 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4552 | u64 ino = btrfs_ino(inode); |
4553 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4554 | |
33345d01 | 4555 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4556 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4557 | } else { | |
33345d01 | 4558 | key.objectid = ino; |
4df27c4d YZ |
4559 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4560 | key.offset = 0; | |
4561 | } | |
4562 | ||
33345d01 | 4563 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4564 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4565 | key.objectid, root->root_key.objectid, | |
33345d01 | 4566 | parent_ino, index, name, name_len); |
4df27c4d | 4567 | } else if (add_backref) { |
33345d01 LZ |
4568 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4569 | parent_ino, index); | |
4df27c4d | 4570 | } |
39279cc3 | 4571 | |
39279cc3 | 4572 | if (ret == 0) { |
4df27c4d | 4573 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
16cdcec7 | 4574 | parent_inode, &key, |
4df27c4d YZ |
4575 | btrfs_inode_type(inode), index); |
4576 | BUG_ON(ret); | |
4577 | ||
dbe674a9 | 4578 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4579 | name_len * 2); |
79c44584 | 4580 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4581 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4582 | } |
4583 | return ret; | |
4584 | } | |
4585 | ||
4586 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4587 | struct inode *dir, struct dentry *dentry, |
4588 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4589 | { |
a1b075d2 JB |
4590 | int err = btrfs_add_link(trans, dir, inode, |
4591 | dentry->d_name.name, dentry->d_name.len, | |
4592 | backref, index); | |
39279cc3 CM |
4593 | if (!err) { |
4594 | d_instantiate(dentry, inode); | |
4595 | return 0; | |
4596 | } | |
4597 | if (err > 0) | |
4598 | err = -EEXIST; | |
4599 | return err; | |
4600 | } | |
4601 | ||
618e21d5 JB |
4602 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4603 | int mode, dev_t rdev) | |
4604 | { | |
4605 | struct btrfs_trans_handle *trans; | |
4606 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4607 | struct inode *inode = NULL; |
618e21d5 JB |
4608 | int err; |
4609 | int drop_inode = 0; | |
4610 | u64 objectid; | |
1832a6d5 | 4611 | unsigned long nr = 0; |
00e4e6b3 | 4612 | u64 index = 0; |
618e21d5 JB |
4613 | |
4614 | if (!new_valid_dev(rdev)) | |
4615 | return -EINVAL; | |
4616 | ||
9ed74f2d JB |
4617 | /* |
4618 | * 2 for inode item and ref | |
4619 | * 2 for dir items | |
4620 | * 1 for xattr if selinux is on | |
4621 | */ | |
a22285a6 YZ |
4622 | trans = btrfs_start_transaction(root, 5); |
4623 | if (IS_ERR(trans)) | |
4624 | return PTR_ERR(trans); | |
1832a6d5 | 4625 | |
581bb050 LZ |
4626 | err = btrfs_find_free_ino(root, &objectid); |
4627 | if (err) | |
4628 | goto out_unlock; | |
4629 | ||
aec7477b | 4630 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4631 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4632 | mode, &index); |
7cf96da3 TI |
4633 | if (IS_ERR(inode)) { |
4634 | err = PTR_ERR(inode); | |
618e21d5 | 4635 | goto out_unlock; |
7cf96da3 | 4636 | } |
618e21d5 | 4637 | |
2a7dba39 | 4638 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4639 | if (err) { |
4640 | drop_inode = 1; | |
4641 | goto out_unlock; | |
4642 | } | |
4643 | ||
a1b075d2 | 4644 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4645 | if (err) |
4646 | drop_inode = 1; | |
4647 | else { | |
4648 | inode->i_op = &btrfs_special_inode_operations; | |
4649 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4650 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4651 | } |
618e21d5 | 4652 | out_unlock: |
d3c2fdcf | 4653 | nr = trans->blocks_used; |
89ce8a63 | 4654 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4655 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4656 | if (drop_inode) { |
4657 | inode_dec_link_count(inode); | |
4658 | iput(inode); | |
4659 | } | |
618e21d5 JB |
4660 | return err; |
4661 | } | |
4662 | ||
39279cc3 CM |
4663 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4664 | int mode, struct nameidata *nd) | |
4665 | { | |
4666 | struct btrfs_trans_handle *trans; | |
4667 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4668 | struct inode *inode = NULL; |
39279cc3 | 4669 | int drop_inode = 0; |
a22285a6 | 4670 | int err; |
1832a6d5 | 4671 | unsigned long nr = 0; |
39279cc3 | 4672 | u64 objectid; |
00e4e6b3 | 4673 | u64 index = 0; |
39279cc3 | 4674 | |
9ed74f2d JB |
4675 | /* |
4676 | * 2 for inode item and ref | |
4677 | * 2 for dir items | |
4678 | * 1 for xattr if selinux is on | |
4679 | */ | |
a22285a6 YZ |
4680 | trans = btrfs_start_transaction(root, 5); |
4681 | if (IS_ERR(trans)) | |
4682 | return PTR_ERR(trans); | |
9ed74f2d | 4683 | |
581bb050 LZ |
4684 | err = btrfs_find_free_ino(root, &objectid); |
4685 | if (err) | |
4686 | goto out_unlock; | |
4687 | ||
aec7477b | 4688 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4689 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4690 | mode, &index); |
7cf96da3 TI |
4691 | if (IS_ERR(inode)) { |
4692 | err = PTR_ERR(inode); | |
39279cc3 | 4693 | goto out_unlock; |
7cf96da3 | 4694 | } |
39279cc3 | 4695 | |
2a7dba39 | 4696 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4697 | if (err) { |
4698 | drop_inode = 1; | |
4699 | goto out_unlock; | |
4700 | } | |
4701 | ||
a1b075d2 | 4702 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4703 | if (err) |
4704 | drop_inode = 1; | |
4705 | else { | |
4706 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4707 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4708 | inode->i_fop = &btrfs_file_operations; |
4709 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4710 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4711 | } |
39279cc3 | 4712 | out_unlock: |
d3c2fdcf | 4713 | nr = trans->blocks_used; |
ab78c84d | 4714 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4715 | if (drop_inode) { |
4716 | inode_dec_link_count(inode); | |
4717 | iput(inode); | |
4718 | } | |
d3c2fdcf | 4719 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4720 | return err; |
4721 | } | |
4722 | ||
4723 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4724 | struct dentry *dentry) | |
4725 | { | |
4726 | struct btrfs_trans_handle *trans; | |
4727 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4728 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4729 | u64 index; |
1832a6d5 | 4730 | unsigned long nr = 0; |
39279cc3 CM |
4731 | int err; |
4732 | int drop_inode = 0; | |
4733 | ||
4a8be425 TH |
4734 | /* do not allow sys_link's with other subvols of the same device */ |
4735 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4736 | return -EXDEV; |
4a8be425 | 4737 | |
c055e99e AV |
4738 | if (inode->i_nlink == ~0U) |
4739 | return -EMLINK; | |
4a8be425 | 4740 | |
3de4586c | 4741 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4742 | if (err) |
4743 | goto fail; | |
4744 | ||
a22285a6 | 4745 | /* |
7e6b6465 | 4746 | * 2 items for inode and inode ref |
a22285a6 | 4747 | * 2 items for dir items |
7e6b6465 | 4748 | * 1 item for parent inode |
a22285a6 | 4749 | */ |
7e6b6465 | 4750 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4751 | if (IS_ERR(trans)) { |
4752 | err = PTR_ERR(trans); | |
4753 | goto fail; | |
4754 | } | |
5f39d397 | 4755 | |
3153495d MX |
4756 | btrfs_inc_nlink(inode); |
4757 | inode->i_ctime = CURRENT_TIME; | |
7de9c6ee | 4758 | ihold(inode); |
aec7477b | 4759 | |
a1b075d2 | 4760 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4761 | |
a5719521 | 4762 | if (err) { |
54aa1f4d | 4763 | drop_inode = 1; |
a5719521 | 4764 | } else { |
6a912213 | 4765 | struct dentry *parent = dget_parent(dentry); |
a5719521 YZ |
4766 | err = btrfs_update_inode(trans, root, inode); |
4767 | BUG_ON(err); | |
6a912213 JB |
4768 | btrfs_log_new_name(trans, inode, NULL, parent); |
4769 | dput(parent); | |
a5719521 | 4770 | } |
39279cc3 | 4771 | |
d3c2fdcf | 4772 | nr = trans->blocks_used; |
ab78c84d | 4773 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4774 | fail: |
39279cc3 CM |
4775 | if (drop_inode) { |
4776 | inode_dec_link_count(inode); | |
4777 | iput(inode); | |
4778 | } | |
d3c2fdcf | 4779 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4780 | return err; |
4781 | } | |
4782 | ||
39279cc3 CM |
4783 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4784 | { | |
b9d86667 | 4785 | struct inode *inode = NULL; |
39279cc3 CM |
4786 | struct btrfs_trans_handle *trans; |
4787 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4788 | int err = 0; | |
4789 | int drop_on_err = 0; | |
b9d86667 | 4790 | u64 objectid = 0; |
00e4e6b3 | 4791 | u64 index = 0; |
d3c2fdcf | 4792 | unsigned long nr = 1; |
39279cc3 | 4793 | |
9ed74f2d JB |
4794 | /* |
4795 | * 2 items for inode and ref | |
4796 | * 2 items for dir items | |
4797 | * 1 for xattr if selinux is on | |
4798 | */ | |
a22285a6 YZ |
4799 | trans = btrfs_start_transaction(root, 5); |
4800 | if (IS_ERR(trans)) | |
4801 | return PTR_ERR(trans); | |
39279cc3 | 4802 | |
581bb050 LZ |
4803 | err = btrfs_find_free_ino(root, &objectid); |
4804 | if (err) | |
4805 | goto out_fail; | |
4806 | ||
aec7477b | 4807 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4808 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4809 | S_IFDIR | mode, &index); |
39279cc3 CM |
4810 | if (IS_ERR(inode)) { |
4811 | err = PTR_ERR(inode); | |
4812 | goto out_fail; | |
4813 | } | |
5f39d397 | 4814 | |
39279cc3 | 4815 | drop_on_err = 1; |
33268eaf | 4816 | |
2a7dba39 | 4817 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4818 | if (err) |
4819 | goto out_fail; | |
4820 | ||
39279cc3 CM |
4821 | inode->i_op = &btrfs_dir_inode_operations; |
4822 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 4823 | |
dbe674a9 | 4824 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4825 | err = btrfs_update_inode(trans, root, inode); |
4826 | if (err) | |
4827 | goto out_fail; | |
5f39d397 | 4828 | |
a1b075d2 JB |
4829 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4830 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4831 | if (err) |
4832 | goto out_fail; | |
5f39d397 | 4833 | |
39279cc3 CM |
4834 | d_instantiate(dentry, inode); |
4835 | drop_on_err = 0; | |
39279cc3 CM |
4836 | |
4837 | out_fail: | |
d3c2fdcf | 4838 | nr = trans->blocks_used; |
ab78c84d | 4839 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4840 | if (drop_on_err) |
4841 | iput(inode); | |
d3c2fdcf | 4842 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4843 | return err; |
4844 | } | |
4845 | ||
d352ac68 CM |
4846 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4847 | * and an extent that you want to insert, deal with overlap and insert | |
4848 | * the new extent into the tree. | |
4849 | */ | |
3b951516 CM |
4850 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4851 | struct extent_map *existing, | |
e6dcd2dc CM |
4852 | struct extent_map *em, |
4853 | u64 map_start, u64 map_len) | |
3b951516 CM |
4854 | { |
4855 | u64 start_diff; | |
3b951516 | 4856 | |
e6dcd2dc CM |
4857 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4858 | start_diff = map_start - em->start; | |
4859 | em->start = map_start; | |
4860 | em->len = map_len; | |
c8b97818 CM |
4861 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4862 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4863 | em->block_start += start_diff; |
c8b97818 CM |
4864 | em->block_len -= start_diff; |
4865 | } | |
e6dcd2dc | 4866 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4867 | } |
4868 | ||
c8b97818 CM |
4869 | static noinline int uncompress_inline(struct btrfs_path *path, |
4870 | struct inode *inode, struct page *page, | |
4871 | size_t pg_offset, u64 extent_offset, | |
4872 | struct btrfs_file_extent_item *item) | |
4873 | { | |
4874 | int ret; | |
4875 | struct extent_buffer *leaf = path->nodes[0]; | |
4876 | char *tmp; | |
4877 | size_t max_size; | |
4878 | unsigned long inline_size; | |
4879 | unsigned long ptr; | |
261507a0 | 4880 | int compress_type; |
c8b97818 CM |
4881 | |
4882 | WARN_ON(pg_offset != 0); | |
261507a0 | 4883 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
4884 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
4885 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4886 | btrfs_item_nr(leaf, path->slots[0])); | |
4887 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
4888 | if (!tmp) |
4889 | return -ENOMEM; | |
c8b97818 CM |
4890 | ptr = btrfs_file_extent_inline_start(item); |
4891 | ||
4892 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4893 | ||
5b050f04 | 4894 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
4895 | ret = btrfs_decompress(compress_type, tmp, page, |
4896 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
4897 | if (ret) { |
4898 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4899 | unsigned long copy_size = min_t(u64, | |
4900 | PAGE_CACHE_SIZE - pg_offset, | |
4901 | max_size - extent_offset); | |
4902 | memset(kaddr + pg_offset, 0, copy_size); | |
4903 | kunmap_atomic(kaddr, KM_USER0); | |
4904 | } | |
4905 | kfree(tmp); | |
4906 | return 0; | |
4907 | } | |
4908 | ||
d352ac68 CM |
4909 | /* |
4910 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4911 | * the ugly parts come from merging extents from the disk with the in-ram |
4912 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4913 | * where the in-ram extents might be locked pending data=ordered completion. |
4914 | * | |
4915 | * This also copies inline extents directly into the page. | |
4916 | */ | |
d397712b | 4917 | |
a52d9a80 | 4918 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4919 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4920 | int create) |
4921 | { | |
4922 | int ret; | |
4923 | int err = 0; | |
db94535d | 4924 | u64 bytenr; |
a52d9a80 CM |
4925 | u64 extent_start = 0; |
4926 | u64 extent_end = 0; | |
33345d01 | 4927 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 4928 | u32 found_type; |
f421950f | 4929 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4930 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4931 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4932 | struct extent_buffer *leaf; |
4933 | struct btrfs_key found_key; | |
a52d9a80 CM |
4934 | struct extent_map *em = NULL; |
4935 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4936 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4937 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 4938 | int compress_type; |
a52d9a80 | 4939 | |
a52d9a80 | 4940 | again: |
890871be | 4941 | read_lock(&em_tree->lock); |
d1310b2e | 4942 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4943 | if (em) |
4944 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4945 | read_unlock(&em_tree->lock); |
d1310b2e | 4946 | |
a52d9a80 | 4947 | if (em) { |
e1c4b745 CM |
4948 | if (em->start > start || em->start + em->len <= start) |
4949 | free_extent_map(em); | |
4950 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4951 | free_extent_map(em); |
4952 | else | |
4953 | goto out; | |
a52d9a80 | 4954 | } |
172ddd60 | 4955 | em = alloc_extent_map(); |
a52d9a80 | 4956 | if (!em) { |
d1310b2e CM |
4957 | err = -ENOMEM; |
4958 | goto out; | |
a52d9a80 | 4959 | } |
e6dcd2dc | 4960 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4961 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4962 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4963 | em->len = (u64)-1; |
c8b97818 | 4964 | em->block_len = (u64)-1; |
f421950f CM |
4965 | |
4966 | if (!path) { | |
4967 | path = btrfs_alloc_path(); | |
026fd317 JB |
4968 | if (!path) { |
4969 | err = -ENOMEM; | |
4970 | goto out; | |
4971 | } | |
4972 | /* | |
4973 | * Chances are we'll be called again, so go ahead and do | |
4974 | * readahead | |
4975 | */ | |
4976 | path->reada = 1; | |
f421950f CM |
4977 | } |
4978 | ||
179e29e4 CM |
4979 | ret = btrfs_lookup_file_extent(trans, root, path, |
4980 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4981 | if (ret < 0) { |
4982 | err = ret; | |
4983 | goto out; | |
4984 | } | |
4985 | ||
4986 | if (ret != 0) { | |
4987 | if (path->slots[0] == 0) | |
4988 | goto not_found; | |
4989 | path->slots[0]--; | |
4990 | } | |
4991 | ||
5f39d397 CM |
4992 | leaf = path->nodes[0]; |
4993 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4994 | struct btrfs_file_extent_item); |
a52d9a80 | 4995 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4996 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4997 | found_type = btrfs_key_type(&found_key); | |
4998 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4999 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5000 | goto not_found; | |
5001 | } | |
5002 | ||
5f39d397 CM |
5003 | found_type = btrfs_file_extent_type(leaf, item); |
5004 | extent_start = found_key.offset; | |
261507a0 | 5005 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5006 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5007 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5008 | extent_end = extent_start + |
db94535d | 5009 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5010 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5011 | size_t size; | |
5012 | size = btrfs_file_extent_inline_len(leaf, item); | |
5013 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5014 | ~((u64)root->sectorsize - 1); | |
5015 | } | |
5016 | ||
5017 | if (start >= extent_end) { | |
5018 | path->slots[0]++; | |
5019 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5020 | ret = btrfs_next_leaf(root, path); | |
5021 | if (ret < 0) { | |
5022 | err = ret; | |
5023 | goto out; | |
a52d9a80 | 5024 | } |
9036c102 YZ |
5025 | if (ret > 0) |
5026 | goto not_found; | |
5027 | leaf = path->nodes[0]; | |
a52d9a80 | 5028 | } |
9036c102 YZ |
5029 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5030 | if (found_key.objectid != objectid || | |
5031 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5032 | goto not_found; | |
5033 | if (start + len <= found_key.offset) | |
5034 | goto not_found; | |
5035 | em->start = start; | |
5036 | em->len = found_key.offset - start; | |
5037 | goto not_found_em; | |
5038 | } | |
5039 | ||
d899e052 YZ |
5040 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5041 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5042 | em->start = extent_start; |
5043 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5044 | em->orig_start = extent_start - |
5045 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5046 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5047 | if (bytenr == 0) { | |
5f39d397 | 5048 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5049 | goto insert; |
5050 | } | |
261507a0 | 5051 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5052 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5053 | em->compress_type = compress_type; |
c8b97818 CM |
5054 | em->block_start = bytenr; |
5055 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5056 | item); | |
5057 | } else { | |
5058 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5059 | em->block_start = bytenr; | |
5060 | em->block_len = em->len; | |
d899e052 YZ |
5061 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5062 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5063 | } |
a52d9a80 CM |
5064 | goto insert; |
5065 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5066 | unsigned long ptr; |
a52d9a80 | 5067 | char *map; |
3326d1b0 CM |
5068 | size_t size; |
5069 | size_t extent_offset; | |
5070 | size_t copy_size; | |
a52d9a80 | 5071 | |
689f9346 | 5072 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5073 | if (!page || create) { |
689f9346 | 5074 | em->start = extent_start; |
9036c102 | 5075 | em->len = extent_end - extent_start; |
689f9346 Y |
5076 | goto out; |
5077 | } | |
5f39d397 | 5078 | |
9036c102 YZ |
5079 | size = btrfs_file_extent_inline_len(leaf, item); |
5080 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5081 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5082 | size - extent_offset); |
3326d1b0 | 5083 | em->start = extent_start + extent_offset; |
70dec807 CM |
5084 | em->len = (copy_size + root->sectorsize - 1) & |
5085 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5086 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5087 | if (compress_type) { |
c8b97818 | 5088 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5089 | em->compress_type = compress_type; |
5090 | } | |
689f9346 | 5091 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5092 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5093 | if (btrfs_file_extent_compression(leaf, item) != |
5094 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5095 | ret = uncompress_inline(path, inode, page, |
5096 | pg_offset, | |
5097 | extent_offset, item); | |
5098 | BUG_ON(ret); | |
5099 | } else { | |
5100 | map = kmap(page); | |
5101 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5102 | copy_size); | |
93c82d57 CM |
5103 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5104 | memset(map + pg_offset + copy_size, 0, | |
5105 | PAGE_CACHE_SIZE - pg_offset - | |
5106 | copy_size); | |
5107 | } | |
c8b97818 CM |
5108 | kunmap(page); |
5109 | } | |
179e29e4 CM |
5110 | flush_dcache_page(page); |
5111 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 5112 | WARN_ON(1); |
179e29e4 CM |
5113 | if (!trans) { |
5114 | kunmap(page); | |
5115 | free_extent_map(em); | |
5116 | em = NULL; | |
ff5714cc | 5117 | |
b3b4aa74 | 5118 | btrfs_release_path(path); |
7a7eaa40 | 5119 | trans = btrfs_join_transaction(root); |
ff5714cc | 5120 | |
3612b495 TI |
5121 | if (IS_ERR(trans)) |
5122 | return ERR_CAST(trans); | |
179e29e4 CM |
5123 | goto again; |
5124 | } | |
c8b97818 | 5125 | map = kmap(page); |
70dec807 | 5126 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5127 | copy_size); |
c8b97818 | 5128 | kunmap(page); |
179e29e4 | 5129 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5130 | } |
d1310b2e | 5131 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5132 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5133 | goto insert; |
5134 | } else { | |
d397712b | 5135 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5136 | WARN_ON(1); |
5137 | } | |
5138 | not_found: | |
5139 | em->start = start; | |
d1310b2e | 5140 | em->len = len; |
a52d9a80 | 5141 | not_found_em: |
5f39d397 | 5142 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5143 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5144 | insert: |
b3b4aa74 | 5145 | btrfs_release_path(path); |
d1310b2e | 5146 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5147 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5148 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5149 | (unsigned long long)em->len, | |
5150 | (unsigned long long)start, | |
5151 | (unsigned long long)len); | |
a52d9a80 CM |
5152 | err = -EIO; |
5153 | goto out; | |
5154 | } | |
d1310b2e CM |
5155 | |
5156 | err = 0; | |
890871be | 5157 | write_lock(&em_tree->lock); |
a52d9a80 | 5158 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5159 | /* it is possible that someone inserted the extent into the tree |
5160 | * while we had the lock dropped. It is also possible that | |
5161 | * an overlapping map exists in the tree | |
5162 | */ | |
a52d9a80 | 5163 | if (ret == -EEXIST) { |
3b951516 | 5164 | struct extent_map *existing; |
e6dcd2dc CM |
5165 | |
5166 | ret = 0; | |
5167 | ||
3b951516 | 5168 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5169 | if (existing && (existing->start > start || |
5170 | existing->start + existing->len <= start)) { | |
5171 | free_extent_map(existing); | |
5172 | existing = NULL; | |
5173 | } | |
3b951516 CM |
5174 | if (!existing) { |
5175 | existing = lookup_extent_mapping(em_tree, em->start, | |
5176 | em->len); | |
5177 | if (existing) { | |
5178 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5179 | em, start, |
5180 | root->sectorsize); | |
3b951516 CM |
5181 | free_extent_map(existing); |
5182 | if (err) { | |
5183 | free_extent_map(em); | |
5184 | em = NULL; | |
5185 | } | |
5186 | } else { | |
5187 | err = -EIO; | |
3b951516 CM |
5188 | free_extent_map(em); |
5189 | em = NULL; | |
5190 | } | |
5191 | } else { | |
5192 | free_extent_map(em); | |
5193 | em = existing; | |
e6dcd2dc | 5194 | err = 0; |
a52d9a80 | 5195 | } |
a52d9a80 | 5196 | } |
890871be | 5197 | write_unlock(&em_tree->lock); |
a52d9a80 | 5198 | out: |
1abe9b8a | 5199 | |
5200 | trace_btrfs_get_extent(root, em); | |
5201 | ||
f421950f CM |
5202 | if (path) |
5203 | btrfs_free_path(path); | |
a52d9a80 CM |
5204 | if (trans) { |
5205 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5206 | if (!err) |
a52d9a80 CM |
5207 | err = ret; |
5208 | } | |
a52d9a80 CM |
5209 | if (err) { |
5210 | free_extent_map(em); | |
a52d9a80 CM |
5211 | return ERR_PTR(err); |
5212 | } | |
5213 | return em; | |
5214 | } | |
5215 | ||
ec29ed5b CM |
5216 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5217 | size_t pg_offset, u64 start, u64 len, | |
5218 | int create) | |
5219 | { | |
5220 | struct extent_map *em; | |
5221 | struct extent_map *hole_em = NULL; | |
5222 | u64 range_start = start; | |
5223 | u64 end; | |
5224 | u64 found; | |
5225 | u64 found_end; | |
5226 | int err = 0; | |
5227 | ||
5228 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5229 | if (IS_ERR(em)) | |
5230 | return em; | |
5231 | if (em) { | |
5232 | /* | |
5233 | * if our em maps to a hole, there might | |
5234 | * actually be delalloc bytes behind it | |
5235 | */ | |
5236 | if (em->block_start != EXTENT_MAP_HOLE) | |
5237 | return em; | |
5238 | else | |
5239 | hole_em = em; | |
5240 | } | |
5241 | ||
5242 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5243 | end = start + len; | |
5244 | if (end < start) | |
5245 | end = (u64)-1; | |
5246 | else | |
5247 | end -= 1; | |
5248 | ||
5249 | em = NULL; | |
5250 | ||
5251 | /* ok, we didn't find anything, lets look for delalloc */ | |
5252 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5253 | end, len, EXTENT_DELALLOC, 1); | |
5254 | found_end = range_start + found; | |
5255 | if (found_end < range_start) | |
5256 | found_end = (u64)-1; | |
5257 | ||
5258 | /* | |
5259 | * we didn't find anything useful, return | |
5260 | * the original results from get_extent() | |
5261 | */ | |
5262 | if (range_start > end || found_end <= start) { | |
5263 | em = hole_em; | |
5264 | hole_em = NULL; | |
5265 | goto out; | |
5266 | } | |
5267 | ||
5268 | /* adjust the range_start to make sure it doesn't | |
5269 | * go backwards from the start they passed in | |
5270 | */ | |
5271 | range_start = max(start,range_start); | |
5272 | found = found_end - range_start; | |
5273 | ||
5274 | if (found > 0) { | |
5275 | u64 hole_start = start; | |
5276 | u64 hole_len = len; | |
5277 | ||
172ddd60 | 5278 | em = alloc_extent_map(); |
ec29ed5b CM |
5279 | if (!em) { |
5280 | err = -ENOMEM; | |
5281 | goto out; | |
5282 | } | |
5283 | /* | |
5284 | * when btrfs_get_extent can't find anything it | |
5285 | * returns one huge hole | |
5286 | * | |
5287 | * make sure what it found really fits our range, and | |
5288 | * adjust to make sure it is based on the start from | |
5289 | * the caller | |
5290 | */ | |
5291 | if (hole_em) { | |
5292 | u64 calc_end = extent_map_end(hole_em); | |
5293 | ||
5294 | if (calc_end <= start || (hole_em->start > end)) { | |
5295 | free_extent_map(hole_em); | |
5296 | hole_em = NULL; | |
5297 | } else { | |
5298 | hole_start = max(hole_em->start, start); | |
5299 | hole_len = calc_end - hole_start; | |
5300 | } | |
5301 | } | |
5302 | em->bdev = NULL; | |
5303 | if (hole_em && range_start > hole_start) { | |
5304 | /* our hole starts before our delalloc, so we | |
5305 | * have to return just the parts of the hole | |
5306 | * that go until the delalloc starts | |
5307 | */ | |
5308 | em->len = min(hole_len, | |
5309 | range_start - hole_start); | |
5310 | em->start = hole_start; | |
5311 | em->orig_start = hole_start; | |
5312 | /* | |
5313 | * don't adjust block start at all, | |
5314 | * it is fixed at EXTENT_MAP_HOLE | |
5315 | */ | |
5316 | em->block_start = hole_em->block_start; | |
5317 | em->block_len = hole_len; | |
5318 | } else { | |
5319 | em->start = range_start; | |
5320 | em->len = found; | |
5321 | em->orig_start = range_start; | |
5322 | em->block_start = EXTENT_MAP_DELALLOC; | |
5323 | em->block_len = found; | |
5324 | } | |
5325 | } else if (hole_em) { | |
5326 | return hole_em; | |
5327 | } | |
5328 | out: | |
5329 | ||
5330 | free_extent_map(hole_em); | |
5331 | if (err) { | |
5332 | free_extent_map(em); | |
5333 | return ERR_PTR(err); | |
5334 | } | |
5335 | return em; | |
5336 | } | |
5337 | ||
4b46fce2 | 5338 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5339 | struct extent_map *em, |
4b46fce2 JB |
5340 | u64 start, u64 len) |
5341 | { | |
5342 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5343 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5344 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5345 | struct btrfs_key ins; | |
5346 | u64 alloc_hint; | |
5347 | int ret; | |
16d299ac | 5348 | bool insert = false; |
4b46fce2 | 5349 | |
16d299ac JB |
5350 | /* |
5351 | * Ok if the extent map we looked up is a hole and is for the exact | |
5352 | * range we want, there is no reason to allocate a new one, however if | |
5353 | * it is not right then we need to free this one and drop the cache for | |
5354 | * our range. | |
5355 | */ | |
5356 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5357 | em->len != len) { | |
5358 | free_extent_map(em); | |
5359 | em = NULL; | |
5360 | insert = true; | |
5361 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5362 | } | |
4b46fce2 | 5363 | |
7a7eaa40 | 5364 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5365 | if (IS_ERR(trans)) |
5366 | return ERR_CAST(trans); | |
4b46fce2 | 5367 | |
4cb5300b CM |
5368 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5369 | btrfs_add_inode_defrag(trans, inode); | |
5370 | ||
4b46fce2 JB |
5371 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5372 | ||
5373 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5374 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5375 | alloc_hint, (u64)-1, &ins, 1); | |
5376 | if (ret) { | |
5377 | em = ERR_PTR(ret); | |
5378 | goto out; | |
5379 | } | |
5380 | ||
4b46fce2 | 5381 | if (!em) { |
172ddd60 | 5382 | em = alloc_extent_map(); |
16d299ac JB |
5383 | if (!em) { |
5384 | em = ERR_PTR(-ENOMEM); | |
5385 | goto out; | |
5386 | } | |
4b46fce2 JB |
5387 | } |
5388 | ||
5389 | em->start = start; | |
5390 | em->orig_start = em->start; | |
5391 | em->len = ins.offset; | |
5392 | ||
5393 | em->block_start = ins.objectid; | |
5394 | em->block_len = ins.offset; | |
5395 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5396 | |
5397 | /* | |
5398 | * We need to do this because if we're using the original em we searched | |
5399 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5400 | */ | |
5401 | em->flags = 0; | |
4b46fce2 JB |
5402 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5403 | ||
16d299ac | 5404 | while (insert) { |
4b46fce2 JB |
5405 | write_lock(&em_tree->lock); |
5406 | ret = add_extent_mapping(em_tree, em); | |
5407 | write_unlock(&em_tree->lock); | |
5408 | if (ret != -EEXIST) | |
5409 | break; | |
5410 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5411 | } | |
5412 | ||
5413 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5414 | ins.offset, ins.offset, 0); | |
5415 | if (ret) { | |
5416 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5417 | em = ERR_PTR(ret); | |
5418 | } | |
5419 | out: | |
5420 | btrfs_end_transaction(trans, root); | |
5421 | return em; | |
5422 | } | |
5423 | ||
46bfbb5c CM |
5424 | /* |
5425 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5426 | * block must be cow'd | |
5427 | */ | |
5428 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5429 | struct inode *inode, u64 offset, u64 len) | |
5430 | { | |
5431 | struct btrfs_path *path; | |
5432 | int ret; | |
5433 | struct extent_buffer *leaf; | |
5434 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5435 | struct btrfs_file_extent_item *fi; | |
5436 | struct btrfs_key key; | |
5437 | u64 disk_bytenr; | |
5438 | u64 backref_offset; | |
5439 | u64 extent_end; | |
5440 | u64 num_bytes; | |
5441 | int slot; | |
5442 | int found_type; | |
5443 | ||
5444 | path = btrfs_alloc_path(); | |
5445 | if (!path) | |
5446 | return -ENOMEM; | |
5447 | ||
33345d01 | 5448 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5449 | offset, 0); |
5450 | if (ret < 0) | |
5451 | goto out; | |
5452 | ||
5453 | slot = path->slots[0]; | |
5454 | if (ret == 1) { | |
5455 | if (slot == 0) { | |
5456 | /* can't find the item, must cow */ | |
5457 | ret = 0; | |
5458 | goto out; | |
5459 | } | |
5460 | slot--; | |
5461 | } | |
5462 | ret = 0; | |
5463 | leaf = path->nodes[0]; | |
5464 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5465 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5466 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5467 | /* not our file or wrong item type, must cow */ | |
5468 | goto out; | |
5469 | } | |
5470 | ||
5471 | if (key.offset > offset) { | |
5472 | /* Wrong offset, must cow */ | |
5473 | goto out; | |
5474 | } | |
5475 | ||
5476 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5477 | found_type = btrfs_file_extent_type(leaf, fi); | |
5478 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5479 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5480 | /* not a regular extent, must cow */ | |
5481 | goto out; | |
5482 | } | |
5483 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5484 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5485 | ||
5486 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5487 | if (extent_end < offset + len) { | |
5488 | /* extent doesn't include our full range, must cow */ | |
5489 | goto out; | |
5490 | } | |
5491 | ||
5492 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5493 | goto out; | |
5494 | ||
5495 | /* | |
5496 | * look for other files referencing this extent, if we | |
5497 | * find any we must cow | |
5498 | */ | |
33345d01 | 5499 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5500 | key.offset - backref_offset, disk_bytenr)) |
5501 | goto out; | |
5502 | ||
5503 | /* | |
5504 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5505 | * in this extent we are about to write. If there | |
5506 | * are any csums in that range we have to cow in order | |
5507 | * to keep the csums correct | |
5508 | */ | |
5509 | disk_bytenr += backref_offset; | |
5510 | disk_bytenr += offset - key.offset; | |
5511 | num_bytes = min(offset + len, extent_end) - offset; | |
5512 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5513 | goto out; | |
5514 | /* | |
5515 | * all of the above have passed, it is safe to overwrite this extent | |
5516 | * without cow | |
5517 | */ | |
5518 | ret = 1; | |
5519 | out: | |
5520 | btrfs_free_path(path); | |
5521 | return ret; | |
5522 | } | |
5523 | ||
4b46fce2 JB |
5524 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5525 | struct buffer_head *bh_result, int create) | |
5526 | { | |
5527 | struct extent_map *em; | |
5528 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5529 | u64 start = iblock << inode->i_blkbits; | |
5530 | u64 len = bh_result->b_size; | |
46bfbb5c | 5531 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5532 | |
5533 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5534 | if (IS_ERR(em)) | |
5535 | return PTR_ERR(em); | |
5536 | ||
5537 | /* | |
5538 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5539 | * io. INLINE is special, and we could probably kludge it in here, but | |
5540 | * it's still buffered so for safety lets just fall back to the generic | |
5541 | * buffered path. | |
5542 | * | |
5543 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5544 | * decompress it, so there will be buffering required no matter what we | |
5545 | * do, so go ahead and fallback to buffered. | |
5546 | * | |
5547 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5548 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5549 | * the generic code. | |
5550 | */ | |
5551 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5552 | em->block_start == EXTENT_MAP_INLINE) { | |
5553 | free_extent_map(em); | |
5554 | return -ENOTBLK; | |
5555 | } | |
5556 | ||
5557 | /* Just a good old fashioned hole, return */ | |
5558 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5559 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5560 | free_extent_map(em); | |
5561 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5562 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5563 | start + root->sectorsize - 1, GFP_NOFS); | |
5564 | return 0; | |
5565 | } | |
5566 | ||
5567 | /* | |
5568 | * We don't allocate a new extent in the following cases | |
5569 | * | |
5570 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5571 | * existing extent. | |
5572 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5573 | * just use the extent. | |
5574 | * | |
5575 | */ | |
46bfbb5c CM |
5576 | if (!create) { |
5577 | len = em->len - (start - em->start); | |
4b46fce2 | 5578 | goto map; |
46bfbb5c | 5579 | } |
4b46fce2 JB |
5580 | |
5581 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5582 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5583 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5584 | int type; |
5585 | int ret; | |
46bfbb5c | 5586 | u64 block_start; |
4b46fce2 JB |
5587 | |
5588 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5589 | type = BTRFS_ORDERED_PREALLOC; | |
5590 | else | |
5591 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5592 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5593 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5594 | |
5595 | /* | |
5596 | * we're not going to log anything, but we do need | |
5597 | * to make sure the current transaction stays open | |
5598 | * while we look for nocow cross refs | |
5599 | */ | |
7a7eaa40 | 5600 | trans = btrfs_join_transaction(root); |
3612b495 | 5601 | if (IS_ERR(trans)) |
46bfbb5c CM |
5602 | goto must_cow; |
5603 | ||
5604 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5605 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5606 | block_start, len, len, type); | |
5607 | btrfs_end_transaction(trans, root); | |
5608 | if (ret) { | |
5609 | free_extent_map(em); | |
5610 | return ret; | |
5611 | } | |
5612 | goto unlock; | |
4b46fce2 | 5613 | } |
46bfbb5c | 5614 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5615 | } |
46bfbb5c CM |
5616 | must_cow: |
5617 | /* | |
5618 | * this will cow the extent, reset the len in case we changed | |
5619 | * it above | |
5620 | */ | |
5621 | len = bh_result->b_size; | |
16d299ac | 5622 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5623 | if (IS_ERR(em)) |
5624 | return PTR_ERR(em); | |
5625 | len = min(len, em->len - (start - em->start)); | |
5626 | unlock: | |
4845e44f CM |
5627 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5628 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5629 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5630 | map: |
5631 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5632 | inode->i_blkbits; | |
46bfbb5c | 5633 | bh_result->b_size = len; |
4b46fce2 JB |
5634 | bh_result->b_bdev = em->bdev; |
5635 | set_buffer_mapped(bh_result); | |
5636 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5637 | set_buffer_new(bh_result); | |
5638 | ||
5639 | free_extent_map(em); | |
5640 | ||
5641 | return 0; | |
5642 | } | |
5643 | ||
5644 | struct btrfs_dio_private { | |
5645 | struct inode *inode; | |
5646 | u64 logical_offset; | |
5647 | u64 disk_bytenr; | |
5648 | u64 bytes; | |
5649 | u32 *csums; | |
5650 | void *private; | |
e65e1535 MX |
5651 | |
5652 | /* number of bios pending for this dio */ | |
5653 | atomic_t pending_bios; | |
5654 | ||
5655 | /* IO errors */ | |
5656 | int errors; | |
5657 | ||
5658 | struct bio *orig_bio; | |
4b46fce2 JB |
5659 | }; |
5660 | ||
5661 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5662 | { | |
e65e1535 | 5663 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5664 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5665 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5666 | struct inode *inode = dip->inode; |
5667 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5668 | u64 start; | |
5669 | u32 *private = dip->csums; | |
5670 | ||
5671 | start = dip->logical_offset; | |
5672 | do { | |
5673 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5674 | struct page *page = bvec->bv_page; | |
5675 | char *kaddr; | |
5676 | u32 csum = ~(u32)0; | |
5677 | unsigned long flags; | |
5678 | ||
5679 | local_irq_save(flags); | |
5680 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5681 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5682 | csum, bvec->bv_len); | |
5683 | btrfs_csum_final(csum, (char *)&csum); | |
5684 | kunmap_atomic(kaddr, KM_IRQ0); | |
5685 | local_irq_restore(flags); | |
5686 | ||
5687 | flush_dcache_page(bvec->bv_page); | |
5688 | if (csum != *private) { | |
33345d01 | 5689 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 5690 | " %llu csum %u private %u\n", |
33345d01 LZ |
5691 | (unsigned long long)btrfs_ino(inode), |
5692 | (unsigned long long)start, | |
4b46fce2 JB |
5693 | csum, *private); |
5694 | err = -EIO; | |
5695 | } | |
5696 | } | |
5697 | ||
5698 | start += bvec->bv_len; | |
5699 | private++; | |
5700 | bvec++; | |
5701 | } while (bvec <= bvec_end); | |
5702 | ||
5703 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5704 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5705 | bio->bi_private = dip->private; | |
5706 | ||
5707 | kfree(dip->csums); | |
5708 | kfree(dip); | |
c0da7aa1 JB |
5709 | |
5710 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5711 | if (err) | |
5712 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5713 | dio_end_io(bio, err); |
5714 | } | |
5715 | ||
5716 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5717 | { | |
5718 | struct btrfs_dio_private *dip = bio->bi_private; | |
5719 | struct inode *inode = dip->inode; | |
5720 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5721 | struct btrfs_trans_handle *trans; | |
5722 | struct btrfs_ordered_extent *ordered = NULL; | |
5723 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5724 | u64 ordered_offset = dip->logical_offset; |
5725 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5726 | int ret; |
5727 | ||
5728 | if (err) | |
5729 | goto out_done; | |
163cf09c CM |
5730 | again: |
5731 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5732 | &ordered_offset, | |
5733 | ordered_bytes); | |
4b46fce2 | 5734 | if (!ret) |
163cf09c | 5735 | goto out_test; |
4b46fce2 JB |
5736 | |
5737 | BUG_ON(!ordered); | |
5738 | ||
7a7eaa40 | 5739 | trans = btrfs_join_transaction(root); |
3612b495 | 5740 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5741 | err = -ENOMEM; |
5742 | goto out; | |
5743 | } | |
5744 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5745 | ||
5746 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5747 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5748 | if (!ret) | |
5749 | ret = btrfs_update_inode(trans, root, inode); | |
5750 | err = ret; | |
5751 | goto out; | |
5752 | } | |
5753 | ||
5754 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5755 | ordered->file_offset + ordered->len - 1, 0, | |
5756 | &cached_state, GFP_NOFS); | |
5757 | ||
5758 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5759 | ret = btrfs_mark_extent_written(trans, inode, | |
5760 | ordered->file_offset, | |
5761 | ordered->file_offset + | |
5762 | ordered->len); | |
5763 | if (ret) { | |
5764 | err = ret; | |
5765 | goto out_unlock; | |
5766 | } | |
5767 | } else { | |
5768 | ret = insert_reserved_file_extent(trans, inode, | |
5769 | ordered->file_offset, | |
5770 | ordered->start, | |
5771 | ordered->disk_len, | |
5772 | ordered->len, | |
5773 | ordered->len, | |
5774 | 0, 0, 0, | |
5775 | BTRFS_FILE_EXTENT_REG); | |
5776 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5777 | ordered->file_offset, ordered->len); | |
5778 | if (ret) { | |
5779 | err = ret; | |
5780 | WARN_ON(1); | |
5781 | goto out_unlock; | |
5782 | } | |
5783 | } | |
5784 | ||
5785 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
1ef30be1 JB |
5786 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); |
5787 | if (!ret) | |
5788 | btrfs_update_inode(trans, root, inode); | |
5789 | ret = 0; | |
4b46fce2 JB |
5790 | out_unlock: |
5791 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5792 | ordered->file_offset + ordered->len - 1, | |
5793 | &cached_state, GFP_NOFS); | |
5794 | out: | |
5795 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5796 | btrfs_end_transaction(trans, root); | |
163cf09c | 5797 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5798 | btrfs_put_ordered_extent(ordered); |
5799 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5800 | |
5801 | out_test: | |
5802 | /* | |
5803 | * our bio might span multiple ordered extents. If we haven't | |
5804 | * completed the accounting for the whole dio, go back and try again | |
5805 | */ | |
5806 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5807 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5808 | ordered_offset; | |
5809 | goto again; | |
5810 | } | |
4b46fce2 JB |
5811 | out_done: |
5812 | bio->bi_private = dip->private; | |
5813 | ||
5814 | kfree(dip->csums); | |
5815 | kfree(dip); | |
c0da7aa1 JB |
5816 | |
5817 | /* If we had an error make sure to clear the uptodate flag */ | |
5818 | if (err) | |
5819 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5820 | dio_end_io(bio, err); |
5821 | } | |
5822 | ||
eaf25d93 CM |
5823 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5824 | struct bio *bio, int mirror_num, | |
5825 | unsigned long bio_flags, u64 offset) | |
5826 | { | |
5827 | int ret; | |
5828 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5829 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5830 | BUG_ON(ret); | |
5831 | return 0; | |
5832 | } | |
5833 | ||
e65e1535 MX |
5834 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5835 | { | |
5836 | struct btrfs_dio_private *dip = bio->bi_private; | |
5837 | ||
5838 | if (err) { | |
33345d01 | 5839 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 5840 | "sector %#Lx len %u err no %d\n", |
33345d01 | 5841 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 5842 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
5843 | dip->errors = 1; |
5844 | ||
5845 | /* | |
5846 | * before atomic variable goto zero, we must make sure | |
5847 | * dip->errors is perceived to be set. | |
5848 | */ | |
5849 | smp_mb__before_atomic_dec(); | |
5850 | } | |
5851 | ||
5852 | /* if there are more bios still pending for this dio, just exit */ | |
5853 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5854 | goto out; | |
5855 | ||
5856 | if (dip->errors) | |
5857 | bio_io_error(dip->orig_bio); | |
5858 | else { | |
5859 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5860 | bio_endio(dip->orig_bio, 0); | |
5861 | } | |
5862 | out: | |
5863 | bio_put(bio); | |
5864 | } | |
5865 | ||
5866 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5867 | u64 first_sector, gfp_t gfp_flags) | |
5868 | { | |
5869 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5870 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5871 | } | |
5872 | ||
5873 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5874 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 5875 | u32 *csums, int async_submit) |
e65e1535 MX |
5876 | { |
5877 | int write = rw & REQ_WRITE; | |
5878 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5879 | int ret; | |
5880 | ||
5881 | bio_get(bio); | |
5882 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5883 | if (ret) | |
5884 | goto err; | |
5885 | ||
1ae39938 JB |
5886 | if (skip_sum) |
5887 | goto map; | |
5888 | ||
5889 | if (write && async_submit) { | |
e65e1535 MX |
5890 | ret = btrfs_wq_submit_bio(root->fs_info, |
5891 | inode, rw, bio, 0, 0, | |
5892 | file_offset, | |
5893 | __btrfs_submit_bio_start_direct_io, | |
5894 | __btrfs_submit_bio_done); | |
5895 | goto err; | |
1ae39938 JB |
5896 | } else if (write) { |
5897 | /* | |
5898 | * If we aren't doing async submit, calculate the csum of the | |
5899 | * bio now. | |
5900 | */ | |
5901 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
5902 | if (ret) | |
5903 | goto err; | |
c2db1073 TI |
5904 | } else if (!skip_sum) { |
5905 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 5906 | file_offset, csums); |
c2db1073 TI |
5907 | if (ret) |
5908 | goto err; | |
5909 | } | |
e65e1535 | 5910 | |
1ae39938 JB |
5911 | map: |
5912 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
5913 | err: |
5914 | bio_put(bio); | |
5915 | return ret; | |
5916 | } | |
5917 | ||
5918 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
5919 | int skip_sum) | |
5920 | { | |
5921 | struct inode *inode = dip->inode; | |
5922 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5923 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5924 | struct bio *bio; | |
5925 | struct bio *orig_bio = dip->orig_bio; | |
5926 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
5927 | u64 start_sector = orig_bio->bi_sector; | |
5928 | u64 file_offset = dip->logical_offset; | |
5929 | u64 submit_len = 0; | |
5930 | u64 map_length; | |
5931 | int nr_pages = 0; | |
5932 | u32 *csums = dip->csums; | |
5933 | int ret = 0; | |
1ae39938 | 5934 | int async_submit = 0; |
98bc3149 | 5935 | int write = rw & REQ_WRITE; |
e65e1535 | 5936 | |
e65e1535 MX |
5937 | map_length = orig_bio->bi_size; |
5938 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
5939 | &map_length, NULL, 0); | |
5940 | if (ret) { | |
64728bbb | 5941 | bio_put(orig_bio); |
e65e1535 MX |
5942 | return -EIO; |
5943 | } | |
5944 | ||
02f57c7a JB |
5945 | if (map_length >= orig_bio->bi_size) { |
5946 | bio = orig_bio; | |
5947 | goto submit; | |
5948 | } | |
5949 | ||
1ae39938 | 5950 | async_submit = 1; |
02f57c7a JB |
5951 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
5952 | if (!bio) | |
5953 | return -ENOMEM; | |
5954 | bio->bi_private = dip; | |
5955 | bio->bi_end_io = btrfs_end_dio_bio; | |
5956 | atomic_inc(&dip->pending_bios); | |
5957 | ||
e65e1535 MX |
5958 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5959 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
5960 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5961 | bvec->bv_offset) < bvec->bv_len)) { | |
5962 | /* | |
5963 | * inc the count before we submit the bio so | |
5964 | * we know the end IO handler won't happen before | |
5965 | * we inc the count. Otherwise, the dip might get freed | |
5966 | * before we're done setting it up | |
5967 | */ | |
5968 | atomic_inc(&dip->pending_bios); | |
5969 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
5970 | file_offset, skip_sum, | |
1ae39938 | 5971 | csums, async_submit); |
e65e1535 MX |
5972 | if (ret) { |
5973 | bio_put(bio); | |
5974 | atomic_dec(&dip->pending_bios); | |
5975 | goto out_err; | |
5976 | } | |
5977 | ||
98bc3149 JB |
5978 | /* Write's use the ordered csums */ |
5979 | if (!write && !skip_sum) | |
e65e1535 MX |
5980 | csums = csums + nr_pages; |
5981 | start_sector += submit_len >> 9; | |
5982 | file_offset += submit_len; | |
5983 | ||
5984 | submit_len = 0; | |
5985 | nr_pages = 0; | |
5986 | ||
5987 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
5988 | start_sector, GFP_NOFS); | |
5989 | if (!bio) | |
5990 | goto out_err; | |
5991 | bio->bi_private = dip; | |
5992 | bio->bi_end_io = btrfs_end_dio_bio; | |
5993 | ||
5994 | map_length = orig_bio->bi_size; | |
5995 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
5996 | &map_length, NULL, 0); | |
5997 | if (ret) { | |
5998 | bio_put(bio); | |
5999 | goto out_err; | |
6000 | } | |
6001 | } else { | |
6002 | submit_len += bvec->bv_len; | |
6003 | nr_pages ++; | |
6004 | bvec++; | |
6005 | } | |
6006 | } | |
6007 | ||
02f57c7a | 6008 | submit: |
e65e1535 | 6009 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6010 | csums, async_submit); |
e65e1535 MX |
6011 | if (!ret) |
6012 | return 0; | |
6013 | ||
6014 | bio_put(bio); | |
6015 | out_err: | |
6016 | dip->errors = 1; | |
6017 | /* | |
6018 | * before atomic variable goto zero, we must | |
6019 | * make sure dip->errors is perceived to be set. | |
6020 | */ | |
6021 | smp_mb__before_atomic_dec(); | |
6022 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6023 | bio_io_error(dip->orig_bio); | |
6024 | ||
6025 | /* bio_end_io() will handle error, so we needn't return it */ | |
6026 | return 0; | |
6027 | } | |
6028 | ||
4b46fce2 JB |
6029 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6030 | loff_t file_offset) | |
6031 | { | |
6032 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6033 | struct btrfs_dio_private *dip; | |
6034 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6035 | int skip_sum; |
7b6d91da | 6036 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6037 | int ret = 0; |
6038 | ||
6039 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6040 | ||
6041 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6042 | if (!dip) { | |
6043 | ret = -ENOMEM; | |
6044 | goto free_ordered; | |
6045 | } | |
6046 | dip->csums = NULL; | |
6047 | ||
98bc3149 JB |
6048 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6049 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6050 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6051 | if (!dip->csums) { | |
b4966b77 | 6052 | kfree(dip); |
4b46fce2 JB |
6053 | ret = -ENOMEM; |
6054 | goto free_ordered; | |
6055 | } | |
6056 | } | |
6057 | ||
6058 | dip->private = bio->bi_private; | |
6059 | dip->inode = inode; | |
6060 | dip->logical_offset = file_offset; | |
6061 | ||
4b46fce2 JB |
6062 | dip->bytes = 0; |
6063 | do { | |
6064 | dip->bytes += bvec->bv_len; | |
6065 | bvec++; | |
6066 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6067 | ||
46bfbb5c | 6068 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6069 | bio->bi_private = dip; |
e65e1535 MX |
6070 | dip->errors = 0; |
6071 | dip->orig_bio = bio; | |
6072 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6073 | |
6074 | if (write) | |
6075 | bio->bi_end_io = btrfs_endio_direct_write; | |
6076 | else | |
6077 | bio->bi_end_io = btrfs_endio_direct_read; | |
6078 | ||
e65e1535 MX |
6079 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6080 | if (!ret) | |
eaf25d93 | 6081 | return; |
4b46fce2 JB |
6082 | free_ordered: |
6083 | /* | |
6084 | * If this is a write, we need to clean up the reserved space and kill | |
6085 | * the ordered extent. | |
6086 | */ | |
6087 | if (write) { | |
6088 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6089 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6090 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6091 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6092 | btrfs_free_reserved_extent(root, ordered->start, | |
6093 | ordered->disk_len); | |
6094 | btrfs_put_ordered_extent(ordered); | |
6095 | btrfs_put_ordered_extent(ordered); | |
6096 | } | |
6097 | bio_endio(bio, ret); | |
6098 | } | |
6099 | ||
5a5f79b5 CM |
6100 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6101 | const struct iovec *iov, loff_t offset, | |
6102 | unsigned long nr_segs) | |
6103 | { | |
6104 | int seg; | |
a1b75f7d | 6105 | int i; |
5a5f79b5 CM |
6106 | size_t size; |
6107 | unsigned long addr; | |
6108 | unsigned blocksize_mask = root->sectorsize - 1; | |
6109 | ssize_t retval = -EINVAL; | |
6110 | loff_t end = offset; | |
6111 | ||
6112 | if (offset & blocksize_mask) | |
6113 | goto out; | |
6114 | ||
6115 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6116 | for (seg = 0; seg < nr_segs; seg++) { | |
6117 | addr = (unsigned long)iov[seg].iov_base; | |
6118 | size = iov[seg].iov_len; | |
6119 | end += size; | |
a1b75f7d | 6120 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6121 | goto out; |
a1b75f7d JB |
6122 | |
6123 | /* If this is a write we don't need to check anymore */ | |
6124 | if (rw & WRITE) | |
6125 | continue; | |
6126 | ||
6127 | /* | |
6128 | * Check to make sure we don't have duplicate iov_base's in this | |
6129 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6130 | * when reading back. | |
6131 | */ | |
6132 | for (i = seg + 1; i < nr_segs; i++) { | |
6133 | if (iov[seg].iov_base == iov[i].iov_base) | |
6134 | goto out; | |
6135 | } | |
5a5f79b5 CM |
6136 | } |
6137 | retval = 0; | |
6138 | out: | |
6139 | return retval; | |
6140 | } | |
16432985 CM |
6141 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6142 | const struct iovec *iov, loff_t offset, | |
6143 | unsigned long nr_segs) | |
6144 | { | |
4b46fce2 JB |
6145 | struct file *file = iocb->ki_filp; |
6146 | struct inode *inode = file->f_mapping->host; | |
6147 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6148 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6149 | u64 lockstart, lockend; |
6150 | ssize_t ret; | |
4845e44f CM |
6151 | int writing = rw & WRITE; |
6152 | int write_bits = 0; | |
3f7c579c | 6153 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6154 | |
5a5f79b5 CM |
6155 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6156 | offset, nr_segs)) { | |
6157 | return 0; | |
6158 | } | |
6159 | ||
4b46fce2 | 6160 | lockstart = offset; |
3f7c579c CM |
6161 | lockend = offset + count - 1; |
6162 | ||
6163 | if (writing) { | |
6164 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6165 | if (ret) | |
6166 | goto out; | |
6167 | } | |
4845e44f | 6168 | |
4b46fce2 | 6169 | while (1) { |
4845e44f CM |
6170 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6171 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6172 | /* |
6173 | * We're concerned with the entire range that we're going to be | |
6174 | * doing DIO to, so we need to make sure theres no ordered | |
6175 | * extents in this range. | |
6176 | */ | |
6177 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6178 | lockend - lockstart + 1); | |
6179 | if (!ordered) | |
6180 | break; | |
4845e44f CM |
6181 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6182 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6183 | btrfs_start_ordered_extent(inode, ordered, 1); |
6184 | btrfs_put_ordered_extent(ordered); | |
6185 | cond_resched(); | |
6186 | } | |
6187 | ||
4845e44f CM |
6188 | /* |
6189 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6190 | * the dirty or uptodate bits | |
6191 | */ | |
6192 | if (writing) { | |
6193 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6194 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6195 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6196 | GFP_NOFS); | |
6197 | if (ret) { | |
6198 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6199 | lockend, EXTENT_LOCKED | write_bits, | |
6200 | 1, 0, &cached_state, GFP_NOFS); | |
6201 | goto out; | |
6202 | } | |
6203 | } | |
6204 | ||
6205 | free_extent_state(cached_state); | |
6206 | cached_state = NULL; | |
6207 | ||
5a5f79b5 CM |
6208 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6209 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6210 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6211 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6212 | |
6213 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6214 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6215 | offset + iov_length(iov, nr_segs) - 1, | |
6216 | EXTENT_LOCKED | write_bits, 1, 0, | |
6217 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6218 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6219 | /* | |
6220 | * We're falling back to buffered, unlock the section we didn't | |
6221 | * do IO on. | |
6222 | */ | |
4845e44f CM |
6223 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6224 | offset + iov_length(iov, nr_segs) - 1, | |
6225 | EXTENT_LOCKED | write_bits, 1, 0, | |
6226 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6227 | } |
4845e44f CM |
6228 | out: |
6229 | free_extent_state(cached_state); | |
4b46fce2 | 6230 | return ret; |
16432985 CM |
6231 | } |
6232 | ||
1506fcc8 YS |
6233 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6234 | __u64 start, __u64 len) | |
6235 | { | |
ec29ed5b | 6236 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6237 | } |
6238 | ||
a52d9a80 | 6239 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6240 | { |
d1310b2e CM |
6241 | struct extent_io_tree *tree; |
6242 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 6243 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 6244 | } |
1832a6d5 | 6245 | |
a52d9a80 | 6246 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6247 | { |
d1310b2e | 6248 | struct extent_io_tree *tree; |
b888db2b CM |
6249 | |
6250 | ||
6251 | if (current->flags & PF_MEMALLOC) { | |
6252 | redirty_page_for_writepage(wbc, page); | |
6253 | unlock_page(page); | |
6254 | return 0; | |
6255 | } | |
d1310b2e | 6256 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6257 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6258 | } |
6259 | ||
f421950f CM |
6260 | int btrfs_writepages(struct address_space *mapping, |
6261 | struct writeback_control *wbc) | |
b293f02e | 6262 | { |
d1310b2e | 6263 | struct extent_io_tree *tree; |
771ed689 | 6264 | |
d1310b2e | 6265 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6266 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6267 | } | |
6268 | ||
3ab2fb5a CM |
6269 | static int |
6270 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6271 | struct list_head *pages, unsigned nr_pages) | |
6272 | { | |
d1310b2e CM |
6273 | struct extent_io_tree *tree; |
6274 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6275 | return extent_readpages(tree, mapping, pages, nr_pages, |
6276 | btrfs_get_extent); | |
6277 | } | |
e6dcd2dc | 6278 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6279 | { |
d1310b2e CM |
6280 | struct extent_io_tree *tree; |
6281 | struct extent_map_tree *map; | |
a52d9a80 | 6282 | int ret; |
8c2383c3 | 6283 | |
d1310b2e CM |
6284 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6285 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6286 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6287 | if (ret == 1) { |
6288 | ClearPagePrivate(page); | |
6289 | set_page_private(page, 0); | |
6290 | page_cache_release(page); | |
39279cc3 | 6291 | } |
a52d9a80 | 6292 | return ret; |
39279cc3 CM |
6293 | } |
6294 | ||
e6dcd2dc CM |
6295 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6296 | { | |
98509cfc CM |
6297 | if (PageWriteback(page) || PageDirty(page)) |
6298 | return 0; | |
b335b003 | 6299 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6300 | } |
6301 | ||
a52d9a80 | 6302 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6303 | { |
d1310b2e | 6304 | struct extent_io_tree *tree; |
e6dcd2dc | 6305 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6306 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6307 | u64 page_start = page_offset(page); |
6308 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6309 | |
8b62b72b CM |
6310 | |
6311 | /* | |
6312 | * we have the page locked, so new writeback can't start, | |
6313 | * and the dirty bit won't be cleared while we are here. | |
6314 | * | |
6315 | * Wait for IO on this page so that we can safely clear | |
6316 | * the PagePrivate2 bit and do ordered accounting | |
6317 | */ | |
e6dcd2dc | 6318 | wait_on_page_writeback(page); |
8b62b72b | 6319 | |
d1310b2e | 6320 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6321 | if (offset) { |
6322 | btrfs_releasepage(page, GFP_NOFS); | |
6323 | return; | |
6324 | } | |
2ac55d41 JB |
6325 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6326 | GFP_NOFS); | |
e6dcd2dc CM |
6327 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6328 | page_offset(page)); | |
6329 | if (ordered) { | |
eb84ae03 CM |
6330 | /* |
6331 | * IO on this page will never be started, so we need | |
6332 | * to account for any ordered extents now | |
6333 | */ | |
e6dcd2dc CM |
6334 | clear_extent_bit(tree, page_start, page_end, |
6335 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6336 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6337 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6338 | /* |
6339 | * whoever cleared the private bit is responsible | |
6340 | * for the finish_ordered_io | |
6341 | */ | |
6342 | if (TestClearPagePrivate2(page)) { | |
6343 | btrfs_finish_ordered_io(page->mapping->host, | |
6344 | page_start, page_end); | |
6345 | } | |
e6dcd2dc | 6346 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6347 | cached_state = NULL; |
6348 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6349 | GFP_NOFS); | |
e6dcd2dc CM |
6350 | } |
6351 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6352 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6353 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6354 | __btrfs_releasepage(page, GFP_NOFS); |
6355 | ||
4a096752 | 6356 | ClearPageChecked(page); |
9ad6b7bc | 6357 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6358 | ClearPagePrivate(page); |
6359 | set_page_private(page, 0); | |
6360 | page_cache_release(page); | |
6361 | } | |
39279cc3 CM |
6362 | } |
6363 | ||
9ebefb18 CM |
6364 | /* |
6365 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6366 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6367 | * be careful to check for EOF conditions here. We set the page up correctly | |
6368 | * for a written page which means we get ENOSPC checking when writing into | |
6369 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6370 | * support these features. | |
6371 | * | |
6372 | * We are not allowed to take the i_mutex here so we have to play games to | |
6373 | * protect against truncate races as the page could now be beyond EOF. Because | |
6374 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6375 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6376 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6377 | * unlock the page. | |
6378 | */ | |
c2ec175c | 6379 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6380 | { |
c2ec175c | 6381 | struct page *page = vmf->page; |
6da6abae | 6382 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6383 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6384 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6385 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6386 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6387 | char *kaddr; |
6388 | unsigned long zero_start; | |
9ebefb18 | 6389 | loff_t size; |
1832a6d5 | 6390 | int ret; |
a52d9a80 | 6391 | u64 page_start; |
e6dcd2dc | 6392 | u64 page_end; |
9ebefb18 | 6393 | |
0ca1f7ce | 6394 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
6395 | if (ret) { |
6396 | if (ret == -ENOMEM) | |
6397 | ret = VM_FAULT_OOM; | |
6398 | else /* -ENOSPC, -EIO, etc */ | |
6399 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 6400 | goto out; |
56a76f82 | 6401 | } |
1832a6d5 | 6402 | |
56a76f82 | 6403 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6404 | again: |
9ebefb18 | 6405 | lock_page(page); |
9ebefb18 | 6406 | size = i_size_read(inode); |
e6dcd2dc CM |
6407 | page_start = page_offset(page); |
6408 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6409 | |
9ebefb18 | 6410 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6411 | (page_start >= size)) { |
9ebefb18 CM |
6412 | /* page got truncated out from underneath us */ |
6413 | goto out_unlock; | |
6414 | } | |
e6dcd2dc CM |
6415 | wait_on_page_writeback(page); |
6416 | ||
2ac55d41 JB |
6417 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6418 | GFP_NOFS); | |
e6dcd2dc CM |
6419 | set_page_extent_mapped(page); |
6420 | ||
eb84ae03 CM |
6421 | /* |
6422 | * we can't set the delalloc bits if there are pending ordered | |
6423 | * extents. Drop our locks and wait for them to finish | |
6424 | */ | |
e6dcd2dc CM |
6425 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6426 | if (ordered) { | |
2ac55d41 JB |
6427 | unlock_extent_cached(io_tree, page_start, page_end, |
6428 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6429 | unlock_page(page); |
eb84ae03 | 6430 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6431 | btrfs_put_ordered_extent(ordered); |
6432 | goto again; | |
6433 | } | |
6434 | ||
fbf19087 JB |
6435 | /* |
6436 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6437 | * if it was already dirty, so for space accounting reasons we need to | |
6438 | * clear any delalloc bits for the range we are fixing to save. There | |
6439 | * is probably a better way to do this, but for now keep consistent with | |
6440 | * prepare_pages in the normal write path. | |
6441 | */ | |
2ac55d41 | 6442 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6443 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6444 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6445 | |
2ac55d41 JB |
6446 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6447 | &cached_state); | |
9ed74f2d | 6448 | if (ret) { |
2ac55d41 JB |
6449 | unlock_extent_cached(io_tree, page_start, page_end, |
6450 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6451 | ret = VM_FAULT_SIGBUS; |
6452 | goto out_unlock; | |
6453 | } | |
e6dcd2dc | 6454 | ret = 0; |
9ebefb18 CM |
6455 | |
6456 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6457 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6458 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6459 | else |
e6dcd2dc | 6460 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6461 | |
e6dcd2dc CM |
6462 | if (zero_start != PAGE_CACHE_SIZE) { |
6463 | kaddr = kmap(page); | |
6464 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6465 | flush_dcache_page(page); | |
6466 | kunmap(page); | |
6467 | } | |
247e743c | 6468 | ClearPageChecked(page); |
e6dcd2dc | 6469 | set_page_dirty(page); |
50a9b214 | 6470 | SetPageUptodate(page); |
5a3f23d5 | 6471 | |
257c62e1 CM |
6472 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6473 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6474 | ||
2ac55d41 | 6475 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6476 | |
6477 | out_unlock: | |
50a9b214 CM |
6478 | if (!ret) |
6479 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6480 | unlock_page(page); |
0ca1f7ce | 6481 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 6482 | out: |
9ebefb18 CM |
6483 | return ret; |
6484 | } | |
6485 | ||
a41ad394 | 6486 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6487 | { |
6488 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6489 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6490 | int ret; |
3893e33b | 6491 | int err = 0; |
39279cc3 | 6492 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6493 | unsigned long nr; |
dbe674a9 | 6494 | u64 mask = root->sectorsize - 1; |
39279cc3 | 6495 | |
5d5e103a JB |
6496 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6497 | if (ret) | |
a41ad394 | 6498 | return ret; |
8082510e | 6499 | |
4a096752 | 6500 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6501 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6502 | |
fcb80c2a JB |
6503 | /* |
6504 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6505 | * 3 things going on here | |
6506 | * | |
6507 | * 1) We need to reserve space for our orphan item and the space to | |
6508 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6509 | * orphan item because we didn't reserve space to remove it. | |
6510 | * | |
6511 | * 2) We need to reserve space to update our inode. | |
6512 | * | |
6513 | * 3) We need to have something to cache all the space that is going to | |
6514 | * be free'd up by the truncate operation, but also have some slack | |
6515 | * space reserved in case it uses space during the truncate (thank you | |
6516 | * very much snapshotting). | |
6517 | * | |
6518 | * And we need these to all be seperate. The fact is we can use alot of | |
6519 | * space doing the truncate, and we have no earthly idea how much space | |
6520 | * we will use, so we need the truncate reservation to be seperate so it | |
6521 | * doesn't end up using space reserved for updating the inode or | |
6522 | * removing the orphan item. We also need to be able to stop the | |
6523 | * transaction and start a new one, which means we need to be able to | |
6524 | * update the inode several times, and we have no idea of knowing how | |
6525 | * many times that will be, so we can't just reserve 1 item for the | |
6526 | * entirety of the opration, so that has to be done seperately as well. | |
6527 | * Then there is the orphan item, which does indeed need to be held on | |
6528 | * to for the whole operation, and we need nobody to touch this reserved | |
6529 | * space except the orphan code. | |
6530 | * | |
6531 | * So that leaves us with | |
6532 | * | |
6533 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6534 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6535 | * transaction reservation. | |
6536 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6537 | * updating the inode. | |
6538 | */ | |
6539 | rsv = btrfs_alloc_block_rsv(root); | |
6540 | if (!rsv) | |
6541 | return -ENOMEM; | |
6542 | btrfs_add_durable_block_rsv(root->fs_info, rsv); | |
6543 | ||
6544 | trans = btrfs_start_transaction(root, 4); | |
6545 | if (IS_ERR(trans)) { | |
6546 | err = PTR_ERR(trans); | |
6547 | goto out; | |
6548 | } | |
f0cd846e | 6549 | |
fcb80c2a JB |
6550 | /* |
6551 | * Reserve space for the truncate process. Truncate should be adding | |
6552 | * space, but if there are snapshots it may end up using space. | |
6553 | */ | |
6554 | ret = btrfs_truncate_reserve_metadata(trans, root, rsv); | |
6555 | BUG_ON(ret); | |
f0cd846e JB |
6556 | |
6557 | ret = btrfs_orphan_add(trans, inode); | |
6558 | if (ret) { | |
6559 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6560 | goto out; |
f0cd846e JB |
6561 | } |
6562 | ||
6563 | nr = trans->blocks_used; | |
6564 | btrfs_end_transaction(trans, root); | |
6565 | btrfs_btree_balance_dirty(root, nr); | |
6566 | ||
fcb80c2a JB |
6567 | /* |
6568 | * Ok so we've already migrated our bytes over for the truncate, so here | |
6569 | * just reserve the one slot we need for updating the inode. | |
6570 | */ | |
6571 | trans = btrfs_start_transaction(root, 1); | |
6572 | if (IS_ERR(trans)) { | |
6573 | err = PTR_ERR(trans); | |
6574 | goto out; | |
6575 | } | |
fcb80c2a | 6576 | trans->block_rsv = rsv; |
5a3f23d5 CM |
6577 | |
6578 | /* | |
6579 | * setattr is responsible for setting the ordered_data_close flag, | |
6580 | * but that is only tested during the last file release. That | |
6581 | * could happen well after the next commit, leaving a great big | |
6582 | * window where new writes may get lost if someone chooses to write | |
6583 | * to this file after truncating to zero | |
6584 | * | |
6585 | * The inode doesn't have any dirty data here, and so if we commit | |
6586 | * this is a noop. If someone immediately starts writing to the inode | |
6587 | * it is very likely we'll catch some of their writes in this | |
6588 | * transaction, and the commit will find this file on the ordered | |
6589 | * data list with good things to send down. | |
6590 | * | |
6591 | * This is a best effort solution, there is still a window where | |
6592 | * using truncate to replace the contents of the file will | |
6593 | * end up with a zero length file after a crash. | |
6594 | */ | |
6595 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6596 | btrfs_add_ordered_operation(trans, root, inode); | |
6597 | ||
8082510e | 6598 | while (1) { |
d68fc57b | 6599 | if (!trans) { |
fcb80c2a JB |
6600 | trans = btrfs_start_transaction(root, 3); |
6601 | if (IS_ERR(trans)) { | |
6602 | err = PTR_ERR(trans); | |
6603 | goto out; | |
6604 | } | |
d68fc57b | 6605 | |
fcb80c2a JB |
6606 | ret = btrfs_truncate_reserve_metadata(trans, root, |
6607 | rsv); | |
6608 | BUG_ON(ret); | |
6609 | ||
fcb80c2a | 6610 | trans->block_rsv = rsv; |
d68fc57b YZ |
6611 | } |
6612 | ||
8082510e YZ |
6613 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6614 | inode->i_size, | |
6615 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6616 | if (ret != -EAGAIN) { |
6617 | err = ret; | |
8082510e | 6618 | break; |
3893e33b | 6619 | } |
39279cc3 | 6620 | |
fcb80c2a | 6621 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6622 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6623 | if (ret) { |
6624 | err = ret; | |
6625 | break; | |
6626 | } | |
5f39d397 | 6627 | |
8082510e YZ |
6628 | nr = trans->blocks_used; |
6629 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6630 | trans = NULL; |
8082510e | 6631 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6632 | } |
6633 | ||
6634 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6635 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6636 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6637 | if (ret) |
6638 | err = ret; | |
ded5db9d JB |
6639 | } else if (ret && inode->i_nlink > 0) { |
6640 | /* | |
6641 | * Failed to do the truncate, remove us from the in memory | |
6642 | * orphan list. | |
6643 | */ | |
6644 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6645 | } |
6646 | ||
fcb80c2a | 6647 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6648 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6649 | if (ret && !err) |
6650 | err = ret; | |
7b128766 | 6651 | |
7b128766 | 6652 | nr = trans->blocks_used; |
89ce8a63 | 6653 | ret = btrfs_end_transaction_throttle(trans, root); |
fcb80c2a JB |
6654 | btrfs_btree_balance_dirty(root, nr); |
6655 | ||
6656 | out: | |
6657 | btrfs_free_block_rsv(root, rsv); | |
6658 | ||
3893e33b JB |
6659 | if (ret && !err) |
6660 | err = ret; | |
a41ad394 | 6661 | |
3893e33b | 6662 | return err; |
39279cc3 CM |
6663 | } |
6664 | ||
d352ac68 CM |
6665 | /* |
6666 | * create a new subvolume directory/inode (helper for the ioctl). | |
6667 | */ | |
d2fb3437 | 6668 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6669 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6670 | { |
39279cc3 | 6671 | struct inode *inode; |
76dda93c | 6672 | int err; |
00e4e6b3 | 6673 | u64 index = 0; |
39279cc3 | 6674 | |
aec7477b | 6675 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d82a6f1d | 6676 | new_dirid, S_IFDIR | 0700, &index); |
54aa1f4d | 6677 | if (IS_ERR(inode)) |
f46b5a66 | 6678 | return PTR_ERR(inode); |
39279cc3 CM |
6679 | inode->i_op = &btrfs_dir_inode_operations; |
6680 | inode->i_fop = &btrfs_dir_file_operations; | |
6681 | ||
39279cc3 | 6682 | inode->i_nlink = 1; |
dbe674a9 | 6683 | btrfs_i_size_write(inode, 0); |
3b96362c | 6684 | |
76dda93c YZ |
6685 | err = btrfs_update_inode(trans, new_root, inode); |
6686 | BUG_ON(err); | |
cb8e7090 | 6687 | |
76dda93c | 6688 | iput(inode); |
cb8e7090 | 6689 | return 0; |
39279cc3 CM |
6690 | } |
6691 | ||
d352ac68 CM |
6692 | /* helper function for file defrag and space balancing. This |
6693 | * forces readahead on a given range of bytes in an inode | |
6694 | */ | |
edbd8d4e | 6695 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
6696 | struct file_ra_state *ra, struct file *file, |
6697 | pgoff_t offset, pgoff_t last_index) | |
6698 | { | |
8e7bf94f | 6699 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 6700 | |
86479a04 CM |
6701 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
6702 | return offset + req_size; | |
86479a04 CM |
6703 | } |
6704 | ||
39279cc3 CM |
6705 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6706 | { | |
6707 | struct btrfs_inode *ei; | |
2ead6ae7 | 6708 | struct inode *inode; |
39279cc3 CM |
6709 | |
6710 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6711 | if (!ei) | |
6712 | return NULL; | |
2ead6ae7 YZ |
6713 | |
6714 | ei->root = NULL; | |
6715 | ei->space_info = NULL; | |
6716 | ei->generation = 0; | |
6717 | ei->sequence = 0; | |
15ee9bc7 | 6718 | ei->last_trans = 0; |
257c62e1 | 6719 | ei->last_sub_trans = 0; |
e02119d5 | 6720 | ei->logged_trans = 0; |
2ead6ae7 YZ |
6721 | ei->delalloc_bytes = 0; |
6722 | ei->reserved_bytes = 0; | |
6723 | ei->disk_i_size = 0; | |
6724 | ei->flags = 0; | |
6725 | ei->index_cnt = (u64)-1; | |
6726 | ei->last_unlink_trans = 0; | |
6727 | ||
0ca1f7ce | 6728 | atomic_set(&ei->outstanding_extents, 0); |
57a45ced | 6729 | atomic_set(&ei->reserved_extents, 0); |
2ead6ae7 YZ |
6730 | |
6731 | ei->ordered_data_close = 0; | |
d68fc57b | 6732 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6733 | ei->dummy_inode = 0; |
4cb5300b | 6734 | ei->in_defrag = 0; |
261507a0 | 6735 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 6736 | |
16cdcec7 MX |
6737 | ei->delayed_node = NULL; |
6738 | ||
2ead6ae7 | 6739 | inode = &ei->vfs_inode; |
a8067e02 | 6740 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
6741 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
6742 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
2ead6ae7 | 6743 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 6744 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6745 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6746 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6747 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6748 | RB_CLEAR_NODE(&ei->rb_node); |
6749 | ||
6750 | return inode; | |
39279cc3 CM |
6751 | } |
6752 | ||
fa0d7e3d NP |
6753 | static void btrfs_i_callback(struct rcu_head *head) |
6754 | { | |
6755 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6756 | INIT_LIST_HEAD(&inode->i_dentry); | |
6757 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6758 | } | |
6759 | ||
39279cc3 CM |
6760 | void btrfs_destroy_inode(struct inode *inode) |
6761 | { | |
e6dcd2dc | 6762 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6763 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6764 | ||
39279cc3 CM |
6765 | WARN_ON(!list_empty(&inode->i_dentry)); |
6766 | WARN_ON(inode->i_data.nrpages); | |
0ca1f7ce | 6767 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents)); |
57a45ced | 6768 | WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents)); |
39279cc3 | 6769 | |
a6dbd429 JB |
6770 | /* |
6771 | * This can happen where we create an inode, but somebody else also | |
6772 | * created the same inode and we need to destroy the one we already | |
6773 | * created. | |
6774 | */ | |
6775 | if (!root) | |
6776 | goto free; | |
6777 | ||
5a3f23d5 CM |
6778 | /* |
6779 | * Make sure we're properly removed from the ordered operation | |
6780 | * lists. | |
6781 | */ | |
6782 | smp_mb(); | |
6783 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6784 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6785 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6786 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6787 | } | |
6788 | ||
d68fc57b | 6789 | spin_lock(&root->orphan_lock); |
7b128766 | 6790 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
33345d01 LZ |
6791 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
6792 | (unsigned long long)btrfs_ino(inode)); | |
8082510e | 6793 | list_del_init(&BTRFS_I(inode)->i_orphan); |
7b128766 | 6794 | } |
d68fc57b | 6795 | spin_unlock(&root->orphan_lock); |
7b128766 | 6796 | |
d397712b | 6797 | while (1) { |
e6dcd2dc CM |
6798 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6799 | if (!ordered) | |
6800 | break; | |
6801 | else { | |
d397712b CM |
6802 | printk(KERN_ERR "btrfs found ordered " |
6803 | "extent %llu %llu on inode cleanup\n", | |
6804 | (unsigned long long)ordered->file_offset, | |
6805 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6806 | btrfs_remove_ordered_extent(inode, ordered); |
6807 | btrfs_put_ordered_extent(ordered); | |
6808 | btrfs_put_ordered_extent(ordered); | |
6809 | } | |
6810 | } | |
5d4f98a2 | 6811 | inode_tree_del(inode); |
5b21f2ed | 6812 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6813 | free: |
16cdcec7 | 6814 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 6815 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6816 | } |
6817 | ||
45321ac5 | 6818 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6819 | { |
6820 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6821 | |
0af3d00b | 6822 | if (btrfs_root_refs(&root->root_item) == 0 && |
82d5902d | 6823 | !is_free_space_inode(root, inode)) |
45321ac5 | 6824 | return 1; |
76dda93c | 6825 | else |
45321ac5 | 6826 | return generic_drop_inode(inode); |
76dda93c YZ |
6827 | } |
6828 | ||
0ee0fda0 | 6829 | static void init_once(void *foo) |
39279cc3 CM |
6830 | { |
6831 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6832 | ||
6833 | inode_init_once(&ei->vfs_inode); | |
6834 | } | |
6835 | ||
6836 | void btrfs_destroy_cachep(void) | |
6837 | { | |
6838 | if (btrfs_inode_cachep) | |
6839 | kmem_cache_destroy(btrfs_inode_cachep); | |
6840 | if (btrfs_trans_handle_cachep) | |
6841 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6842 | if (btrfs_transaction_cachep) | |
6843 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6844 | if (btrfs_path_cachep) |
6845 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6846 | if (btrfs_free_space_cachep) |
6847 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6848 | } |
6849 | ||
6850 | int btrfs_init_cachep(void) | |
6851 | { | |
9601e3f6 CH |
6852 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6853 | sizeof(struct btrfs_inode), 0, | |
6854 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6855 | if (!btrfs_inode_cachep) |
6856 | goto fail; | |
9601e3f6 CH |
6857 | |
6858 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6859 | sizeof(struct btrfs_trans_handle), 0, | |
6860 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6861 | if (!btrfs_trans_handle_cachep) |
6862 | goto fail; | |
9601e3f6 CH |
6863 | |
6864 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6865 | sizeof(struct btrfs_transaction), 0, | |
6866 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6867 | if (!btrfs_transaction_cachep) |
6868 | goto fail; | |
9601e3f6 CH |
6869 | |
6870 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6871 | sizeof(struct btrfs_path), 0, | |
6872 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6873 | if (!btrfs_path_cachep) |
6874 | goto fail; | |
9601e3f6 | 6875 | |
dc89e982 JB |
6876 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6877 | sizeof(struct btrfs_free_space), 0, | |
6878 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6879 | if (!btrfs_free_space_cachep) | |
6880 | goto fail; | |
6881 | ||
39279cc3 CM |
6882 | return 0; |
6883 | fail: | |
6884 | btrfs_destroy_cachep(); | |
6885 | return -ENOMEM; | |
6886 | } | |
6887 | ||
6888 | static int btrfs_getattr(struct vfsmount *mnt, | |
6889 | struct dentry *dentry, struct kstat *stat) | |
6890 | { | |
6891 | struct inode *inode = dentry->d_inode; | |
6892 | generic_fillattr(inode, stat); | |
3394e160 | 6893 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 6894 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
6895 | stat->blocks = (inode_get_bytes(inode) + |
6896 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
6897 | return 0; |
6898 | } | |
6899 | ||
75e7cb7f LB |
6900 | /* |
6901 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6902 | * directory. | |
6903 | */ | |
6904 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6905 | { | |
6906 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6907 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6908 | ||
6909 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6910 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6911 | else | |
6912 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6913 | ||
6914 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6915 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6916 | else | |
6917 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6918 | } | |
6919 | ||
d397712b CM |
6920 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6921 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6922 | { |
6923 | struct btrfs_trans_handle *trans; | |
6924 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6925 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6926 | struct inode *new_inode = new_dentry->d_inode; |
6927 | struct inode *old_inode = old_dentry->d_inode; | |
6928 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6929 | u64 index = 0; |
4df27c4d | 6930 | u64 root_objectid; |
39279cc3 | 6931 | int ret; |
33345d01 | 6932 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 6933 | |
33345d01 | 6934 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
6935 | return -EPERM; |
6936 | ||
4df27c4d | 6937 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 6938 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
6939 | return -EXDEV; |
6940 | ||
33345d01 LZ |
6941 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
6942 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 6943 | return -ENOTEMPTY; |
5f39d397 | 6944 | |
4df27c4d YZ |
6945 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
6946 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
6947 | return -ENOTEMPTY; | |
5a3f23d5 CM |
6948 | /* |
6949 | * we're using rename to replace one file with another. | |
6950 | * and the replacement file is large. Start IO on it now so | |
6951 | * we don't add too much work to the end of the transaction | |
6952 | */ | |
4baf8c92 | 6953 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
6954 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
6955 | filemap_flush(old_inode->i_mapping); | |
6956 | ||
76dda93c | 6957 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 6958 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 6959 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
6960 | /* |
6961 | * We want to reserve the absolute worst case amount of items. So if | |
6962 | * both inodes are subvols and we need to unlink them then that would | |
6963 | * require 4 item modifications, but if they are both normal inodes it | |
6964 | * would require 5 item modifications, so we'll assume their normal | |
6965 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
6966 | * should cover the worst case number of items we'll modify. | |
6967 | */ | |
6968 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
6969 | if (IS_ERR(trans)) { |
6970 | ret = PTR_ERR(trans); | |
6971 | goto out_notrans; | |
6972 | } | |
76dda93c | 6973 | |
4df27c4d YZ |
6974 | if (dest != root) |
6975 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 6976 | |
a5719521 YZ |
6977 | ret = btrfs_set_inode_index(new_dir, &index); |
6978 | if (ret) | |
6979 | goto out_fail; | |
5a3f23d5 | 6980 | |
33345d01 | 6981 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6982 | /* force full log commit if subvolume involved. */ |
6983 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
6984 | } else { | |
a5719521 YZ |
6985 | ret = btrfs_insert_inode_ref(trans, dest, |
6986 | new_dentry->d_name.name, | |
6987 | new_dentry->d_name.len, | |
33345d01 LZ |
6988 | old_ino, |
6989 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
6990 | if (ret) |
6991 | goto out_fail; | |
4df27c4d YZ |
6992 | /* |
6993 | * this is an ugly little race, but the rename is required | |
6994 | * to make sure that if we crash, the inode is either at the | |
6995 | * old name or the new one. pinning the log transaction lets | |
6996 | * us make sure we don't allow a log commit to come in after | |
6997 | * we unlink the name but before we add the new name back in. | |
6998 | */ | |
6999 | btrfs_pin_log_trans(root); | |
7000 | } | |
5a3f23d5 CM |
7001 | /* |
7002 | * make sure the inode gets flushed if it is replacing | |
7003 | * something. | |
7004 | */ | |
33345d01 | 7005 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7006 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7007 | |
39279cc3 CM |
7008 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7009 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7010 | old_inode->i_ctime = ctime; | |
5f39d397 | 7011 | |
12fcfd22 CM |
7012 | if (old_dentry->d_parent != new_dentry->d_parent) |
7013 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7014 | ||
33345d01 | 7015 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7016 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7017 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7018 | old_dentry->d_name.name, | |
7019 | old_dentry->d_name.len); | |
7020 | } else { | |
92986796 AV |
7021 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7022 | old_dentry->d_inode, | |
7023 | old_dentry->d_name.name, | |
7024 | old_dentry->d_name.len); | |
7025 | if (!ret) | |
7026 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
7027 | } |
7028 | BUG_ON(ret); | |
39279cc3 CM |
7029 | |
7030 | if (new_inode) { | |
7031 | new_inode->i_ctime = CURRENT_TIME; | |
33345d01 | 7032 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7033 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7034 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7035 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7036 | root_objectid, | |
7037 | new_dentry->d_name.name, | |
7038 | new_dentry->d_name.len); | |
7039 | BUG_ON(new_inode->i_nlink == 0); | |
7040 | } else { | |
7041 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7042 | new_dentry->d_inode, | |
7043 | new_dentry->d_name.name, | |
7044 | new_dentry->d_name.len); | |
7045 | } | |
7046 | BUG_ON(ret); | |
7b128766 | 7047 | if (new_inode->i_nlink == 0) { |
e02119d5 | 7048 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7049 | BUG_ON(ret); |
7b128766 | 7050 | } |
39279cc3 | 7051 | } |
aec7477b | 7052 | |
75e7cb7f LB |
7053 | fixup_inode_flags(new_dir, old_inode); |
7054 | ||
4df27c4d YZ |
7055 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7056 | new_dentry->d_name.name, | |
a5719521 | 7057 | new_dentry->d_name.len, 0, index); |
4df27c4d | 7058 | BUG_ON(ret); |
39279cc3 | 7059 | |
33345d01 | 7060 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
6a912213 JB |
7061 | struct dentry *parent = dget_parent(new_dentry); |
7062 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
7063 | dput(parent); | |
4df27c4d YZ |
7064 | btrfs_end_log_trans(root); |
7065 | } | |
39279cc3 | 7066 | out_fail: |
ab78c84d | 7067 | btrfs_end_transaction_throttle(trans, root); |
b44c59a8 | 7068 | out_notrans: |
33345d01 | 7069 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7070 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7071 | |
39279cc3 CM |
7072 | return ret; |
7073 | } | |
7074 | ||
d352ac68 CM |
7075 | /* |
7076 | * some fairly slow code that needs optimization. This walks the list | |
7077 | * of all the inodes with pending delalloc and forces them to disk. | |
7078 | */ | |
24bbcf04 | 7079 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7080 | { |
7081 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7082 | struct btrfs_inode *binode; | |
5b21f2ed | 7083 | struct inode *inode; |
ea8c2819 | 7084 | |
c146afad YZ |
7085 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7086 | return -EROFS; | |
7087 | ||
75eff68e | 7088 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7089 | while (!list_empty(head)) { |
ea8c2819 CM |
7090 | binode = list_entry(head->next, struct btrfs_inode, |
7091 | delalloc_inodes); | |
5b21f2ed ZY |
7092 | inode = igrab(&binode->vfs_inode); |
7093 | if (!inode) | |
7094 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7095 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7096 | if (inode) { |
8c8bee1d | 7097 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7098 | if (delay_iput) |
7099 | btrfs_add_delayed_iput(inode); | |
7100 | else | |
7101 | iput(inode); | |
5b21f2ed ZY |
7102 | } |
7103 | cond_resched(); | |
75eff68e | 7104 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7105 | } |
75eff68e | 7106 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7107 | |
7108 | /* the filemap_flush will queue IO into the worker threads, but | |
7109 | * we have to make sure the IO is actually started and that | |
7110 | * ordered extents get created before we return | |
7111 | */ | |
7112 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7113 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7114 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7115 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7116 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7117 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7118 | } |
7119 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7120 | return 0; |
7121 | } | |
7122 | ||
39279cc3 CM |
7123 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7124 | const char *symname) | |
7125 | { | |
7126 | struct btrfs_trans_handle *trans; | |
7127 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7128 | struct btrfs_path *path; | |
7129 | struct btrfs_key key; | |
1832a6d5 | 7130 | struct inode *inode = NULL; |
39279cc3 CM |
7131 | int err; |
7132 | int drop_inode = 0; | |
7133 | u64 objectid; | |
00e4e6b3 | 7134 | u64 index = 0 ; |
39279cc3 CM |
7135 | int name_len; |
7136 | int datasize; | |
5f39d397 | 7137 | unsigned long ptr; |
39279cc3 | 7138 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7139 | struct extent_buffer *leaf; |
1832a6d5 | 7140 | unsigned long nr = 0; |
39279cc3 CM |
7141 | |
7142 | name_len = strlen(symname) + 1; | |
7143 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7144 | return -ENAMETOOLONG; | |
1832a6d5 | 7145 | |
9ed74f2d JB |
7146 | /* |
7147 | * 2 items for inode item and ref | |
7148 | * 2 items for dir items | |
7149 | * 1 item for xattr if selinux is on | |
7150 | */ | |
a22285a6 YZ |
7151 | trans = btrfs_start_transaction(root, 5); |
7152 | if (IS_ERR(trans)) | |
7153 | return PTR_ERR(trans); | |
1832a6d5 | 7154 | |
581bb050 LZ |
7155 | err = btrfs_find_free_ino(root, &objectid); |
7156 | if (err) | |
7157 | goto out_unlock; | |
7158 | ||
aec7477b | 7159 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7160 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7161 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7162 | if (IS_ERR(inode)) { |
7163 | err = PTR_ERR(inode); | |
39279cc3 | 7164 | goto out_unlock; |
7cf96da3 | 7165 | } |
39279cc3 | 7166 | |
2a7dba39 | 7167 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7168 | if (err) { |
7169 | drop_inode = 1; | |
7170 | goto out_unlock; | |
7171 | } | |
7172 | ||
a1b075d2 | 7173 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7174 | if (err) |
7175 | drop_inode = 1; | |
7176 | else { | |
7177 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7178 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
7179 | inode->i_fop = &btrfs_file_operations; |
7180 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 7181 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7182 | } |
39279cc3 CM |
7183 | if (drop_inode) |
7184 | goto out_unlock; | |
7185 | ||
7186 | path = btrfs_alloc_path(); | |
7187 | BUG_ON(!path); | |
33345d01 | 7188 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7189 | key.offset = 0; |
39279cc3 CM |
7190 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7191 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7192 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7193 | datasize); | |
54aa1f4d CM |
7194 | if (err) { |
7195 | drop_inode = 1; | |
b0839166 | 7196 | btrfs_free_path(path); |
54aa1f4d CM |
7197 | goto out_unlock; |
7198 | } | |
5f39d397 CM |
7199 | leaf = path->nodes[0]; |
7200 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7201 | struct btrfs_file_extent_item); | |
7202 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7203 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7204 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7205 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7206 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7207 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7208 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7209 | ||
39279cc3 | 7210 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7211 | write_extent_buffer(leaf, symname, ptr, name_len); |
7212 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7213 | btrfs_free_path(path); |
5f39d397 | 7214 | |
39279cc3 CM |
7215 | inode->i_op = &btrfs_symlink_inode_operations; |
7216 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7217 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7218 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7219 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7220 | err = btrfs_update_inode(trans, root, inode); |
7221 | if (err) | |
7222 | drop_inode = 1; | |
39279cc3 CM |
7223 | |
7224 | out_unlock: | |
d3c2fdcf | 7225 | nr = trans->blocks_used; |
ab78c84d | 7226 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
7227 | if (drop_inode) { |
7228 | inode_dec_link_count(inode); | |
7229 | iput(inode); | |
7230 | } | |
d3c2fdcf | 7231 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7232 | return err; |
7233 | } | |
16432985 | 7234 | |
0af3d00b JB |
7235 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7236 | u64 start, u64 num_bytes, u64 min_size, | |
7237 | loff_t actual_len, u64 *alloc_hint, | |
7238 | struct btrfs_trans_handle *trans) | |
d899e052 | 7239 | { |
d899e052 YZ |
7240 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7241 | struct btrfs_key ins; | |
d899e052 | 7242 | u64 cur_offset = start; |
55a61d1d | 7243 | u64 i_size; |
d899e052 | 7244 | int ret = 0; |
0af3d00b | 7245 | bool own_trans = true; |
d899e052 | 7246 | |
0af3d00b JB |
7247 | if (trans) |
7248 | own_trans = false; | |
d899e052 | 7249 | while (num_bytes > 0) { |
0af3d00b JB |
7250 | if (own_trans) { |
7251 | trans = btrfs_start_transaction(root, 3); | |
7252 | if (IS_ERR(trans)) { | |
7253 | ret = PTR_ERR(trans); | |
7254 | break; | |
7255 | } | |
5a303d5d YZ |
7256 | } |
7257 | ||
efa56464 YZ |
7258 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7259 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7260 | if (ret) { |
0af3d00b JB |
7261 | if (own_trans) |
7262 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7263 | break; |
d899e052 | 7264 | } |
5a303d5d | 7265 | |
d899e052 YZ |
7266 | ret = insert_reserved_file_extent(trans, inode, |
7267 | cur_offset, ins.objectid, | |
7268 | ins.offset, ins.offset, | |
920bbbfb | 7269 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7270 | BTRFS_FILE_EXTENT_PREALLOC); |
7271 | BUG_ON(ret); | |
a1ed835e CM |
7272 | btrfs_drop_extent_cache(inode, cur_offset, |
7273 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7274 | |
d899e052 YZ |
7275 | num_bytes -= ins.offset; |
7276 | cur_offset += ins.offset; | |
efa56464 | 7277 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7278 | |
d899e052 | 7279 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7280 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7281 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7282 | (actual_len > inode->i_size) && |
7283 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7284 | if (cur_offset > actual_len) |
55a61d1d | 7285 | i_size = actual_len; |
d1ea6a61 | 7286 | else |
55a61d1d JB |
7287 | i_size = cur_offset; |
7288 | i_size_write(inode, i_size); | |
7289 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7290 | } |
7291 | ||
d899e052 YZ |
7292 | ret = btrfs_update_inode(trans, root, inode); |
7293 | BUG_ON(ret); | |
d899e052 | 7294 | |
0af3d00b JB |
7295 | if (own_trans) |
7296 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7297 | } |
d899e052 YZ |
7298 | return ret; |
7299 | } | |
7300 | ||
0af3d00b JB |
7301 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7302 | u64 start, u64 num_bytes, u64 min_size, | |
7303 | loff_t actual_len, u64 *alloc_hint) | |
7304 | { | |
7305 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7306 | min_size, actual_len, alloc_hint, | |
7307 | NULL); | |
7308 | } | |
7309 | ||
7310 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7311 | struct btrfs_trans_handle *trans, int mode, | |
7312 | u64 start, u64 num_bytes, u64 min_size, | |
7313 | loff_t actual_len, u64 *alloc_hint) | |
7314 | { | |
7315 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7316 | min_size, actual_len, alloc_hint, trans); | |
7317 | } | |
7318 | ||
e6dcd2dc CM |
7319 | static int btrfs_set_page_dirty(struct page *page) |
7320 | { | |
e6dcd2dc CM |
7321 | return __set_page_dirty_nobuffers(page); |
7322 | } | |
7323 | ||
b74c79e9 | 7324 | static int btrfs_permission(struct inode *inode, int mask, unsigned int flags) |
fdebe2bd | 7325 | { |
b83cc969 LZ |
7326 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7327 | ||
7328 | if (btrfs_root_readonly(root) && (mask & MAY_WRITE)) | |
7329 | return -EROFS; | |
6cbff00f | 7330 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 7331 | return -EACCES; |
b74c79e9 | 7332 | return generic_permission(inode, mask, flags, btrfs_check_acl); |
fdebe2bd | 7333 | } |
39279cc3 | 7334 | |
6e1d5dcc | 7335 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7336 | .getattr = btrfs_getattr, |
39279cc3 CM |
7337 | .lookup = btrfs_lookup, |
7338 | .create = btrfs_create, | |
7339 | .unlink = btrfs_unlink, | |
7340 | .link = btrfs_link, | |
7341 | .mkdir = btrfs_mkdir, | |
7342 | .rmdir = btrfs_rmdir, | |
7343 | .rename = btrfs_rename, | |
7344 | .symlink = btrfs_symlink, | |
7345 | .setattr = btrfs_setattr, | |
618e21d5 | 7346 | .mknod = btrfs_mknod, |
95819c05 CH |
7347 | .setxattr = btrfs_setxattr, |
7348 | .getxattr = btrfs_getxattr, | |
5103e947 | 7349 | .listxattr = btrfs_listxattr, |
95819c05 | 7350 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7351 | .permission = btrfs_permission, |
39279cc3 | 7352 | }; |
6e1d5dcc | 7353 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7354 | .lookup = btrfs_lookup, |
fdebe2bd | 7355 | .permission = btrfs_permission, |
39279cc3 | 7356 | }; |
76dda93c | 7357 | |
828c0950 | 7358 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7359 | .llseek = generic_file_llseek, |
7360 | .read = generic_read_dir, | |
cbdf5a24 | 7361 | .readdir = btrfs_real_readdir, |
34287aa3 | 7362 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7363 | #ifdef CONFIG_COMPAT |
34287aa3 | 7364 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7365 | #endif |
6bf13c0c | 7366 | .release = btrfs_release_file, |
e02119d5 | 7367 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7368 | }; |
7369 | ||
d1310b2e | 7370 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7371 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7372 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7373 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7374 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7375 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7376 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 7377 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
7378 | .set_bit_hook = btrfs_set_bit_hook, |
7379 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7380 | .merge_extent_hook = btrfs_merge_extent_hook, |
7381 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7382 | }; |
7383 | ||
35054394 CM |
7384 | /* |
7385 | * btrfs doesn't support the bmap operation because swapfiles | |
7386 | * use bmap to make a mapping of extents in the file. They assume | |
7387 | * these extents won't change over the life of the file and they | |
7388 | * use the bmap result to do IO directly to the drive. | |
7389 | * | |
7390 | * the btrfs bmap call would return logical addresses that aren't | |
7391 | * suitable for IO and they also will change frequently as COW | |
7392 | * operations happen. So, swapfile + btrfs == corruption. | |
7393 | * | |
7394 | * For now we're avoiding this by dropping bmap. | |
7395 | */ | |
7f09410b | 7396 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7397 | .readpage = btrfs_readpage, |
7398 | .writepage = btrfs_writepage, | |
b293f02e | 7399 | .writepages = btrfs_writepages, |
3ab2fb5a | 7400 | .readpages = btrfs_readpages, |
16432985 | 7401 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7402 | .invalidatepage = btrfs_invalidatepage, |
7403 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7404 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7405 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7406 | }; |
7407 | ||
7f09410b | 7408 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7409 | .readpage = btrfs_readpage, |
7410 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7411 | .invalidatepage = btrfs_invalidatepage, |
7412 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7413 | }; |
7414 | ||
6e1d5dcc | 7415 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7416 | .getattr = btrfs_getattr, |
7417 | .setattr = btrfs_setattr, | |
95819c05 CH |
7418 | .setxattr = btrfs_setxattr, |
7419 | .getxattr = btrfs_getxattr, | |
5103e947 | 7420 | .listxattr = btrfs_listxattr, |
95819c05 | 7421 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7422 | .permission = btrfs_permission, |
1506fcc8 | 7423 | .fiemap = btrfs_fiemap, |
39279cc3 | 7424 | }; |
6e1d5dcc | 7425 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7426 | .getattr = btrfs_getattr, |
7427 | .setattr = btrfs_setattr, | |
fdebe2bd | 7428 | .permission = btrfs_permission, |
95819c05 CH |
7429 | .setxattr = btrfs_setxattr, |
7430 | .getxattr = btrfs_getxattr, | |
33268eaf | 7431 | .listxattr = btrfs_listxattr, |
95819c05 | 7432 | .removexattr = btrfs_removexattr, |
618e21d5 | 7433 | }; |
6e1d5dcc | 7434 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7435 | .readlink = generic_readlink, |
7436 | .follow_link = page_follow_link_light, | |
7437 | .put_link = page_put_link, | |
f209561a | 7438 | .getattr = btrfs_getattr, |
fdebe2bd | 7439 | .permission = btrfs_permission, |
0279b4cd JO |
7440 | .setxattr = btrfs_setxattr, |
7441 | .getxattr = btrfs_getxattr, | |
7442 | .listxattr = btrfs_listxattr, | |
7443 | .removexattr = btrfs_removexattr, | |
39279cc3 | 7444 | }; |
76dda93c | 7445 | |
82d339d9 | 7446 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
7447 | .d_delete = btrfs_dentry_delete, |
7448 | }; |