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" | |
e6dcd2dc | 48 | #include "ordered-data.h" |
95819c05 | 49 | #include "xattr.h" |
e02119d5 | 50 | #include "tree-log.h" |
4a54c8c1 | 51 | #include "volumes.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); |
560f7d75 LZ |
396 | if (!pages) { |
397 | /* just bail out to the uncompressed code */ | |
398 | goto cont; | |
399 | } | |
c8b97818 | 400 | |
261507a0 LZ |
401 | if (BTRFS_I(inode)->force_compress) |
402 | compress_type = BTRFS_I(inode)->force_compress; | |
403 | ||
404 | ret = btrfs_compress_pages(compress_type, | |
405 | inode->i_mapping, start, | |
406 | total_compressed, pages, | |
407 | nr_pages, &nr_pages_ret, | |
408 | &total_in, | |
409 | &total_compressed, | |
410 | max_compressed); | |
c8b97818 CM |
411 | |
412 | if (!ret) { | |
413 | unsigned long offset = total_compressed & | |
414 | (PAGE_CACHE_SIZE - 1); | |
415 | struct page *page = pages[nr_pages_ret - 1]; | |
416 | char *kaddr; | |
417 | ||
418 | /* zero the tail end of the last page, we might be | |
419 | * sending it down to disk | |
420 | */ | |
421 | if (offset) { | |
422 | kaddr = kmap_atomic(page, KM_USER0); | |
423 | memset(kaddr + offset, 0, | |
424 | PAGE_CACHE_SIZE - offset); | |
425 | kunmap_atomic(kaddr, KM_USER0); | |
426 | } | |
427 | will_compress = 1; | |
428 | } | |
429 | } | |
560f7d75 | 430 | cont: |
c8b97818 | 431 | if (start == 0) { |
7a7eaa40 | 432 | trans = btrfs_join_transaction(root); |
3612b495 | 433 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 434 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 435 | |
c8b97818 | 436 | /* lets try to make an inline extent */ |
771ed689 | 437 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 438 | /* we didn't compress the entire range, try |
771ed689 | 439 | * to make an uncompressed inline extent. |
c8b97818 CM |
440 | */ |
441 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 442 | start, end, 0, 0, NULL); |
c8b97818 | 443 | } else { |
771ed689 | 444 | /* try making a compressed inline extent */ |
c8b97818 CM |
445 | ret = cow_file_range_inline(trans, root, inode, |
446 | start, end, | |
fe3f566c LZ |
447 | total_compressed, |
448 | compress_type, pages); | |
c8b97818 CM |
449 | } |
450 | if (ret == 0) { | |
771ed689 CM |
451 | /* |
452 | * inline extent creation worked, we don't need | |
453 | * to create any more async work items. Unlock | |
454 | * and free up our temp pages. | |
455 | */ | |
c8b97818 | 456 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
457 | &BTRFS_I(inode)->io_tree, |
458 | start, end, NULL, | |
459 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 460 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 461 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
462 | |
463 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
464 | goto free_pages_out; |
465 | } | |
c2167754 | 466 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
467 | } |
468 | ||
469 | if (will_compress) { | |
470 | /* | |
471 | * we aren't doing an inline extent round the compressed size | |
472 | * up to a block size boundary so the allocator does sane | |
473 | * things | |
474 | */ | |
475 | total_compressed = (total_compressed + blocksize - 1) & | |
476 | ~(blocksize - 1); | |
477 | ||
478 | /* | |
479 | * one last check to make sure the compression is really a | |
480 | * win, compare the page count read with the blocks on disk | |
481 | */ | |
482 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
483 | ~(PAGE_CACHE_SIZE - 1); | |
484 | if (total_compressed >= total_in) { | |
485 | will_compress = 0; | |
486 | } else { | |
c8b97818 CM |
487 | num_bytes = total_in; |
488 | } | |
489 | } | |
490 | if (!will_compress && pages) { | |
491 | /* | |
492 | * the compression code ran but failed to make things smaller, | |
493 | * free any pages it allocated and our page pointer array | |
494 | */ | |
495 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 496 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
497 | page_cache_release(pages[i]); |
498 | } | |
499 | kfree(pages); | |
500 | pages = NULL; | |
501 | total_compressed = 0; | |
502 | nr_pages_ret = 0; | |
503 | ||
504 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
505 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
506 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 507 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 508 | } |
c8b97818 | 509 | } |
771ed689 CM |
510 | if (will_compress) { |
511 | *num_added += 1; | |
c8b97818 | 512 | |
771ed689 CM |
513 | /* the async work queues will take care of doing actual |
514 | * allocation on disk for these compressed pages, | |
515 | * and will submit them to the elevator. | |
516 | */ | |
517 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
518 | total_compressed, pages, nr_pages_ret, |
519 | compress_type); | |
179e29e4 | 520 | |
24ae6365 | 521 | if (start + num_bytes < end) { |
771ed689 CM |
522 | start += num_bytes; |
523 | pages = NULL; | |
524 | cond_resched(); | |
525 | goto again; | |
526 | } | |
527 | } else { | |
f03d9301 | 528 | cleanup_and_bail_uncompressed: |
771ed689 CM |
529 | /* |
530 | * No compression, but we still need to write the pages in | |
531 | * the file we've been given so far. redirty the locked | |
532 | * page if it corresponds to our extent and set things up | |
533 | * for the async work queue to run cow_file_range to do | |
534 | * the normal delalloc dance | |
535 | */ | |
536 | if (page_offset(locked_page) >= start && | |
537 | page_offset(locked_page) <= end) { | |
538 | __set_page_dirty_nobuffers(locked_page); | |
539 | /* unlocked later on in the async handlers */ | |
540 | } | |
261507a0 LZ |
541 | add_async_extent(async_cow, start, end - start + 1, |
542 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
543 | *num_added += 1; |
544 | } | |
3b951516 | 545 | |
771ed689 CM |
546 | out: |
547 | return 0; | |
548 | ||
549 | free_pages_out: | |
550 | for (i = 0; i < nr_pages_ret; i++) { | |
551 | WARN_ON(pages[i]->mapping); | |
552 | page_cache_release(pages[i]); | |
553 | } | |
d397712b | 554 | kfree(pages); |
771ed689 CM |
555 | |
556 | goto out; | |
557 | } | |
558 | ||
559 | /* | |
560 | * phase two of compressed writeback. This is the ordered portion | |
561 | * of the code, which only gets called in the order the work was | |
562 | * queued. We walk all the async extents created by compress_file_range | |
563 | * and send them down to the disk. | |
564 | */ | |
565 | static noinline int submit_compressed_extents(struct inode *inode, | |
566 | struct async_cow *async_cow) | |
567 | { | |
568 | struct async_extent *async_extent; | |
569 | u64 alloc_hint = 0; | |
570 | struct btrfs_trans_handle *trans; | |
571 | struct btrfs_key ins; | |
572 | struct extent_map *em; | |
573 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
574 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
575 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 576 | int ret = 0; |
771ed689 CM |
577 | |
578 | if (list_empty(&async_cow->extents)) | |
579 | return 0; | |
580 | ||
771ed689 | 581 | |
d397712b | 582 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
583 | async_extent = list_entry(async_cow->extents.next, |
584 | struct async_extent, list); | |
585 | list_del(&async_extent->list); | |
c8b97818 | 586 | |
771ed689 CM |
587 | io_tree = &BTRFS_I(inode)->io_tree; |
588 | ||
f5a84ee3 | 589 | retry: |
771ed689 CM |
590 | /* did the compression code fall back to uncompressed IO? */ |
591 | if (!async_extent->pages) { | |
592 | int page_started = 0; | |
593 | unsigned long nr_written = 0; | |
594 | ||
595 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 JB |
596 | async_extent->start + |
597 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
598 | |
599 | /* allocate blocks */ | |
f5a84ee3 JB |
600 | ret = cow_file_range(inode, async_cow->locked_page, |
601 | async_extent->start, | |
602 | async_extent->start + | |
603 | async_extent->ram_size - 1, | |
604 | &page_started, &nr_written, 0); | |
771ed689 CM |
605 | |
606 | /* | |
607 | * if page_started, cow_file_range inserted an | |
608 | * inline extent and took care of all the unlocking | |
609 | * and IO for us. Otherwise, we need to submit | |
610 | * all those pages down to the drive. | |
611 | */ | |
f5a84ee3 | 612 | if (!page_started && !ret) |
771ed689 CM |
613 | extent_write_locked_range(io_tree, |
614 | inode, async_extent->start, | |
d397712b | 615 | async_extent->start + |
771ed689 CM |
616 | async_extent->ram_size - 1, |
617 | btrfs_get_extent, | |
618 | WB_SYNC_ALL); | |
619 | kfree(async_extent); | |
620 | cond_resched(); | |
621 | continue; | |
622 | } | |
623 | ||
624 | lock_extent(io_tree, async_extent->start, | |
625 | async_extent->start + async_extent->ram_size - 1, | |
626 | GFP_NOFS); | |
771ed689 | 627 | |
7a7eaa40 | 628 | trans = btrfs_join_transaction(root); |
3612b495 | 629 | BUG_ON(IS_ERR(trans)); |
74b21075 | 630 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 CM |
631 | ret = btrfs_reserve_extent(trans, root, |
632 | async_extent->compressed_size, | |
633 | async_extent->compressed_size, | |
634 | 0, alloc_hint, | |
635 | (u64)-1, &ins, 1); | |
c2167754 YZ |
636 | btrfs_end_transaction(trans, root); |
637 | ||
f5a84ee3 JB |
638 | if (ret) { |
639 | int i; | |
640 | for (i = 0; i < async_extent->nr_pages; i++) { | |
641 | WARN_ON(async_extent->pages[i]->mapping); | |
642 | page_cache_release(async_extent->pages[i]); | |
643 | } | |
644 | kfree(async_extent->pages); | |
645 | async_extent->nr_pages = 0; | |
646 | async_extent->pages = NULL; | |
647 | unlock_extent(io_tree, async_extent->start, | |
648 | async_extent->start + | |
649 | async_extent->ram_size - 1, GFP_NOFS); | |
650 | goto retry; | |
651 | } | |
652 | ||
c2167754 YZ |
653 | /* |
654 | * here we're doing allocation and writeback of the | |
655 | * compressed pages | |
656 | */ | |
657 | btrfs_drop_extent_cache(inode, async_extent->start, | |
658 | async_extent->start + | |
659 | async_extent->ram_size - 1, 0); | |
660 | ||
172ddd60 | 661 | em = alloc_extent_map(); |
c26a9203 | 662 | BUG_ON(!em); |
771ed689 CM |
663 | em->start = async_extent->start; |
664 | em->len = async_extent->ram_size; | |
445a6944 | 665 | em->orig_start = em->start; |
c8b97818 | 666 | |
771ed689 CM |
667 | em->block_start = ins.objectid; |
668 | em->block_len = ins.offset; | |
669 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 670 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
671 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
672 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
673 | ||
d397712b | 674 | while (1) { |
890871be | 675 | write_lock(&em_tree->lock); |
771ed689 | 676 | ret = add_extent_mapping(em_tree, em); |
890871be | 677 | write_unlock(&em_tree->lock); |
771ed689 CM |
678 | if (ret != -EEXIST) { |
679 | free_extent_map(em); | |
680 | break; | |
681 | } | |
682 | btrfs_drop_extent_cache(inode, async_extent->start, | |
683 | async_extent->start + | |
684 | async_extent->ram_size - 1, 0); | |
685 | } | |
686 | ||
261507a0 LZ |
687 | ret = btrfs_add_ordered_extent_compress(inode, |
688 | async_extent->start, | |
689 | ins.objectid, | |
690 | async_extent->ram_size, | |
691 | ins.offset, | |
692 | BTRFS_ORDERED_COMPRESSED, | |
693 | async_extent->compress_type); | |
771ed689 CM |
694 | BUG_ON(ret); |
695 | ||
771ed689 CM |
696 | /* |
697 | * clear dirty, set writeback and unlock the pages. | |
698 | */ | |
699 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
700 | &BTRFS_I(inode)->io_tree, |
701 | async_extent->start, | |
702 | async_extent->start + | |
703 | async_extent->ram_size - 1, | |
704 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
705 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 706 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 707 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
708 | |
709 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
710 | async_extent->start, |
711 | async_extent->ram_size, | |
712 | ins.objectid, | |
713 | ins.offset, async_extent->pages, | |
714 | async_extent->nr_pages); | |
771ed689 CM |
715 | |
716 | BUG_ON(ret); | |
771ed689 CM |
717 | alloc_hint = ins.objectid + ins.offset; |
718 | kfree(async_extent); | |
719 | cond_resched(); | |
720 | } | |
721 | ||
771ed689 CM |
722 | return 0; |
723 | } | |
724 | ||
4b46fce2 JB |
725 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
726 | u64 num_bytes) | |
727 | { | |
728 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
729 | struct extent_map *em; | |
730 | u64 alloc_hint = 0; | |
731 | ||
732 | read_lock(&em_tree->lock); | |
733 | em = search_extent_mapping(em_tree, start, num_bytes); | |
734 | if (em) { | |
735 | /* | |
736 | * if block start isn't an actual block number then find the | |
737 | * first block in this inode and use that as a hint. If that | |
738 | * block is also bogus then just don't worry about it. | |
739 | */ | |
740 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
741 | free_extent_map(em); | |
742 | em = search_extent_mapping(em_tree, 0, 0); | |
743 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
744 | alloc_hint = em->block_start; | |
745 | if (em) | |
746 | free_extent_map(em); | |
747 | } else { | |
748 | alloc_hint = em->block_start; | |
749 | free_extent_map(em); | |
750 | } | |
751 | } | |
752 | read_unlock(&em_tree->lock); | |
753 | ||
754 | return alloc_hint; | |
755 | } | |
756 | ||
771ed689 CM |
757 | /* |
758 | * when extent_io.c finds a delayed allocation range in the file, | |
759 | * the call backs end up in this code. The basic idea is to | |
760 | * allocate extents on disk for the range, and create ordered data structs | |
761 | * in ram to track those extents. | |
762 | * | |
763 | * locked_page is the page that writepage had locked already. We use | |
764 | * it to make sure we don't do extra locks or unlocks. | |
765 | * | |
766 | * *page_started is set to one if we unlock locked_page and do everything | |
767 | * required to start IO on it. It may be clean and already done with | |
768 | * IO when we return. | |
769 | */ | |
770 | static noinline int cow_file_range(struct inode *inode, | |
771 | struct page *locked_page, | |
772 | u64 start, u64 end, int *page_started, | |
773 | unsigned long *nr_written, | |
774 | int unlock) | |
775 | { | |
776 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
777 | struct btrfs_trans_handle *trans; | |
778 | u64 alloc_hint = 0; | |
779 | u64 num_bytes; | |
780 | unsigned long ram_size; | |
781 | u64 disk_num_bytes; | |
782 | u64 cur_alloc_size; | |
783 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
784 | struct btrfs_key ins; |
785 | struct extent_map *em; | |
786 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
787 | int ret = 0; | |
788 | ||
2cf8572d | 789 | BUG_ON(btrfs_is_free_space_inode(root, inode)); |
7a7eaa40 | 790 | trans = btrfs_join_transaction(root); |
3612b495 | 791 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 792 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 793 | |
771ed689 CM |
794 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
795 | num_bytes = max(blocksize, num_bytes); | |
796 | disk_num_bytes = num_bytes; | |
797 | ret = 0; | |
798 | ||
4cb5300b CM |
799 | /* if this is a small write inside eof, kick off defrag */ |
800 | if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024) | |
801 | btrfs_add_inode_defrag(trans, inode); | |
802 | ||
771ed689 CM |
803 | if (start == 0) { |
804 | /* lets try to make an inline extent */ | |
805 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 806 | start, end, 0, 0, NULL); |
771ed689 CM |
807 | if (ret == 0) { |
808 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
809 | &BTRFS_I(inode)->io_tree, |
810 | start, end, NULL, | |
811 | EXTENT_CLEAR_UNLOCK_PAGE | | |
812 | EXTENT_CLEAR_UNLOCK | | |
813 | EXTENT_CLEAR_DELALLOC | | |
814 | EXTENT_CLEAR_DIRTY | | |
815 | EXTENT_SET_WRITEBACK | | |
816 | EXTENT_END_WRITEBACK); | |
c2167754 | 817 | |
771ed689 CM |
818 | *nr_written = *nr_written + |
819 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
820 | *page_started = 1; | |
821 | ret = 0; | |
822 | goto out; | |
823 | } | |
824 | } | |
825 | ||
826 | BUG_ON(disk_num_bytes > | |
6c41761f | 827 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 828 | |
4b46fce2 | 829 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
830 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
831 | ||
d397712b | 832 | while (disk_num_bytes > 0) { |
a791e35e CM |
833 | unsigned long op; |
834 | ||
287a0ab9 | 835 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 836 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 837 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 838 | (u64)-1, &ins, 1); |
d397712b CM |
839 | BUG_ON(ret); |
840 | ||
172ddd60 | 841 | em = alloc_extent_map(); |
c26a9203 | 842 | BUG_ON(!em); |
e6dcd2dc | 843 | em->start = start; |
445a6944 | 844 | em->orig_start = em->start; |
771ed689 CM |
845 | ram_size = ins.offset; |
846 | em->len = ins.offset; | |
c8b97818 | 847 | |
e6dcd2dc | 848 | em->block_start = ins.objectid; |
c8b97818 | 849 | em->block_len = ins.offset; |
e6dcd2dc | 850 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 851 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 852 | |
d397712b | 853 | while (1) { |
890871be | 854 | write_lock(&em_tree->lock); |
e6dcd2dc | 855 | ret = add_extent_mapping(em_tree, em); |
890871be | 856 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
857 | if (ret != -EEXIST) { |
858 | free_extent_map(em); | |
859 | break; | |
860 | } | |
861 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 862 | start + ram_size - 1, 0); |
e6dcd2dc CM |
863 | } |
864 | ||
98d20f67 | 865 | cur_alloc_size = ins.offset; |
e6dcd2dc | 866 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 867 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 868 | BUG_ON(ret); |
c8b97818 | 869 | |
17d217fe YZ |
870 | if (root->root_key.objectid == |
871 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
872 | ret = btrfs_reloc_clone_csums(inode, start, | |
873 | cur_alloc_size); | |
874 | BUG_ON(ret); | |
875 | } | |
876 | ||
d397712b | 877 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 878 | break; |
d397712b | 879 | |
c8b97818 CM |
880 | /* we're not doing compressed IO, don't unlock the first |
881 | * page (which the caller expects to stay locked), don't | |
882 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
883 | * |
884 | * Do set the Private2 bit so we know this page was properly | |
885 | * setup for writepage | |
c8b97818 | 886 | */ |
a791e35e CM |
887 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
888 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
889 | EXTENT_SET_PRIVATE2; | |
890 | ||
c8b97818 CM |
891 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
892 | start, start + ram_size - 1, | |
a791e35e | 893 | locked_page, op); |
c8b97818 | 894 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
895 | num_bytes -= cur_alloc_size; |
896 | alloc_hint = ins.objectid + ins.offset; | |
897 | start += cur_alloc_size; | |
b888db2b | 898 | } |
b888db2b | 899 | out: |
771ed689 | 900 | ret = 0; |
b888db2b | 901 | btrfs_end_transaction(trans, root); |
c8b97818 | 902 | |
be20aa9d | 903 | return ret; |
771ed689 | 904 | } |
c8b97818 | 905 | |
771ed689 CM |
906 | /* |
907 | * work queue call back to started compression on a file and pages | |
908 | */ | |
909 | static noinline void async_cow_start(struct btrfs_work *work) | |
910 | { | |
911 | struct async_cow *async_cow; | |
912 | int num_added = 0; | |
913 | async_cow = container_of(work, struct async_cow, work); | |
914 | ||
915 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
916 | async_cow->start, async_cow->end, async_cow, | |
917 | &num_added); | |
918 | if (num_added == 0) | |
919 | async_cow->inode = NULL; | |
920 | } | |
921 | ||
922 | /* | |
923 | * work queue call back to submit previously compressed pages | |
924 | */ | |
925 | static noinline void async_cow_submit(struct btrfs_work *work) | |
926 | { | |
927 | struct async_cow *async_cow; | |
928 | struct btrfs_root *root; | |
929 | unsigned long nr_pages; | |
930 | ||
931 | async_cow = container_of(work, struct async_cow, work); | |
932 | ||
933 | root = async_cow->root; | |
934 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
935 | PAGE_CACHE_SHIFT; | |
936 | ||
937 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
938 | ||
939 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
940 | 5 * 1042 * 1024 && | |
941 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
942 | wake_up(&root->fs_info->async_submit_wait); | |
943 | ||
d397712b | 944 | if (async_cow->inode) |
771ed689 | 945 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 946 | } |
c8b97818 | 947 | |
771ed689 CM |
948 | static noinline void async_cow_free(struct btrfs_work *work) |
949 | { | |
950 | struct async_cow *async_cow; | |
951 | async_cow = container_of(work, struct async_cow, work); | |
952 | kfree(async_cow); | |
953 | } | |
954 | ||
955 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
956 | u64 start, u64 end, int *page_started, | |
957 | unsigned long *nr_written) | |
958 | { | |
959 | struct async_cow *async_cow; | |
960 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
961 | unsigned long nr_pages; | |
962 | u64 cur_end; | |
963 | int limit = 10 * 1024 * 1042; | |
964 | ||
a3429ab7 CM |
965 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
966 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 967 | while (start < end) { |
771ed689 | 968 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
8d413713 | 969 | BUG_ON(!async_cow); |
771ed689 CM |
970 | async_cow->inode = inode; |
971 | async_cow->root = root; | |
972 | async_cow->locked_page = locked_page; | |
973 | async_cow->start = start; | |
974 | ||
6cbff00f | 975 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
976 | cur_end = end; |
977 | else | |
978 | cur_end = min(end, start + 512 * 1024 - 1); | |
979 | ||
980 | async_cow->end = cur_end; | |
981 | INIT_LIST_HEAD(&async_cow->extents); | |
982 | ||
983 | async_cow->work.func = async_cow_start; | |
984 | async_cow->work.ordered_func = async_cow_submit; | |
985 | async_cow->work.ordered_free = async_cow_free; | |
986 | async_cow->work.flags = 0; | |
987 | ||
771ed689 CM |
988 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
989 | PAGE_CACHE_SHIFT; | |
990 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
991 | ||
992 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
993 | &async_cow->work); | |
994 | ||
995 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
996 | wait_event(root->fs_info->async_submit_wait, | |
997 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
998 | limit)); | |
999 | } | |
1000 | ||
d397712b | 1001 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1002 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1003 | wait_event(root->fs_info->async_submit_wait, | |
1004 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1005 | 0)); | |
1006 | } | |
1007 | ||
1008 | *nr_written += nr_pages; | |
1009 | start = cur_end + 1; | |
1010 | } | |
1011 | *page_started = 1; | |
1012 | return 0; | |
be20aa9d CM |
1013 | } |
1014 | ||
d397712b | 1015 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1016 | u64 bytenr, u64 num_bytes) |
1017 | { | |
1018 | int ret; | |
1019 | struct btrfs_ordered_sum *sums; | |
1020 | LIST_HEAD(list); | |
1021 | ||
07d400a6 | 1022 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1023 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1024 | if (ret == 0 && list_empty(&list)) |
1025 | return 0; | |
1026 | ||
1027 | while (!list_empty(&list)) { | |
1028 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1029 | list_del(&sums->list); | |
1030 | kfree(sums); | |
1031 | } | |
1032 | return 1; | |
1033 | } | |
1034 | ||
d352ac68 CM |
1035 | /* |
1036 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1037 | * of the extents that exist in the file, and COWs the file as required. | |
1038 | * | |
1039 | * If no cow copies or snapshots exist, we write directly to the existing | |
1040 | * blocks on disk | |
1041 | */ | |
7f366cfe CM |
1042 | static noinline int run_delalloc_nocow(struct inode *inode, |
1043 | struct page *locked_page, | |
771ed689 CM |
1044 | u64 start, u64 end, int *page_started, int force, |
1045 | unsigned long *nr_written) | |
be20aa9d | 1046 | { |
be20aa9d | 1047 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1048 | struct btrfs_trans_handle *trans; |
be20aa9d | 1049 | struct extent_buffer *leaf; |
be20aa9d | 1050 | struct btrfs_path *path; |
80ff3856 | 1051 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1052 | struct btrfs_key found_key; |
80ff3856 YZ |
1053 | u64 cow_start; |
1054 | u64 cur_offset; | |
1055 | u64 extent_end; | |
5d4f98a2 | 1056 | u64 extent_offset; |
80ff3856 YZ |
1057 | u64 disk_bytenr; |
1058 | u64 num_bytes; | |
1059 | int extent_type; | |
1060 | int ret; | |
d899e052 | 1061 | int type; |
80ff3856 YZ |
1062 | int nocow; |
1063 | int check_prev = 1; | |
82d5902d | 1064 | bool nolock; |
33345d01 | 1065 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1066 | |
1067 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1068 | if (!path) |
1069 | return -ENOMEM; | |
82d5902d | 1070 | |
2cf8572d | 1071 | nolock = btrfs_is_free_space_inode(root, inode); |
82d5902d LZ |
1072 | |
1073 | if (nolock) | |
7a7eaa40 | 1074 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1075 | else |
7a7eaa40 | 1076 | trans = btrfs_join_transaction(root); |
ff5714cc | 1077 | |
3612b495 | 1078 | BUG_ON(IS_ERR(trans)); |
74b21075 | 1079 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1080 | |
80ff3856 YZ |
1081 | cow_start = (u64)-1; |
1082 | cur_offset = start; | |
1083 | while (1) { | |
33345d01 | 1084 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 YZ |
1085 | cur_offset, 0); |
1086 | BUG_ON(ret < 0); | |
1087 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1088 | leaf = path->nodes[0]; | |
1089 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1090 | path->slots[0] - 1); | |
33345d01 | 1091 | if (found_key.objectid == ino && |
80ff3856 YZ |
1092 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1093 | path->slots[0]--; | |
1094 | } | |
1095 | check_prev = 0; | |
1096 | next_slot: | |
1097 | leaf = path->nodes[0]; | |
1098 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1099 | ret = btrfs_next_leaf(root, path); | |
1100 | if (ret < 0) | |
1101 | BUG_ON(1); | |
1102 | if (ret > 0) | |
1103 | break; | |
1104 | leaf = path->nodes[0]; | |
1105 | } | |
be20aa9d | 1106 | |
80ff3856 YZ |
1107 | nocow = 0; |
1108 | disk_bytenr = 0; | |
17d217fe | 1109 | num_bytes = 0; |
80ff3856 YZ |
1110 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1111 | ||
33345d01 | 1112 | if (found_key.objectid > ino || |
80ff3856 YZ |
1113 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1114 | found_key.offset > end) | |
1115 | break; | |
1116 | ||
1117 | if (found_key.offset > cur_offset) { | |
1118 | extent_end = found_key.offset; | |
e9061e21 | 1119 | extent_type = 0; |
80ff3856 YZ |
1120 | goto out_check; |
1121 | } | |
1122 | ||
1123 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1124 | struct btrfs_file_extent_item); | |
1125 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1126 | ||
d899e052 YZ |
1127 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1128 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1129 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1130 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1131 | extent_end = found_key.offset + |
1132 | btrfs_file_extent_num_bytes(leaf, fi); | |
1133 | if (extent_end <= start) { | |
1134 | path->slots[0]++; | |
1135 | goto next_slot; | |
1136 | } | |
17d217fe YZ |
1137 | if (disk_bytenr == 0) |
1138 | goto out_check; | |
80ff3856 YZ |
1139 | if (btrfs_file_extent_compression(leaf, fi) || |
1140 | btrfs_file_extent_encryption(leaf, fi) || | |
1141 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1142 | goto out_check; | |
d899e052 YZ |
1143 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1144 | goto out_check; | |
d2fb3437 | 1145 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1146 | goto out_check; |
33345d01 | 1147 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1148 | found_key.offset - |
1149 | extent_offset, disk_bytenr)) | |
17d217fe | 1150 | goto out_check; |
5d4f98a2 | 1151 | disk_bytenr += extent_offset; |
17d217fe YZ |
1152 | disk_bytenr += cur_offset - found_key.offset; |
1153 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1154 | /* | |
1155 | * force cow if csum exists in the range. | |
1156 | * this ensure that csum for a given extent are | |
1157 | * either valid or do not exist. | |
1158 | */ | |
1159 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1160 | goto out_check; | |
80ff3856 YZ |
1161 | nocow = 1; |
1162 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1163 | extent_end = found_key.offset + | |
1164 | btrfs_file_extent_inline_len(leaf, fi); | |
1165 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1166 | } else { | |
1167 | BUG_ON(1); | |
1168 | } | |
1169 | out_check: | |
1170 | if (extent_end <= start) { | |
1171 | path->slots[0]++; | |
1172 | goto next_slot; | |
1173 | } | |
1174 | if (!nocow) { | |
1175 | if (cow_start == (u64)-1) | |
1176 | cow_start = cur_offset; | |
1177 | cur_offset = extent_end; | |
1178 | if (cur_offset > end) | |
1179 | break; | |
1180 | path->slots[0]++; | |
1181 | goto next_slot; | |
7ea394f1 YZ |
1182 | } |
1183 | ||
b3b4aa74 | 1184 | btrfs_release_path(path); |
80ff3856 YZ |
1185 | if (cow_start != (u64)-1) { |
1186 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1187 | found_key.offset - 1, page_started, |
1188 | nr_written, 1); | |
80ff3856 YZ |
1189 | BUG_ON(ret); |
1190 | cow_start = (u64)-1; | |
7ea394f1 | 1191 | } |
80ff3856 | 1192 | |
d899e052 YZ |
1193 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1194 | struct extent_map *em; | |
1195 | struct extent_map_tree *em_tree; | |
1196 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1197 | em = alloc_extent_map(); |
c26a9203 | 1198 | BUG_ON(!em); |
d899e052 | 1199 | em->start = cur_offset; |
445a6944 | 1200 | em->orig_start = em->start; |
d899e052 YZ |
1201 | em->len = num_bytes; |
1202 | em->block_len = num_bytes; | |
1203 | em->block_start = disk_bytenr; | |
1204 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1205 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1206 | while (1) { | |
890871be | 1207 | write_lock(&em_tree->lock); |
d899e052 | 1208 | ret = add_extent_mapping(em_tree, em); |
890871be | 1209 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1210 | if (ret != -EEXIST) { |
1211 | free_extent_map(em); | |
1212 | break; | |
1213 | } | |
1214 | btrfs_drop_extent_cache(inode, em->start, | |
1215 | em->start + em->len - 1, 0); | |
1216 | } | |
1217 | type = BTRFS_ORDERED_PREALLOC; | |
1218 | } else { | |
1219 | type = BTRFS_ORDERED_NOCOW; | |
1220 | } | |
80ff3856 YZ |
1221 | |
1222 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1223 | num_bytes, num_bytes, type); |
1224 | BUG_ON(ret); | |
771ed689 | 1225 | |
efa56464 YZ |
1226 | if (root->root_key.objectid == |
1227 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1228 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1229 | num_bytes); | |
1230 | BUG_ON(ret); | |
1231 | } | |
1232 | ||
d899e052 | 1233 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1234 | cur_offset, cur_offset + num_bytes - 1, |
1235 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1236 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1237 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1238 | cur_offset = extent_end; |
1239 | if (cur_offset > end) | |
1240 | break; | |
be20aa9d | 1241 | } |
b3b4aa74 | 1242 | btrfs_release_path(path); |
80ff3856 YZ |
1243 | |
1244 | if (cur_offset <= end && cow_start == (u64)-1) | |
1245 | cow_start = cur_offset; | |
1246 | if (cow_start != (u64)-1) { | |
1247 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1248 | page_started, nr_written, 1); |
80ff3856 YZ |
1249 | BUG_ON(ret); |
1250 | } | |
1251 | ||
0cb59c99 JB |
1252 | if (nolock) { |
1253 | ret = btrfs_end_transaction_nolock(trans, root); | |
1254 | BUG_ON(ret); | |
1255 | } else { | |
1256 | ret = btrfs_end_transaction(trans, root); | |
1257 | BUG_ON(ret); | |
1258 | } | |
7ea394f1 | 1259 | btrfs_free_path(path); |
80ff3856 | 1260 | return 0; |
be20aa9d CM |
1261 | } |
1262 | ||
d352ac68 CM |
1263 | /* |
1264 | * extent_io.c call back to do delayed allocation processing | |
1265 | */ | |
c8b97818 | 1266 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1267 | u64 start, u64 end, int *page_started, |
1268 | unsigned long *nr_written) | |
be20aa9d | 1269 | { |
be20aa9d | 1270 | int ret; |
7f366cfe | 1271 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1272 | |
6cbff00f | 1273 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1274 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1275 | page_started, 1, nr_written); |
6cbff00f | 1276 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1277 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1278 | page_started, 0, nr_written); |
1e701a32 | 1279 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1280 | !(BTRFS_I(inode)->force_compress) && |
1281 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1282 | ret = cow_file_range(inode, locked_page, start, end, |
1283 | page_started, nr_written, 1); | |
be20aa9d | 1284 | else |
771ed689 | 1285 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1286 | page_started, nr_written); |
b888db2b CM |
1287 | return ret; |
1288 | } | |
1289 | ||
1bf85046 JM |
1290 | static void btrfs_split_extent_hook(struct inode *inode, |
1291 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1292 | { |
0ca1f7ce | 1293 | /* not delalloc, ignore it */ |
9ed74f2d | 1294 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1295 | return; |
9ed74f2d | 1296 | |
9e0baf60 JB |
1297 | spin_lock(&BTRFS_I(inode)->lock); |
1298 | BTRFS_I(inode)->outstanding_extents++; | |
1299 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1304 | * extents so we can keep track of new extents that are just merged onto old | |
1305 | * extents, such as when we are doing sequential writes, so we can properly | |
1306 | * account for the metadata space we'll need. | |
1307 | */ | |
1bf85046 JM |
1308 | static void btrfs_merge_extent_hook(struct inode *inode, |
1309 | struct extent_state *new, | |
1310 | struct extent_state *other) | |
9ed74f2d | 1311 | { |
9ed74f2d JB |
1312 | /* not delalloc, ignore it */ |
1313 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1314 | return; |
9ed74f2d | 1315 | |
9e0baf60 JB |
1316 | spin_lock(&BTRFS_I(inode)->lock); |
1317 | BTRFS_I(inode)->outstanding_extents--; | |
1318 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1319 | } |
1320 | ||
d352ac68 CM |
1321 | /* |
1322 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1323 | * bytes in this file, and to maintain the list of inodes that | |
1324 | * have pending delalloc work to be done. | |
1325 | */ | |
1bf85046 JM |
1326 | static void btrfs_set_bit_hook(struct inode *inode, |
1327 | struct extent_state *state, int *bits) | |
291d673e | 1328 | { |
9ed74f2d | 1329 | |
75eff68e CM |
1330 | /* |
1331 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1332 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1333 | * bit, which is only set or cleared with irqs on |
1334 | */ | |
0ca1f7ce | 1335 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1336 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1337 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1338 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
9ed74f2d | 1339 | |
9e0baf60 | 1340 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1341 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1342 | } else { |
1343 | spin_lock(&BTRFS_I(inode)->lock); | |
1344 | BTRFS_I(inode)->outstanding_extents++; | |
1345 | spin_unlock(&BTRFS_I(inode)->lock); | |
1346 | } | |
287a0ab9 | 1347 | |
75eff68e | 1348 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1349 | BTRFS_I(inode)->delalloc_bytes += len; |
1350 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1351 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1352 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1353 | &root->fs_info->delalloc_inodes); | |
1354 | } | |
75eff68e | 1355 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1356 | } |
291d673e CM |
1357 | } |
1358 | ||
d352ac68 CM |
1359 | /* |
1360 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1361 | */ | |
1bf85046 JM |
1362 | static void btrfs_clear_bit_hook(struct inode *inode, |
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; |
2cf8572d | 1373 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
bcbfce8a | 1374 | |
9e0baf60 | 1375 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1376 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1377 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1378 | spin_lock(&BTRFS_I(inode)->lock); | |
1379 | BTRFS_I(inode)->outstanding_extents--; | |
1380 | spin_unlock(&BTRFS_I(inode)->lock); | |
1381 | } | |
0ca1f7ce YZ |
1382 | |
1383 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1384 | btrfs_delalloc_release_metadata(inode, len); | |
1385 | ||
0cb59c99 JB |
1386 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1387 | && do_list) | |
0ca1f7ce | 1388 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1389 | |
75eff68e | 1390 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1391 | root->fs_info->delalloc_bytes -= len; |
1392 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1393 | ||
0cb59c99 | 1394 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1395 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1396 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1397 | } | |
75eff68e | 1398 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1399 | } |
291d673e CM |
1400 | } |
1401 | ||
d352ac68 CM |
1402 | /* |
1403 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1404 | * we don't create bios that span stripes or chunks | |
1405 | */ | |
239b14b3 | 1406 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1407 | size_t size, struct bio *bio, |
1408 | unsigned long bio_flags) | |
239b14b3 CM |
1409 | { |
1410 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1411 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1412 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1413 | u64 length = 0; |
1414 | u64 map_length; | |
239b14b3 CM |
1415 | int ret; |
1416 | ||
771ed689 CM |
1417 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1418 | return 0; | |
1419 | ||
f2d8d74d | 1420 | length = bio->bi_size; |
239b14b3 CM |
1421 | map_tree = &root->fs_info->mapping_tree; |
1422 | map_length = length; | |
cea9e445 | 1423 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1424 | &map_length, NULL, 0); |
cea9e445 | 1425 | |
d397712b | 1426 | if (map_length < length + size) |
239b14b3 | 1427 | return 1; |
411fc6bc | 1428 | return ret; |
239b14b3 CM |
1429 | } |
1430 | ||
d352ac68 CM |
1431 | /* |
1432 | * in order to insert checksums into the metadata in large chunks, | |
1433 | * we wait until bio submission time. All the pages in the bio are | |
1434 | * checksummed and sums are attached onto the ordered extent record. | |
1435 | * | |
1436 | * At IO completion time the cums attached on the ordered extent record | |
1437 | * are inserted into the btree | |
1438 | */ | |
d397712b CM |
1439 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1440 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1441 | unsigned long bio_flags, |
1442 | u64 bio_offset) | |
065631f6 | 1443 | { |
065631f6 | 1444 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1445 | int ret = 0; |
e015640f | 1446 | |
d20f7043 | 1447 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1448 | BUG_ON(ret); |
4a69a410 CM |
1449 | return 0; |
1450 | } | |
e015640f | 1451 | |
4a69a410 CM |
1452 | /* |
1453 | * in order to insert checksums into the metadata in large chunks, | |
1454 | * we wait until bio submission time. All the pages in the bio are | |
1455 | * checksummed and sums are attached onto the ordered extent record. | |
1456 | * | |
1457 | * At IO completion time the cums attached on the ordered extent record | |
1458 | * are inserted into the btree | |
1459 | */ | |
b2950863 | 1460 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1461 | int mirror_num, unsigned long bio_flags, |
1462 | u64 bio_offset) | |
4a69a410 CM |
1463 | { |
1464 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1465 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1466 | } |
1467 | ||
d352ac68 | 1468 | /* |
cad321ad CM |
1469 | * extent_io.c submission hook. This does the right thing for csum calculation |
1470 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1471 | */ |
b2950863 | 1472 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1473 | int mirror_num, unsigned long bio_flags, |
1474 | u64 bio_offset) | |
44b8bd7e CM |
1475 | { |
1476 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1477 | int ret = 0; | |
19b9bdb0 | 1478 | int skip_sum; |
44b8bd7e | 1479 | |
6cbff00f | 1480 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1481 | |
2cf8572d | 1482 | if (btrfs_is_free_space_inode(root, inode)) |
0cb59c99 JB |
1483 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2); |
1484 | else | |
1485 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
e6dcd2dc | 1486 | BUG_ON(ret); |
065631f6 | 1487 | |
7b6d91da | 1488 | if (!(rw & REQ_WRITE)) { |
d20f7043 | 1489 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1490 | return btrfs_submit_compressed_read(inode, bio, |
1491 | mirror_num, bio_flags); | |
c2db1073 TI |
1492 | } else if (!skip_sum) { |
1493 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1494 | if (ret) | |
1495 | return ret; | |
1496 | } | |
4d1b5fb4 | 1497 | goto mapit; |
19b9bdb0 | 1498 | } else if (!skip_sum) { |
17d217fe YZ |
1499 | /* csum items have already been cloned */ |
1500 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1501 | goto mapit; | |
19b9bdb0 CM |
1502 | /* we're doing a write, do the async checksumming */ |
1503 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1504 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1505 | bio_flags, bio_offset, |
1506 | __btrfs_submit_bio_start, | |
4a69a410 | 1507 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1508 | } |
1509 | ||
0b86a832 | 1510 | mapit: |
8b712842 | 1511 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1512 | } |
6885f308 | 1513 | |
d352ac68 CM |
1514 | /* |
1515 | * given a list of ordered sums record them in the inode. This happens | |
1516 | * at IO completion time based on sums calculated at bio submission time. | |
1517 | */ | |
ba1da2f4 | 1518 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1519 | struct inode *inode, u64 file_offset, |
1520 | struct list_head *list) | |
1521 | { | |
e6dcd2dc CM |
1522 | struct btrfs_ordered_sum *sum; |
1523 | ||
c6e30871 | 1524 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1525 | btrfs_csum_file_blocks(trans, |
1526 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1527 | } |
1528 | return 0; | |
1529 | } | |
1530 | ||
2ac55d41 JB |
1531 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1532 | struct extent_state **cached_state) | |
ea8c2819 | 1533 | { |
d397712b | 1534 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1535 | WARN_ON(1); |
ea8c2819 | 1536 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1537 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1538 | } |
1539 | ||
d352ac68 | 1540 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1541 | struct btrfs_writepage_fixup { |
1542 | struct page *page; | |
1543 | struct btrfs_work work; | |
1544 | }; | |
1545 | ||
b2950863 | 1546 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1547 | { |
1548 | struct btrfs_writepage_fixup *fixup; | |
1549 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1550 | struct extent_state *cached_state = NULL; |
247e743c CM |
1551 | struct page *page; |
1552 | struct inode *inode; | |
1553 | u64 page_start; | |
1554 | u64 page_end; | |
1555 | ||
1556 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1557 | page = fixup->page; | |
4a096752 | 1558 | again: |
247e743c CM |
1559 | lock_page(page); |
1560 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1561 | ClearPageChecked(page); | |
1562 | goto out_page; | |
1563 | } | |
1564 | ||
1565 | inode = page->mapping->host; | |
1566 | page_start = page_offset(page); | |
1567 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1568 | ||
2ac55d41 JB |
1569 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1570 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1571 | |
1572 | /* already ordered? We're done */ | |
8b62b72b | 1573 | if (PagePrivate2(page)) |
247e743c | 1574 | goto out; |
4a096752 CM |
1575 | |
1576 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1577 | if (ordered) { | |
2ac55d41 JB |
1578 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1579 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1580 | unlock_page(page); |
1581 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1582 | goto again; | |
1583 | } | |
247e743c | 1584 | |
0ca1f7ce | 1585 | BUG(); |
2ac55d41 | 1586 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c CM |
1587 | ClearPageChecked(page); |
1588 | out: | |
2ac55d41 JB |
1589 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1590 | &cached_state, GFP_NOFS); | |
247e743c CM |
1591 | out_page: |
1592 | unlock_page(page); | |
1593 | page_cache_release(page); | |
b897abec | 1594 | kfree(fixup); |
247e743c CM |
1595 | } |
1596 | ||
1597 | /* | |
1598 | * There are a few paths in the higher layers of the kernel that directly | |
1599 | * set the page dirty bit without asking the filesystem if it is a | |
1600 | * good idea. This causes problems because we want to make sure COW | |
1601 | * properly happens and the data=ordered rules are followed. | |
1602 | * | |
c8b97818 | 1603 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1604 | * hasn't been properly setup for IO. We kick off an async process |
1605 | * to fix it up. The async helper will wait for ordered extents, set | |
1606 | * the delalloc bit and make it safe to write the page. | |
1607 | */ | |
b2950863 | 1608 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1609 | { |
1610 | struct inode *inode = page->mapping->host; | |
1611 | struct btrfs_writepage_fixup *fixup; | |
1612 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1613 | |
8b62b72b CM |
1614 | /* this page is properly in the ordered list */ |
1615 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1616 | return 0; |
1617 | ||
1618 | if (PageChecked(page)) | |
1619 | return -EAGAIN; | |
1620 | ||
1621 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1622 | if (!fixup) | |
1623 | return -EAGAIN; | |
f421950f | 1624 | |
247e743c CM |
1625 | SetPageChecked(page); |
1626 | page_cache_get(page); | |
1627 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1628 | fixup->page = page; | |
1629 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1630 | return -EAGAIN; | |
1631 | } | |
1632 | ||
d899e052 YZ |
1633 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1634 | struct inode *inode, u64 file_pos, | |
1635 | u64 disk_bytenr, u64 disk_num_bytes, | |
1636 | u64 num_bytes, u64 ram_bytes, | |
1637 | u8 compression, u8 encryption, | |
1638 | u16 other_encoding, int extent_type) | |
1639 | { | |
1640 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1641 | struct btrfs_file_extent_item *fi; | |
1642 | struct btrfs_path *path; | |
1643 | struct extent_buffer *leaf; | |
1644 | struct btrfs_key ins; | |
1645 | u64 hint; | |
1646 | int ret; | |
1647 | ||
1648 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1649 | if (!path) |
1650 | return -ENOMEM; | |
d899e052 | 1651 | |
b9473439 | 1652 | path->leave_spinning = 1; |
a1ed835e CM |
1653 | |
1654 | /* | |
1655 | * we may be replacing one extent in the tree with another. | |
1656 | * The new extent is pinned in the extent map, and we don't want | |
1657 | * to drop it from the cache until it is completely in the btree. | |
1658 | * | |
1659 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1660 | * the caller is expected to unpin it and allow it to be merged | |
1661 | * with the others. | |
1662 | */ | |
920bbbfb YZ |
1663 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1664 | &hint, 0); | |
d899e052 YZ |
1665 | BUG_ON(ret); |
1666 | ||
33345d01 | 1667 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1668 | ins.offset = file_pos; |
1669 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1670 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1671 | BUG_ON(ret); | |
1672 | leaf = path->nodes[0]; | |
1673 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1674 | struct btrfs_file_extent_item); | |
1675 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1676 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1677 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1678 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1679 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1680 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1681 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1682 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1683 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1684 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1685 | |
1686 | btrfs_unlock_up_safe(path, 1); | |
1687 | btrfs_set_lock_blocking(leaf); | |
1688 | ||
d899e052 YZ |
1689 | btrfs_mark_buffer_dirty(leaf); |
1690 | ||
1691 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1692 | |
1693 | ins.objectid = disk_bytenr; | |
1694 | ins.offset = disk_num_bytes; | |
1695 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1696 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1697 | root->root_key.objectid, | |
33345d01 | 1698 | btrfs_ino(inode), file_pos, &ins); |
d899e052 | 1699 | BUG_ON(ret); |
d899e052 | 1700 | btrfs_free_path(path); |
b9473439 | 1701 | |
d899e052 YZ |
1702 | return 0; |
1703 | } | |
1704 | ||
5d13a98f CM |
1705 | /* |
1706 | * helper function for btrfs_finish_ordered_io, this | |
1707 | * just reads in some of the csum leaves to prime them into ram | |
1708 | * before we start the transaction. It limits the amount of btree | |
1709 | * reads required while inside the transaction. | |
1710 | */ | |
d352ac68 CM |
1711 | /* as ordered data IO finishes, this gets called so we can finish |
1712 | * an ordered extent if the range of bytes in the file it covers are | |
1713 | * fully written. | |
1714 | */ | |
211f90e6 | 1715 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1716 | { |
e6dcd2dc | 1717 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1718 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1719 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1720 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1721 | struct extent_state *cached_state = NULL; |
261507a0 | 1722 | int compress_type = 0; |
e6dcd2dc | 1723 | int ret; |
82d5902d | 1724 | bool nolock; |
e6dcd2dc | 1725 | |
5a1a3df1 JB |
1726 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1727 | end - start + 1); | |
ba1da2f4 | 1728 | if (!ret) |
e6dcd2dc | 1729 | return 0; |
e6dcd2dc | 1730 | BUG_ON(!ordered_extent); |
efd049fb | 1731 | |
2cf8572d | 1732 | nolock = btrfs_is_free_space_inode(root, inode); |
0cb59c99 | 1733 | |
c2167754 YZ |
1734 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1735 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1736 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1737 | if (!ret) { | |
0cb59c99 | 1738 | if (nolock) |
7a7eaa40 | 1739 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1740 | else |
7a7eaa40 | 1741 | trans = btrfs_join_transaction(root); |
3612b495 | 1742 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1743 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 YZ |
1744 | ret = btrfs_update_inode(trans, root, inode); |
1745 | BUG_ON(ret); | |
c2167754 YZ |
1746 | } |
1747 | goto out; | |
1748 | } | |
e6dcd2dc | 1749 | |
2ac55d41 JB |
1750 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1751 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1752 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1753 | |
0cb59c99 | 1754 | if (nolock) |
7a7eaa40 | 1755 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1756 | else |
7a7eaa40 | 1757 | trans = btrfs_join_transaction(root); |
3612b495 | 1758 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1759 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1760 | |
c8b97818 | 1761 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1762 | compress_type = ordered_extent->compress_type; |
d899e052 | 1763 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1764 | BUG_ON(compress_type); |
920bbbfb | 1765 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1766 | ordered_extent->file_offset, |
1767 | ordered_extent->file_offset + | |
1768 | ordered_extent->len); | |
1769 | BUG_ON(ret); | |
1770 | } else { | |
0af3d00b | 1771 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1772 | ret = insert_reserved_file_extent(trans, inode, |
1773 | ordered_extent->file_offset, | |
1774 | ordered_extent->start, | |
1775 | ordered_extent->disk_len, | |
1776 | ordered_extent->len, | |
1777 | ordered_extent->len, | |
261507a0 | 1778 | compress_type, 0, 0, |
d899e052 | 1779 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1780 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1781 | ordered_extent->file_offset, | |
1782 | ordered_extent->len); | |
d899e052 YZ |
1783 | BUG_ON(ret); |
1784 | } | |
2ac55d41 JB |
1785 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1786 | ordered_extent->file_offset + | |
1787 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1788 | ||
e6dcd2dc CM |
1789 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1790 | &ordered_extent->list); | |
1791 | ||
1ef30be1 | 1792 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1793 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
1ef30be1 JB |
1794 | ret = btrfs_update_inode(trans, root, inode); |
1795 | BUG_ON(ret); | |
1796 | } | |
1797 | ret = 0; | |
c2167754 | 1798 | out: |
5b0e95bf JB |
1799 | if (root != root->fs_info->tree_root) |
1800 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); | |
1801 | if (trans) { | |
1802 | if (nolock) | |
0cb59c99 | 1803 | btrfs_end_transaction_nolock(trans, root); |
5b0e95bf | 1804 | else |
0cb59c99 JB |
1805 | btrfs_end_transaction(trans, root); |
1806 | } | |
1807 | ||
e6dcd2dc CM |
1808 | /* once for us */ |
1809 | btrfs_put_ordered_extent(ordered_extent); | |
1810 | /* once for the tree */ | |
1811 | btrfs_put_ordered_extent(ordered_extent); | |
1812 | ||
e6dcd2dc CM |
1813 | return 0; |
1814 | } | |
1815 | ||
b2950863 | 1816 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1817 | struct extent_state *state, int uptodate) |
1818 | { | |
1abe9b8a | 1819 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1820 | ||
8b62b72b | 1821 | ClearPagePrivate2(page); |
211f90e6 CM |
1822 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1823 | } | |
1824 | ||
d352ac68 CM |
1825 | /* |
1826 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
1827 | * if there's a match, we allow the bio to finish. If not, the code in |
1828 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 1829 | */ |
b2950863 | 1830 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1831 | struct extent_state *state) |
07157aac | 1832 | { |
35ebb934 | 1833 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1834 | struct inode *inode = page->mapping->host; |
d1310b2e | 1835 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1836 | char *kaddr; |
aadfeb6e | 1837 | u64 private = ~(u32)0; |
07157aac | 1838 | int ret; |
ff79f819 CM |
1839 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1840 | u32 csum = ~(u32)0; | |
d1310b2e | 1841 | |
d20f7043 CM |
1842 | if (PageChecked(page)) { |
1843 | ClearPageChecked(page); | |
1844 | goto good; | |
1845 | } | |
6cbff00f CH |
1846 | |
1847 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 1848 | goto good; |
17d217fe YZ |
1849 | |
1850 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1851 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1852 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1853 | GFP_NOFS); | |
b6cda9bc | 1854 | return 0; |
17d217fe | 1855 | } |
d20f7043 | 1856 | |
c2e639f0 | 1857 | if (state && state->start == start) { |
70dec807 CM |
1858 | private = state->private; |
1859 | ret = 0; | |
1860 | } else { | |
1861 | ret = get_state_private(io_tree, start, &private); | |
1862 | } | |
9ab86c8e | 1863 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1864 | if (ret) |
07157aac | 1865 | goto zeroit; |
d397712b | 1866 | |
ff79f819 CM |
1867 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1868 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1869 | if (csum != private) |
07157aac | 1870 | goto zeroit; |
d397712b | 1871 | |
9ab86c8e | 1872 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1873 | good: |
07157aac CM |
1874 | return 0; |
1875 | ||
1876 | zeroit: | |
945d8962 | 1877 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
1878 | "private %llu\n", |
1879 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
1880 | (unsigned long long)start, csum, |
1881 | (unsigned long long)private); | |
db94535d CM |
1882 | memset(kaddr + offset, 1, end - start + 1); |
1883 | flush_dcache_page(page); | |
9ab86c8e | 1884 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1885 | if (private == 0) |
1886 | return 0; | |
7e38326f | 1887 | return -EIO; |
07157aac | 1888 | } |
b888db2b | 1889 | |
24bbcf04 YZ |
1890 | struct delayed_iput { |
1891 | struct list_head list; | |
1892 | struct inode *inode; | |
1893 | }; | |
1894 | ||
1895 | void btrfs_add_delayed_iput(struct inode *inode) | |
1896 | { | |
1897 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
1898 | struct delayed_iput *delayed; | |
1899 | ||
1900 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
1901 | return; | |
1902 | ||
1903 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
1904 | delayed->inode = inode; | |
1905 | ||
1906 | spin_lock(&fs_info->delayed_iput_lock); | |
1907 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
1908 | spin_unlock(&fs_info->delayed_iput_lock); | |
1909 | } | |
1910 | ||
1911 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
1912 | { | |
1913 | LIST_HEAD(list); | |
1914 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1915 | struct delayed_iput *delayed; | |
1916 | int empty; | |
1917 | ||
1918 | spin_lock(&fs_info->delayed_iput_lock); | |
1919 | empty = list_empty(&fs_info->delayed_iputs); | |
1920 | spin_unlock(&fs_info->delayed_iput_lock); | |
1921 | if (empty) | |
1922 | return; | |
1923 | ||
1924 | down_read(&root->fs_info->cleanup_work_sem); | |
1925 | spin_lock(&fs_info->delayed_iput_lock); | |
1926 | list_splice_init(&fs_info->delayed_iputs, &list); | |
1927 | spin_unlock(&fs_info->delayed_iput_lock); | |
1928 | ||
1929 | while (!list_empty(&list)) { | |
1930 | delayed = list_entry(list.next, struct delayed_iput, list); | |
1931 | list_del(&delayed->list); | |
1932 | iput(delayed->inode); | |
1933 | kfree(delayed); | |
1934 | } | |
1935 | up_read(&root->fs_info->cleanup_work_sem); | |
1936 | } | |
1937 | ||
d68fc57b YZ |
1938 | enum btrfs_orphan_cleanup_state { |
1939 | ORPHAN_CLEANUP_STARTED = 1, | |
1940 | ORPHAN_CLEANUP_DONE = 2, | |
1941 | }; | |
1942 | ||
1943 | /* | |
1944 | * This is called in transaction commmit time. If there are no orphan | |
1945 | * files in the subvolume, it removes orphan item and frees block_rsv | |
1946 | * structure. | |
1947 | */ | |
1948 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
1949 | struct btrfs_root *root) | |
1950 | { | |
1951 | int ret; | |
1952 | ||
1953 | if (!list_empty(&root->orphan_list) || | |
1954 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
1955 | return; | |
1956 | ||
1957 | if (root->orphan_item_inserted && | |
1958 | btrfs_root_refs(&root->root_item) > 0) { | |
1959 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
1960 | root->root_key.objectid); | |
1961 | BUG_ON(ret); | |
1962 | root->orphan_item_inserted = 0; | |
1963 | } | |
1964 | ||
1965 | if (root->orphan_block_rsv) { | |
1966 | WARN_ON(root->orphan_block_rsv->size > 0); | |
1967 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
1968 | root->orphan_block_rsv = NULL; | |
1969 | } | |
1970 | } | |
1971 | ||
7b128766 JB |
1972 | /* |
1973 | * This creates an orphan entry for the given inode in case something goes | |
1974 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
1975 | * |
1976 | * NOTE: caller of this function should reserve 5 units of metadata for | |
1977 | * this function. | |
7b128766 JB |
1978 | */ |
1979 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1980 | { | |
1981 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
1982 | struct btrfs_block_rsv *block_rsv = NULL; |
1983 | int reserve = 0; | |
1984 | int insert = 0; | |
1985 | int ret; | |
7b128766 | 1986 | |
d68fc57b YZ |
1987 | if (!root->orphan_block_rsv) { |
1988 | block_rsv = btrfs_alloc_block_rsv(root); | |
b532402e TI |
1989 | if (!block_rsv) |
1990 | return -ENOMEM; | |
d68fc57b | 1991 | } |
7b128766 | 1992 | |
d68fc57b YZ |
1993 | spin_lock(&root->orphan_lock); |
1994 | if (!root->orphan_block_rsv) { | |
1995 | root->orphan_block_rsv = block_rsv; | |
1996 | } else if (block_rsv) { | |
1997 | btrfs_free_block_rsv(root, block_rsv); | |
1998 | block_rsv = NULL; | |
7b128766 | 1999 | } |
7b128766 | 2000 | |
d68fc57b YZ |
2001 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2002 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2003 | #if 0 | |
2004 | /* | |
2005 | * For proper ENOSPC handling, we should do orphan | |
2006 | * cleanup when mounting. But this introduces backward | |
2007 | * compatibility issue. | |
2008 | */ | |
2009 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2010 | insert = 2; | |
2011 | else | |
2012 | insert = 1; | |
2013 | #endif | |
2014 | insert = 1; | |
7b128766 JB |
2015 | } |
2016 | ||
d68fc57b YZ |
2017 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2018 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2019 | reserve = 1; | |
2020 | } | |
2021 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2022 | |
d68fc57b YZ |
2023 | /* grab metadata reservation from transaction handle */ |
2024 | if (reserve) { | |
2025 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2026 | BUG_ON(ret); | |
2027 | } | |
7b128766 | 2028 | |
d68fc57b YZ |
2029 | /* insert an orphan item to track this unlinked/truncated file */ |
2030 | if (insert >= 1) { | |
33345d01 | 2031 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
d68fc57b YZ |
2032 | BUG_ON(ret); |
2033 | } | |
2034 | ||
2035 | /* insert an orphan item to track subvolume contains orphan files */ | |
2036 | if (insert >= 2) { | |
2037 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2038 | root->root_key.objectid); | |
2039 | BUG_ON(ret); | |
2040 | } | |
2041 | return 0; | |
7b128766 JB |
2042 | } |
2043 | ||
2044 | /* | |
2045 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2046 | * item for this particular inode. | |
2047 | */ | |
2048 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2049 | { | |
2050 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2051 | int delete_item = 0; |
2052 | int release_rsv = 0; | |
7b128766 JB |
2053 | int ret = 0; |
2054 | ||
d68fc57b YZ |
2055 | spin_lock(&root->orphan_lock); |
2056 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2057 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2058 | delete_item = 1; | |
7b128766 JB |
2059 | } |
2060 | ||
d68fc57b YZ |
2061 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2062 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2063 | release_rsv = 1; | |
7b128766 | 2064 | } |
d68fc57b | 2065 | spin_unlock(&root->orphan_lock); |
7b128766 | 2066 | |
d68fc57b | 2067 | if (trans && delete_item) { |
33345d01 | 2068 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
d68fc57b YZ |
2069 | BUG_ON(ret); |
2070 | } | |
7b128766 | 2071 | |
d68fc57b YZ |
2072 | if (release_rsv) |
2073 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2074 | |
d68fc57b | 2075 | return 0; |
7b128766 JB |
2076 | } |
2077 | ||
2078 | /* | |
2079 | * this cleans up any orphans that may be left on the list from the last use | |
2080 | * of this root. | |
2081 | */ | |
66b4ffd1 | 2082 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2083 | { |
2084 | struct btrfs_path *path; | |
2085 | struct extent_buffer *leaf; | |
7b128766 JB |
2086 | struct btrfs_key key, found_key; |
2087 | struct btrfs_trans_handle *trans; | |
2088 | struct inode *inode; | |
8f6d7f4f | 2089 | u64 last_objectid = 0; |
7b128766 JB |
2090 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2091 | ||
d68fc57b | 2092 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2093 | return 0; |
c71bf099 YZ |
2094 | |
2095 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2096 | if (!path) { |
2097 | ret = -ENOMEM; | |
2098 | goto out; | |
2099 | } | |
7b128766 JB |
2100 | path->reada = -1; |
2101 | ||
2102 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2103 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2104 | key.offset = (u64)-1; | |
2105 | ||
7b128766 JB |
2106 | while (1) { |
2107 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2108 | if (ret < 0) |
2109 | goto out; | |
7b128766 JB |
2110 | |
2111 | /* | |
2112 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2113 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2114 | * find the key and see if we have stuff that matches |
2115 | */ | |
2116 | if (ret > 0) { | |
66b4ffd1 | 2117 | ret = 0; |
7b128766 JB |
2118 | if (path->slots[0] == 0) |
2119 | break; | |
2120 | path->slots[0]--; | |
2121 | } | |
2122 | ||
2123 | /* pull out the item */ | |
2124 | leaf = path->nodes[0]; | |
7b128766 JB |
2125 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2126 | ||
2127 | /* make sure the item matches what we want */ | |
2128 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2129 | break; | |
2130 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2131 | break; | |
2132 | ||
2133 | /* release the path since we're done with it */ | |
b3b4aa74 | 2134 | btrfs_release_path(path); |
7b128766 JB |
2135 | |
2136 | /* | |
2137 | * this is where we are basically btrfs_lookup, without the | |
2138 | * crossing root thing. we store the inode number in the | |
2139 | * offset of the orphan item. | |
2140 | */ | |
8f6d7f4f JB |
2141 | |
2142 | if (found_key.offset == last_objectid) { | |
2143 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2144 | "stopping orphan cleanup\n"); | |
2145 | ret = -EINVAL; | |
2146 | goto out; | |
2147 | } | |
2148 | ||
2149 | last_objectid = found_key.offset; | |
2150 | ||
5d4f98a2 YZ |
2151 | found_key.objectid = found_key.offset; |
2152 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2153 | found_key.offset = 0; | |
73f73415 | 2154 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2155 | ret = PTR_RET(inode); |
2156 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2157 | goto out; |
7b128766 JB |
2158 | |
2159 | /* | |
a8c9e576 JB |
2160 | * Inode is already gone but the orphan item is still there, |
2161 | * kill the orphan item. | |
7b128766 | 2162 | */ |
a8c9e576 JB |
2163 | if (ret == -ESTALE) { |
2164 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2165 | if (IS_ERR(trans)) { |
2166 | ret = PTR_ERR(trans); | |
2167 | goto out; | |
2168 | } | |
a8c9e576 JB |
2169 | ret = btrfs_del_orphan_item(trans, root, |
2170 | found_key.objectid); | |
2171 | BUG_ON(ret); | |
5b21f2ed | 2172 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2173 | continue; |
2174 | } | |
2175 | ||
a8c9e576 JB |
2176 | /* |
2177 | * add this inode to the orphan list so btrfs_orphan_del does | |
2178 | * the proper thing when we hit it | |
2179 | */ | |
2180 | spin_lock(&root->orphan_lock); | |
2181 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2182 | spin_unlock(&root->orphan_lock); | |
2183 | ||
7b128766 JB |
2184 | /* if we have links, this was a truncate, lets do that */ |
2185 | if (inode->i_nlink) { | |
a41ad394 JB |
2186 | if (!S_ISREG(inode->i_mode)) { |
2187 | WARN_ON(1); | |
2188 | iput(inode); | |
2189 | continue; | |
2190 | } | |
7b128766 | 2191 | nr_truncate++; |
66b4ffd1 | 2192 | ret = btrfs_truncate(inode); |
7b128766 JB |
2193 | } else { |
2194 | nr_unlink++; | |
2195 | } | |
2196 | ||
2197 | /* this will do delete_inode and everything for us */ | |
2198 | iput(inode); | |
66b4ffd1 JB |
2199 | if (ret) |
2200 | goto out; | |
7b128766 | 2201 | } |
d68fc57b YZ |
2202 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2203 | ||
2204 | if (root->orphan_block_rsv) | |
2205 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2206 | (u64)-1); | |
2207 | ||
2208 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2209 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2210 | if (!IS_ERR(trans)) |
2211 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2212 | } |
7b128766 JB |
2213 | |
2214 | if (nr_unlink) | |
2215 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2216 | if (nr_truncate) | |
2217 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2218 | |
2219 | out: | |
2220 | if (ret) | |
2221 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2222 | btrfs_free_path(path); | |
2223 | return ret; | |
7b128766 JB |
2224 | } |
2225 | ||
46a53cca CM |
2226 | /* |
2227 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2228 | * don't find any xattrs, we know there can't be any acls. | |
2229 | * | |
2230 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2231 | */ | |
2232 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2233 | int slot, u64 objectid) | |
2234 | { | |
2235 | u32 nritems = btrfs_header_nritems(leaf); | |
2236 | struct btrfs_key found_key; | |
2237 | int scanned = 0; | |
2238 | ||
2239 | slot++; | |
2240 | while (slot < nritems) { | |
2241 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2242 | ||
2243 | /* we found a different objectid, there must not be acls */ | |
2244 | if (found_key.objectid != objectid) | |
2245 | return 0; | |
2246 | ||
2247 | /* we found an xattr, assume we've got an acl */ | |
2248 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2249 | return 1; | |
2250 | ||
2251 | /* | |
2252 | * we found a key greater than an xattr key, there can't | |
2253 | * be any acls later on | |
2254 | */ | |
2255 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2256 | return 0; | |
2257 | ||
2258 | slot++; | |
2259 | scanned++; | |
2260 | ||
2261 | /* | |
2262 | * it goes inode, inode backrefs, xattrs, extents, | |
2263 | * so if there are a ton of hard links to an inode there can | |
2264 | * be a lot of backrefs. Don't waste time searching too hard, | |
2265 | * this is just an optimization | |
2266 | */ | |
2267 | if (scanned >= 8) | |
2268 | break; | |
2269 | } | |
2270 | /* we hit the end of the leaf before we found an xattr or | |
2271 | * something larger than an xattr. We have to assume the inode | |
2272 | * has acls | |
2273 | */ | |
2274 | return 1; | |
2275 | } | |
2276 | ||
d352ac68 CM |
2277 | /* |
2278 | * read an inode from the btree into the in-memory inode | |
2279 | */ | |
5d4f98a2 | 2280 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2281 | { |
2282 | struct btrfs_path *path; | |
5f39d397 | 2283 | struct extent_buffer *leaf; |
39279cc3 | 2284 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2285 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2286 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2287 | struct btrfs_key location; | |
46a53cca | 2288 | int maybe_acls; |
618e21d5 | 2289 | u32 rdev; |
39279cc3 | 2290 | int ret; |
2f7e33d4 MX |
2291 | bool filled = false; |
2292 | ||
2293 | ret = btrfs_fill_inode(inode, &rdev); | |
2294 | if (!ret) | |
2295 | filled = true; | |
39279cc3 CM |
2296 | |
2297 | path = btrfs_alloc_path(); | |
1748f843 MF |
2298 | if (!path) |
2299 | goto make_bad; | |
2300 | ||
d90c7321 | 2301 | path->leave_spinning = 1; |
39279cc3 | 2302 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2303 | |
39279cc3 | 2304 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2305 | if (ret) |
39279cc3 | 2306 | goto make_bad; |
39279cc3 | 2307 | |
5f39d397 | 2308 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2309 | |
2310 | if (filled) | |
2311 | goto cache_acl; | |
2312 | ||
5f39d397 CM |
2313 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2314 | struct btrfs_inode_item); | |
5f39d397 CM |
2315 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
2316 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2317 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2318 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2319 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2320 | |
2321 | tspec = btrfs_inode_atime(inode_item); | |
2322 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2323 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2324 | ||
2325 | tspec = btrfs_inode_mtime(inode_item); | |
2326 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2327 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2328 | ||
2329 | tspec = btrfs_inode_ctime(inode_item); | |
2330 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2331 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2332 | ||
a76a3cd4 | 2333 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2334 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2335 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2336 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2337 | inode->i_rdev = 0; |
5f39d397 CM |
2338 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2339 | ||
aec7477b | 2340 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2341 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2342 | cache_acl: |
46a53cca CM |
2343 | /* |
2344 | * try to precache a NULL acl entry for files that don't have | |
2345 | * any xattrs or acls | |
2346 | */ | |
33345d01 LZ |
2347 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2348 | btrfs_ino(inode)); | |
72c04902 AV |
2349 | if (!maybe_acls) |
2350 | cache_no_acl(inode); | |
46a53cca | 2351 | |
39279cc3 | 2352 | btrfs_free_path(path); |
39279cc3 | 2353 | |
39279cc3 | 2354 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2355 | case S_IFREG: |
2356 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2357 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2358 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2359 | inode->i_fop = &btrfs_file_operations; |
2360 | inode->i_op = &btrfs_file_inode_operations; | |
2361 | break; | |
2362 | case S_IFDIR: | |
2363 | inode->i_fop = &btrfs_dir_file_operations; | |
2364 | if (root == root->fs_info->tree_root) | |
2365 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2366 | else | |
2367 | inode->i_op = &btrfs_dir_inode_operations; | |
2368 | break; | |
2369 | case S_IFLNK: | |
2370 | inode->i_op = &btrfs_symlink_inode_operations; | |
2371 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2372 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2373 | break; |
618e21d5 | 2374 | default: |
0279b4cd | 2375 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2376 | init_special_inode(inode, inode->i_mode, rdev); |
2377 | break; | |
39279cc3 | 2378 | } |
6cbff00f CH |
2379 | |
2380 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2381 | return; |
2382 | ||
2383 | make_bad: | |
39279cc3 | 2384 | btrfs_free_path(path); |
39279cc3 CM |
2385 | make_bad_inode(inode); |
2386 | } | |
2387 | ||
d352ac68 CM |
2388 | /* |
2389 | * given a leaf and an inode, copy the inode fields into the leaf | |
2390 | */ | |
e02119d5 CM |
2391 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2392 | struct extent_buffer *leaf, | |
5f39d397 | 2393 | struct btrfs_inode_item *item, |
39279cc3 CM |
2394 | struct inode *inode) |
2395 | { | |
5f39d397 CM |
2396 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2397 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2398 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2399 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2400 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2401 | ||
2402 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2403 | inode->i_atime.tv_sec); | |
2404 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2405 | inode->i_atime.tv_nsec); | |
2406 | ||
2407 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2408 | inode->i_mtime.tv_sec); | |
2409 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2410 | inode->i_mtime.tv_nsec); | |
2411 | ||
2412 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2413 | inode->i_ctime.tv_sec); | |
2414 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2415 | inode->i_ctime.tv_nsec); | |
2416 | ||
a76a3cd4 | 2417 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2418 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2419 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2420 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2421 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2422 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2423 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2424 | } |
2425 | ||
d352ac68 CM |
2426 | /* |
2427 | * copy everything in the in-memory inode into the btree. | |
2428 | */ | |
d397712b CM |
2429 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2430 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2431 | { |
2432 | struct btrfs_inode_item *inode_item; | |
2433 | struct btrfs_path *path; | |
5f39d397 | 2434 | struct extent_buffer *leaf; |
39279cc3 CM |
2435 | int ret; |
2436 | ||
16cdcec7 | 2437 | /* |
149e2d76 MX |
2438 | * If the inode is a free space inode, we can deadlock during commit |
2439 | * if we put it into the delayed code. | |
2440 | * | |
2441 | * The data relocation inode should also be directly updated | |
2442 | * without delay | |
16cdcec7 | 2443 | */ |
2cf8572d | 2444 | if (!btrfs_is_free_space_inode(root, inode) |
149e2d76 | 2445 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
16cdcec7 MX |
2446 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2447 | if (!ret) | |
2448 | btrfs_set_inode_last_trans(trans, inode); | |
2449 | return ret; | |
2450 | } | |
2451 | ||
39279cc3 | 2452 | path = btrfs_alloc_path(); |
16cdcec7 MX |
2453 | if (!path) |
2454 | return -ENOMEM; | |
2455 | ||
b9473439 | 2456 | path->leave_spinning = 1; |
16cdcec7 MX |
2457 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2458 | 1); | |
39279cc3 CM |
2459 | if (ret) { |
2460 | if (ret > 0) | |
2461 | ret = -ENOENT; | |
2462 | goto failed; | |
2463 | } | |
2464 | ||
b4ce94de | 2465 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2466 | leaf = path->nodes[0]; |
2467 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2468 | struct btrfs_inode_item); |
39279cc3 | 2469 | |
e02119d5 | 2470 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2471 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2472 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2473 | ret = 0; |
2474 | failed: | |
39279cc3 CM |
2475 | btrfs_free_path(path); |
2476 | return ret; | |
2477 | } | |
2478 | ||
d352ac68 CM |
2479 | /* |
2480 | * unlink helper that gets used here in inode.c and in the tree logging | |
2481 | * recovery code. It remove a link in a directory with a given name, and | |
2482 | * also drops the back refs in the inode to the directory | |
2483 | */ | |
92986796 AV |
2484 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2485 | struct btrfs_root *root, | |
2486 | struct inode *dir, struct inode *inode, | |
2487 | const char *name, int name_len) | |
39279cc3 CM |
2488 | { |
2489 | struct btrfs_path *path; | |
39279cc3 | 2490 | int ret = 0; |
5f39d397 | 2491 | struct extent_buffer *leaf; |
39279cc3 | 2492 | struct btrfs_dir_item *di; |
5f39d397 | 2493 | struct btrfs_key key; |
aec7477b | 2494 | u64 index; |
33345d01 LZ |
2495 | u64 ino = btrfs_ino(inode); |
2496 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2497 | |
2498 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2499 | if (!path) { |
2500 | ret = -ENOMEM; | |
554233a6 | 2501 | goto out; |
54aa1f4d CM |
2502 | } |
2503 | ||
b9473439 | 2504 | path->leave_spinning = 1; |
33345d01 | 2505 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2506 | name, name_len, -1); |
2507 | if (IS_ERR(di)) { | |
2508 | ret = PTR_ERR(di); | |
2509 | goto err; | |
2510 | } | |
2511 | if (!di) { | |
2512 | ret = -ENOENT; | |
2513 | goto err; | |
2514 | } | |
5f39d397 CM |
2515 | leaf = path->nodes[0]; |
2516 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2517 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2518 | if (ret) |
2519 | goto err; | |
b3b4aa74 | 2520 | btrfs_release_path(path); |
39279cc3 | 2521 | |
33345d01 LZ |
2522 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2523 | dir_ino, &index); | |
aec7477b | 2524 | if (ret) { |
d397712b | 2525 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2526 | "inode %llu parent %llu\n", name_len, name, |
2527 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
aec7477b JB |
2528 | goto err; |
2529 | } | |
2530 | ||
16cdcec7 MX |
2531 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
2532 | if (ret) | |
39279cc3 | 2533 | goto err; |
39279cc3 | 2534 | |
e02119d5 | 2535 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2536 | inode, dir_ino); |
49eb7e46 | 2537 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2538 | |
2539 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2540 | dir, index); | |
6418c961 CM |
2541 | if (ret == -ENOENT) |
2542 | ret = 0; | |
39279cc3 CM |
2543 | err: |
2544 | btrfs_free_path(path); | |
e02119d5 CM |
2545 | if (ret) |
2546 | goto out; | |
2547 | ||
2548 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2549 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2550 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2551 | out: |
39279cc3 CM |
2552 | return ret; |
2553 | } | |
2554 | ||
92986796 AV |
2555 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2556 | struct btrfs_root *root, | |
2557 | struct inode *dir, struct inode *inode, | |
2558 | const char *name, int name_len) | |
2559 | { | |
2560 | int ret; | |
2561 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2562 | if (!ret) { | |
2563 | btrfs_drop_nlink(inode); | |
2564 | ret = btrfs_update_inode(trans, root, inode); | |
2565 | } | |
2566 | return ret; | |
2567 | } | |
2568 | ||
2569 | ||
a22285a6 YZ |
2570 | /* helper to check if there is any shared block in the path */ |
2571 | static int check_path_shared(struct btrfs_root *root, | |
2572 | struct btrfs_path *path) | |
39279cc3 | 2573 | { |
a22285a6 YZ |
2574 | struct extent_buffer *eb; |
2575 | int level; | |
0e4dcbef | 2576 | u64 refs = 1; |
5df6a9f6 | 2577 | |
a22285a6 | 2578 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2579 | int ret; |
2580 | ||
a22285a6 YZ |
2581 | if (!path->nodes[level]) |
2582 | break; | |
2583 | eb = path->nodes[level]; | |
2584 | if (!btrfs_block_can_be_shared(root, eb)) | |
2585 | continue; | |
2586 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2587 | &refs, NULL); | |
2588 | if (refs > 1) | |
2589 | return 1; | |
5df6a9f6 | 2590 | } |
dedefd72 | 2591 | return 0; |
39279cc3 CM |
2592 | } |
2593 | ||
a22285a6 YZ |
2594 | /* |
2595 | * helper to start transaction for unlink and rmdir. | |
2596 | * | |
2597 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2598 | * so in enospc case, we should make sure they will free space before | |
2599 | * allowing them to use the global metadata reservation. | |
2600 | */ | |
2601 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2602 | struct dentry *dentry) | |
4df27c4d | 2603 | { |
39279cc3 | 2604 | struct btrfs_trans_handle *trans; |
a22285a6 | 2605 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2606 | struct btrfs_path *path; |
a22285a6 | 2607 | struct btrfs_inode_ref *ref; |
4df27c4d | 2608 | struct btrfs_dir_item *di; |
7b128766 | 2609 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2610 | u64 index; |
a22285a6 YZ |
2611 | int check_link = 1; |
2612 | int err = -ENOSPC; | |
4df27c4d | 2613 | int ret; |
33345d01 LZ |
2614 | u64 ino = btrfs_ino(inode); |
2615 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2616 | |
e70bea5f JB |
2617 | /* |
2618 | * 1 for the possible orphan item | |
2619 | * 1 for the dir item | |
2620 | * 1 for the dir index | |
2621 | * 1 for the inode ref | |
2622 | * 1 for the inode ref in the tree log | |
2623 | * 2 for the dir entries in the log | |
2624 | * 1 for the inode | |
2625 | */ | |
2626 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2627 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2628 | return trans; | |
4df27c4d | 2629 | |
33345d01 | 2630 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2631 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2632 | |
a22285a6 YZ |
2633 | /* check if there is someone else holds reference */ |
2634 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2635 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2636 | |
a22285a6 YZ |
2637 | if (atomic_read(&inode->i_count) > 2) |
2638 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2639 | |
a22285a6 YZ |
2640 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2641 | return ERR_PTR(-ENOSPC); | |
2642 | ||
2643 | path = btrfs_alloc_path(); | |
2644 | if (!path) { | |
2645 | root->fs_info->enospc_unlink = 0; | |
2646 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2647 | } |
2648 | ||
3880a1b4 JB |
2649 | /* 1 for the orphan item */ |
2650 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2651 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2652 | btrfs_free_path(path); |
2653 | root->fs_info->enospc_unlink = 0; | |
2654 | return trans; | |
2655 | } | |
4df27c4d | 2656 | |
a22285a6 YZ |
2657 | path->skip_locking = 1; |
2658 | path->search_commit_root = 1; | |
4df27c4d | 2659 | |
a22285a6 YZ |
2660 | ret = btrfs_lookup_inode(trans, root, path, |
2661 | &BTRFS_I(dir)->location, 0); | |
2662 | if (ret < 0) { | |
2663 | err = ret; | |
2664 | goto out; | |
2665 | } | |
2666 | if (ret == 0) { | |
2667 | if (check_path_shared(root, path)) | |
2668 | goto out; | |
2669 | } else { | |
2670 | check_link = 0; | |
5df6a9f6 | 2671 | } |
b3b4aa74 | 2672 | btrfs_release_path(path); |
a22285a6 YZ |
2673 | |
2674 | ret = btrfs_lookup_inode(trans, root, path, | |
2675 | &BTRFS_I(inode)->location, 0); | |
2676 | if (ret < 0) { | |
2677 | err = ret; | |
2678 | goto out; | |
2679 | } | |
2680 | if (ret == 0) { | |
2681 | if (check_path_shared(root, path)) | |
2682 | goto out; | |
2683 | } else { | |
2684 | check_link = 0; | |
2685 | } | |
b3b4aa74 | 2686 | btrfs_release_path(path); |
a22285a6 YZ |
2687 | |
2688 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2689 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2690 | ino, (u64)-1, 0); |
a22285a6 YZ |
2691 | if (ret < 0) { |
2692 | err = ret; | |
2693 | goto out; | |
2694 | } | |
2695 | BUG_ON(ret == 0); | |
2696 | if (check_path_shared(root, path)) | |
2697 | goto out; | |
b3b4aa74 | 2698 | btrfs_release_path(path); |
a22285a6 YZ |
2699 | } |
2700 | ||
2701 | if (!check_link) { | |
2702 | err = 0; | |
2703 | goto out; | |
2704 | } | |
2705 | ||
33345d01 | 2706 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2707 | dentry->d_name.name, dentry->d_name.len, 0); |
2708 | if (IS_ERR(di)) { | |
2709 | err = PTR_ERR(di); | |
2710 | goto out; | |
2711 | } | |
2712 | if (di) { | |
2713 | if (check_path_shared(root, path)) | |
2714 | goto out; | |
2715 | } else { | |
2716 | err = 0; | |
2717 | goto out; | |
2718 | } | |
b3b4aa74 | 2719 | btrfs_release_path(path); |
a22285a6 YZ |
2720 | |
2721 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2722 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 2723 | ino, dir_ino, 0); |
a22285a6 YZ |
2724 | if (IS_ERR(ref)) { |
2725 | err = PTR_ERR(ref); | |
2726 | goto out; | |
2727 | } | |
2728 | BUG_ON(!ref); | |
2729 | if (check_path_shared(root, path)) | |
2730 | goto out; | |
2731 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 2732 | btrfs_release_path(path); |
a22285a6 | 2733 | |
16cdcec7 MX |
2734 | /* |
2735 | * This is a commit root search, if we can lookup inode item and other | |
2736 | * relative items in the commit root, it means the transaction of | |
2737 | * dir/file creation has been committed, and the dir index item that we | |
2738 | * delay to insert has also been inserted into the commit root. So | |
2739 | * we needn't worry about the delayed insertion of the dir index item | |
2740 | * here. | |
2741 | */ | |
33345d01 | 2742 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
2743 | dentry->d_name.name, dentry->d_name.len, 0); |
2744 | if (IS_ERR(di)) { | |
2745 | err = PTR_ERR(di); | |
2746 | goto out; | |
2747 | } | |
2748 | BUG_ON(ret == -ENOENT); | |
2749 | if (check_path_shared(root, path)) | |
2750 | goto out; | |
2751 | ||
2752 | err = 0; | |
2753 | out: | |
2754 | btrfs_free_path(path); | |
3880a1b4 JB |
2755 | /* Migrate the orphan reservation over */ |
2756 | if (!err) | |
2757 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
2758 | &root->fs_info->global_block_rsv, | |
5a77d76c | 2759 | trans->bytes_reserved); |
3880a1b4 | 2760 | |
a22285a6 YZ |
2761 | if (err) { |
2762 | btrfs_end_transaction(trans, root); | |
2763 | root->fs_info->enospc_unlink = 0; | |
2764 | return ERR_PTR(err); | |
2765 | } | |
2766 | ||
2767 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2768 | return trans; | |
2769 | } | |
2770 | ||
2771 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2772 | struct btrfs_root *root) | |
2773 | { | |
2774 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
5a77d76c JB |
2775 | btrfs_block_rsv_release(root, trans->block_rsv, |
2776 | trans->bytes_reserved); | |
2777 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
2778 | BUG_ON(!root->fs_info->enospc_unlink); |
2779 | root->fs_info->enospc_unlink = 0; | |
2780 | } | |
2781 | btrfs_end_transaction_throttle(trans, root); | |
2782 | } | |
2783 | ||
2784 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2785 | { | |
2786 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2787 | struct btrfs_trans_handle *trans; | |
2788 | struct inode *inode = dentry->d_inode; | |
2789 | int ret; | |
2790 | unsigned long nr = 0; | |
2791 | ||
2792 | trans = __unlink_start_trans(dir, dentry); | |
2793 | if (IS_ERR(trans)) | |
2794 | return PTR_ERR(trans); | |
5f39d397 | 2795 | |
12fcfd22 CM |
2796 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
2797 | ||
e02119d5 CM |
2798 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2799 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
2800 | if (ret) |
2801 | goto out; | |
7b128766 | 2802 | |
a22285a6 | 2803 | if (inode->i_nlink == 0) { |
7b128766 | 2804 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
2805 | if (ret) |
2806 | goto out; | |
a22285a6 | 2807 | } |
7b128766 | 2808 | |
b532402e | 2809 | out: |
d3c2fdcf | 2810 | nr = trans->blocks_used; |
a22285a6 | 2811 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2812 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2813 | return ret; |
2814 | } | |
2815 | ||
4df27c4d YZ |
2816 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2817 | struct btrfs_root *root, | |
2818 | struct inode *dir, u64 objectid, | |
2819 | const char *name, int name_len) | |
2820 | { | |
2821 | struct btrfs_path *path; | |
2822 | struct extent_buffer *leaf; | |
2823 | struct btrfs_dir_item *di; | |
2824 | struct btrfs_key key; | |
2825 | u64 index; | |
2826 | int ret; | |
33345d01 | 2827 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
2828 | |
2829 | path = btrfs_alloc_path(); | |
2830 | if (!path) | |
2831 | return -ENOMEM; | |
2832 | ||
33345d01 | 2833 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 2834 | name, name_len, -1); |
c704005d | 2835 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
2836 | |
2837 | leaf = path->nodes[0]; | |
2838 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2839 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2840 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2841 | BUG_ON(ret); | |
b3b4aa74 | 2842 | btrfs_release_path(path); |
4df27c4d YZ |
2843 | |
2844 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2845 | objectid, root->root_key.objectid, | |
33345d01 | 2846 | dir_ino, &index, name, name_len); |
4df27c4d YZ |
2847 | if (ret < 0) { |
2848 | BUG_ON(ret != -ENOENT); | |
33345d01 | 2849 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 2850 | name, name_len); |
c704005d | 2851 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
2852 | |
2853 | leaf = path->nodes[0]; | |
2854 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2855 | btrfs_release_path(path); |
4df27c4d YZ |
2856 | index = key.offset; |
2857 | } | |
945d8962 | 2858 | btrfs_release_path(path); |
4df27c4d | 2859 | |
16cdcec7 | 2860 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
4df27c4d | 2861 | BUG_ON(ret); |
4df27c4d YZ |
2862 | |
2863 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2864 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2865 | ret = btrfs_update_inode(trans, root, dir); | |
2866 | BUG_ON(ret); | |
4df27c4d | 2867 | |
71d7aed0 | 2868 | btrfs_free_path(path); |
4df27c4d YZ |
2869 | return 0; |
2870 | } | |
2871 | ||
39279cc3 CM |
2872 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2873 | { | |
2874 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2875 | int err = 0; |
39279cc3 | 2876 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 2877 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2878 | unsigned long nr = 0; |
39279cc3 | 2879 | |
3394e160 | 2880 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 2881 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2882 | return -ENOTEMPTY; |
2883 | ||
a22285a6 YZ |
2884 | trans = __unlink_start_trans(dir, dentry); |
2885 | if (IS_ERR(trans)) | |
5df6a9f6 | 2886 | return PTR_ERR(trans); |
5df6a9f6 | 2887 | |
33345d01 | 2888 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
2889 | err = btrfs_unlink_subvol(trans, root, dir, |
2890 | BTRFS_I(inode)->location.objectid, | |
2891 | dentry->d_name.name, | |
2892 | dentry->d_name.len); | |
2893 | goto out; | |
2894 | } | |
2895 | ||
7b128766 JB |
2896 | err = btrfs_orphan_add(trans, inode); |
2897 | if (err) | |
4df27c4d | 2898 | goto out; |
7b128766 | 2899 | |
39279cc3 | 2900 | /* now the directory is empty */ |
e02119d5 CM |
2901 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2902 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2903 | if (!err) |
dbe674a9 | 2904 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2905 | out: |
d3c2fdcf | 2906 | nr = trans->blocks_used; |
a22285a6 | 2907 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2908 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2909 | |
39279cc3 CM |
2910 | return err; |
2911 | } | |
2912 | ||
39279cc3 CM |
2913 | /* |
2914 | * this can truncate away extent items, csum items and directory items. | |
2915 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2916 | * any higher than new_size |
39279cc3 CM |
2917 | * |
2918 | * csum items that cross the new i_size are truncated to the new size | |
2919 | * as well. | |
7b128766 JB |
2920 | * |
2921 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2922 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2923 | */ |
8082510e YZ |
2924 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2925 | struct btrfs_root *root, | |
2926 | struct inode *inode, | |
2927 | u64 new_size, u32 min_type) | |
39279cc3 | 2928 | { |
39279cc3 | 2929 | struct btrfs_path *path; |
5f39d397 | 2930 | struct extent_buffer *leaf; |
39279cc3 | 2931 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
2932 | struct btrfs_key key; |
2933 | struct btrfs_key found_key; | |
39279cc3 | 2934 | u64 extent_start = 0; |
db94535d | 2935 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2936 | u64 extent_offset = 0; |
39279cc3 | 2937 | u64 item_end = 0; |
8082510e YZ |
2938 | u64 mask = root->sectorsize - 1; |
2939 | u32 found_type = (u8)-1; | |
39279cc3 CM |
2940 | int found_extent; |
2941 | int del_item; | |
85e21bac CM |
2942 | int pending_del_nr = 0; |
2943 | int pending_del_slot = 0; | |
179e29e4 | 2944 | int extent_type = -1; |
771ed689 | 2945 | int encoding; |
8082510e YZ |
2946 | int ret; |
2947 | int err = 0; | |
33345d01 | 2948 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
2949 | |
2950 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 2951 | |
0eb0e19c MF |
2952 | path = btrfs_alloc_path(); |
2953 | if (!path) | |
2954 | return -ENOMEM; | |
2955 | path->reada = -1; | |
2956 | ||
0af3d00b | 2957 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 2958 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 2959 | |
16cdcec7 MX |
2960 | /* |
2961 | * This function is also used to drop the items in the log tree before | |
2962 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
2963 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
2964 | * items. | |
2965 | */ | |
2966 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
2967 | btrfs_kill_delayed_inode_items(inode); | |
2968 | ||
33345d01 | 2969 | key.objectid = ino; |
39279cc3 | 2970 | key.offset = (u64)-1; |
5f39d397 CM |
2971 | key.type = (u8)-1; |
2972 | ||
85e21bac | 2973 | search_again: |
b9473439 | 2974 | path->leave_spinning = 1; |
85e21bac | 2975 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
2976 | if (ret < 0) { |
2977 | err = ret; | |
2978 | goto out; | |
2979 | } | |
d397712b | 2980 | |
85e21bac | 2981 | if (ret > 0) { |
e02119d5 CM |
2982 | /* there are no items in the tree for us to truncate, we're |
2983 | * done | |
2984 | */ | |
8082510e YZ |
2985 | if (path->slots[0] == 0) |
2986 | goto out; | |
85e21bac CM |
2987 | path->slots[0]--; |
2988 | } | |
2989 | ||
d397712b | 2990 | while (1) { |
39279cc3 | 2991 | fi = NULL; |
5f39d397 CM |
2992 | leaf = path->nodes[0]; |
2993 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2994 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2995 | encoding = 0; |
39279cc3 | 2996 | |
33345d01 | 2997 | if (found_key.objectid != ino) |
39279cc3 | 2998 | break; |
5f39d397 | 2999 | |
85e21bac | 3000 | if (found_type < min_type) |
39279cc3 CM |
3001 | break; |
3002 | ||
5f39d397 | 3003 | item_end = found_key.offset; |
39279cc3 | 3004 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3005 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3006 | struct btrfs_file_extent_item); |
179e29e4 | 3007 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3008 | encoding = btrfs_file_extent_compression(leaf, fi); |
3009 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3010 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3011 | ||
179e29e4 | 3012 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3013 | item_end += |
db94535d | 3014 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3015 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3016 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3017 | fi); |
39279cc3 | 3018 | } |
008630c1 | 3019 | item_end--; |
39279cc3 | 3020 | } |
8082510e YZ |
3021 | if (found_type > min_type) { |
3022 | del_item = 1; | |
3023 | } else { | |
3024 | if (item_end < new_size) | |
b888db2b | 3025 | break; |
8082510e YZ |
3026 | if (found_key.offset >= new_size) |
3027 | del_item = 1; | |
3028 | else | |
3029 | del_item = 0; | |
39279cc3 | 3030 | } |
39279cc3 | 3031 | found_extent = 0; |
39279cc3 | 3032 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3033 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3034 | goto delete; | |
3035 | ||
3036 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3037 | u64 num_dec; |
db94535d | 3038 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3039 | if (!del_item && !encoding) { |
db94535d CM |
3040 | u64 orig_num_bytes = |
3041 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3042 | extent_num_bytes = new_size - |
5f39d397 | 3043 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3044 | extent_num_bytes = extent_num_bytes & |
3045 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3046 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3047 | extent_num_bytes); | |
3048 | num_dec = (orig_num_bytes - | |
9069218d | 3049 | extent_num_bytes); |
e02119d5 | 3050 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3051 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3052 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3053 | } else { |
db94535d CM |
3054 | extent_num_bytes = |
3055 | btrfs_file_extent_disk_num_bytes(leaf, | |
3056 | fi); | |
5d4f98a2 YZ |
3057 | extent_offset = found_key.offset - |
3058 | btrfs_file_extent_offset(leaf, fi); | |
3059 | ||
39279cc3 | 3060 | /* FIXME blocksize != 4096 */ |
9069218d | 3061 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3062 | if (extent_start != 0) { |
3063 | found_extent = 1; | |
e02119d5 | 3064 | if (root->ref_cows) |
a76a3cd4 | 3065 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3066 | } |
39279cc3 | 3067 | } |
9069218d | 3068 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3069 | /* |
3070 | * we can't truncate inline items that have had | |
3071 | * special encodings | |
3072 | */ | |
3073 | if (!del_item && | |
3074 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3075 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3076 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3077 | u32 size = new_size - found_key.offset; |
3078 | ||
3079 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3080 | inode_sub_bytes(inode, item_end + 1 - |
3081 | new_size); | |
e02119d5 CM |
3082 | } |
3083 | size = | |
3084 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3085 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3086 | size, 1); |
e02119d5 | 3087 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3088 | inode_sub_bytes(inode, item_end + 1 - |
3089 | found_key.offset); | |
9069218d | 3090 | } |
39279cc3 | 3091 | } |
179e29e4 | 3092 | delete: |
39279cc3 | 3093 | if (del_item) { |
85e21bac CM |
3094 | if (!pending_del_nr) { |
3095 | /* no pending yet, add ourselves */ | |
3096 | pending_del_slot = path->slots[0]; | |
3097 | pending_del_nr = 1; | |
3098 | } else if (pending_del_nr && | |
3099 | path->slots[0] + 1 == pending_del_slot) { | |
3100 | /* hop on the pending chunk */ | |
3101 | pending_del_nr++; | |
3102 | pending_del_slot = path->slots[0]; | |
3103 | } else { | |
d397712b | 3104 | BUG(); |
85e21bac | 3105 | } |
39279cc3 CM |
3106 | } else { |
3107 | break; | |
3108 | } | |
0af3d00b JB |
3109 | if (found_extent && (root->ref_cows || |
3110 | root == root->fs_info->tree_root)) { | |
b9473439 | 3111 | btrfs_set_path_blocking(path); |
39279cc3 | 3112 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3113 | extent_num_bytes, 0, |
3114 | btrfs_header_owner(leaf), | |
33345d01 | 3115 | ino, extent_offset); |
39279cc3 CM |
3116 | BUG_ON(ret); |
3117 | } | |
85e21bac | 3118 | |
8082510e YZ |
3119 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3120 | break; | |
3121 | ||
3122 | if (path->slots[0] == 0 || | |
3123 | path->slots[0] != pending_del_slot) { | |
82d5902d LZ |
3124 | if (root->ref_cows && |
3125 | BTRFS_I(inode)->location.objectid != | |
3126 | BTRFS_FREE_INO_OBJECTID) { | |
8082510e YZ |
3127 | err = -EAGAIN; |
3128 | goto out; | |
3129 | } | |
3130 | if (pending_del_nr) { | |
3131 | ret = btrfs_del_items(trans, root, path, | |
3132 | pending_del_slot, | |
3133 | pending_del_nr); | |
3134 | BUG_ON(ret); | |
3135 | pending_del_nr = 0; | |
3136 | } | |
b3b4aa74 | 3137 | btrfs_release_path(path); |
85e21bac | 3138 | goto search_again; |
8082510e YZ |
3139 | } else { |
3140 | path->slots[0]--; | |
85e21bac | 3141 | } |
39279cc3 | 3142 | } |
8082510e | 3143 | out: |
85e21bac CM |
3144 | if (pending_del_nr) { |
3145 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3146 | pending_del_nr); | |
d68fc57b | 3147 | BUG_ON(ret); |
85e21bac | 3148 | } |
39279cc3 | 3149 | btrfs_free_path(path); |
8082510e | 3150 | return err; |
39279cc3 CM |
3151 | } |
3152 | ||
3153 | /* | |
3154 | * taken from block_truncate_page, but does cow as it zeros out | |
3155 | * any bytes left in the last page in the file. | |
3156 | */ | |
3157 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3158 | { | |
3159 | struct inode *inode = mapping->host; | |
db94535d | 3160 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3161 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3162 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3163 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3164 | char *kaddr; |
db94535d | 3165 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3166 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3167 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3168 | struct page *page; | |
3b16a4e3 | 3169 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3170 | int ret = 0; |
a52d9a80 | 3171 | u64 page_start; |
e6dcd2dc | 3172 | u64 page_end; |
39279cc3 CM |
3173 | |
3174 | if ((offset & (blocksize - 1)) == 0) | |
3175 | goto out; | |
0ca1f7ce | 3176 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3177 | if (ret) |
3178 | goto out; | |
39279cc3 CM |
3179 | |
3180 | ret = -ENOMEM; | |
211c17f5 | 3181 | again: |
3b16a4e3 | 3182 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3183 | if (!page) { |
0ca1f7ce | 3184 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3185 | goto out; |
5d5e103a | 3186 | } |
e6dcd2dc CM |
3187 | |
3188 | page_start = page_offset(page); | |
3189 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3190 | ||
39279cc3 | 3191 | if (!PageUptodate(page)) { |
9ebefb18 | 3192 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3193 | lock_page(page); |
211c17f5 CM |
3194 | if (page->mapping != mapping) { |
3195 | unlock_page(page); | |
3196 | page_cache_release(page); | |
3197 | goto again; | |
3198 | } | |
39279cc3 CM |
3199 | if (!PageUptodate(page)) { |
3200 | ret = -EIO; | |
89642229 | 3201 | goto out_unlock; |
39279cc3 CM |
3202 | } |
3203 | } | |
211c17f5 | 3204 | wait_on_page_writeback(page); |
e6dcd2dc | 3205 | |
2ac55d41 JB |
3206 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3207 | GFP_NOFS); | |
e6dcd2dc CM |
3208 | set_page_extent_mapped(page); |
3209 | ||
3210 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3211 | if (ordered) { | |
2ac55d41 JB |
3212 | unlock_extent_cached(io_tree, page_start, page_end, |
3213 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3214 | unlock_page(page); |
3215 | page_cache_release(page); | |
eb84ae03 | 3216 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3217 | btrfs_put_ordered_extent(ordered); |
3218 | goto again; | |
3219 | } | |
3220 | ||
2ac55d41 | 3221 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3222 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3223 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3224 | |
2ac55d41 JB |
3225 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3226 | &cached_state); | |
9ed74f2d | 3227 | if (ret) { |
2ac55d41 JB |
3228 | unlock_extent_cached(io_tree, page_start, page_end, |
3229 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3230 | goto out_unlock; |
3231 | } | |
3232 | ||
e6dcd2dc CM |
3233 | ret = 0; |
3234 | if (offset != PAGE_CACHE_SIZE) { | |
3235 | kaddr = kmap(page); | |
3236 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3237 | flush_dcache_page(page); | |
3238 | kunmap(page); | |
3239 | } | |
247e743c | 3240 | ClearPageChecked(page); |
e6dcd2dc | 3241 | set_page_dirty(page); |
2ac55d41 JB |
3242 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3243 | GFP_NOFS); | |
39279cc3 | 3244 | |
89642229 | 3245 | out_unlock: |
5d5e103a | 3246 | if (ret) |
0ca1f7ce | 3247 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3248 | unlock_page(page); |
3249 | page_cache_release(page); | |
3250 | out: | |
3251 | return ret; | |
3252 | } | |
3253 | ||
695a0d0d JB |
3254 | /* |
3255 | * This function puts in dummy file extents for the area we're creating a hole | |
3256 | * for. So if we are truncating this file to a larger size we need to insert | |
3257 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3258 | * the range between oldsize and size | |
3259 | */ | |
a41ad394 | 3260 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3261 | { |
9036c102 YZ |
3262 | struct btrfs_trans_handle *trans; |
3263 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3264 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3265 | struct extent_map *em = NULL; |
2ac55d41 | 3266 | struct extent_state *cached_state = NULL; |
9036c102 | 3267 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3268 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3269 | u64 block_end = (size + mask) & ~mask; |
3270 | u64 last_byte; | |
3271 | u64 cur_offset; | |
3272 | u64 hole_size; | |
9ed74f2d | 3273 | int err = 0; |
39279cc3 | 3274 | |
9036c102 YZ |
3275 | if (size <= hole_start) |
3276 | return 0; | |
3277 | ||
9036c102 YZ |
3278 | while (1) { |
3279 | struct btrfs_ordered_extent *ordered; | |
3280 | btrfs_wait_ordered_range(inode, hole_start, | |
3281 | block_end - hole_start); | |
2ac55d41 JB |
3282 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3283 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3284 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3285 | if (!ordered) | |
3286 | break; | |
2ac55d41 JB |
3287 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3288 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3289 | btrfs_put_ordered_extent(ordered); |
3290 | } | |
39279cc3 | 3291 | |
9036c102 YZ |
3292 | cur_offset = hole_start; |
3293 | while (1) { | |
3294 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3295 | block_end - cur_offset, 0); | |
c704005d | 3296 | BUG_ON(IS_ERR_OR_NULL(em)); |
9036c102 YZ |
3297 | last_byte = min(extent_map_end(em), block_end); |
3298 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3299 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3300 | u64 hint_byte = 0; |
9036c102 | 3301 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3302 | |
a22285a6 YZ |
3303 | trans = btrfs_start_transaction(root, 2); |
3304 | if (IS_ERR(trans)) { | |
3305 | err = PTR_ERR(trans); | |
9ed74f2d | 3306 | break; |
a22285a6 | 3307 | } |
8082510e YZ |
3308 | |
3309 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3310 | cur_offset + hole_size, | |
3311 | &hint_byte, 1); | |
5b397377 MX |
3312 | if (err) { |
3313 | btrfs_end_transaction(trans, root); | |
3893e33b | 3314 | break; |
5b397377 | 3315 | } |
8082510e | 3316 | |
9036c102 | 3317 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3318 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3319 | 0, hole_size, 0, hole_size, |
3320 | 0, 0, 0); | |
5b397377 MX |
3321 | if (err) { |
3322 | btrfs_end_transaction(trans, root); | |
3893e33b | 3323 | break; |
5b397377 | 3324 | } |
8082510e | 3325 | |
9036c102 YZ |
3326 | btrfs_drop_extent_cache(inode, hole_start, |
3327 | last_byte - 1, 0); | |
8082510e YZ |
3328 | |
3329 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3330 | } |
3331 | free_extent_map(em); | |
a22285a6 | 3332 | em = NULL; |
9036c102 | 3333 | cur_offset = last_byte; |
8082510e | 3334 | if (cur_offset >= block_end) |
9036c102 YZ |
3335 | break; |
3336 | } | |
1832a6d5 | 3337 | |
a22285a6 | 3338 | free_extent_map(em); |
2ac55d41 JB |
3339 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3340 | GFP_NOFS); | |
9036c102 YZ |
3341 | return err; |
3342 | } | |
39279cc3 | 3343 | |
a41ad394 | 3344 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3345 | { |
a41ad394 | 3346 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3347 | int ret; |
3348 | ||
a41ad394 | 3349 | if (newsize == oldsize) |
8082510e YZ |
3350 | return 0; |
3351 | ||
a41ad394 JB |
3352 | if (newsize > oldsize) { |
3353 | i_size_write(inode, newsize); | |
3354 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3355 | truncate_pagecache(inode, oldsize, newsize); | |
3356 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
8082510e | 3357 | if (ret) { |
a41ad394 | 3358 | btrfs_setsize(inode, oldsize); |
8082510e YZ |
3359 | return ret; |
3360 | } | |
3361 | ||
930f028a | 3362 | mark_inode_dirty(inode); |
a41ad394 | 3363 | } else { |
8082510e | 3364 | |
a41ad394 JB |
3365 | /* |
3366 | * We're truncating a file that used to have good data down to | |
3367 | * zero. Make sure it gets into the ordered flush list so that | |
3368 | * any new writes get down to disk quickly. | |
3369 | */ | |
3370 | if (newsize == 0) | |
3371 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3372 | |
a41ad394 JB |
3373 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3374 | truncate_setsize(inode, newsize); | |
3375 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3376 | } |
3377 | ||
a41ad394 | 3378 | return ret; |
8082510e YZ |
3379 | } |
3380 | ||
9036c102 YZ |
3381 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3382 | { | |
3383 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3384 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3385 | int err; |
39279cc3 | 3386 | |
b83cc969 LZ |
3387 | if (btrfs_root_readonly(root)) |
3388 | return -EROFS; | |
3389 | ||
9036c102 YZ |
3390 | err = inode_change_ok(inode, attr); |
3391 | if (err) | |
3392 | return err; | |
2bf5a725 | 3393 | |
5a3f23d5 | 3394 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3395 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3396 | if (err) |
3397 | return err; | |
39279cc3 | 3398 | } |
9036c102 | 3399 | |
1025774c CH |
3400 | if (attr->ia_valid) { |
3401 | setattr_copy(inode, attr); | |
3402 | mark_inode_dirty(inode); | |
3403 | ||
3404 | if (attr->ia_valid & ATTR_MODE) | |
3405 | err = btrfs_acl_chmod(inode); | |
3406 | } | |
33268eaf | 3407 | |
39279cc3 CM |
3408 | return err; |
3409 | } | |
61295eb8 | 3410 | |
bd555975 | 3411 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3412 | { |
3413 | struct btrfs_trans_handle *trans; | |
3414 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3415 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3416 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3417 | unsigned long nr; |
39279cc3 CM |
3418 | int ret; |
3419 | ||
1abe9b8a | 3420 | trace_btrfs_inode_evict(inode); |
3421 | ||
39279cc3 | 3422 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3423 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
2cf8572d | 3424 | btrfs_is_free_space_inode(root, inode))) |
bd555975 AV |
3425 | goto no_delete; |
3426 | ||
39279cc3 | 3427 | if (is_bad_inode(inode)) { |
7b128766 | 3428 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3429 | goto no_delete; |
3430 | } | |
bd555975 | 3431 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3432 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3433 | |
c71bf099 YZ |
3434 | if (root->fs_info->log_root_recovering) { |
3435 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3436 | goto no_delete; | |
3437 | } | |
3438 | ||
76dda93c YZ |
3439 | if (inode->i_nlink > 0) { |
3440 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3441 | goto no_delete; | |
3442 | } | |
3443 | ||
4289a667 JB |
3444 | rsv = btrfs_alloc_block_rsv(root); |
3445 | if (!rsv) { | |
3446 | btrfs_orphan_del(NULL, inode); | |
3447 | goto no_delete; | |
3448 | } | |
4a338542 | 3449 | rsv->size = min_size; |
726c35fa | 3450 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3451 | |
dbe674a9 | 3452 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3453 | |
4289a667 JB |
3454 | /* |
3455 | * This is a bit simpler than btrfs_truncate since | |
3456 | * | |
3457 | * 1) We've already reserved our space for our orphan item in the | |
3458 | * unlink. | |
3459 | * 2) We're going to delete the inode item, so we don't need to update | |
3460 | * it at all. | |
3461 | * | |
3462 | * So we just need to reserve some slack space in case we add bytes when | |
3463 | * doing the truncate. | |
3464 | */ | |
8082510e | 3465 | while (1) { |
36ba022a | 3466 | ret = btrfs_block_rsv_refill(root, rsv, min_size); |
726c35fa JB |
3467 | |
3468 | /* | |
3469 | * Try and steal from the global reserve since we will | |
3470 | * likely not use this space anyway, we want to try as | |
3471 | * hard as possible to get this to work. | |
3472 | */ | |
3473 | if (ret) | |
3474 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
3475 | ||
d68fc57b | 3476 | if (ret) { |
4289a667 | 3477 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3478 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3479 | btrfs_orphan_del(NULL, inode); |
3480 | btrfs_free_block_rsv(root, rsv); | |
3481 | goto no_delete; | |
3482 | } | |
3483 | ||
3484 | trans = btrfs_start_transaction(root, 0); | |
3485 | if (IS_ERR(trans)) { | |
3486 | btrfs_orphan_del(NULL, inode); | |
3487 | btrfs_free_block_rsv(root, rsv); | |
3488 | goto no_delete; | |
d68fc57b | 3489 | } |
7b128766 | 3490 | |
4289a667 JB |
3491 | trans->block_rsv = rsv; |
3492 | ||
d68fc57b | 3493 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3494 | if (ret != -EAGAIN) |
3495 | break; | |
85e21bac | 3496 | |
8082510e YZ |
3497 | nr = trans->blocks_used; |
3498 | btrfs_end_transaction(trans, root); | |
3499 | trans = NULL; | |
3500 | btrfs_btree_balance_dirty(root, nr); | |
3501 | } | |
5f39d397 | 3502 | |
4289a667 JB |
3503 | btrfs_free_block_rsv(root, rsv); |
3504 | ||
8082510e | 3505 | if (ret == 0) { |
4289a667 | 3506 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3507 | ret = btrfs_orphan_del(trans, inode); |
3508 | BUG_ON(ret); | |
3509 | } | |
54aa1f4d | 3510 | |
4289a667 | 3511 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3512 | if (!(root == root->fs_info->tree_root || |
3513 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3514 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3515 | |
d3c2fdcf | 3516 | nr = trans->blocks_used; |
54aa1f4d | 3517 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3518 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3519 | no_delete: |
bd555975 | 3520 | end_writeback(inode); |
8082510e | 3521 | return; |
39279cc3 CM |
3522 | } |
3523 | ||
3524 | /* | |
3525 | * this returns the key found in the dir entry in the location pointer. | |
3526 | * If no dir entries were found, location->objectid is 0. | |
3527 | */ | |
3528 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3529 | struct btrfs_key *location) | |
3530 | { | |
3531 | const char *name = dentry->d_name.name; | |
3532 | int namelen = dentry->d_name.len; | |
3533 | struct btrfs_dir_item *di; | |
3534 | struct btrfs_path *path; | |
3535 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3536 | int ret = 0; |
39279cc3 CM |
3537 | |
3538 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3539 | if (!path) |
3540 | return -ENOMEM; | |
3954401f | 3541 | |
33345d01 | 3542 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3543 | namelen, 0); |
0d9f7f3e Y |
3544 | if (IS_ERR(di)) |
3545 | ret = PTR_ERR(di); | |
d397712b | 3546 | |
c704005d | 3547 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3548 | goto out_err; |
d397712b | 3549 | |
5f39d397 | 3550 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3551 | out: |
39279cc3 CM |
3552 | btrfs_free_path(path); |
3553 | return ret; | |
3954401f CM |
3554 | out_err: |
3555 | location->objectid = 0; | |
3556 | goto out; | |
39279cc3 CM |
3557 | } |
3558 | ||
3559 | /* | |
3560 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3561 | * needs to be changed to reflect the root directory of the tree root. This | |
3562 | * is kind of like crossing a mount point. | |
3563 | */ | |
3564 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3565 | struct inode *dir, |
3566 | struct dentry *dentry, | |
3567 | struct btrfs_key *location, | |
3568 | struct btrfs_root **sub_root) | |
39279cc3 | 3569 | { |
4df27c4d YZ |
3570 | struct btrfs_path *path; |
3571 | struct btrfs_root *new_root; | |
3572 | struct btrfs_root_ref *ref; | |
3573 | struct extent_buffer *leaf; | |
3574 | int ret; | |
3575 | int err = 0; | |
39279cc3 | 3576 | |
4df27c4d YZ |
3577 | path = btrfs_alloc_path(); |
3578 | if (!path) { | |
3579 | err = -ENOMEM; | |
3580 | goto out; | |
3581 | } | |
39279cc3 | 3582 | |
4df27c4d YZ |
3583 | err = -ENOENT; |
3584 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3585 | BTRFS_I(dir)->root->root_key.objectid, | |
3586 | location->objectid); | |
3587 | if (ret) { | |
3588 | if (ret < 0) | |
3589 | err = ret; | |
3590 | goto out; | |
3591 | } | |
39279cc3 | 3592 | |
4df27c4d YZ |
3593 | leaf = path->nodes[0]; |
3594 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3595 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3596 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3597 | goto out; | |
39279cc3 | 3598 | |
4df27c4d YZ |
3599 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3600 | (unsigned long)(ref + 1), | |
3601 | dentry->d_name.len); | |
3602 | if (ret) | |
3603 | goto out; | |
3604 | ||
b3b4aa74 | 3605 | btrfs_release_path(path); |
4df27c4d YZ |
3606 | |
3607 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3608 | if (IS_ERR(new_root)) { | |
3609 | err = PTR_ERR(new_root); | |
3610 | goto out; | |
3611 | } | |
3612 | ||
3613 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3614 | err = -ENOENT; | |
3615 | goto out; | |
3616 | } | |
3617 | ||
3618 | *sub_root = new_root; | |
3619 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3620 | location->type = BTRFS_INODE_ITEM_KEY; | |
3621 | location->offset = 0; | |
3622 | err = 0; | |
3623 | out: | |
3624 | btrfs_free_path(path); | |
3625 | return err; | |
39279cc3 CM |
3626 | } |
3627 | ||
5d4f98a2 YZ |
3628 | static void inode_tree_add(struct inode *inode) |
3629 | { | |
3630 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3631 | struct btrfs_inode *entry; | |
03e860bd FNP |
3632 | struct rb_node **p; |
3633 | struct rb_node *parent; | |
33345d01 | 3634 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
3635 | again: |
3636 | p = &root->inode_tree.rb_node; | |
3637 | parent = NULL; | |
5d4f98a2 | 3638 | |
1d3382cb | 3639 | if (inode_unhashed(inode)) |
76dda93c YZ |
3640 | return; |
3641 | ||
5d4f98a2 YZ |
3642 | spin_lock(&root->inode_lock); |
3643 | while (*p) { | |
3644 | parent = *p; | |
3645 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3646 | ||
33345d01 | 3647 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3648 | p = &parent->rb_left; |
33345d01 | 3649 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3650 | p = &parent->rb_right; |
5d4f98a2 YZ |
3651 | else { |
3652 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3653 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3654 | rb_erase(parent, &root->inode_tree); |
3655 | RB_CLEAR_NODE(parent); | |
3656 | spin_unlock(&root->inode_lock); | |
3657 | goto again; | |
5d4f98a2 YZ |
3658 | } |
3659 | } | |
3660 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3661 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3662 | spin_unlock(&root->inode_lock); | |
3663 | } | |
3664 | ||
3665 | static void inode_tree_del(struct inode *inode) | |
3666 | { | |
3667 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3668 | int empty = 0; |
5d4f98a2 | 3669 | |
03e860bd | 3670 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3671 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3672 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3673 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3674 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3675 | } |
03e860bd | 3676 | spin_unlock(&root->inode_lock); |
76dda93c | 3677 | |
0af3d00b JB |
3678 | /* |
3679 | * Free space cache has inodes in the tree root, but the tree root has a | |
3680 | * root_refs of 0, so this could end up dropping the tree root as a | |
3681 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3682 | * make sure we don't drop it. | |
3683 | */ | |
3684 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3685 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3686 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3687 | spin_lock(&root->inode_lock); | |
3688 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3689 | spin_unlock(&root->inode_lock); | |
3690 | if (empty) | |
3691 | btrfs_add_dead_root(root); | |
3692 | } | |
3693 | } | |
3694 | ||
3695 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3696 | { | |
3697 | struct rb_node *node; | |
3698 | struct rb_node *prev; | |
3699 | struct btrfs_inode *entry; | |
3700 | struct inode *inode; | |
3701 | u64 objectid = 0; | |
3702 | ||
3703 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3704 | ||
3705 | spin_lock(&root->inode_lock); | |
3706 | again: | |
3707 | node = root->inode_tree.rb_node; | |
3708 | prev = NULL; | |
3709 | while (node) { | |
3710 | prev = node; | |
3711 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3712 | ||
33345d01 | 3713 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 3714 | node = node->rb_left; |
33345d01 | 3715 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
3716 | node = node->rb_right; |
3717 | else | |
3718 | break; | |
3719 | } | |
3720 | if (!node) { | |
3721 | while (prev) { | |
3722 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 3723 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
3724 | node = prev; |
3725 | break; | |
3726 | } | |
3727 | prev = rb_next(prev); | |
3728 | } | |
3729 | } | |
3730 | while (node) { | |
3731 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 3732 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
3733 | inode = igrab(&entry->vfs_inode); |
3734 | if (inode) { | |
3735 | spin_unlock(&root->inode_lock); | |
3736 | if (atomic_read(&inode->i_count) > 1) | |
3737 | d_prune_aliases(inode); | |
3738 | /* | |
45321ac5 | 3739 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
3740 | * the inode cache when its usage count |
3741 | * hits zero. | |
3742 | */ | |
3743 | iput(inode); | |
3744 | cond_resched(); | |
3745 | spin_lock(&root->inode_lock); | |
3746 | goto again; | |
3747 | } | |
3748 | ||
3749 | if (cond_resched_lock(&root->inode_lock)) | |
3750 | goto again; | |
3751 | ||
3752 | node = rb_next(node); | |
3753 | } | |
3754 | spin_unlock(&root->inode_lock); | |
3755 | return 0; | |
5d4f98a2 YZ |
3756 | } |
3757 | ||
e02119d5 CM |
3758 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
3759 | { | |
3760 | struct btrfs_iget_args *args = p; | |
3761 | inode->i_ino = args->ino; | |
e02119d5 | 3762 | BTRFS_I(inode)->root = args->root; |
6a63209f | 3763 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3764 | return 0; |
3765 | } | |
3766 | ||
3767 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3768 | { | |
3769 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 3770 | return args->ino == btrfs_ino(inode) && |
d397712b | 3771 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
3772 | } |
3773 | ||
5d4f98a2 YZ |
3774 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3775 | u64 objectid, | |
3776 | struct btrfs_root *root) | |
39279cc3 CM |
3777 | { |
3778 | struct inode *inode; | |
3779 | struct btrfs_iget_args args; | |
3780 | args.ino = objectid; | |
3781 | args.root = root; | |
3782 | ||
3783 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3784 | btrfs_init_locked_inode, | |
3785 | (void *)&args); | |
3786 | return inode; | |
3787 | } | |
3788 | ||
1a54ef8c BR |
3789 | /* Get an inode object given its location and corresponding root. |
3790 | * Returns in *is_new if the inode was read from disk | |
3791 | */ | |
3792 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 3793 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
3794 | { |
3795 | struct inode *inode; | |
3796 | ||
3797 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3798 | if (!inode) | |
5d4f98a2 | 3799 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3800 | |
3801 | if (inode->i_state & I_NEW) { | |
3802 | BTRFS_I(inode)->root = root; | |
3803 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3804 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
3805 | if (!is_bad_inode(inode)) { |
3806 | inode_tree_add(inode); | |
3807 | unlock_new_inode(inode); | |
3808 | if (new) | |
3809 | *new = 1; | |
3810 | } else { | |
e0b6d65b ST |
3811 | unlock_new_inode(inode); |
3812 | iput(inode); | |
3813 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
3814 | } |
3815 | } | |
3816 | ||
1a54ef8c BR |
3817 | return inode; |
3818 | } | |
3819 | ||
4df27c4d YZ |
3820 | static struct inode *new_simple_dir(struct super_block *s, |
3821 | struct btrfs_key *key, | |
3822 | struct btrfs_root *root) | |
3823 | { | |
3824 | struct inode *inode = new_inode(s); | |
3825 | ||
3826 | if (!inode) | |
3827 | return ERR_PTR(-ENOMEM); | |
3828 | ||
4df27c4d YZ |
3829 | BTRFS_I(inode)->root = root; |
3830 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3831 | BTRFS_I(inode)->dummy_inode = 1; | |
3832 | ||
3833 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3834 | inode->i_op = &simple_dir_inode_operations; | |
3835 | inode->i_fop = &simple_dir_operations; | |
3836 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3837 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3838 | ||
3839 | return inode; | |
3840 | } | |
3841 | ||
3de4586c | 3842 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3843 | { |
d397712b | 3844 | struct inode *inode; |
4df27c4d | 3845 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3846 | struct btrfs_root *sub_root = root; |
3847 | struct btrfs_key location; | |
76dda93c | 3848 | int index; |
b4aff1f8 | 3849 | int ret = 0; |
39279cc3 CM |
3850 | |
3851 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3852 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3853 | |
b4aff1f8 JB |
3854 | if (unlikely(d_need_lookup(dentry))) { |
3855 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
3856 | kfree(dentry->d_fsdata); | |
3857 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
3858 | /* This thing is hashed, drop it for now */ |
3859 | d_drop(dentry); | |
b4aff1f8 JB |
3860 | } else { |
3861 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
3862 | } | |
5f39d397 | 3863 | |
39279cc3 CM |
3864 | if (ret < 0) |
3865 | return ERR_PTR(ret); | |
5f39d397 | 3866 | |
4df27c4d YZ |
3867 | if (location.objectid == 0) |
3868 | return NULL; | |
3869 | ||
3870 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 3871 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
3872 | return inode; |
3873 | } | |
3874 | ||
3875 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3876 | ||
76dda93c | 3877 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3878 | ret = fixup_tree_root_location(root, dir, dentry, |
3879 | &location, &sub_root); | |
3880 | if (ret < 0) { | |
3881 | if (ret != -ENOENT) | |
3882 | inode = ERR_PTR(ret); | |
3883 | else | |
3884 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3885 | } else { | |
73f73415 | 3886 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 3887 | } |
76dda93c YZ |
3888 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3889 | ||
34d19bad | 3890 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
3891 | down_read(&root->fs_info->cleanup_work_sem); |
3892 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 3893 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 3894 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
3895 | if (ret) |
3896 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
3897 | } |
3898 | ||
3de4586c CM |
3899 | return inode; |
3900 | } | |
3901 | ||
fe15ce44 | 3902 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
3903 | { |
3904 | struct btrfs_root *root; | |
3905 | ||
efefb143 YZ |
3906 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3907 | dentry = dentry->d_parent; | |
76dda93c | 3908 | |
efefb143 YZ |
3909 | if (dentry->d_inode) { |
3910 | root = BTRFS_I(dentry->d_inode)->root; | |
3911 | if (btrfs_root_refs(&root->root_item) == 0) | |
3912 | return 1; | |
3913 | } | |
76dda93c YZ |
3914 | return 0; |
3915 | } | |
3916 | ||
b4aff1f8 JB |
3917 | static void btrfs_dentry_release(struct dentry *dentry) |
3918 | { | |
3919 | if (dentry->d_fsdata) | |
3920 | kfree(dentry->d_fsdata); | |
3921 | } | |
3922 | ||
3de4586c CM |
3923 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3924 | struct nameidata *nd) | |
3925 | { | |
a66e7cc6 JB |
3926 | struct dentry *ret; |
3927 | ||
3928 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
3929 | if (unlikely(d_need_lookup(dentry))) { | |
3930 | spin_lock(&dentry->d_lock); | |
3931 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
3932 | spin_unlock(&dentry->d_lock); | |
3933 | } | |
3934 | return ret; | |
39279cc3 CM |
3935 | } |
3936 | ||
16cdcec7 | 3937 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
3938 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
3939 | }; | |
3940 | ||
cbdf5a24 DW |
3941 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3942 | filldir_t filldir) | |
39279cc3 | 3943 | { |
6da6abae | 3944 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3945 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3946 | struct btrfs_item *item; | |
3947 | struct btrfs_dir_item *di; | |
3948 | struct btrfs_key key; | |
5f39d397 | 3949 | struct btrfs_key found_key; |
39279cc3 | 3950 | struct btrfs_path *path; |
16cdcec7 MX |
3951 | struct list_head ins_list; |
3952 | struct list_head del_list; | |
b4aff1f8 | 3953 | struct qstr q; |
39279cc3 | 3954 | int ret; |
5f39d397 | 3955 | struct extent_buffer *leaf; |
39279cc3 | 3956 | int slot; |
39279cc3 CM |
3957 | unsigned char d_type; |
3958 | int over = 0; | |
3959 | u32 di_cur; | |
3960 | u32 di_total; | |
3961 | u32 di_len; | |
3962 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3963 | char tmp_name[32]; |
3964 | char *name_ptr; | |
3965 | int name_len; | |
16cdcec7 | 3966 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
3967 | |
3968 | /* FIXME, use a real flag for deciding about the key type */ | |
3969 | if (root->fs_info->tree_root == root) | |
3970 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3971 | |
3954401f CM |
3972 | /* special case for "." */ |
3973 | if (filp->f_pos == 0) { | |
3765fefa HS |
3974 | over = filldir(dirent, ".", 1, |
3975 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
3976 | if (over) |
3977 | return 0; | |
3978 | filp->f_pos = 1; | |
3979 | } | |
3954401f CM |
3980 | /* special case for .., just use the back ref */ |
3981 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3982 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3983 | over = filldir(dirent, "..", 2, |
3765fefa | 3984 | filp->f_pos, pino, DT_DIR); |
3954401f | 3985 | if (over) |
49593bfa | 3986 | return 0; |
3954401f CM |
3987 | filp->f_pos = 2; |
3988 | } | |
49593bfa | 3989 | path = btrfs_alloc_path(); |
16cdcec7 MX |
3990 | if (!path) |
3991 | return -ENOMEM; | |
ff5714cc | 3992 | |
026fd317 | 3993 | path->reada = 1; |
49593bfa | 3994 | |
16cdcec7 MX |
3995 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
3996 | INIT_LIST_HEAD(&ins_list); | |
3997 | INIT_LIST_HEAD(&del_list); | |
3998 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
3999 | } | |
4000 | ||
39279cc3 CM |
4001 | btrfs_set_key_type(&key, key_type); |
4002 | key.offset = filp->f_pos; | |
33345d01 | 4003 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4004 | |
39279cc3 CM |
4005 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4006 | if (ret < 0) | |
4007 | goto err; | |
49593bfa DW |
4008 | |
4009 | while (1) { | |
5f39d397 | 4010 | leaf = path->nodes[0]; |
39279cc3 | 4011 | slot = path->slots[0]; |
b9e03af0 LZ |
4012 | if (slot >= btrfs_header_nritems(leaf)) { |
4013 | ret = btrfs_next_leaf(root, path); | |
4014 | if (ret < 0) | |
4015 | goto err; | |
4016 | else if (ret > 0) | |
4017 | break; | |
4018 | continue; | |
39279cc3 | 4019 | } |
3de4586c | 4020 | |
5f39d397 CM |
4021 | item = btrfs_item_nr(leaf, slot); |
4022 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4023 | ||
4024 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4025 | break; |
5f39d397 | 4026 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4027 | break; |
5f39d397 | 4028 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4029 | goto next; |
16cdcec7 MX |
4030 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4031 | btrfs_should_delete_dir_index(&del_list, | |
4032 | found_key.offset)) | |
4033 | goto next; | |
5f39d397 CM |
4034 | |
4035 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4036 | is_curr = 1; |
49593bfa | 4037 | |
39279cc3 CM |
4038 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4039 | di_cur = 0; | |
5f39d397 | 4040 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4041 | |
4042 | while (di_cur < di_total) { | |
5f39d397 | 4043 | struct btrfs_key location; |
b4aff1f8 | 4044 | struct dentry *tmp; |
5f39d397 | 4045 | |
22a94d44 JB |
4046 | if (verify_dir_item(root, leaf, di)) |
4047 | break; | |
4048 | ||
5f39d397 | 4049 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4050 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4051 | name_ptr = tmp_name; |
4052 | } else { | |
4053 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4054 | if (!name_ptr) { |
4055 | ret = -ENOMEM; | |
4056 | goto err; | |
4057 | } | |
5f39d397 CM |
4058 | } |
4059 | read_extent_buffer(leaf, name_ptr, | |
4060 | (unsigned long)(di + 1), name_len); | |
4061 | ||
4062 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4063 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4064 | |
b4aff1f8 JB |
4065 | q.name = name_ptr; |
4066 | q.len = name_len; | |
4067 | q.hash = full_name_hash(q.name, q.len); | |
4068 | tmp = d_lookup(filp->f_dentry, &q); | |
4069 | if (!tmp) { | |
4070 | struct btrfs_key *newkey; | |
4071 | ||
4072 | newkey = kzalloc(sizeof(struct btrfs_key), | |
4073 | GFP_NOFS); | |
4074 | if (!newkey) | |
4075 | goto no_dentry; | |
4076 | tmp = d_alloc(filp->f_dentry, &q); | |
4077 | if (!tmp) { | |
4078 | kfree(newkey); | |
4079 | dput(tmp); | |
4080 | goto no_dentry; | |
4081 | } | |
4082 | memcpy(newkey, &location, | |
4083 | sizeof(struct btrfs_key)); | |
4084 | tmp->d_fsdata = newkey; | |
4085 | tmp->d_flags |= DCACHE_NEED_LOOKUP; | |
4086 | d_rehash(tmp); | |
4087 | dput(tmp); | |
4088 | } else { | |
4089 | dput(tmp); | |
4090 | } | |
4091 | no_dentry: | |
3de4586c CM |
4092 | /* is this a reference to our own snapshot? If so |
4093 | * skip it | |
4094 | */ | |
4095 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4096 | location.objectid == root->root_key.objectid) { | |
4097 | over = 0; | |
4098 | goto skip; | |
4099 | } | |
5f39d397 | 4100 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4101 | found_key.offset, location.objectid, |
39279cc3 | 4102 | d_type); |
5f39d397 | 4103 | |
3de4586c | 4104 | skip: |
5f39d397 CM |
4105 | if (name_ptr != tmp_name) |
4106 | kfree(name_ptr); | |
4107 | ||
39279cc3 CM |
4108 | if (over) |
4109 | goto nopos; | |
5103e947 | 4110 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4111 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4112 | di_cur += di_len; |
4113 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4114 | } | |
b9e03af0 LZ |
4115 | next: |
4116 | path->slots[0]++; | |
39279cc3 | 4117 | } |
49593bfa | 4118 | |
16cdcec7 MX |
4119 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4120 | if (is_curr) | |
4121 | filp->f_pos++; | |
4122 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4123 | &ins_list); | |
4124 | if (ret) | |
4125 | goto nopos; | |
4126 | } | |
4127 | ||
49593bfa | 4128 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4129 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4130 | /* |
4131 | * 32-bit glibc will use getdents64, but then strtol - | |
4132 | * so the last number we can serve is this. | |
4133 | */ | |
4134 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4135 | else |
4136 | filp->f_pos++; | |
39279cc3 CM |
4137 | nopos: |
4138 | ret = 0; | |
4139 | err: | |
16cdcec7 MX |
4140 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4141 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4142 | btrfs_free_path(path); |
39279cc3 CM |
4143 | return ret; |
4144 | } | |
4145 | ||
a9185b41 | 4146 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4147 | { |
4148 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4149 | struct btrfs_trans_handle *trans; | |
4150 | int ret = 0; | |
0af3d00b | 4151 | bool nolock = false; |
39279cc3 | 4152 | |
8929ecfa | 4153 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4154 | return 0; |
4155 | ||
2cf8572d | 4156 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode)) |
82d5902d | 4157 | nolock = true; |
0af3d00b | 4158 | |
a9185b41 | 4159 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4160 | if (nolock) |
7a7eaa40 | 4161 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4162 | else |
7a7eaa40 | 4163 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4164 | if (IS_ERR(trans)) |
4165 | return PTR_ERR(trans); | |
0af3d00b JB |
4166 | if (nolock) |
4167 | ret = btrfs_end_transaction_nolock(trans, root); | |
4168 | else | |
4169 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4170 | } |
4171 | return ret; | |
4172 | } | |
4173 | ||
4174 | /* | |
54aa1f4d | 4175 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4176 | * inode changes. But, it is most likely to find the inode in cache. |
4177 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4178 | * to keep or drop this code. | |
4179 | */ | |
aa385729 | 4180 | void btrfs_dirty_inode(struct inode *inode, int flags) |
39279cc3 CM |
4181 | { |
4182 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4183 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4184 | int ret; |
4185 | ||
4186 | if (BTRFS_I(inode)->dummy_inode) | |
4187 | return; | |
39279cc3 | 4188 | |
7a7eaa40 | 4189 | trans = btrfs_join_transaction(root); |
3612b495 | 4190 | BUG_ON(IS_ERR(trans)); |
8929ecfa YZ |
4191 | |
4192 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4193 | if (ret && ret == -ENOSPC) { |
4194 | /* whoops, lets try again with the full transaction */ | |
4195 | btrfs_end_transaction(trans, root); | |
4196 | trans = btrfs_start_transaction(root, 1); | |
9aeead73 | 4197 | if (IS_ERR(trans)) { |
7a36ddec | 4198 | printk_ratelimited(KERN_ERR "btrfs: fail to " |
33345d01 LZ |
4199 | "dirty inode %llu error %ld\n", |
4200 | (unsigned long long)btrfs_ino(inode), | |
4201 | PTR_ERR(trans)); | |
9aeead73 CM |
4202 | return; |
4203 | } | |
8929ecfa | 4204 | |
94b60442 CM |
4205 | ret = btrfs_update_inode(trans, root, inode); |
4206 | if (ret) { | |
7a36ddec | 4207 | printk_ratelimited(KERN_ERR "btrfs: fail to " |
33345d01 LZ |
4208 | "dirty inode %llu error %d\n", |
4209 | (unsigned long long)btrfs_ino(inode), | |
4210 | ret); | |
94b60442 CM |
4211 | } |
4212 | } | |
39279cc3 | 4213 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4214 | if (BTRFS_I(inode)->delayed_node) |
4215 | btrfs_balance_delayed_items(root); | |
39279cc3 CM |
4216 | } |
4217 | ||
d352ac68 CM |
4218 | /* |
4219 | * find the highest existing sequence number in a directory | |
4220 | * and then set the in-memory index_cnt variable to reflect | |
4221 | * free sequence numbers | |
4222 | */ | |
aec7477b JB |
4223 | static int btrfs_set_inode_index_count(struct inode *inode) |
4224 | { | |
4225 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4226 | struct btrfs_key key, found_key; | |
4227 | struct btrfs_path *path; | |
4228 | struct extent_buffer *leaf; | |
4229 | int ret; | |
4230 | ||
33345d01 | 4231 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4232 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4233 | key.offset = (u64)-1; | |
4234 | ||
4235 | path = btrfs_alloc_path(); | |
4236 | if (!path) | |
4237 | return -ENOMEM; | |
4238 | ||
4239 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4240 | if (ret < 0) | |
4241 | goto out; | |
4242 | /* FIXME: we should be able to handle this */ | |
4243 | if (ret == 0) | |
4244 | goto out; | |
4245 | ret = 0; | |
4246 | ||
4247 | /* | |
4248 | * MAGIC NUMBER EXPLANATION: | |
4249 | * since we search a directory based on f_pos we have to start at 2 | |
4250 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4251 | * else has to start at 2 | |
4252 | */ | |
4253 | if (path->slots[0] == 0) { | |
4254 | BTRFS_I(inode)->index_cnt = 2; | |
4255 | goto out; | |
4256 | } | |
4257 | ||
4258 | path->slots[0]--; | |
4259 | ||
4260 | leaf = path->nodes[0]; | |
4261 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4262 | ||
33345d01 | 4263 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4264 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4265 | BTRFS_I(inode)->index_cnt = 2; | |
4266 | goto out; | |
4267 | } | |
4268 | ||
4269 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4270 | out: | |
4271 | btrfs_free_path(path); | |
4272 | return ret; | |
4273 | } | |
4274 | ||
d352ac68 CM |
4275 | /* |
4276 | * helper to find a free sequence number in a given directory. This current | |
4277 | * code is very simple, later versions will do smarter things in the btree | |
4278 | */ | |
3de4586c | 4279 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4280 | { |
4281 | int ret = 0; | |
4282 | ||
4283 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4284 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4285 | if (ret) { | |
4286 | ret = btrfs_set_inode_index_count(dir); | |
4287 | if (ret) | |
4288 | return ret; | |
4289 | } | |
aec7477b JB |
4290 | } |
4291 | ||
00e4e6b3 | 4292 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4293 | BTRFS_I(dir)->index_cnt++; |
4294 | ||
4295 | return ret; | |
4296 | } | |
4297 | ||
39279cc3 CM |
4298 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4299 | struct btrfs_root *root, | |
aec7477b | 4300 | struct inode *dir, |
9c58309d | 4301 | const char *name, int name_len, |
d82a6f1d JB |
4302 | u64 ref_objectid, u64 objectid, int mode, |
4303 | u64 *index) | |
39279cc3 CM |
4304 | { |
4305 | struct inode *inode; | |
5f39d397 | 4306 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4307 | struct btrfs_key *location; |
5f39d397 | 4308 | struct btrfs_path *path; |
9c58309d CM |
4309 | struct btrfs_inode_ref *ref; |
4310 | struct btrfs_key key[2]; | |
4311 | u32 sizes[2]; | |
4312 | unsigned long ptr; | |
39279cc3 CM |
4313 | int ret; |
4314 | int owner; | |
4315 | ||
5f39d397 | 4316 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4317 | if (!path) |
4318 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4319 | |
39279cc3 | 4320 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4321 | if (!inode) { |
4322 | btrfs_free_path(path); | |
39279cc3 | 4323 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4324 | } |
39279cc3 | 4325 | |
581bb050 LZ |
4326 | /* |
4327 | * we have to initialize this early, so we can reclaim the inode | |
4328 | * number if we fail afterwards in this function. | |
4329 | */ | |
4330 | inode->i_ino = objectid; | |
4331 | ||
aec7477b | 4332 | if (dir) { |
1abe9b8a | 4333 | trace_btrfs_inode_request(dir); |
4334 | ||
3de4586c | 4335 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4336 | if (ret) { |
8fb27640 | 4337 | btrfs_free_path(path); |
09771430 | 4338 | iput(inode); |
aec7477b | 4339 | return ERR_PTR(ret); |
09771430 | 4340 | } |
aec7477b JB |
4341 | } |
4342 | /* | |
4343 | * index_cnt is ignored for everything but a dir, | |
4344 | * btrfs_get_inode_index_count has an explanation for the magic | |
4345 | * number | |
4346 | */ | |
4347 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4348 | BTRFS_I(inode)->root = root; |
e02119d5 | 4349 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4350 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4351 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4352 | |
569254b0 | 4353 | if (S_ISDIR(mode)) |
39279cc3 CM |
4354 | owner = 0; |
4355 | else | |
4356 | owner = 1; | |
9c58309d CM |
4357 | |
4358 | key[0].objectid = objectid; | |
4359 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4360 | key[0].offset = 0; | |
4361 | ||
4362 | key[1].objectid = objectid; | |
4363 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4364 | key[1].offset = ref_objectid; | |
4365 | ||
4366 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4367 | sizes[1] = name_len + sizeof(*ref); | |
4368 | ||
b9473439 | 4369 | path->leave_spinning = 1; |
9c58309d CM |
4370 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4371 | if (ret != 0) | |
5f39d397 CM |
4372 | goto fail; |
4373 | ||
ecc11fab | 4374 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4375 | inode_set_bytes(inode, 0); |
39279cc3 | 4376 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4377 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4378 | struct btrfs_inode_item); | |
e02119d5 | 4379 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4380 | |
4381 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4382 | struct btrfs_inode_ref); | |
4383 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4384 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4385 | ptr = (unsigned long)(ref + 1); |
4386 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4387 | ||
5f39d397 CM |
4388 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4389 | btrfs_free_path(path); | |
4390 | ||
39279cc3 CM |
4391 | location = &BTRFS_I(inode)->location; |
4392 | location->objectid = objectid; | |
39279cc3 CM |
4393 | location->offset = 0; |
4394 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4395 | ||
6cbff00f CH |
4396 | btrfs_inherit_iflags(inode, dir); |
4397 | ||
569254b0 | 4398 | if (S_ISREG(mode)) { |
94272164 CM |
4399 | if (btrfs_test_opt(root, NODATASUM)) |
4400 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4401 | if (btrfs_test_opt(root, NODATACOW) || |
4402 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4403 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4404 | } | |
4405 | ||
39279cc3 | 4406 | insert_inode_hash(inode); |
5d4f98a2 | 4407 | inode_tree_add(inode); |
1abe9b8a | 4408 | |
4409 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4410 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4411 | |
39279cc3 | 4412 | return inode; |
5f39d397 | 4413 | fail: |
aec7477b JB |
4414 | if (dir) |
4415 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4416 | btrfs_free_path(path); |
09771430 | 4417 | iput(inode); |
5f39d397 | 4418 | return ERR_PTR(ret); |
39279cc3 CM |
4419 | } |
4420 | ||
4421 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4422 | { | |
4423 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4424 | } | |
4425 | ||
d352ac68 CM |
4426 | /* |
4427 | * utility function to add 'inode' into 'parent_inode' with | |
4428 | * a give name and a given sequence number. | |
4429 | * if 'add_backref' is true, also insert a backref from the | |
4430 | * inode to the parent directory. | |
4431 | */ | |
e02119d5 CM |
4432 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4433 | struct inode *parent_inode, struct inode *inode, | |
4434 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4435 | { |
4df27c4d | 4436 | int ret = 0; |
39279cc3 | 4437 | struct btrfs_key key; |
e02119d5 | 4438 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4439 | u64 ino = btrfs_ino(inode); |
4440 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4441 | |
33345d01 | 4442 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4443 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4444 | } else { | |
33345d01 | 4445 | key.objectid = ino; |
4df27c4d YZ |
4446 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4447 | key.offset = 0; | |
4448 | } | |
4449 | ||
33345d01 | 4450 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4451 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4452 | key.objectid, root->root_key.objectid, | |
33345d01 | 4453 | parent_ino, index, name, name_len); |
4df27c4d | 4454 | } else if (add_backref) { |
33345d01 LZ |
4455 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4456 | parent_ino, index); | |
4df27c4d | 4457 | } |
39279cc3 | 4458 | |
39279cc3 | 4459 | if (ret == 0) { |
4df27c4d | 4460 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
16cdcec7 | 4461 | parent_inode, &key, |
4df27c4d YZ |
4462 | btrfs_inode_type(inode), index); |
4463 | BUG_ON(ret); | |
4464 | ||
dbe674a9 | 4465 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4466 | name_len * 2); |
79c44584 | 4467 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4468 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4469 | } |
4470 | return ret; | |
4471 | } | |
4472 | ||
4473 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4474 | struct inode *dir, struct dentry *dentry, |
4475 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4476 | { |
a1b075d2 JB |
4477 | int err = btrfs_add_link(trans, dir, inode, |
4478 | dentry->d_name.name, dentry->d_name.len, | |
4479 | backref, index); | |
39279cc3 CM |
4480 | if (!err) { |
4481 | d_instantiate(dentry, inode); | |
4482 | return 0; | |
4483 | } | |
4484 | if (err > 0) | |
4485 | err = -EEXIST; | |
4486 | return err; | |
4487 | } | |
4488 | ||
618e21d5 JB |
4489 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4490 | int mode, dev_t rdev) | |
4491 | { | |
4492 | struct btrfs_trans_handle *trans; | |
4493 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4494 | struct inode *inode = NULL; |
618e21d5 JB |
4495 | int err; |
4496 | int drop_inode = 0; | |
4497 | u64 objectid; | |
1832a6d5 | 4498 | unsigned long nr = 0; |
00e4e6b3 | 4499 | u64 index = 0; |
618e21d5 JB |
4500 | |
4501 | if (!new_valid_dev(rdev)) | |
4502 | return -EINVAL; | |
4503 | ||
9ed74f2d JB |
4504 | /* |
4505 | * 2 for inode item and ref | |
4506 | * 2 for dir items | |
4507 | * 1 for xattr if selinux is on | |
4508 | */ | |
a22285a6 YZ |
4509 | trans = btrfs_start_transaction(root, 5); |
4510 | if (IS_ERR(trans)) | |
4511 | return PTR_ERR(trans); | |
1832a6d5 | 4512 | |
581bb050 LZ |
4513 | err = btrfs_find_free_ino(root, &objectid); |
4514 | if (err) | |
4515 | goto out_unlock; | |
4516 | ||
aec7477b | 4517 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4518 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4519 | mode, &index); |
7cf96da3 TI |
4520 | if (IS_ERR(inode)) { |
4521 | err = PTR_ERR(inode); | |
618e21d5 | 4522 | goto out_unlock; |
7cf96da3 | 4523 | } |
618e21d5 | 4524 | |
2a7dba39 | 4525 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4526 | if (err) { |
4527 | drop_inode = 1; | |
4528 | goto out_unlock; | |
4529 | } | |
4530 | ||
a1b075d2 | 4531 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4532 | if (err) |
4533 | drop_inode = 1; | |
4534 | else { | |
4535 | inode->i_op = &btrfs_special_inode_operations; | |
4536 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4537 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4538 | } |
618e21d5 | 4539 | out_unlock: |
d3c2fdcf | 4540 | nr = trans->blocks_used; |
89ce8a63 | 4541 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4542 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4543 | if (drop_inode) { |
4544 | inode_dec_link_count(inode); | |
4545 | iput(inode); | |
4546 | } | |
618e21d5 JB |
4547 | return err; |
4548 | } | |
4549 | ||
39279cc3 CM |
4550 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4551 | int mode, struct nameidata *nd) | |
4552 | { | |
4553 | struct btrfs_trans_handle *trans; | |
4554 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4555 | struct inode *inode = NULL; |
39279cc3 | 4556 | int drop_inode = 0; |
a22285a6 | 4557 | int err; |
1832a6d5 | 4558 | unsigned long nr = 0; |
39279cc3 | 4559 | u64 objectid; |
00e4e6b3 | 4560 | u64 index = 0; |
39279cc3 | 4561 | |
9ed74f2d JB |
4562 | /* |
4563 | * 2 for inode item and ref | |
4564 | * 2 for dir items | |
4565 | * 1 for xattr if selinux is on | |
4566 | */ | |
a22285a6 YZ |
4567 | trans = btrfs_start_transaction(root, 5); |
4568 | if (IS_ERR(trans)) | |
4569 | return PTR_ERR(trans); | |
9ed74f2d | 4570 | |
581bb050 LZ |
4571 | err = btrfs_find_free_ino(root, &objectid); |
4572 | if (err) | |
4573 | goto out_unlock; | |
4574 | ||
aec7477b | 4575 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4576 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4577 | mode, &index); |
7cf96da3 TI |
4578 | if (IS_ERR(inode)) { |
4579 | err = PTR_ERR(inode); | |
39279cc3 | 4580 | goto out_unlock; |
7cf96da3 | 4581 | } |
39279cc3 | 4582 | |
2a7dba39 | 4583 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4584 | if (err) { |
4585 | drop_inode = 1; | |
4586 | goto out_unlock; | |
4587 | } | |
4588 | ||
a1b075d2 | 4589 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4590 | if (err) |
4591 | drop_inode = 1; | |
4592 | else { | |
4593 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4594 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4595 | inode->i_fop = &btrfs_file_operations; |
4596 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4597 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4598 | } |
39279cc3 | 4599 | out_unlock: |
d3c2fdcf | 4600 | nr = trans->blocks_used; |
ab78c84d | 4601 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4602 | if (drop_inode) { |
4603 | inode_dec_link_count(inode); | |
4604 | iput(inode); | |
4605 | } | |
d3c2fdcf | 4606 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4607 | return err; |
4608 | } | |
4609 | ||
4610 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4611 | struct dentry *dentry) | |
4612 | { | |
4613 | struct btrfs_trans_handle *trans; | |
4614 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4615 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4616 | u64 index; |
1832a6d5 | 4617 | unsigned long nr = 0; |
39279cc3 CM |
4618 | int err; |
4619 | int drop_inode = 0; | |
4620 | ||
4a8be425 TH |
4621 | /* do not allow sys_link's with other subvols of the same device */ |
4622 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4623 | return -EXDEV; |
4a8be425 | 4624 | |
c055e99e AV |
4625 | if (inode->i_nlink == ~0U) |
4626 | return -EMLINK; | |
4a8be425 | 4627 | |
3de4586c | 4628 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4629 | if (err) |
4630 | goto fail; | |
4631 | ||
a22285a6 | 4632 | /* |
7e6b6465 | 4633 | * 2 items for inode and inode ref |
a22285a6 | 4634 | * 2 items for dir items |
7e6b6465 | 4635 | * 1 item for parent inode |
a22285a6 | 4636 | */ |
7e6b6465 | 4637 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4638 | if (IS_ERR(trans)) { |
4639 | err = PTR_ERR(trans); | |
4640 | goto fail; | |
4641 | } | |
5f39d397 | 4642 | |
3153495d MX |
4643 | btrfs_inc_nlink(inode); |
4644 | inode->i_ctime = CURRENT_TIME; | |
7de9c6ee | 4645 | ihold(inode); |
aec7477b | 4646 | |
a1b075d2 | 4647 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4648 | |
a5719521 | 4649 | if (err) { |
54aa1f4d | 4650 | drop_inode = 1; |
a5719521 | 4651 | } else { |
10d9f309 | 4652 | struct dentry *parent = dentry->d_parent; |
a5719521 YZ |
4653 | err = btrfs_update_inode(trans, root, inode); |
4654 | BUG_ON(err); | |
6a912213 | 4655 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 4656 | } |
39279cc3 | 4657 | |
d3c2fdcf | 4658 | nr = trans->blocks_used; |
ab78c84d | 4659 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4660 | fail: |
39279cc3 CM |
4661 | if (drop_inode) { |
4662 | inode_dec_link_count(inode); | |
4663 | iput(inode); | |
4664 | } | |
d3c2fdcf | 4665 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4666 | return err; |
4667 | } | |
4668 | ||
39279cc3 CM |
4669 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4670 | { | |
b9d86667 | 4671 | struct inode *inode = NULL; |
39279cc3 CM |
4672 | struct btrfs_trans_handle *trans; |
4673 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4674 | int err = 0; | |
4675 | int drop_on_err = 0; | |
b9d86667 | 4676 | u64 objectid = 0; |
00e4e6b3 | 4677 | u64 index = 0; |
d3c2fdcf | 4678 | unsigned long nr = 1; |
39279cc3 | 4679 | |
9ed74f2d JB |
4680 | /* |
4681 | * 2 items for inode and ref | |
4682 | * 2 items for dir items | |
4683 | * 1 for xattr if selinux is on | |
4684 | */ | |
a22285a6 YZ |
4685 | trans = btrfs_start_transaction(root, 5); |
4686 | if (IS_ERR(trans)) | |
4687 | return PTR_ERR(trans); | |
39279cc3 | 4688 | |
581bb050 LZ |
4689 | err = btrfs_find_free_ino(root, &objectid); |
4690 | if (err) | |
4691 | goto out_fail; | |
4692 | ||
aec7477b | 4693 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4694 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4695 | S_IFDIR | mode, &index); |
39279cc3 CM |
4696 | if (IS_ERR(inode)) { |
4697 | err = PTR_ERR(inode); | |
4698 | goto out_fail; | |
4699 | } | |
5f39d397 | 4700 | |
39279cc3 | 4701 | drop_on_err = 1; |
33268eaf | 4702 | |
2a7dba39 | 4703 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4704 | if (err) |
4705 | goto out_fail; | |
4706 | ||
39279cc3 CM |
4707 | inode->i_op = &btrfs_dir_inode_operations; |
4708 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 4709 | |
dbe674a9 | 4710 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4711 | err = btrfs_update_inode(trans, root, inode); |
4712 | if (err) | |
4713 | goto out_fail; | |
5f39d397 | 4714 | |
a1b075d2 JB |
4715 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4716 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4717 | if (err) |
4718 | goto out_fail; | |
5f39d397 | 4719 | |
39279cc3 CM |
4720 | d_instantiate(dentry, inode); |
4721 | drop_on_err = 0; | |
39279cc3 CM |
4722 | |
4723 | out_fail: | |
d3c2fdcf | 4724 | nr = trans->blocks_used; |
ab78c84d | 4725 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4726 | if (drop_on_err) |
4727 | iput(inode); | |
d3c2fdcf | 4728 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4729 | return err; |
4730 | } | |
4731 | ||
d352ac68 CM |
4732 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4733 | * and an extent that you want to insert, deal with overlap and insert | |
4734 | * the new extent into the tree. | |
4735 | */ | |
3b951516 CM |
4736 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4737 | struct extent_map *existing, | |
e6dcd2dc CM |
4738 | struct extent_map *em, |
4739 | u64 map_start, u64 map_len) | |
3b951516 CM |
4740 | { |
4741 | u64 start_diff; | |
3b951516 | 4742 | |
e6dcd2dc CM |
4743 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4744 | start_diff = map_start - em->start; | |
4745 | em->start = map_start; | |
4746 | em->len = map_len; | |
c8b97818 CM |
4747 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4748 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4749 | em->block_start += start_diff; |
c8b97818 CM |
4750 | em->block_len -= start_diff; |
4751 | } | |
e6dcd2dc | 4752 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4753 | } |
4754 | ||
c8b97818 CM |
4755 | static noinline int uncompress_inline(struct btrfs_path *path, |
4756 | struct inode *inode, struct page *page, | |
4757 | size_t pg_offset, u64 extent_offset, | |
4758 | struct btrfs_file_extent_item *item) | |
4759 | { | |
4760 | int ret; | |
4761 | struct extent_buffer *leaf = path->nodes[0]; | |
4762 | char *tmp; | |
4763 | size_t max_size; | |
4764 | unsigned long inline_size; | |
4765 | unsigned long ptr; | |
261507a0 | 4766 | int compress_type; |
c8b97818 CM |
4767 | |
4768 | WARN_ON(pg_offset != 0); | |
261507a0 | 4769 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
4770 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
4771 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4772 | btrfs_item_nr(leaf, path->slots[0])); | |
4773 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
4774 | if (!tmp) |
4775 | return -ENOMEM; | |
c8b97818 CM |
4776 | ptr = btrfs_file_extent_inline_start(item); |
4777 | ||
4778 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4779 | ||
5b050f04 | 4780 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
4781 | ret = btrfs_decompress(compress_type, tmp, page, |
4782 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
4783 | if (ret) { |
4784 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4785 | unsigned long copy_size = min_t(u64, | |
4786 | PAGE_CACHE_SIZE - pg_offset, | |
4787 | max_size - extent_offset); | |
4788 | memset(kaddr + pg_offset, 0, copy_size); | |
4789 | kunmap_atomic(kaddr, KM_USER0); | |
4790 | } | |
4791 | kfree(tmp); | |
4792 | return 0; | |
4793 | } | |
4794 | ||
d352ac68 CM |
4795 | /* |
4796 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4797 | * the ugly parts come from merging extents from the disk with the in-ram |
4798 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4799 | * where the in-ram extents might be locked pending data=ordered completion. |
4800 | * | |
4801 | * This also copies inline extents directly into the page. | |
4802 | */ | |
d397712b | 4803 | |
a52d9a80 | 4804 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4805 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4806 | int create) |
4807 | { | |
4808 | int ret; | |
4809 | int err = 0; | |
db94535d | 4810 | u64 bytenr; |
a52d9a80 CM |
4811 | u64 extent_start = 0; |
4812 | u64 extent_end = 0; | |
33345d01 | 4813 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 4814 | u32 found_type; |
f421950f | 4815 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4816 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4817 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4818 | struct extent_buffer *leaf; |
4819 | struct btrfs_key found_key; | |
a52d9a80 CM |
4820 | struct extent_map *em = NULL; |
4821 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4822 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4823 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 4824 | int compress_type; |
a52d9a80 | 4825 | |
a52d9a80 | 4826 | again: |
890871be | 4827 | read_lock(&em_tree->lock); |
d1310b2e | 4828 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4829 | if (em) |
4830 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4831 | read_unlock(&em_tree->lock); |
d1310b2e | 4832 | |
a52d9a80 | 4833 | if (em) { |
e1c4b745 CM |
4834 | if (em->start > start || em->start + em->len <= start) |
4835 | free_extent_map(em); | |
4836 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4837 | free_extent_map(em); |
4838 | else | |
4839 | goto out; | |
a52d9a80 | 4840 | } |
172ddd60 | 4841 | em = alloc_extent_map(); |
a52d9a80 | 4842 | if (!em) { |
d1310b2e CM |
4843 | err = -ENOMEM; |
4844 | goto out; | |
a52d9a80 | 4845 | } |
e6dcd2dc | 4846 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4847 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4848 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4849 | em->len = (u64)-1; |
c8b97818 | 4850 | em->block_len = (u64)-1; |
f421950f CM |
4851 | |
4852 | if (!path) { | |
4853 | path = btrfs_alloc_path(); | |
026fd317 JB |
4854 | if (!path) { |
4855 | err = -ENOMEM; | |
4856 | goto out; | |
4857 | } | |
4858 | /* | |
4859 | * Chances are we'll be called again, so go ahead and do | |
4860 | * readahead | |
4861 | */ | |
4862 | path->reada = 1; | |
f421950f CM |
4863 | } |
4864 | ||
179e29e4 CM |
4865 | ret = btrfs_lookup_file_extent(trans, root, path, |
4866 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4867 | if (ret < 0) { |
4868 | err = ret; | |
4869 | goto out; | |
4870 | } | |
4871 | ||
4872 | if (ret != 0) { | |
4873 | if (path->slots[0] == 0) | |
4874 | goto not_found; | |
4875 | path->slots[0]--; | |
4876 | } | |
4877 | ||
5f39d397 CM |
4878 | leaf = path->nodes[0]; |
4879 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4880 | struct btrfs_file_extent_item); |
a52d9a80 | 4881 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4882 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4883 | found_type = btrfs_key_type(&found_key); | |
4884 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4885 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4886 | goto not_found; | |
4887 | } | |
4888 | ||
5f39d397 CM |
4889 | found_type = btrfs_file_extent_type(leaf, item); |
4890 | extent_start = found_key.offset; | |
261507a0 | 4891 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4892 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4893 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4894 | extent_end = extent_start + |
db94535d | 4895 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4896 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4897 | size_t size; | |
4898 | size = btrfs_file_extent_inline_len(leaf, item); | |
4899 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4900 | ~((u64)root->sectorsize - 1); | |
4901 | } | |
4902 | ||
4903 | if (start >= extent_end) { | |
4904 | path->slots[0]++; | |
4905 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4906 | ret = btrfs_next_leaf(root, path); | |
4907 | if (ret < 0) { | |
4908 | err = ret; | |
4909 | goto out; | |
a52d9a80 | 4910 | } |
9036c102 YZ |
4911 | if (ret > 0) |
4912 | goto not_found; | |
4913 | leaf = path->nodes[0]; | |
a52d9a80 | 4914 | } |
9036c102 YZ |
4915 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4916 | if (found_key.objectid != objectid || | |
4917 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4918 | goto not_found; | |
4919 | if (start + len <= found_key.offset) | |
4920 | goto not_found; | |
4921 | em->start = start; | |
4922 | em->len = found_key.offset - start; | |
4923 | goto not_found_em; | |
4924 | } | |
4925 | ||
d899e052 YZ |
4926 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4927 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4928 | em->start = extent_start; |
4929 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4930 | em->orig_start = extent_start - |
4931 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4932 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4933 | if (bytenr == 0) { | |
5f39d397 | 4934 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4935 | goto insert; |
4936 | } | |
261507a0 | 4937 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 4938 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 4939 | em->compress_type = compress_type; |
c8b97818 CM |
4940 | em->block_start = bytenr; |
4941 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4942 | item); | |
4943 | } else { | |
4944 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4945 | em->block_start = bytenr; | |
4946 | em->block_len = em->len; | |
d899e052 YZ |
4947 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4948 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4949 | } |
a52d9a80 CM |
4950 | goto insert; |
4951 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4952 | unsigned long ptr; |
a52d9a80 | 4953 | char *map; |
3326d1b0 CM |
4954 | size_t size; |
4955 | size_t extent_offset; | |
4956 | size_t copy_size; | |
a52d9a80 | 4957 | |
689f9346 | 4958 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4959 | if (!page || create) { |
689f9346 | 4960 | em->start = extent_start; |
9036c102 | 4961 | em->len = extent_end - extent_start; |
689f9346 Y |
4962 | goto out; |
4963 | } | |
5f39d397 | 4964 | |
9036c102 YZ |
4965 | size = btrfs_file_extent_inline_len(leaf, item); |
4966 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4967 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4968 | size - extent_offset); |
3326d1b0 | 4969 | em->start = extent_start + extent_offset; |
70dec807 CM |
4970 | em->len = (copy_size + root->sectorsize - 1) & |
4971 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4972 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 4973 | if (compress_type) { |
c8b97818 | 4974 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
4975 | em->compress_type = compress_type; |
4976 | } | |
689f9346 | 4977 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4978 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
4979 | if (btrfs_file_extent_compression(leaf, item) != |
4980 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
4981 | ret = uncompress_inline(path, inode, page, |
4982 | pg_offset, | |
4983 | extent_offset, item); | |
4984 | BUG_ON(ret); | |
4985 | } else { | |
4986 | map = kmap(page); | |
4987 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4988 | copy_size); | |
93c82d57 CM |
4989 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4990 | memset(map + pg_offset + copy_size, 0, | |
4991 | PAGE_CACHE_SIZE - pg_offset - | |
4992 | copy_size); | |
4993 | } | |
c8b97818 CM |
4994 | kunmap(page); |
4995 | } | |
179e29e4 CM |
4996 | flush_dcache_page(page); |
4997 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 4998 | WARN_ON(1); |
179e29e4 CM |
4999 | if (!trans) { |
5000 | kunmap(page); | |
5001 | free_extent_map(em); | |
5002 | em = NULL; | |
ff5714cc | 5003 | |
b3b4aa74 | 5004 | btrfs_release_path(path); |
7a7eaa40 | 5005 | trans = btrfs_join_transaction(root); |
ff5714cc | 5006 | |
3612b495 TI |
5007 | if (IS_ERR(trans)) |
5008 | return ERR_CAST(trans); | |
179e29e4 CM |
5009 | goto again; |
5010 | } | |
c8b97818 | 5011 | map = kmap(page); |
70dec807 | 5012 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5013 | copy_size); |
c8b97818 | 5014 | kunmap(page); |
179e29e4 | 5015 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5016 | } |
d1310b2e | 5017 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5018 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5019 | goto insert; |
5020 | } else { | |
d397712b | 5021 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5022 | WARN_ON(1); |
5023 | } | |
5024 | not_found: | |
5025 | em->start = start; | |
d1310b2e | 5026 | em->len = len; |
a52d9a80 | 5027 | not_found_em: |
5f39d397 | 5028 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5029 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5030 | insert: |
b3b4aa74 | 5031 | btrfs_release_path(path); |
d1310b2e | 5032 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5033 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5034 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5035 | (unsigned long long)em->len, | |
5036 | (unsigned long long)start, | |
5037 | (unsigned long long)len); | |
a52d9a80 CM |
5038 | err = -EIO; |
5039 | goto out; | |
5040 | } | |
d1310b2e CM |
5041 | |
5042 | err = 0; | |
890871be | 5043 | write_lock(&em_tree->lock); |
a52d9a80 | 5044 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5045 | /* it is possible that someone inserted the extent into the tree |
5046 | * while we had the lock dropped. It is also possible that | |
5047 | * an overlapping map exists in the tree | |
5048 | */ | |
a52d9a80 | 5049 | if (ret == -EEXIST) { |
3b951516 | 5050 | struct extent_map *existing; |
e6dcd2dc CM |
5051 | |
5052 | ret = 0; | |
5053 | ||
3b951516 | 5054 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5055 | if (existing && (existing->start > start || |
5056 | existing->start + existing->len <= start)) { | |
5057 | free_extent_map(existing); | |
5058 | existing = NULL; | |
5059 | } | |
3b951516 CM |
5060 | if (!existing) { |
5061 | existing = lookup_extent_mapping(em_tree, em->start, | |
5062 | em->len); | |
5063 | if (existing) { | |
5064 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5065 | em, start, |
5066 | root->sectorsize); | |
3b951516 CM |
5067 | free_extent_map(existing); |
5068 | if (err) { | |
5069 | free_extent_map(em); | |
5070 | em = NULL; | |
5071 | } | |
5072 | } else { | |
5073 | err = -EIO; | |
3b951516 CM |
5074 | free_extent_map(em); |
5075 | em = NULL; | |
5076 | } | |
5077 | } else { | |
5078 | free_extent_map(em); | |
5079 | em = existing; | |
e6dcd2dc | 5080 | err = 0; |
a52d9a80 | 5081 | } |
a52d9a80 | 5082 | } |
890871be | 5083 | write_unlock(&em_tree->lock); |
a52d9a80 | 5084 | out: |
1abe9b8a | 5085 | |
5086 | trace_btrfs_get_extent(root, em); | |
5087 | ||
f421950f CM |
5088 | if (path) |
5089 | btrfs_free_path(path); | |
a52d9a80 CM |
5090 | if (trans) { |
5091 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5092 | if (!err) |
a52d9a80 CM |
5093 | err = ret; |
5094 | } | |
a52d9a80 CM |
5095 | if (err) { |
5096 | free_extent_map(em); | |
a52d9a80 CM |
5097 | return ERR_PTR(err); |
5098 | } | |
5099 | return em; | |
5100 | } | |
5101 | ||
ec29ed5b CM |
5102 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5103 | size_t pg_offset, u64 start, u64 len, | |
5104 | int create) | |
5105 | { | |
5106 | struct extent_map *em; | |
5107 | struct extent_map *hole_em = NULL; | |
5108 | u64 range_start = start; | |
5109 | u64 end; | |
5110 | u64 found; | |
5111 | u64 found_end; | |
5112 | int err = 0; | |
5113 | ||
5114 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5115 | if (IS_ERR(em)) | |
5116 | return em; | |
5117 | if (em) { | |
5118 | /* | |
5119 | * if our em maps to a hole, there might | |
5120 | * actually be delalloc bytes behind it | |
5121 | */ | |
5122 | if (em->block_start != EXTENT_MAP_HOLE) | |
5123 | return em; | |
5124 | else | |
5125 | hole_em = em; | |
5126 | } | |
5127 | ||
5128 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5129 | end = start + len; | |
5130 | if (end < start) | |
5131 | end = (u64)-1; | |
5132 | else | |
5133 | end -= 1; | |
5134 | ||
5135 | em = NULL; | |
5136 | ||
5137 | /* ok, we didn't find anything, lets look for delalloc */ | |
5138 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5139 | end, len, EXTENT_DELALLOC, 1); | |
5140 | found_end = range_start + found; | |
5141 | if (found_end < range_start) | |
5142 | found_end = (u64)-1; | |
5143 | ||
5144 | /* | |
5145 | * we didn't find anything useful, return | |
5146 | * the original results from get_extent() | |
5147 | */ | |
5148 | if (range_start > end || found_end <= start) { | |
5149 | em = hole_em; | |
5150 | hole_em = NULL; | |
5151 | goto out; | |
5152 | } | |
5153 | ||
5154 | /* adjust the range_start to make sure it doesn't | |
5155 | * go backwards from the start they passed in | |
5156 | */ | |
5157 | range_start = max(start,range_start); | |
5158 | found = found_end - range_start; | |
5159 | ||
5160 | if (found > 0) { | |
5161 | u64 hole_start = start; | |
5162 | u64 hole_len = len; | |
5163 | ||
172ddd60 | 5164 | em = alloc_extent_map(); |
ec29ed5b CM |
5165 | if (!em) { |
5166 | err = -ENOMEM; | |
5167 | goto out; | |
5168 | } | |
5169 | /* | |
5170 | * when btrfs_get_extent can't find anything it | |
5171 | * returns one huge hole | |
5172 | * | |
5173 | * make sure what it found really fits our range, and | |
5174 | * adjust to make sure it is based on the start from | |
5175 | * the caller | |
5176 | */ | |
5177 | if (hole_em) { | |
5178 | u64 calc_end = extent_map_end(hole_em); | |
5179 | ||
5180 | if (calc_end <= start || (hole_em->start > end)) { | |
5181 | free_extent_map(hole_em); | |
5182 | hole_em = NULL; | |
5183 | } else { | |
5184 | hole_start = max(hole_em->start, start); | |
5185 | hole_len = calc_end - hole_start; | |
5186 | } | |
5187 | } | |
5188 | em->bdev = NULL; | |
5189 | if (hole_em && range_start > hole_start) { | |
5190 | /* our hole starts before our delalloc, so we | |
5191 | * have to return just the parts of the hole | |
5192 | * that go until the delalloc starts | |
5193 | */ | |
5194 | em->len = min(hole_len, | |
5195 | range_start - hole_start); | |
5196 | em->start = hole_start; | |
5197 | em->orig_start = hole_start; | |
5198 | /* | |
5199 | * don't adjust block start at all, | |
5200 | * it is fixed at EXTENT_MAP_HOLE | |
5201 | */ | |
5202 | em->block_start = hole_em->block_start; | |
5203 | em->block_len = hole_len; | |
5204 | } else { | |
5205 | em->start = range_start; | |
5206 | em->len = found; | |
5207 | em->orig_start = range_start; | |
5208 | em->block_start = EXTENT_MAP_DELALLOC; | |
5209 | em->block_len = found; | |
5210 | } | |
5211 | } else if (hole_em) { | |
5212 | return hole_em; | |
5213 | } | |
5214 | out: | |
5215 | ||
5216 | free_extent_map(hole_em); | |
5217 | if (err) { | |
5218 | free_extent_map(em); | |
5219 | return ERR_PTR(err); | |
5220 | } | |
5221 | return em; | |
5222 | } | |
5223 | ||
4b46fce2 | 5224 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5225 | struct extent_map *em, |
4b46fce2 JB |
5226 | u64 start, u64 len) |
5227 | { | |
5228 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5229 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5230 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5231 | struct btrfs_key ins; | |
5232 | u64 alloc_hint; | |
5233 | int ret; | |
16d299ac | 5234 | bool insert = false; |
4b46fce2 | 5235 | |
16d299ac JB |
5236 | /* |
5237 | * Ok if the extent map we looked up is a hole and is for the exact | |
5238 | * range we want, there is no reason to allocate a new one, however if | |
5239 | * it is not right then we need to free this one and drop the cache for | |
5240 | * our range. | |
5241 | */ | |
5242 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5243 | em->len != len) { | |
5244 | free_extent_map(em); | |
5245 | em = NULL; | |
5246 | insert = true; | |
5247 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5248 | } | |
4b46fce2 | 5249 | |
7a7eaa40 | 5250 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5251 | if (IS_ERR(trans)) |
5252 | return ERR_CAST(trans); | |
4b46fce2 | 5253 | |
4cb5300b CM |
5254 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5255 | btrfs_add_inode_defrag(trans, inode); | |
5256 | ||
4b46fce2 JB |
5257 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5258 | ||
5259 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5260 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5261 | alloc_hint, (u64)-1, &ins, 1); | |
5262 | if (ret) { | |
5263 | em = ERR_PTR(ret); | |
5264 | goto out; | |
5265 | } | |
5266 | ||
4b46fce2 | 5267 | if (!em) { |
172ddd60 | 5268 | em = alloc_extent_map(); |
16d299ac JB |
5269 | if (!em) { |
5270 | em = ERR_PTR(-ENOMEM); | |
5271 | goto out; | |
5272 | } | |
4b46fce2 JB |
5273 | } |
5274 | ||
5275 | em->start = start; | |
5276 | em->orig_start = em->start; | |
5277 | em->len = ins.offset; | |
5278 | ||
5279 | em->block_start = ins.objectid; | |
5280 | em->block_len = ins.offset; | |
5281 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5282 | |
5283 | /* | |
5284 | * We need to do this because if we're using the original em we searched | |
5285 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5286 | */ | |
5287 | em->flags = 0; | |
4b46fce2 JB |
5288 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5289 | ||
16d299ac | 5290 | while (insert) { |
4b46fce2 JB |
5291 | write_lock(&em_tree->lock); |
5292 | ret = add_extent_mapping(em_tree, em); | |
5293 | write_unlock(&em_tree->lock); | |
5294 | if (ret != -EEXIST) | |
5295 | break; | |
5296 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5297 | } | |
5298 | ||
5299 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5300 | ins.offset, ins.offset, 0); | |
5301 | if (ret) { | |
5302 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5303 | em = ERR_PTR(ret); | |
5304 | } | |
5305 | out: | |
5306 | btrfs_end_transaction(trans, root); | |
5307 | return em; | |
5308 | } | |
5309 | ||
46bfbb5c CM |
5310 | /* |
5311 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5312 | * block must be cow'd | |
5313 | */ | |
5314 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5315 | struct inode *inode, u64 offset, u64 len) | |
5316 | { | |
5317 | struct btrfs_path *path; | |
5318 | int ret; | |
5319 | struct extent_buffer *leaf; | |
5320 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5321 | struct btrfs_file_extent_item *fi; | |
5322 | struct btrfs_key key; | |
5323 | u64 disk_bytenr; | |
5324 | u64 backref_offset; | |
5325 | u64 extent_end; | |
5326 | u64 num_bytes; | |
5327 | int slot; | |
5328 | int found_type; | |
5329 | ||
5330 | path = btrfs_alloc_path(); | |
5331 | if (!path) | |
5332 | return -ENOMEM; | |
5333 | ||
33345d01 | 5334 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5335 | offset, 0); |
5336 | if (ret < 0) | |
5337 | goto out; | |
5338 | ||
5339 | slot = path->slots[0]; | |
5340 | if (ret == 1) { | |
5341 | if (slot == 0) { | |
5342 | /* can't find the item, must cow */ | |
5343 | ret = 0; | |
5344 | goto out; | |
5345 | } | |
5346 | slot--; | |
5347 | } | |
5348 | ret = 0; | |
5349 | leaf = path->nodes[0]; | |
5350 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5351 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5352 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5353 | /* not our file or wrong item type, must cow */ | |
5354 | goto out; | |
5355 | } | |
5356 | ||
5357 | if (key.offset > offset) { | |
5358 | /* Wrong offset, must cow */ | |
5359 | goto out; | |
5360 | } | |
5361 | ||
5362 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5363 | found_type = btrfs_file_extent_type(leaf, fi); | |
5364 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5365 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5366 | /* not a regular extent, must cow */ | |
5367 | goto out; | |
5368 | } | |
5369 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5370 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5371 | ||
5372 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5373 | if (extent_end < offset + len) { | |
5374 | /* extent doesn't include our full range, must cow */ | |
5375 | goto out; | |
5376 | } | |
5377 | ||
5378 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5379 | goto out; | |
5380 | ||
5381 | /* | |
5382 | * look for other files referencing this extent, if we | |
5383 | * find any we must cow | |
5384 | */ | |
33345d01 | 5385 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5386 | key.offset - backref_offset, disk_bytenr)) |
5387 | goto out; | |
5388 | ||
5389 | /* | |
5390 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5391 | * in this extent we are about to write. If there | |
5392 | * are any csums in that range we have to cow in order | |
5393 | * to keep the csums correct | |
5394 | */ | |
5395 | disk_bytenr += backref_offset; | |
5396 | disk_bytenr += offset - key.offset; | |
5397 | num_bytes = min(offset + len, extent_end) - offset; | |
5398 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5399 | goto out; | |
5400 | /* | |
5401 | * all of the above have passed, it is safe to overwrite this extent | |
5402 | * without cow | |
5403 | */ | |
5404 | ret = 1; | |
5405 | out: | |
5406 | btrfs_free_path(path); | |
5407 | return ret; | |
5408 | } | |
5409 | ||
4b46fce2 JB |
5410 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5411 | struct buffer_head *bh_result, int create) | |
5412 | { | |
5413 | struct extent_map *em; | |
5414 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5415 | u64 start = iblock << inode->i_blkbits; | |
5416 | u64 len = bh_result->b_size; | |
46bfbb5c | 5417 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5418 | |
5419 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5420 | if (IS_ERR(em)) | |
5421 | return PTR_ERR(em); | |
5422 | ||
5423 | /* | |
5424 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5425 | * io. INLINE is special, and we could probably kludge it in here, but | |
5426 | * it's still buffered so for safety lets just fall back to the generic | |
5427 | * buffered path. | |
5428 | * | |
5429 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5430 | * decompress it, so there will be buffering required no matter what we | |
5431 | * do, so go ahead and fallback to buffered. | |
5432 | * | |
5433 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5434 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5435 | * the generic code. | |
5436 | */ | |
5437 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5438 | em->block_start == EXTENT_MAP_INLINE) { | |
5439 | free_extent_map(em); | |
5440 | return -ENOTBLK; | |
5441 | } | |
5442 | ||
5443 | /* Just a good old fashioned hole, return */ | |
5444 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5445 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5446 | free_extent_map(em); | |
5447 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5448 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5449 | start + root->sectorsize - 1, GFP_NOFS); | |
5450 | return 0; | |
5451 | } | |
5452 | ||
5453 | /* | |
5454 | * We don't allocate a new extent in the following cases | |
5455 | * | |
5456 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5457 | * existing extent. | |
5458 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5459 | * just use the extent. | |
5460 | * | |
5461 | */ | |
46bfbb5c CM |
5462 | if (!create) { |
5463 | len = em->len - (start - em->start); | |
4b46fce2 | 5464 | goto map; |
46bfbb5c | 5465 | } |
4b46fce2 JB |
5466 | |
5467 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5468 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5469 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5470 | int type; |
5471 | int ret; | |
46bfbb5c | 5472 | u64 block_start; |
4b46fce2 JB |
5473 | |
5474 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5475 | type = BTRFS_ORDERED_PREALLOC; | |
5476 | else | |
5477 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5478 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5479 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5480 | |
5481 | /* | |
5482 | * we're not going to log anything, but we do need | |
5483 | * to make sure the current transaction stays open | |
5484 | * while we look for nocow cross refs | |
5485 | */ | |
7a7eaa40 | 5486 | trans = btrfs_join_transaction(root); |
3612b495 | 5487 | if (IS_ERR(trans)) |
46bfbb5c CM |
5488 | goto must_cow; |
5489 | ||
5490 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5491 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5492 | block_start, len, len, type); | |
5493 | btrfs_end_transaction(trans, root); | |
5494 | if (ret) { | |
5495 | free_extent_map(em); | |
5496 | return ret; | |
5497 | } | |
5498 | goto unlock; | |
4b46fce2 | 5499 | } |
46bfbb5c | 5500 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5501 | } |
46bfbb5c CM |
5502 | must_cow: |
5503 | /* | |
5504 | * this will cow the extent, reset the len in case we changed | |
5505 | * it above | |
5506 | */ | |
5507 | len = bh_result->b_size; | |
16d299ac | 5508 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5509 | if (IS_ERR(em)) |
5510 | return PTR_ERR(em); | |
5511 | len = min(len, em->len - (start - em->start)); | |
5512 | unlock: | |
4845e44f CM |
5513 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5514 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5515 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5516 | map: |
5517 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5518 | inode->i_blkbits; | |
46bfbb5c | 5519 | bh_result->b_size = len; |
4b46fce2 JB |
5520 | bh_result->b_bdev = em->bdev; |
5521 | set_buffer_mapped(bh_result); | |
5522 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5523 | set_buffer_new(bh_result); | |
5524 | ||
5525 | free_extent_map(em); | |
5526 | ||
5527 | return 0; | |
5528 | } | |
5529 | ||
5530 | struct btrfs_dio_private { | |
5531 | struct inode *inode; | |
5532 | u64 logical_offset; | |
5533 | u64 disk_bytenr; | |
5534 | u64 bytes; | |
5535 | u32 *csums; | |
5536 | void *private; | |
e65e1535 MX |
5537 | |
5538 | /* number of bios pending for this dio */ | |
5539 | atomic_t pending_bios; | |
5540 | ||
5541 | /* IO errors */ | |
5542 | int errors; | |
5543 | ||
5544 | struct bio *orig_bio; | |
4b46fce2 JB |
5545 | }; |
5546 | ||
5547 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5548 | { | |
e65e1535 | 5549 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5550 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5551 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5552 | struct inode *inode = dip->inode; |
5553 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5554 | u64 start; | |
5555 | u32 *private = dip->csums; | |
5556 | ||
5557 | start = dip->logical_offset; | |
5558 | do { | |
5559 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5560 | struct page *page = bvec->bv_page; | |
5561 | char *kaddr; | |
5562 | u32 csum = ~(u32)0; | |
5563 | unsigned long flags; | |
5564 | ||
5565 | local_irq_save(flags); | |
5566 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5567 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5568 | csum, bvec->bv_len); | |
5569 | btrfs_csum_final(csum, (char *)&csum); | |
5570 | kunmap_atomic(kaddr, KM_IRQ0); | |
5571 | local_irq_restore(flags); | |
5572 | ||
5573 | flush_dcache_page(bvec->bv_page); | |
5574 | if (csum != *private) { | |
33345d01 | 5575 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 5576 | " %llu csum %u private %u\n", |
33345d01 LZ |
5577 | (unsigned long long)btrfs_ino(inode), |
5578 | (unsigned long long)start, | |
4b46fce2 JB |
5579 | csum, *private); |
5580 | err = -EIO; | |
5581 | } | |
5582 | } | |
5583 | ||
5584 | start += bvec->bv_len; | |
5585 | private++; | |
5586 | bvec++; | |
5587 | } while (bvec <= bvec_end); | |
5588 | ||
5589 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5590 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5591 | bio->bi_private = dip->private; | |
5592 | ||
5593 | kfree(dip->csums); | |
5594 | kfree(dip); | |
c0da7aa1 JB |
5595 | |
5596 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5597 | if (err) | |
5598 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5599 | dio_end_io(bio, err); |
5600 | } | |
5601 | ||
5602 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5603 | { | |
5604 | struct btrfs_dio_private *dip = bio->bi_private; | |
5605 | struct inode *inode = dip->inode; | |
5606 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5607 | struct btrfs_trans_handle *trans; | |
5608 | struct btrfs_ordered_extent *ordered = NULL; | |
5609 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5610 | u64 ordered_offset = dip->logical_offset; |
5611 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5612 | int ret; |
5613 | ||
5614 | if (err) | |
5615 | goto out_done; | |
163cf09c CM |
5616 | again: |
5617 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5618 | &ordered_offset, | |
5619 | ordered_bytes); | |
4b46fce2 | 5620 | if (!ret) |
163cf09c | 5621 | goto out_test; |
4b46fce2 JB |
5622 | |
5623 | BUG_ON(!ordered); | |
5624 | ||
7a7eaa40 | 5625 | trans = btrfs_join_transaction(root); |
3612b495 | 5626 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5627 | err = -ENOMEM; |
5628 | goto out; | |
5629 | } | |
5630 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5631 | ||
5632 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5633 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5634 | if (!ret) | |
f0dd9592 | 5635 | err = btrfs_update_inode(trans, root, inode); |
4b46fce2 JB |
5636 | goto out; |
5637 | } | |
5638 | ||
5639 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5640 | ordered->file_offset + ordered->len - 1, 0, | |
5641 | &cached_state, GFP_NOFS); | |
5642 | ||
5643 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5644 | ret = btrfs_mark_extent_written(trans, inode, | |
5645 | ordered->file_offset, | |
5646 | ordered->file_offset + | |
5647 | ordered->len); | |
5648 | if (ret) { | |
5649 | err = ret; | |
5650 | goto out_unlock; | |
5651 | } | |
5652 | } else { | |
5653 | ret = insert_reserved_file_extent(trans, inode, | |
5654 | ordered->file_offset, | |
5655 | ordered->start, | |
5656 | ordered->disk_len, | |
5657 | ordered->len, | |
5658 | ordered->len, | |
5659 | 0, 0, 0, | |
5660 | BTRFS_FILE_EXTENT_REG); | |
5661 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5662 | ordered->file_offset, ordered->len); | |
5663 | if (ret) { | |
5664 | err = ret; | |
5665 | WARN_ON(1); | |
5666 | goto out_unlock; | |
5667 | } | |
5668 | } | |
5669 | ||
5670 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
1ef30be1 | 5671 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); |
a39f7521 | 5672 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) |
1ef30be1 JB |
5673 | btrfs_update_inode(trans, root, inode); |
5674 | ret = 0; | |
4b46fce2 JB |
5675 | out_unlock: |
5676 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5677 | ordered->file_offset + ordered->len - 1, | |
5678 | &cached_state, GFP_NOFS); | |
5679 | out: | |
5680 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5681 | btrfs_end_transaction(trans, root); | |
163cf09c | 5682 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5683 | btrfs_put_ordered_extent(ordered); |
5684 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5685 | |
5686 | out_test: | |
5687 | /* | |
5688 | * our bio might span multiple ordered extents. If we haven't | |
5689 | * completed the accounting for the whole dio, go back and try again | |
5690 | */ | |
5691 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5692 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5693 | ordered_offset; | |
5694 | goto again; | |
5695 | } | |
4b46fce2 JB |
5696 | out_done: |
5697 | bio->bi_private = dip->private; | |
5698 | ||
5699 | kfree(dip->csums); | |
5700 | kfree(dip); | |
c0da7aa1 JB |
5701 | |
5702 | /* If we had an error make sure to clear the uptodate flag */ | |
5703 | if (err) | |
5704 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5705 | dio_end_io(bio, err); |
5706 | } | |
5707 | ||
eaf25d93 CM |
5708 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5709 | struct bio *bio, int mirror_num, | |
5710 | unsigned long bio_flags, u64 offset) | |
5711 | { | |
5712 | int ret; | |
5713 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5714 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5715 | BUG_ON(ret); | |
5716 | return 0; | |
5717 | } | |
5718 | ||
e65e1535 MX |
5719 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5720 | { | |
5721 | struct btrfs_dio_private *dip = bio->bi_private; | |
5722 | ||
5723 | if (err) { | |
33345d01 | 5724 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 5725 | "sector %#Lx len %u err no %d\n", |
33345d01 | 5726 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 5727 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
5728 | dip->errors = 1; |
5729 | ||
5730 | /* | |
5731 | * before atomic variable goto zero, we must make sure | |
5732 | * dip->errors is perceived to be set. | |
5733 | */ | |
5734 | smp_mb__before_atomic_dec(); | |
5735 | } | |
5736 | ||
5737 | /* if there are more bios still pending for this dio, just exit */ | |
5738 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5739 | goto out; | |
5740 | ||
5741 | if (dip->errors) | |
5742 | bio_io_error(dip->orig_bio); | |
5743 | else { | |
5744 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5745 | bio_endio(dip->orig_bio, 0); | |
5746 | } | |
5747 | out: | |
5748 | bio_put(bio); | |
5749 | } | |
5750 | ||
5751 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5752 | u64 first_sector, gfp_t gfp_flags) | |
5753 | { | |
5754 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5755 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5756 | } | |
5757 | ||
5758 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5759 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 5760 | u32 *csums, int async_submit) |
e65e1535 MX |
5761 | { |
5762 | int write = rw & REQ_WRITE; | |
5763 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5764 | int ret; | |
5765 | ||
5766 | bio_get(bio); | |
5767 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5768 | if (ret) | |
5769 | goto err; | |
5770 | ||
1ae39938 JB |
5771 | if (skip_sum) |
5772 | goto map; | |
5773 | ||
5774 | if (write && async_submit) { | |
e65e1535 MX |
5775 | ret = btrfs_wq_submit_bio(root->fs_info, |
5776 | inode, rw, bio, 0, 0, | |
5777 | file_offset, | |
5778 | __btrfs_submit_bio_start_direct_io, | |
5779 | __btrfs_submit_bio_done); | |
5780 | goto err; | |
1ae39938 JB |
5781 | } else if (write) { |
5782 | /* | |
5783 | * If we aren't doing async submit, calculate the csum of the | |
5784 | * bio now. | |
5785 | */ | |
5786 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
5787 | if (ret) | |
5788 | goto err; | |
c2db1073 TI |
5789 | } else if (!skip_sum) { |
5790 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 5791 | file_offset, csums); |
c2db1073 TI |
5792 | if (ret) |
5793 | goto err; | |
5794 | } | |
e65e1535 | 5795 | |
1ae39938 JB |
5796 | map: |
5797 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
5798 | err: |
5799 | bio_put(bio); | |
5800 | return ret; | |
5801 | } | |
5802 | ||
5803 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
5804 | int skip_sum) | |
5805 | { | |
5806 | struct inode *inode = dip->inode; | |
5807 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5808 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5809 | struct bio *bio; | |
5810 | struct bio *orig_bio = dip->orig_bio; | |
5811 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
5812 | u64 start_sector = orig_bio->bi_sector; | |
5813 | u64 file_offset = dip->logical_offset; | |
5814 | u64 submit_len = 0; | |
5815 | u64 map_length; | |
5816 | int nr_pages = 0; | |
5817 | u32 *csums = dip->csums; | |
5818 | int ret = 0; | |
1ae39938 | 5819 | int async_submit = 0; |
98bc3149 | 5820 | int write = rw & REQ_WRITE; |
e65e1535 | 5821 | |
e65e1535 MX |
5822 | map_length = orig_bio->bi_size; |
5823 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
5824 | &map_length, NULL, 0); | |
5825 | if (ret) { | |
64728bbb | 5826 | bio_put(orig_bio); |
e65e1535 MX |
5827 | return -EIO; |
5828 | } | |
5829 | ||
02f57c7a JB |
5830 | if (map_length >= orig_bio->bi_size) { |
5831 | bio = orig_bio; | |
5832 | goto submit; | |
5833 | } | |
5834 | ||
1ae39938 | 5835 | async_submit = 1; |
02f57c7a JB |
5836 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
5837 | if (!bio) | |
5838 | return -ENOMEM; | |
5839 | bio->bi_private = dip; | |
5840 | bio->bi_end_io = btrfs_end_dio_bio; | |
5841 | atomic_inc(&dip->pending_bios); | |
5842 | ||
e65e1535 MX |
5843 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5844 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
5845 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5846 | bvec->bv_offset) < bvec->bv_len)) { | |
5847 | /* | |
5848 | * inc the count before we submit the bio so | |
5849 | * we know the end IO handler won't happen before | |
5850 | * we inc the count. Otherwise, the dip might get freed | |
5851 | * before we're done setting it up | |
5852 | */ | |
5853 | atomic_inc(&dip->pending_bios); | |
5854 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
5855 | file_offset, skip_sum, | |
1ae39938 | 5856 | csums, async_submit); |
e65e1535 MX |
5857 | if (ret) { |
5858 | bio_put(bio); | |
5859 | atomic_dec(&dip->pending_bios); | |
5860 | goto out_err; | |
5861 | } | |
5862 | ||
98bc3149 JB |
5863 | /* Write's use the ordered csums */ |
5864 | if (!write && !skip_sum) | |
e65e1535 MX |
5865 | csums = csums + nr_pages; |
5866 | start_sector += submit_len >> 9; | |
5867 | file_offset += submit_len; | |
5868 | ||
5869 | submit_len = 0; | |
5870 | nr_pages = 0; | |
5871 | ||
5872 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
5873 | start_sector, GFP_NOFS); | |
5874 | if (!bio) | |
5875 | goto out_err; | |
5876 | bio->bi_private = dip; | |
5877 | bio->bi_end_io = btrfs_end_dio_bio; | |
5878 | ||
5879 | map_length = orig_bio->bi_size; | |
5880 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
5881 | &map_length, NULL, 0); | |
5882 | if (ret) { | |
5883 | bio_put(bio); | |
5884 | goto out_err; | |
5885 | } | |
5886 | } else { | |
5887 | submit_len += bvec->bv_len; | |
5888 | nr_pages ++; | |
5889 | bvec++; | |
5890 | } | |
5891 | } | |
5892 | ||
02f57c7a | 5893 | submit: |
e65e1535 | 5894 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 5895 | csums, async_submit); |
e65e1535 MX |
5896 | if (!ret) |
5897 | return 0; | |
5898 | ||
5899 | bio_put(bio); | |
5900 | out_err: | |
5901 | dip->errors = 1; | |
5902 | /* | |
5903 | * before atomic variable goto zero, we must | |
5904 | * make sure dip->errors is perceived to be set. | |
5905 | */ | |
5906 | smp_mb__before_atomic_dec(); | |
5907 | if (atomic_dec_and_test(&dip->pending_bios)) | |
5908 | bio_io_error(dip->orig_bio); | |
5909 | ||
5910 | /* bio_end_io() will handle error, so we needn't return it */ | |
5911 | return 0; | |
5912 | } | |
5913 | ||
4b46fce2 JB |
5914 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
5915 | loff_t file_offset) | |
5916 | { | |
5917 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5918 | struct btrfs_dio_private *dip; | |
5919 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 5920 | int skip_sum; |
7b6d91da | 5921 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
5922 | int ret = 0; |
5923 | ||
5924 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
5925 | ||
5926 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
5927 | if (!dip) { | |
5928 | ret = -ENOMEM; | |
5929 | goto free_ordered; | |
5930 | } | |
5931 | dip->csums = NULL; | |
5932 | ||
98bc3149 JB |
5933 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
5934 | if (!write && !skip_sum) { | |
4b46fce2 JB |
5935 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
5936 | if (!dip->csums) { | |
b4966b77 | 5937 | kfree(dip); |
4b46fce2 JB |
5938 | ret = -ENOMEM; |
5939 | goto free_ordered; | |
5940 | } | |
5941 | } | |
5942 | ||
5943 | dip->private = bio->bi_private; | |
5944 | dip->inode = inode; | |
5945 | dip->logical_offset = file_offset; | |
5946 | ||
4b46fce2 JB |
5947 | dip->bytes = 0; |
5948 | do { | |
5949 | dip->bytes += bvec->bv_len; | |
5950 | bvec++; | |
5951 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
5952 | ||
46bfbb5c | 5953 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 5954 | bio->bi_private = dip; |
e65e1535 MX |
5955 | dip->errors = 0; |
5956 | dip->orig_bio = bio; | |
5957 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
5958 | |
5959 | if (write) | |
5960 | bio->bi_end_io = btrfs_endio_direct_write; | |
5961 | else | |
5962 | bio->bi_end_io = btrfs_endio_direct_read; | |
5963 | ||
e65e1535 MX |
5964 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
5965 | if (!ret) | |
eaf25d93 | 5966 | return; |
4b46fce2 JB |
5967 | free_ordered: |
5968 | /* | |
5969 | * If this is a write, we need to clean up the reserved space and kill | |
5970 | * the ordered extent. | |
5971 | */ | |
5972 | if (write) { | |
5973 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 5974 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
5975 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
5976 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
5977 | btrfs_free_reserved_extent(root, ordered->start, | |
5978 | ordered->disk_len); | |
5979 | btrfs_put_ordered_extent(ordered); | |
5980 | btrfs_put_ordered_extent(ordered); | |
5981 | } | |
5982 | bio_endio(bio, ret); | |
5983 | } | |
5984 | ||
5a5f79b5 CM |
5985 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
5986 | const struct iovec *iov, loff_t offset, | |
5987 | unsigned long nr_segs) | |
5988 | { | |
5989 | int seg; | |
a1b75f7d | 5990 | int i; |
5a5f79b5 CM |
5991 | size_t size; |
5992 | unsigned long addr; | |
5993 | unsigned blocksize_mask = root->sectorsize - 1; | |
5994 | ssize_t retval = -EINVAL; | |
5995 | loff_t end = offset; | |
5996 | ||
5997 | if (offset & blocksize_mask) | |
5998 | goto out; | |
5999 | ||
6000 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6001 | for (seg = 0; seg < nr_segs; seg++) { | |
6002 | addr = (unsigned long)iov[seg].iov_base; | |
6003 | size = iov[seg].iov_len; | |
6004 | end += size; | |
a1b75f7d | 6005 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6006 | goto out; |
a1b75f7d JB |
6007 | |
6008 | /* If this is a write we don't need to check anymore */ | |
6009 | if (rw & WRITE) | |
6010 | continue; | |
6011 | ||
6012 | /* | |
6013 | * Check to make sure we don't have duplicate iov_base's in this | |
6014 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6015 | * when reading back. | |
6016 | */ | |
6017 | for (i = seg + 1; i < nr_segs; i++) { | |
6018 | if (iov[seg].iov_base == iov[i].iov_base) | |
6019 | goto out; | |
6020 | } | |
5a5f79b5 CM |
6021 | } |
6022 | retval = 0; | |
6023 | out: | |
6024 | return retval; | |
6025 | } | |
16432985 CM |
6026 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6027 | const struct iovec *iov, loff_t offset, | |
6028 | unsigned long nr_segs) | |
6029 | { | |
4b46fce2 JB |
6030 | struct file *file = iocb->ki_filp; |
6031 | struct inode *inode = file->f_mapping->host; | |
6032 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6033 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6034 | u64 lockstart, lockend; |
6035 | ssize_t ret; | |
4845e44f CM |
6036 | int writing = rw & WRITE; |
6037 | int write_bits = 0; | |
3f7c579c | 6038 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6039 | |
5a5f79b5 CM |
6040 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6041 | offset, nr_segs)) { | |
6042 | return 0; | |
6043 | } | |
6044 | ||
4b46fce2 | 6045 | lockstart = offset; |
3f7c579c CM |
6046 | lockend = offset + count - 1; |
6047 | ||
6048 | if (writing) { | |
6049 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6050 | if (ret) | |
6051 | goto out; | |
6052 | } | |
4845e44f | 6053 | |
4b46fce2 | 6054 | while (1) { |
4845e44f CM |
6055 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6056 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6057 | /* |
6058 | * We're concerned with the entire range that we're going to be | |
6059 | * doing DIO to, so we need to make sure theres no ordered | |
6060 | * extents in this range. | |
6061 | */ | |
6062 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6063 | lockend - lockstart + 1); | |
6064 | if (!ordered) | |
6065 | break; | |
4845e44f CM |
6066 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6067 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6068 | btrfs_start_ordered_extent(inode, ordered, 1); |
6069 | btrfs_put_ordered_extent(ordered); | |
6070 | cond_resched(); | |
6071 | } | |
6072 | ||
4845e44f CM |
6073 | /* |
6074 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6075 | * the dirty or uptodate bits | |
6076 | */ | |
6077 | if (writing) { | |
6078 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6079 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6080 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6081 | GFP_NOFS); | |
6082 | if (ret) { | |
6083 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6084 | lockend, EXTENT_LOCKED | write_bits, | |
6085 | 1, 0, &cached_state, GFP_NOFS); | |
6086 | goto out; | |
6087 | } | |
6088 | } | |
6089 | ||
6090 | free_extent_state(cached_state); | |
6091 | cached_state = NULL; | |
6092 | ||
5a5f79b5 CM |
6093 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6094 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6095 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6096 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6097 | |
6098 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6099 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6100 | offset + iov_length(iov, nr_segs) - 1, | |
6101 | EXTENT_LOCKED | write_bits, 1, 0, | |
6102 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6103 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6104 | /* | |
6105 | * We're falling back to buffered, unlock the section we didn't | |
6106 | * do IO on. | |
6107 | */ | |
4845e44f CM |
6108 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6109 | offset + iov_length(iov, nr_segs) - 1, | |
6110 | EXTENT_LOCKED | write_bits, 1, 0, | |
6111 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6112 | } |
4845e44f CM |
6113 | out: |
6114 | free_extent_state(cached_state); | |
4b46fce2 | 6115 | return ret; |
16432985 CM |
6116 | } |
6117 | ||
1506fcc8 YS |
6118 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6119 | __u64 start, __u64 len) | |
6120 | { | |
ec29ed5b | 6121 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6122 | } |
6123 | ||
a52d9a80 | 6124 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6125 | { |
d1310b2e CM |
6126 | struct extent_io_tree *tree; |
6127 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6128 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6129 | } |
1832a6d5 | 6130 | |
a52d9a80 | 6131 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6132 | { |
d1310b2e | 6133 | struct extent_io_tree *tree; |
b888db2b CM |
6134 | |
6135 | ||
6136 | if (current->flags & PF_MEMALLOC) { | |
6137 | redirty_page_for_writepage(wbc, page); | |
6138 | unlock_page(page); | |
6139 | return 0; | |
6140 | } | |
d1310b2e | 6141 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6142 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6143 | } |
6144 | ||
f421950f CM |
6145 | int btrfs_writepages(struct address_space *mapping, |
6146 | struct writeback_control *wbc) | |
b293f02e | 6147 | { |
d1310b2e | 6148 | struct extent_io_tree *tree; |
771ed689 | 6149 | |
d1310b2e | 6150 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6151 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6152 | } | |
6153 | ||
3ab2fb5a CM |
6154 | static int |
6155 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6156 | struct list_head *pages, unsigned nr_pages) | |
6157 | { | |
d1310b2e CM |
6158 | struct extent_io_tree *tree; |
6159 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6160 | return extent_readpages(tree, mapping, pages, nr_pages, |
6161 | btrfs_get_extent); | |
6162 | } | |
e6dcd2dc | 6163 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6164 | { |
d1310b2e CM |
6165 | struct extent_io_tree *tree; |
6166 | struct extent_map_tree *map; | |
a52d9a80 | 6167 | int ret; |
8c2383c3 | 6168 | |
d1310b2e CM |
6169 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6170 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6171 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6172 | if (ret == 1) { |
6173 | ClearPagePrivate(page); | |
6174 | set_page_private(page, 0); | |
6175 | page_cache_release(page); | |
39279cc3 | 6176 | } |
a52d9a80 | 6177 | return ret; |
39279cc3 CM |
6178 | } |
6179 | ||
e6dcd2dc CM |
6180 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6181 | { | |
98509cfc CM |
6182 | if (PageWriteback(page) || PageDirty(page)) |
6183 | return 0; | |
b335b003 | 6184 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6185 | } |
6186 | ||
a52d9a80 | 6187 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6188 | { |
d1310b2e | 6189 | struct extent_io_tree *tree; |
e6dcd2dc | 6190 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6191 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6192 | u64 page_start = page_offset(page); |
6193 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6194 | |
8b62b72b CM |
6195 | |
6196 | /* | |
6197 | * we have the page locked, so new writeback can't start, | |
6198 | * and the dirty bit won't be cleared while we are here. | |
6199 | * | |
6200 | * Wait for IO on this page so that we can safely clear | |
6201 | * the PagePrivate2 bit and do ordered accounting | |
6202 | */ | |
e6dcd2dc | 6203 | wait_on_page_writeback(page); |
8b62b72b | 6204 | |
d1310b2e | 6205 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6206 | if (offset) { |
6207 | btrfs_releasepage(page, GFP_NOFS); | |
6208 | return; | |
6209 | } | |
2ac55d41 JB |
6210 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6211 | GFP_NOFS); | |
e6dcd2dc CM |
6212 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6213 | page_offset(page)); | |
6214 | if (ordered) { | |
eb84ae03 CM |
6215 | /* |
6216 | * IO on this page will never be started, so we need | |
6217 | * to account for any ordered extents now | |
6218 | */ | |
e6dcd2dc CM |
6219 | clear_extent_bit(tree, page_start, page_end, |
6220 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6221 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6222 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6223 | /* |
6224 | * whoever cleared the private bit is responsible | |
6225 | * for the finish_ordered_io | |
6226 | */ | |
6227 | if (TestClearPagePrivate2(page)) { | |
6228 | btrfs_finish_ordered_io(page->mapping->host, | |
6229 | page_start, page_end); | |
6230 | } | |
e6dcd2dc | 6231 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6232 | cached_state = NULL; |
6233 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6234 | GFP_NOFS); | |
e6dcd2dc CM |
6235 | } |
6236 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6237 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6238 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6239 | __btrfs_releasepage(page, GFP_NOFS); |
6240 | ||
4a096752 | 6241 | ClearPageChecked(page); |
9ad6b7bc | 6242 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6243 | ClearPagePrivate(page); |
6244 | set_page_private(page, 0); | |
6245 | page_cache_release(page); | |
6246 | } | |
39279cc3 CM |
6247 | } |
6248 | ||
9ebefb18 CM |
6249 | /* |
6250 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6251 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6252 | * be careful to check for EOF conditions here. We set the page up correctly | |
6253 | * for a written page which means we get ENOSPC checking when writing into | |
6254 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6255 | * support these features. | |
6256 | * | |
6257 | * We are not allowed to take the i_mutex here so we have to play games to | |
6258 | * protect against truncate races as the page could now be beyond EOF. Because | |
6259 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6260 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6261 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6262 | * unlock the page. | |
6263 | */ | |
c2ec175c | 6264 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6265 | { |
c2ec175c | 6266 | struct page *page = vmf->page; |
6da6abae | 6267 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6268 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6269 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6270 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6271 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6272 | char *kaddr; |
6273 | unsigned long zero_start; | |
9ebefb18 | 6274 | loff_t size; |
1832a6d5 | 6275 | int ret; |
a52d9a80 | 6276 | u64 page_start; |
e6dcd2dc | 6277 | u64 page_end; |
9ebefb18 | 6278 | |
0ca1f7ce | 6279 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
6280 | if (ret) { |
6281 | if (ret == -ENOMEM) | |
6282 | ret = VM_FAULT_OOM; | |
6283 | else /* -ENOSPC, -EIO, etc */ | |
6284 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 6285 | goto out; |
56a76f82 | 6286 | } |
1832a6d5 | 6287 | |
56a76f82 | 6288 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6289 | again: |
9ebefb18 | 6290 | lock_page(page); |
9ebefb18 | 6291 | size = i_size_read(inode); |
e6dcd2dc CM |
6292 | page_start = page_offset(page); |
6293 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6294 | |
9ebefb18 | 6295 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6296 | (page_start >= size)) { |
9ebefb18 CM |
6297 | /* page got truncated out from underneath us */ |
6298 | goto out_unlock; | |
6299 | } | |
e6dcd2dc CM |
6300 | wait_on_page_writeback(page); |
6301 | ||
2ac55d41 JB |
6302 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6303 | GFP_NOFS); | |
e6dcd2dc CM |
6304 | set_page_extent_mapped(page); |
6305 | ||
eb84ae03 CM |
6306 | /* |
6307 | * we can't set the delalloc bits if there are pending ordered | |
6308 | * extents. Drop our locks and wait for them to finish | |
6309 | */ | |
e6dcd2dc CM |
6310 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6311 | if (ordered) { | |
2ac55d41 JB |
6312 | unlock_extent_cached(io_tree, page_start, page_end, |
6313 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6314 | unlock_page(page); |
eb84ae03 | 6315 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6316 | btrfs_put_ordered_extent(ordered); |
6317 | goto again; | |
6318 | } | |
6319 | ||
fbf19087 JB |
6320 | /* |
6321 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6322 | * if it was already dirty, so for space accounting reasons we need to | |
6323 | * clear any delalloc bits for the range we are fixing to save. There | |
6324 | * is probably a better way to do this, but for now keep consistent with | |
6325 | * prepare_pages in the normal write path. | |
6326 | */ | |
2ac55d41 | 6327 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6328 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6329 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6330 | |
2ac55d41 JB |
6331 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6332 | &cached_state); | |
9ed74f2d | 6333 | if (ret) { |
2ac55d41 JB |
6334 | unlock_extent_cached(io_tree, page_start, page_end, |
6335 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6336 | ret = VM_FAULT_SIGBUS; |
6337 | goto out_unlock; | |
6338 | } | |
e6dcd2dc | 6339 | ret = 0; |
9ebefb18 CM |
6340 | |
6341 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6342 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6343 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6344 | else |
e6dcd2dc | 6345 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6346 | |
e6dcd2dc CM |
6347 | if (zero_start != PAGE_CACHE_SIZE) { |
6348 | kaddr = kmap(page); | |
6349 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6350 | flush_dcache_page(page); | |
6351 | kunmap(page); | |
6352 | } | |
247e743c | 6353 | ClearPageChecked(page); |
e6dcd2dc | 6354 | set_page_dirty(page); |
50a9b214 | 6355 | SetPageUptodate(page); |
5a3f23d5 | 6356 | |
257c62e1 CM |
6357 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6358 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6359 | ||
2ac55d41 | 6360 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6361 | |
6362 | out_unlock: | |
50a9b214 CM |
6363 | if (!ret) |
6364 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6365 | unlock_page(page); |
0ca1f7ce | 6366 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 6367 | out: |
9ebefb18 CM |
6368 | return ret; |
6369 | } | |
6370 | ||
a41ad394 | 6371 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6372 | { |
6373 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6374 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6375 | int ret; |
3893e33b | 6376 | int err = 0; |
39279cc3 | 6377 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6378 | unsigned long nr; |
dbe674a9 | 6379 | u64 mask = root->sectorsize - 1; |
07127184 | 6380 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6381 | |
5d5e103a JB |
6382 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6383 | if (ret) | |
a41ad394 | 6384 | return ret; |
8082510e | 6385 | |
4a096752 | 6386 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6387 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6388 | |
fcb80c2a JB |
6389 | /* |
6390 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6391 | * 3 things going on here | |
6392 | * | |
6393 | * 1) We need to reserve space for our orphan item and the space to | |
6394 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6395 | * orphan item because we didn't reserve space to remove it. | |
6396 | * | |
6397 | * 2) We need to reserve space to update our inode. | |
6398 | * | |
6399 | * 3) We need to have something to cache all the space that is going to | |
6400 | * be free'd up by the truncate operation, but also have some slack | |
6401 | * space reserved in case it uses space during the truncate (thank you | |
6402 | * very much snapshotting). | |
6403 | * | |
6404 | * And we need these to all be seperate. The fact is we can use alot of | |
6405 | * space doing the truncate, and we have no earthly idea how much space | |
6406 | * we will use, so we need the truncate reservation to be seperate so it | |
6407 | * doesn't end up using space reserved for updating the inode or | |
6408 | * removing the orphan item. We also need to be able to stop the | |
6409 | * transaction and start a new one, which means we need to be able to | |
6410 | * update the inode several times, and we have no idea of knowing how | |
6411 | * many times that will be, so we can't just reserve 1 item for the | |
6412 | * entirety of the opration, so that has to be done seperately as well. | |
6413 | * Then there is the orphan item, which does indeed need to be held on | |
6414 | * to for the whole operation, and we need nobody to touch this reserved | |
6415 | * space except the orphan code. | |
6416 | * | |
6417 | * So that leaves us with | |
6418 | * | |
6419 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6420 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6421 | * transaction reservation. | |
6422 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6423 | * updating the inode. | |
6424 | */ | |
6425 | rsv = btrfs_alloc_block_rsv(root); | |
6426 | if (!rsv) | |
6427 | return -ENOMEM; | |
4a338542 | 6428 | rsv->size = min_size; |
f0cd846e | 6429 | |
907cbceb | 6430 | /* |
07127184 | 6431 | * 1 for the truncate slack space |
907cbceb JB |
6432 | * 1 for the orphan item we're going to add |
6433 | * 1 for the orphan item deletion | |
6434 | * 1 for updating the inode. | |
6435 | */ | |
07127184 | 6436 | trans = btrfs_start_transaction(root, 4); |
fcb80c2a JB |
6437 | if (IS_ERR(trans)) { |
6438 | err = PTR_ERR(trans); | |
6439 | goto out; | |
6440 | } | |
f0cd846e | 6441 | |
907cbceb JB |
6442 | /* Migrate the slack space for the truncate to our reserve */ |
6443 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6444 | min_size); | |
fcb80c2a | 6445 | BUG_ON(ret); |
f0cd846e JB |
6446 | |
6447 | ret = btrfs_orphan_add(trans, inode); | |
6448 | if (ret) { | |
6449 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6450 | goto out; |
f0cd846e JB |
6451 | } |
6452 | ||
5a3f23d5 CM |
6453 | /* |
6454 | * setattr is responsible for setting the ordered_data_close flag, | |
6455 | * but that is only tested during the last file release. That | |
6456 | * could happen well after the next commit, leaving a great big | |
6457 | * window where new writes may get lost if someone chooses to write | |
6458 | * to this file after truncating to zero | |
6459 | * | |
6460 | * The inode doesn't have any dirty data here, and so if we commit | |
6461 | * this is a noop. If someone immediately starts writing to the inode | |
6462 | * it is very likely we'll catch some of their writes in this | |
6463 | * transaction, and the commit will find this file on the ordered | |
6464 | * data list with good things to send down. | |
6465 | * | |
6466 | * This is a best effort solution, there is still a window where | |
6467 | * using truncate to replace the contents of the file will | |
6468 | * end up with a zero length file after a crash. | |
6469 | */ | |
6470 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6471 | btrfs_add_ordered_operation(trans, root, inode); | |
6472 | ||
8082510e | 6473 | while (1) { |
36ba022a | 6474 | ret = btrfs_block_rsv_refill(root, rsv, min_size); |
907cbceb JB |
6475 | if (ret) { |
6476 | /* | |
6477 | * This can only happen with the original transaction we | |
6478 | * started above, every other time we shouldn't have a | |
6479 | * transaction started yet. | |
6480 | */ | |
6481 | if (ret == -EAGAIN) | |
6482 | goto end_trans; | |
6483 | err = ret; | |
6484 | break; | |
6485 | } | |
6486 | ||
d68fc57b | 6487 | if (!trans) { |
907cbceb JB |
6488 | /* Just need the 1 for updating the inode */ |
6489 | trans = btrfs_start_transaction(root, 1); | |
fcb80c2a JB |
6490 | if (IS_ERR(trans)) { |
6491 | err = PTR_ERR(trans); | |
6492 | goto out; | |
6493 | } | |
d68fc57b YZ |
6494 | } |
6495 | ||
907cbceb JB |
6496 | trans->block_rsv = rsv; |
6497 | ||
8082510e YZ |
6498 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6499 | inode->i_size, | |
6500 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6501 | if (ret != -EAGAIN) { |
6502 | err = ret; | |
8082510e | 6503 | break; |
3893e33b | 6504 | } |
39279cc3 | 6505 | |
fcb80c2a | 6506 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6507 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6508 | if (ret) { |
6509 | err = ret; | |
6510 | break; | |
6511 | } | |
907cbceb | 6512 | end_trans: |
8082510e YZ |
6513 | nr = trans->blocks_used; |
6514 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6515 | trans = NULL; |
8082510e | 6516 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6517 | } |
6518 | ||
6519 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6520 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6521 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6522 | if (ret) |
6523 | err = ret; | |
ded5db9d JB |
6524 | } else if (ret && inode->i_nlink > 0) { |
6525 | /* | |
6526 | * Failed to do the truncate, remove us from the in memory | |
6527 | * orphan list. | |
6528 | */ | |
6529 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6530 | } |
6531 | ||
fcb80c2a | 6532 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6533 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6534 | if (ret && !err) |
6535 | err = ret; | |
7b128766 | 6536 | |
7b128766 | 6537 | nr = trans->blocks_used; |
89ce8a63 | 6538 | ret = btrfs_end_transaction_throttle(trans, root); |
fcb80c2a JB |
6539 | btrfs_btree_balance_dirty(root, nr); |
6540 | ||
6541 | out: | |
6542 | btrfs_free_block_rsv(root, rsv); | |
6543 | ||
3893e33b JB |
6544 | if (ret && !err) |
6545 | err = ret; | |
a41ad394 | 6546 | |
3893e33b | 6547 | return err; |
39279cc3 CM |
6548 | } |
6549 | ||
d352ac68 CM |
6550 | /* |
6551 | * create a new subvolume directory/inode (helper for the ioctl). | |
6552 | */ | |
d2fb3437 | 6553 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6554 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6555 | { |
39279cc3 | 6556 | struct inode *inode; |
76dda93c | 6557 | int err; |
00e4e6b3 | 6558 | u64 index = 0; |
39279cc3 | 6559 | |
aec7477b | 6560 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d82a6f1d | 6561 | new_dirid, S_IFDIR | 0700, &index); |
54aa1f4d | 6562 | if (IS_ERR(inode)) |
f46b5a66 | 6563 | return PTR_ERR(inode); |
39279cc3 CM |
6564 | inode->i_op = &btrfs_dir_inode_operations; |
6565 | inode->i_fop = &btrfs_dir_file_operations; | |
6566 | ||
39279cc3 | 6567 | inode->i_nlink = 1; |
dbe674a9 | 6568 | btrfs_i_size_write(inode, 0); |
3b96362c | 6569 | |
76dda93c YZ |
6570 | err = btrfs_update_inode(trans, new_root, inode); |
6571 | BUG_ON(err); | |
cb8e7090 | 6572 | |
76dda93c | 6573 | iput(inode); |
cb8e7090 | 6574 | return 0; |
39279cc3 CM |
6575 | } |
6576 | ||
39279cc3 CM |
6577 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6578 | { | |
6579 | struct btrfs_inode *ei; | |
2ead6ae7 | 6580 | struct inode *inode; |
39279cc3 CM |
6581 | |
6582 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6583 | if (!ei) | |
6584 | return NULL; | |
2ead6ae7 YZ |
6585 | |
6586 | ei->root = NULL; | |
6587 | ei->space_info = NULL; | |
6588 | ei->generation = 0; | |
6589 | ei->sequence = 0; | |
15ee9bc7 | 6590 | ei->last_trans = 0; |
257c62e1 | 6591 | ei->last_sub_trans = 0; |
e02119d5 | 6592 | ei->logged_trans = 0; |
2ead6ae7 | 6593 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
6594 | ei->disk_i_size = 0; |
6595 | ei->flags = 0; | |
7709cde3 | 6596 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
6597 | ei->index_cnt = (u64)-1; |
6598 | ei->last_unlink_trans = 0; | |
6599 | ||
9e0baf60 JB |
6600 | spin_lock_init(&ei->lock); |
6601 | ei->outstanding_extents = 0; | |
6602 | ei->reserved_extents = 0; | |
2ead6ae7 YZ |
6603 | |
6604 | ei->ordered_data_close = 0; | |
d68fc57b | 6605 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6606 | ei->dummy_inode = 0; |
4cb5300b | 6607 | ei->in_defrag = 0; |
261507a0 | 6608 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 6609 | |
16cdcec7 MX |
6610 | ei->delayed_node = NULL; |
6611 | ||
2ead6ae7 | 6612 | inode = &ei->vfs_inode; |
a8067e02 | 6613 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
6614 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
6615 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
2ead6ae7 | 6616 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 6617 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6618 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6619 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6620 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6621 | RB_CLEAR_NODE(&ei->rb_node); |
6622 | ||
6623 | return inode; | |
39279cc3 CM |
6624 | } |
6625 | ||
fa0d7e3d NP |
6626 | static void btrfs_i_callback(struct rcu_head *head) |
6627 | { | |
6628 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6629 | INIT_LIST_HEAD(&inode->i_dentry); | |
6630 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6631 | } | |
6632 | ||
39279cc3 CM |
6633 | void btrfs_destroy_inode(struct inode *inode) |
6634 | { | |
e6dcd2dc | 6635 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6636 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6637 | ||
39279cc3 CM |
6638 | WARN_ON(!list_empty(&inode->i_dentry)); |
6639 | WARN_ON(inode->i_data.nrpages); | |
9e0baf60 JB |
6640 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
6641 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
6642 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
6643 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 6644 | |
a6dbd429 JB |
6645 | /* |
6646 | * This can happen where we create an inode, but somebody else also | |
6647 | * created the same inode and we need to destroy the one we already | |
6648 | * created. | |
6649 | */ | |
6650 | if (!root) | |
6651 | goto free; | |
6652 | ||
5a3f23d5 CM |
6653 | /* |
6654 | * Make sure we're properly removed from the ordered operation | |
6655 | * lists. | |
6656 | */ | |
6657 | smp_mb(); | |
6658 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6659 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6660 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6661 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6662 | } | |
6663 | ||
d68fc57b | 6664 | spin_lock(&root->orphan_lock); |
7b128766 | 6665 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
33345d01 LZ |
6666 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
6667 | (unsigned long long)btrfs_ino(inode)); | |
8082510e | 6668 | list_del_init(&BTRFS_I(inode)->i_orphan); |
7b128766 | 6669 | } |
d68fc57b | 6670 | spin_unlock(&root->orphan_lock); |
7b128766 | 6671 | |
d397712b | 6672 | while (1) { |
e6dcd2dc CM |
6673 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6674 | if (!ordered) | |
6675 | break; | |
6676 | else { | |
d397712b CM |
6677 | printk(KERN_ERR "btrfs found ordered " |
6678 | "extent %llu %llu on inode cleanup\n", | |
6679 | (unsigned long long)ordered->file_offset, | |
6680 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6681 | btrfs_remove_ordered_extent(inode, ordered); |
6682 | btrfs_put_ordered_extent(ordered); | |
6683 | btrfs_put_ordered_extent(ordered); | |
6684 | } | |
6685 | } | |
5d4f98a2 | 6686 | inode_tree_del(inode); |
5b21f2ed | 6687 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6688 | free: |
16cdcec7 | 6689 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 6690 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6691 | } |
6692 | ||
45321ac5 | 6693 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6694 | { |
6695 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6696 | |
0af3d00b | 6697 | if (btrfs_root_refs(&root->root_item) == 0 && |
2cf8572d | 6698 | !btrfs_is_free_space_inode(root, inode)) |
45321ac5 | 6699 | return 1; |
76dda93c | 6700 | else |
45321ac5 | 6701 | return generic_drop_inode(inode); |
76dda93c YZ |
6702 | } |
6703 | ||
0ee0fda0 | 6704 | static void init_once(void *foo) |
39279cc3 CM |
6705 | { |
6706 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6707 | ||
6708 | inode_init_once(&ei->vfs_inode); | |
6709 | } | |
6710 | ||
6711 | void btrfs_destroy_cachep(void) | |
6712 | { | |
6713 | if (btrfs_inode_cachep) | |
6714 | kmem_cache_destroy(btrfs_inode_cachep); | |
6715 | if (btrfs_trans_handle_cachep) | |
6716 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6717 | if (btrfs_transaction_cachep) | |
6718 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6719 | if (btrfs_path_cachep) |
6720 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6721 | if (btrfs_free_space_cachep) |
6722 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6723 | } |
6724 | ||
6725 | int btrfs_init_cachep(void) | |
6726 | { | |
9601e3f6 CH |
6727 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6728 | sizeof(struct btrfs_inode), 0, | |
6729 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6730 | if (!btrfs_inode_cachep) |
6731 | goto fail; | |
9601e3f6 CH |
6732 | |
6733 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6734 | sizeof(struct btrfs_trans_handle), 0, | |
6735 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6736 | if (!btrfs_trans_handle_cachep) |
6737 | goto fail; | |
9601e3f6 CH |
6738 | |
6739 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6740 | sizeof(struct btrfs_transaction), 0, | |
6741 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6742 | if (!btrfs_transaction_cachep) |
6743 | goto fail; | |
9601e3f6 CH |
6744 | |
6745 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6746 | sizeof(struct btrfs_path), 0, | |
6747 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6748 | if (!btrfs_path_cachep) |
6749 | goto fail; | |
9601e3f6 | 6750 | |
dc89e982 JB |
6751 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6752 | sizeof(struct btrfs_free_space), 0, | |
6753 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6754 | if (!btrfs_free_space_cachep) | |
6755 | goto fail; | |
6756 | ||
39279cc3 CM |
6757 | return 0; |
6758 | fail: | |
6759 | btrfs_destroy_cachep(); | |
6760 | return -ENOMEM; | |
6761 | } | |
6762 | ||
6763 | static int btrfs_getattr(struct vfsmount *mnt, | |
6764 | struct dentry *dentry, struct kstat *stat) | |
6765 | { | |
6766 | struct inode *inode = dentry->d_inode; | |
6767 | generic_fillattr(inode, stat); | |
0ee5dc67 | 6768 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 6769 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
6770 | stat->blocks = (inode_get_bytes(inode) + |
6771 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
6772 | return 0; |
6773 | } | |
6774 | ||
75e7cb7f LB |
6775 | /* |
6776 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6777 | * directory. | |
6778 | */ | |
6779 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6780 | { | |
6781 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6782 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6783 | ||
6784 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6785 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6786 | else | |
6787 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6788 | ||
6789 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6790 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6791 | else | |
6792 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6793 | } | |
6794 | ||
d397712b CM |
6795 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6796 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6797 | { |
6798 | struct btrfs_trans_handle *trans; | |
6799 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6800 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6801 | struct inode *new_inode = new_dentry->d_inode; |
6802 | struct inode *old_inode = old_dentry->d_inode; | |
6803 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6804 | u64 index = 0; |
4df27c4d | 6805 | u64 root_objectid; |
39279cc3 | 6806 | int ret; |
33345d01 | 6807 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 6808 | |
33345d01 | 6809 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
6810 | return -EPERM; |
6811 | ||
4df27c4d | 6812 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 6813 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
6814 | return -EXDEV; |
6815 | ||
33345d01 LZ |
6816 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
6817 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 6818 | return -ENOTEMPTY; |
5f39d397 | 6819 | |
4df27c4d YZ |
6820 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
6821 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
6822 | return -ENOTEMPTY; | |
5a3f23d5 CM |
6823 | /* |
6824 | * we're using rename to replace one file with another. | |
6825 | * and the replacement file is large. Start IO on it now so | |
6826 | * we don't add too much work to the end of the transaction | |
6827 | */ | |
4baf8c92 | 6828 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
6829 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
6830 | filemap_flush(old_inode->i_mapping); | |
6831 | ||
76dda93c | 6832 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 6833 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 6834 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
6835 | /* |
6836 | * We want to reserve the absolute worst case amount of items. So if | |
6837 | * both inodes are subvols and we need to unlink them then that would | |
6838 | * require 4 item modifications, but if they are both normal inodes it | |
6839 | * would require 5 item modifications, so we'll assume their normal | |
6840 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
6841 | * should cover the worst case number of items we'll modify. | |
6842 | */ | |
6843 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
6844 | if (IS_ERR(trans)) { |
6845 | ret = PTR_ERR(trans); | |
6846 | goto out_notrans; | |
6847 | } | |
76dda93c | 6848 | |
4df27c4d YZ |
6849 | if (dest != root) |
6850 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 6851 | |
a5719521 YZ |
6852 | ret = btrfs_set_inode_index(new_dir, &index); |
6853 | if (ret) | |
6854 | goto out_fail; | |
5a3f23d5 | 6855 | |
33345d01 | 6856 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6857 | /* force full log commit if subvolume involved. */ |
6858 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
6859 | } else { | |
a5719521 YZ |
6860 | ret = btrfs_insert_inode_ref(trans, dest, |
6861 | new_dentry->d_name.name, | |
6862 | new_dentry->d_name.len, | |
33345d01 LZ |
6863 | old_ino, |
6864 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
6865 | if (ret) |
6866 | goto out_fail; | |
4df27c4d YZ |
6867 | /* |
6868 | * this is an ugly little race, but the rename is required | |
6869 | * to make sure that if we crash, the inode is either at the | |
6870 | * old name or the new one. pinning the log transaction lets | |
6871 | * us make sure we don't allow a log commit to come in after | |
6872 | * we unlink the name but before we add the new name back in. | |
6873 | */ | |
6874 | btrfs_pin_log_trans(root); | |
6875 | } | |
5a3f23d5 CM |
6876 | /* |
6877 | * make sure the inode gets flushed if it is replacing | |
6878 | * something. | |
6879 | */ | |
33345d01 | 6880 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 6881 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 6882 | |
39279cc3 CM |
6883 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
6884 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
6885 | old_inode->i_ctime = ctime; | |
5f39d397 | 6886 | |
12fcfd22 CM |
6887 | if (old_dentry->d_parent != new_dentry->d_parent) |
6888 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
6889 | ||
33345d01 | 6890 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6891 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
6892 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
6893 | old_dentry->d_name.name, | |
6894 | old_dentry->d_name.len); | |
6895 | } else { | |
92986796 AV |
6896 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
6897 | old_dentry->d_inode, | |
6898 | old_dentry->d_name.name, | |
6899 | old_dentry->d_name.len); | |
6900 | if (!ret) | |
6901 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
6902 | } |
6903 | BUG_ON(ret); | |
39279cc3 CM |
6904 | |
6905 | if (new_inode) { | |
6906 | new_inode->i_ctime = CURRENT_TIME; | |
33345d01 | 6907 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
6908 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
6909 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
6910 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
6911 | root_objectid, | |
6912 | new_dentry->d_name.name, | |
6913 | new_dentry->d_name.len); | |
6914 | BUG_ON(new_inode->i_nlink == 0); | |
6915 | } else { | |
6916 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
6917 | new_dentry->d_inode, | |
6918 | new_dentry->d_name.name, | |
6919 | new_dentry->d_name.len); | |
6920 | } | |
6921 | BUG_ON(ret); | |
7b128766 | 6922 | if (new_inode->i_nlink == 0) { |
e02119d5 | 6923 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 6924 | BUG_ON(ret); |
7b128766 | 6925 | } |
39279cc3 | 6926 | } |
aec7477b | 6927 | |
75e7cb7f LB |
6928 | fixup_inode_flags(new_dir, old_inode); |
6929 | ||
4df27c4d YZ |
6930 | ret = btrfs_add_link(trans, new_dir, old_inode, |
6931 | new_dentry->d_name.name, | |
a5719521 | 6932 | new_dentry->d_name.len, 0, index); |
4df27c4d | 6933 | BUG_ON(ret); |
39279cc3 | 6934 | |
33345d01 | 6935 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 6936 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 6937 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
6938 | btrfs_end_log_trans(root); |
6939 | } | |
39279cc3 | 6940 | out_fail: |
ab78c84d | 6941 | btrfs_end_transaction_throttle(trans, root); |
b44c59a8 | 6942 | out_notrans: |
33345d01 | 6943 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 6944 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 6945 | |
39279cc3 CM |
6946 | return ret; |
6947 | } | |
6948 | ||
d352ac68 CM |
6949 | /* |
6950 | * some fairly slow code that needs optimization. This walks the list | |
6951 | * of all the inodes with pending delalloc and forces them to disk. | |
6952 | */ | |
24bbcf04 | 6953 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
6954 | { |
6955 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
6956 | struct btrfs_inode *binode; | |
5b21f2ed | 6957 | struct inode *inode; |
ea8c2819 | 6958 | |
c146afad YZ |
6959 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
6960 | return -EROFS; | |
6961 | ||
75eff68e | 6962 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 6963 | while (!list_empty(head)) { |
ea8c2819 CM |
6964 | binode = list_entry(head->next, struct btrfs_inode, |
6965 | delalloc_inodes); | |
5b21f2ed ZY |
6966 | inode = igrab(&binode->vfs_inode); |
6967 | if (!inode) | |
6968 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 6969 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 6970 | if (inode) { |
8c8bee1d | 6971 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
6972 | if (delay_iput) |
6973 | btrfs_add_delayed_iput(inode); | |
6974 | else | |
6975 | iput(inode); | |
5b21f2ed ZY |
6976 | } |
6977 | cond_resched(); | |
75eff68e | 6978 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 6979 | } |
75eff68e | 6980 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
6981 | |
6982 | /* the filemap_flush will queue IO into the worker threads, but | |
6983 | * we have to make sure the IO is actually started and that | |
6984 | * ordered extents get created before we return | |
6985 | */ | |
6986 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 6987 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 6988 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 6989 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
6990 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
6991 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
6992 | } |
6993 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
6994 | return 0; |
6995 | } | |
6996 | ||
39279cc3 CM |
6997 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
6998 | const char *symname) | |
6999 | { | |
7000 | struct btrfs_trans_handle *trans; | |
7001 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7002 | struct btrfs_path *path; | |
7003 | struct btrfs_key key; | |
1832a6d5 | 7004 | struct inode *inode = NULL; |
39279cc3 CM |
7005 | int err; |
7006 | int drop_inode = 0; | |
7007 | u64 objectid; | |
00e4e6b3 | 7008 | u64 index = 0 ; |
39279cc3 CM |
7009 | int name_len; |
7010 | int datasize; | |
5f39d397 | 7011 | unsigned long ptr; |
39279cc3 | 7012 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7013 | struct extent_buffer *leaf; |
1832a6d5 | 7014 | unsigned long nr = 0; |
39279cc3 CM |
7015 | |
7016 | name_len = strlen(symname) + 1; | |
7017 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7018 | return -ENAMETOOLONG; | |
1832a6d5 | 7019 | |
9ed74f2d JB |
7020 | /* |
7021 | * 2 items for inode item and ref | |
7022 | * 2 items for dir items | |
7023 | * 1 item for xattr if selinux is on | |
7024 | */ | |
a22285a6 YZ |
7025 | trans = btrfs_start_transaction(root, 5); |
7026 | if (IS_ERR(trans)) | |
7027 | return PTR_ERR(trans); | |
1832a6d5 | 7028 | |
581bb050 LZ |
7029 | err = btrfs_find_free_ino(root, &objectid); |
7030 | if (err) | |
7031 | goto out_unlock; | |
7032 | ||
aec7477b | 7033 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7034 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7035 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7036 | if (IS_ERR(inode)) { |
7037 | err = PTR_ERR(inode); | |
39279cc3 | 7038 | goto out_unlock; |
7cf96da3 | 7039 | } |
39279cc3 | 7040 | |
2a7dba39 | 7041 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7042 | if (err) { |
7043 | drop_inode = 1; | |
7044 | goto out_unlock; | |
7045 | } | |
7046 | ||
a1b075d2 | 7047 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7048 | if (err) |
7049 | drop_inode = 1; | |
7050 | else { | |
7051 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7052 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
7053 | inode->i_fop = &btrfs_file_operations; |
7054 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 7055 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7056 | } |
39279cc3 CM |
7057 | if (drop_inode) |
7058 | goto out_unlock; | |
7059 | ||
7060 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7061 | if (!path) { |
7062 | err = -ENOMEM; | |
7063 | drop_inode = 1; | |
7064 | goto out_unlock; | |
7065 | } | |
33345d01 | 7066 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7067 | key.offset = 0; |
39279cc3 CM |
7068 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7069 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7070 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7071 | datasize); | |
54aa1f4d CM |
7072 | if (err) { |
7073 | drop_inode = 1; | |
b0839166 | 7074 | btrfs_free_path(path); |
54aa1f4d CM |
7075 | goto out_unlock; |
7076 | } | |
5f39d397 CM |
7077 | leaf = path->nodes[0]; |
7078 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7079 | struct btrfs_file_extent_item); | |
7080 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7081 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7082 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7083 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7084 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7085 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7086 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7087 | ||
39279cc3 | 7088 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7089 | write_extent_buffer(leaf, symname, ptr, name_len); |
7090 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7091 | btrfs_free_path(path); |
5f39d397 | 7092 | |
39279cc3 CM |
7093 | inode->i_op = &btrfs_symlink_inode_operations; |
7094 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7095 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7096 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7097 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7098 | err = btrfs_update_inode(trans, root, inode); |
7099 | if (err) | |
7100 | drop_inode = 1; | |
39279cc3 CM |
7101 | |
7102 | out_unlock: | |
d3c2fdcf | 7103 | nr = trans->blocks_used; |
ab78c84d | 7104 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
7105 | if (drop_inode) { |
7106 | inode_dec_link_count(inode); | |
7107 | iput(inode); | |
7108 | } | |
d3c2fdcf | 7109 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7110 | return err; |
7111 | } | |
16432985 | 7112 | |
0af3d00b JB |
7113 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7114 | u64 start, u64 num_bytes, u64 min_size, | |
7115 | loff_t actual_len, u64 *alloc_hint, | |
7116 | struct btrfs_trans_handle *trans) | |
d899e052 | 7117 | { |
d899e052 YZ |
7118 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7119 | struct btrfs_key ins; | |
d899e052 | 7120 | u64 cur_offset = start; |
55a61d1d | 7121 | u64 i_size; |
d899e052 | 7122 | int ret = 0; |
0af3d00b | 7123 | bool own_trans = true; |
d899e052 | 7124 | |
0af3d00b JB |
7125 | if (trans) |
7126 | own_trans = false; | |
d899e052 | 7127 | while (num_bytes > 0) { |
0af3d00b JB |
7128 | if (own_trans) { |
7129 | trans = btrfs_start_transaction(root, 3); | |
7130 | if (IS_ERR(trans)) { | |
7131 | ret = PTR_ERR(trans); | |
7132 | break; | |
7133 | } | |
5a303d5d YZ |
7134 | } |
7135 | ||
efa56464 YZ |
7136 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7137 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7138 | if (ret) { |
0af3d00b JB |
7139 | if (own_trans) |
7140 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7141 | break; |
d899e052 | 7142 | } |
5a303d5d | 7143 | |
d899e052 YZ |
7144 | ret = insert_reserved_file_extent(trans, inode, |
7145 | cur_offset, ins.objectid, | |
7146 | ins.offset, ins.offset, | |
920bbbfb | 7147 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7148 | BTRFS_FILE_EXTENT_PREALLOC); |
7149 | BUG_ON(ret); | |
a1ed835e CM |
7150 | btrfs_drop_extent_cache(inode, cur_offset, |
7151 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7152 | |
d899e052 YZ |
7153 | num_bytes -= ins.offset; |
7154 | cur_offset += ins.offset; | |
efa56464 | 7155 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7156 | |
d899e052 | 7157 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7158 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7159 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7160 | (actual_len > inode->i_size) && |
7161 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7162 | if (cur_offset > actual_len) |
55a61d1d | 7163 | i_size = actual_len; |
d1ea6a61 | 7164 | else |
55a61d1d JB |
7165 | i_size = cur_offset; |
7166 | i_size_write(inode, i_size); | |
7167 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7168 | } |
7169 | ||
d899e052 YZ |
7170 | ret = btrfs_update_inode(trans, root, inode); |
7171 | BUG_ON(ret); | |
d899e052 | 7172 | |
0af3d00b JB |
7173 | if (own_trans) |
7174 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7175 | } |
d899e052 YZ |
7176 | return ret; |
7177 | } | |
7178 | ||
0af3d00b JB |
7179 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7180 | u64 start, u64 num_bytes, u64 min_size, | |
7181 | loff_t actual_len, u64 *alloc_hint) | |
7182 | { | |
7183 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7184 | min_size, actual_len, alloc_hint, | |
7185 | NULL); | |
7186 | } | |
7187 | ||
7188 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7189 | struct btrfs_trans_handle *trans, int mode, | |
7190 | u64 start, u64 num_bytes, u64 min_size, | |
7191 | loff_t actual_len, u64 *alloc_hint) | |
7192 | { | |
7193 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7194 | min_size, actual_len, alloc_hint, trans); | |
7195 | } | |
7196 | ||
e6dcd2dc CM |
7197 | static int btrfs_set_page_dirty(struct page *page) |
7198 | { | |
e6dcd2dc CM |
7199 | return __set_page_dirty_nobuffers(page); |
7200 | } | |
7201 | ||
10556cb2 | 7202 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7203 | { |
b83cc969 | 7204 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7205 | umode_t mode = inode->i_mode; |
b83cc969 | 7206 | |
cb6db4e5 JM |
7207 | if (mask & MAY_WRITE && |
7208 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7209 | if (btrfs_root_readonly(root)) | |
7210 | return -EROFS; | |
7211 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7212 | return -EACCES; | |
7213 | } | |
2830ba7f | 7214 | return generic_permission(inode, mask); |
fdebe2bd | 7215 | } |
39279cc3 | 7216 | |
6e1d5dcc | 7217 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7218 | .getattr = btrfs_getattr, |
39279cc3 CM |
7219 | .lookup = btrfs_lookup, |
7220 | .create = btrfs_create, | |
7221 | .unlink = btrfs_unlink, | |
7222 | .link = btrfs_link, | |
7223 | .mkdir = btrfs_mkdir, | |
7224 | .rmdir = btrfs_rmdir, | |
7225 | .rename = btrfs_rename, | |
7226 | .symlink = btrfs_symlink, | |
7227 | .setattr = btrfs_setattr, | |
618e21d5 | 7228 | .mknod = btrfs_mknod, |
95819c05 CH |
7229 | .setxattr = btrfs_setxattr, |
7230 | .getxattr = btrfs_getxattr, | |
5103e947 | 7231 | .listxattr = btrfs_listxattr, |
95819c05 | 7232 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7233 | .permission = btrfs_permission, |
4e34e719 | 7234 | .get_acl = btrfs_get_acl, |
39279cc3 | 7235 | }; |
6e1d5dcc | 7236 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7237 | .lookup = btrfs_lookup, |
fdebe2bd | 7238 | .permission = btrfs_permission, |
4e34e719 | 7239 | .get_acl = btrfs_get_acl, |
39279cc3 | 7240 | }; |
76dda93c | 7241 | |
828c0950 | 7242 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7243 | .llseek = generic_file_llseek, |
7244 | .read = generic_read_dir, | |
cbdf5a24 | 7245 | .readdir = btrfs_real_readdir, |
34287aa3 | 7246 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7247 | #ifdef CONFIG_COMPAT |
34287aa3 | 7248 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7249 | #endif |
6bf13c0c | 7250 | .release = btrfs_release_file, |
e02119d5 | 7251 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7252 | }; |
7253 | ||
d1310b2e | 7254 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7255 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7256 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7257 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7258 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7259 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7260 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7261 | .set_bit_hook = btrfs_set_bit_hook, |
7262 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7263 | .merge_extent_hook = btrfs_merge_extent_hook, |
7264 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7265 | }; |
7266 | ||
35054394 CM |
7267 | /* |
7268 | * btrfs doesn't support the bmap operation because swapfiles | |
7269 | * use bmap to make a mapping of extents in the file. They assume | |
7270 | * these extents won't change over the life of the file and they | |
7271 | * use the bmap result to do IO directly to the drive. | |
7272 | * | |
7273 | * the btrfs bmap call would return logical addresses that aren't | |
7274 | * suitable for IO and they also will change frequently as COW | |
7275 | * operations happen. So, swapfile + btrfs == corruption. | |
7276 | * | |
7277 | * For now we're avoiding this by dropping bmap. | |
7278 | */ | |
7f09410b | 7279 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7280 | .readpage = btrfs_readpage, |
7281 | .writepage = btrfs_writepage, | |
b293f02e | 7282 | .writepages = btrfs_writepages, |
3ab2fb5a | 7283 | .readpages = btrfs_readpages, |
16432985 | 7284 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7285 | .invalidatepage = btrfs_invalidatepage, |
7286 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7287 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7288 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7289 | }; |
7290 | ||
7f09410b | 7291 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7292 | .readpage = btrfs_readpage, |
7293 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7294 | .invalidatepage = btrfs_invalidatepage, |
7295 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7296 | }; |
7297 | ||
6e1d5dcc | 7298 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7299 | .getattr = btrfs_getattr, |
7300 | .setattr = btrfs_setattr, | |
95819c05 CH |
7301 | .setxattr = btrfs_setxattr, |
7302 | .getxattr = btrfs_getxattr, | |
5103e947 | 7303 | .listxattr = btrfs_listxattr, |
95819c05 | 7304 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7305 | .permission = btrfs_permission, |
1506fcc8 | 7306 | .fiemap = btrfs_fiemap, |
4e34e719 | 7307 | .get_acl = btrfs_get_acl, |
39279cc3 | 7308 | }; |
6e1d5dcc | 7309 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7310 | .getattr = btrfs_getattr, |
7311 | .setattr = btrfs_setattr, | |
fdebe2bd | 7312 | .permission = btrfs_permission, |
95819c05 CH |
7313 | .setxattr = btrfs_setxattr, |
7314 | .getxattr = btrfs_getxattr, | |
33268eaf | 7315 | .listxattr = btrfs_listxattr, |
95819c05 | 7316 | .removexattr = btrfs_removexattr, |
4e34e719 | 7317 | .get_acl = btrfs_get_acl, |
618e21d5 | 7318 | }; |
6e1d5dcc | 7319 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7320 | .readlink = generic_readlink, |
7321 | .follow_link = page_follow_link_light, | |
7322 | .put_link = page_put_link, | |
f209561a | 7323 | .getattr = btrfs_getattr, |
fdebe2bd | 7324 | .permission = btrfs_permission, |
0279b4cd JO |
7325 | .setxattr = btrfs_setxattr, |
7326 | .getxattr = btrfs_getxattr, | |
7327 | .listxattr = btrfs_listxattr, | |
7328 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7329 | .get_acl = btrfs_get_acl, |
39279cc3 | 7330 | }; |
76dda93c | 7331 | |
82d339d9 | 7332 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7333 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7334 | .d_release = btrfs_dentry_release, |
76dda93c | 7335 | }; |