]>
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> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
4b4e25f2 | 43 | #include "compat.h" |
39279cc3 CM |
44 | #include "ctree.h" |
45 | #include "disk-io.h" | |
46 | #include "transaction.h" | |
47 | #include "btrfs_inode.h" | |
39279cc3 | 48 | #include "print-tree.h" |
e6dcd2dc | 49 | #include "ordered-data.h" |
95819c05 | 50 | #include "xattr.h" |
e02119d5 | 51 | #include "tree-log.h" |
4a54c8c1 | 52 | #include "volumes.h" |
c8b97818 | 53 | #include "compression.h" |
b4ce94de | 54 | #include "locking.h" |
dc89e982 | 55 | #include "free-space-cache.h" |
581bb050 | 56 | #include "inode-map.h" |
39279cc3 CM |
57 | |
58 | struct btrfs_iget_args { | |
59 | u64 ino; | |
60 | struct btrfs_root *root; | |
61 | }; | |
62 | ||
6e1d5dcc AD |
63 | static const struct inode_operations btrfs_dir_inode_operations; |
64 | static const struct inode_operations btrfs_symlink_inode_operations; | |
65 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
66 | static const struct inode_operations btrfs_special_inode_operations; | |
67 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
68 | static const struct address_space_operations btrfs_aops; |
69 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 70 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 71 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
72 | |
73 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 74 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
75 | struct kmem_cache *btrfs_trans_handle_cachep; |
76 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 77 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 78 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
79 | |
80 | #define S_SHIFT 12 | |
81 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
82 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
83 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
84 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
85 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
86 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
87 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
88 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
89 | }; | |
90 | ||
3972f260 | 91 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 92 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 93 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
94 | static noinline int cow_file_range(struct inode *inode, |
95 | struct page *locked_page, | |
96 | u64 start, u64 end, int *page_started, | |
97 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
98 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
99 | u64 len, u64 orig_start, | |
100 | u64 block_start, u64 block_len, | |
101 | u64 orig_block_len, int type); | |
7b128766 | 102 | |
f34f57a3 | 103 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
104 | struct inode *inode, struct inode *dir, |
105 | const struct qstr *qstr) | |
0279b4cd JO |
106 | { |
107 | int err; | |
108 | ||
f34f57a3 | 109 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 110 | if (!err) |
2a7dba39 | 111 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
112 | return err; |
113 | } | |
114 | ||
c8b97818 CM |
115 | /* |
116 | * this does all the hard work for inserting an inline extent into | |
117 | * the btree. The caller should have done a btrfs_drop_extents so that | |
118 | * no overlapping inline items exist in the btree | |
119 | */ | |
d397712b | 120 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
121 | struct btrfs_root *root, struct inode *inode, |
122 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 123 | int compress_type, |
c8b97818 CM |
124 | struct page **compressed_pages) |
125 | { | |
126 | struct btrfs_key key; | |
127 | struct btrfs_path *path; | |
128 | struct extent_buffer *leaf; | |
129 | struct page *page = NULL; | |
130 | char *kaddr; | |
131 | unsigned long ptr; | |
132 | struct btrfs_file_extent_item *ei; | |
133 | int err = 0; | |
134 | int ret; | |
135 | size_t cur_size = size; | |
136 | size_t datasize; | |
137 | unsigned long offset; | |
c8b97818 | 138 | |
fe3f566c | 139 | if (compressed_size && compressed_pages) |
c8b97818 | 140 | cur_size = compressed_size; |
c8b97818 | 141 | |
d397712b CM |
142 | path = btrfs_alloc_path(); |
143 | if (!path) | |
c8b97818 CM |
144 | return -ENOMEM; |
145 | ||
b9473439 | 146 | path->leave_spinning = 1; |
c8b97818 | 147 | |
33345d01 | 148 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
149 | key.offset = start; |
150 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
151 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
152 | ||
153 | inode_add_bytes(inode, size); | |
154 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
155 | datasize); | |
c8b97818 CM |
156 | if (ret) { |
157 | err = ret; | |
c8b97818 CM |
158 | goto fail; |
159 | } | |
160 | leaf = path->nodes[0]; | |
161 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
162 | struct btrfs_file_extent_item); | |
163 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
164 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
165 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
166 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
167 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
168 | ptr = btrfs_file_extent_inline_start(ei); | |
169 | ||
261507a0 | 170 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
171 | struct page *cpage; |
172 | int i = 0; | |
d397712b | 173 | while (compressed_size > 0) { |
c8b97818 | 174 | cpage = compressed_pages[i]; |
5b050f04 | 175 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
176 | PAGE_CACHE_SIZE); |
177 | ||
7ac687d9 | 178 | kaddr = kmap_atomic(cpage); |
c8b97818 | 179 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 180 | kunmap_atomic(kaddr); |
c8b97818 CM |
181 | |
182 | i++; | |
183 | ptr += cur_size; | |
184 | compressed_size -= cur_size; | |
185 | } | |
186 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 187 | compress_type); |
c8b97818 CM |
188 | } else { |
189 | page = find_get_page(inode->i_mapping, | |
190 | start >> PAGE_CACHE_SHIFT); | |
191 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 192 | kaddr = kmap_atomic(page); |
c8b97818 CM |
193 | offset = start & (PAGE_CACHE_SIZE - 1); |
194 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 195 | kunmap_atomic(kaddr); |
c8b97818 CM |
196 | page_cache_release(page); |
197 | } | |
198 | btrfs_mark_buffer_dirty(leaf); | |
199 | btrfs_free_path(path); | |
200 | ||
c2167754 YZ |
201 | /* |
202 | * we're an inline extent, so nobody can | |
203 | * extend the file past i_size without locking | |
204 | * a page we already have locked. | |
205 | * | |
206 | * We must do any isize and inode updates | |
207 | * before we unlock the pages. Otherwise we | |
208 | * could end up racing with unlink. | |
209 | */ | |
c8b97818 | 210 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 211 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 212 | |
79787eaa | 213 | return ret; |
c8b97818 CM |
214 | fail: |
215 | btrfs_free_path(path); | |
216 | return err; | |
217 | } | |
218 | ||
219 | ||
220 | /* | |
221 | * conditionally insert an inline extent into the file. This | |
222 | * does the checks required to make sure the data is small enough | |
223 | * to fit as an inline extent. | |
224 | */ | |
7f366cfe | 225 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
226 | struct btrfs_root *root, |
227 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 228 | size_t compressed_size, int compress_type, |
c8b97818 CM |
229 | struct page **compressed_pages) |
230 | { | |
231 | u64 isize = i_size_read(inode); | |
232 | u64 actual_end = min(end + 1, isize); | |
233 | u64 inline_len = actual_end - start; | |
234 | u64 aligned_end = (end + root->sectorsize - 1) & | |
235 | ~((u64)root->sectorsize - 1); | |
c8b97818 CM |
236 | u64 data_len = inline_len; |
237 | int ret; | |
238 | ||
239 | if (compressed_size) | |
240 | data_len = compressed_size; | |
241 | ||
242 | if (start > 0 || | |
70b99e69 | 243 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
244 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
245 | (!compressed_size && | |
246 | (actual_end & (root->sectorsize - 1)) == 0) || | |
247 | end + 1 < isize || | |
248 | data_len > root->fs_info->max_inline) { | |
249 | return 1; | |
250 | } | |
251 | ||
2671485d | 252 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
253 | if (ret) |
254 | return ret; | |
c8b97818 CM |
255 | |
256 | if (isize > actual_end) | |
257 | inline_len = min_t(u64, isize, actual_end); | |
258 | ret = insert_inline_extent(trans, root, inode, start, | |
259 | inline_len, compressed_size, | |
fe3f566c | 260 | compress_type, compressed_pages); |
2adcac1a | 261 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
262 | btrfs_abort_transaction(trans, root, ret); |
263 | return ret; | |
2adcac1a JB |
264 | } else if (ret == -ENOSPC) { |
265 | return 1; | |
79787eaa | 266 | } |
2adcac1a | 267 | |
0ca1f7ce | 268 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 269 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
270 | return 0; |
271 | } | |
272 | ||
771ed689 CM |
273 | struct async_extent { |
274 | u64 start; | |
275 | u64 ram_size; | |
276 | u64 compressed_size; | |
277 | struct page **pages; | |
278 | unsigned long nr_pages; | |
261507a0 | 279 | int compress_type; |
771ed689 CM |
280 | struct list_head list; |
281 | }; | |
282 | ||
283 | struct async_cow { | |
284 | struct inode *inode; | |
285 | struct btrfs_root *root; | |
286 | struct page *locked_page; | |
287 | u64 start; | |
288 | u64 end; | |
289 | struct list_head extents; | |
290 | struct btrfs_work work; | |
291 | }; | |
292 | ||
293 | static noinline int add_async_extent(struct async_cow *cow, | |
294 | u64 start, u64 ram_size, | |
295 | u64 compressed_size, | |
296 | struct page **pages, | |
261507a0 LZ |
297 | unsigned long nr_pages, |
298 | int compress_type) | |
771ed689 CM |
299 | { |
300 | struct async_extent *async_extent; | |
301 | ||
302 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 303 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
304 | async_extent->start = start; |
305 | async_extent->ram_size = ram_size; | |
306 | async_extent->compressed_size = compressed_size; | |
307 | async_extent->pages = pages; | |
308 | async_extent->nr_pages = nr_pages; | |
261507a0 | 309 | async_extent->compress_type = compress_type; |
771ed689 CM |
310 | list_add_tail(&async_extent->list, &cow->extents); |
311 | return 0; | |
312 | } | |
313 | ||
d352ac68 | 314 | /* |
771ed689 CM |
315 | * we create compressed extents in two phases. The first |
316 | * phase compresses a range of pages that have already been | |
317 | * locked (both pages and state bits are locked). | |
c8b97818 | 318 | * |
771ed689 CM |
319 | * This is done inside an ordered work queue, and the compression |
320 | * is spread across many cpus. The actual IO submission is step | |
321 | * two, and the ordered work queue takes care of making sure that | |
322 | * happens in the same order things were put onto the queue by | |
323 | * writepages and friends. | |
c8b97818 | 324 | * |
771ed689 CM |
325 | * If this code finds it can't get good compression, it puts an |
326 | * entry onto the work queue to write the uncompressed bytes. This | |
327 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
328 | * are written in the same order that the flusher thread sent them |
329 | * down. | |
d352ac68 | 330 | */ |
771ed689 CM |
331 | static noinline int compress_file_range(struct inode *inode, |
332 | struct page *locked_page, | |
333 | u64 start, u64 end, | |
334 | struct async_cow *async_cow, | |
335 | int *num_added) | |
b888db2b CM |
336 | { |
337 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
338 | struct btrfs_trans_handle *trans; | |
db94535d | 339 | u64 num_bytes; |
db94535d | 340 | u64 blocksize = root->sectorsize; |
c8b97818 | 341 | u64 actual_end; |
42dc7bab | 342 | u64 isize = i_size_read(inode); |
e6dcd2dc | 343 | int ret = 0; |
c8b97818 CM |
344 | struct page **pages = NULL; |
345 | unsigned long nr_pages; | |
346 | unsigned long nr_pages_ret = 0; | |
347 | unsigned long total_compressed = 0; | |
348 | unsigned long total_in = 0; | |
349 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 350 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
351 | int i; |
352 | int will_compress; | |
261507a0 | 353 | int compress_type = root->fs_info->compress_type; |
b888db2b | 354 | |
4cb13e5d LB |
355 | /* if this is a small write inside eof, kick off a defrag */ |
356 | if ((end - start + 1) < 16 * 1024 && | |
357 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
358 | btrfs_add_inode_defrag(NULL, inode); |
359 | ||
42dc7bab | 360 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
361 | again: |
362 | will_compress = 0; | |
363 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
364 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 365 | |
f03d9301 CM |
366 | /* |
367 | * we don't want to send crud past the end of i_size through | |
368 | * compression, that's just a waste of CPU time. So, if the | |
369 | * end of the file is before the start of our current | |
370 | * requested range of bytes, we bail out to the uncompressed | |
371 | * cleanup code that can deal with all of this. | |
372 | * | |
373 | * It isn't really the fastest way to fix things, but this is a | |
374 | * very uncommon corner. | |
375 | */ | |
376 | if (actual_end <= start) | |
377 | goto cleanup_and_bail_uncompressed; | |
378 | ||
c8b97818 CM |
379 | total_compressed = actual_end - start; |
380 | ||
381 | /* we want to make sure that amount of ram required to uncompress | |
382 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
383 | * of a compressed extent to 128k. This is a crucial number |
384 | * because it also controls how easily we can spread reads across | |
385 | * cpus for decompression. | |
386 | * | |
387 | * We also want to make sure the amount of IO required to do | |
388 | * a random read is reasonably small, so we limit the size of | |
389 | * a compressed extent to 128k. | |
c8b97818 CM |
390 | */ |
391 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 392 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 393 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
394 | total_in = 0; |
395 | ret = 0; | |
db94535d | 396 | |
771ed689 CM |
397 | /* |
398 | * we do compression for mount -o compress and when the | |
399 | * inode has not been flagged as nocompress. This flag can | |
400 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 401 | */ |
6cbff00f | 402 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 403 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
404 | (BTRFS_I(inode)->force_compress) || |
405 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 406 | WARN_ON(pages); |
cfbc246e | 407 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
408 | if (!pages) { |
409 | /* just bail out to the uncompressed code */ | |
410 | goto cont; | |
411 | } | |
c8b97818 | 412 | |
261507a0 LZ |
413 | if (BTRFS_I(inode)->force_compress) |
414 | compress_type = BTRFS_I(inode)->force_compress; | |
415 | ||
416 | ret = btrfs_compress_pages(compress_type, | |
417 | inode->i_mapping, start, | |
418 | total_compressed, pages, | |
419 | nr_pages, &nr_pages_ret, | |
420 | &total_in, | |
421 | &total_compressed, | |
422 | max_compressed); | |
c8b97818 CM |
423 | |
424 | if (!ret) { | |
425 | unsigned long offset = total_compressed & | |
426 | (PAGE_CACHE_SIZE - 1); | |
427 | struct page *page = pages[nr_pages_ret - 1]; | |
428 | char *kaddr; | |
429 | ||
430 | /* zero the tail end of the last page, we might be | |
431 | * sending it down to disk | |
432 | */ | |
433 | if (offset) { | |
7ac687d9 | 434 | kaddr = kmap_atomic(page); |
c8b97818 CM |
435 | memset(kaddr + offset, 0, |
436 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 437 | kunmap_atomic(kaddr); |
c8b97818 CM |
438 | } |
439 | will_compress = 1; | |
440 | } | |
441 | } | |
560f7d75 | 442 | cont: |
c8b97818 | 443 | if (start == 0) { |
7a7eaa40 | 444 | trans = btrfs_join_transaction(root); |
79787eaa JM |
445 | if (IS_ERR(trans)) { |
446 | ret = PTR_ERR(trans); | |
447 | trans = NULL; | |
448 | goto cleanup_and_out; | |
449 | } | |
0ca1f7ce | 450 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 451 | |
c8b97818 | 452 | /* lets try to make an inline extent */ |
771ed689 | 453 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 454 | /* we didn't compress the entire range, try |
771ed689 | 455 | * to make an uncompressed inline extent. |
c8b97818 CM |
456 | */ |
457 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 458 | start, end, 0, 0, NULL); |
c8b97818 | 459 | } else { |
771ed689 | 460 | /* try making a compressed inline extent */ |
c8b97818 CM |
461 | ret = cow_file_range_inline(trans, root, inode, |
462 | start, end, | |
fe3f566c LZ |
463 | total_compressed, |
464 | compress_type, pages); | |
c8b97818 | 465 | } |
79787eaa | 466 | if (ret <= 0) { |
771ed689 | 467 | /* |
79787eaa JM |
468 | * inline extent creation worked or returned error, |
469 | * we don't need to create any more async work items. | |
470 | * Unlock and free up our temp pages. | |
771ed689 | 471 | */ |
c8b97818 | 472 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
473 | &BTRFS_I(inode)->io_tree, |
474 | start, end, NULL, | |
475 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 476 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 477 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
478 | |
479 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
480 | goto free_pages_out; |
481 | } | |
c2167754 | 482 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
483 | } |
484 | ||
485 | if (will_compress) { | |
486 | /* | |
487 | * we aren't doing an inline extent round the compressed size | |
488 | * up to a block size boundary so the allocator does sane | |
489 | * things | |
490 | */ | |
491 | total_compressed = (total_compressed + blocksize - 1) & | |
492 | ~(blocksize - 1); | |
493 | ||
494 | /* | |
495 | * one last check to make sure the compression is really a | |
496 | * win, compare the page count read with the blocks on disk | |
497 | */ | |
498 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
499 | ~(PAGE_CACHE_SIZE - 1); | |
500 | if (total_compressed >= total_in) { | |
501 | will_compress = 0; | |
502 | } else { | |
c8b97818 CM |
503 | num_bytes = total_in; |
504 | } | |
505 | } | |
506 | if (!will_compress && pages) { | |
507 | /* | |
508 | * the compression code ran but failed to make things smaller, | |
509 | * free any pages it allocated and our page pointer array | |
510 | */ | |
511 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 512 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
513 | page_cache_release(pages[i]); |
514 | } | |
515 | kfree(pages); | |
516 | pages = NULL; | |
517 | total_compressed = 0; | |
518 | nr_pages_ret = 0; | |
519 | ||
520 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
521 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
522 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 523 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 524 | } |
c8b97818 | 525 | } |
771ed689 CM |
526 | if (will_compress) { |
527 | *num_added += 1; | |
c8b97818 | 528 | |
771ed689 CM |
529 | /* the async work queues will take care of doing actual |
530 | * allocation on disk for these compressed pages, | |
531 | * and will submit them to the elevator. | |
532 | */ | |
533 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
534 | total_compressed, pages, nr_pages_ret, |
535 | compress_type); | |
179e29e4 | 536 | |
24ae6365 | 537 | if (start + num_bytes < end) { |
771ed689 CM |
538 | start += num_bytes; |
539 | pages = NULL; | |
540 | cond_resched(); | |
541 | goto again; | |
542 | } | |
543 | } else { | |
f03d9301 | 544 | cleanup_and_bail_uncompressed: |
771ed689 CM |
545 | /* |
546 | * No compression, but we still need to write the pages in | |
547 | * the file we've been given so far. redirty the locked | |
548 | * page if it corresponds to our extent and set things up | |
549 | * for the async work queue to run cow_file_range to do | |
550 | * the normal delalloc dance | |
551 | */ | |
552 | if (page_offset(locked_page) >= start && | |
553 | page_offset(locked_page) <= end) { | |
554 | __set_page_dirty_nobuffers(locked_page); | |
555 | /* unlocked later on in the async handlers */ | |
556 | } | |
261507a0 LZ |
557 | add_async_extent(async_cow, start, end - start + 1, |
558 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
559 | *num_added += 1; |
560 | } | |
3b951516 | 561 | |
771ed689 | 562 | out: |
79787eaa | 563 | return ret; |
771ed689 CM |
564 | |
565 | free_pages_out: | |
566 | for (i = 0; i < nr_pages_ret; i++) { | |
567 | WARN_ON(pages[i]->mapping); | |
568 | page_cache_release(pages[i]); | |
569 | } | |
d397712b | 570 | kfree(pages); |
771ed689 CM |
571 | |
572 | goto out; | |
79787eaa JM |
573 | |
574 | cleanup_and_out: | |
575 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
576 | start, end, NULL, | |
577 | EXTENT_CLEAR_UNLOCK_PAGE | | |
578 | EXTENT_CLEAR_DIRTY | | |
579 | EXTENT_CLEAR_DELALLOC | | |
580 | EXTENT_SET_WRITEBACK | | |
581 | EXTENT_END_WRITEBACK); | |
582 | if (!trans || IS_ERR(trans)) | |
583 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
584 | else | |
585 | btrfs_abort_transaction(trans, root, ret); | |
586 | goto free_pages_out; | |
771ed689 CM |
587 | } |
588 | ||
589 | /* | |
590 | * phase two of compressed writeback. This is the ordered portion | |
591 | * of the code, which only gets called in the order the work was | |
592 | * queued. We walk all the async extents created by compress_file_range | |
593 | * and send them down to the disk. | |
594 | */ | |
595 | static noinline int submit_compressed_extents(struct inode *inode, | |
596 | struct async_cow *async_cow) | |
597 | { | |
598 | struct async_extent *async_extent; | |
599 | u64 alloc_hint = 0; | |
600 | struct btrfs_trans_handle *trans; | |
601 | struct btrfs_key ins; | |
602 | struct extent_map *em; | |
603 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
604 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
605 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 606 | int ret = 0; |
771ed689 CM |
607 | |
608 | if (list_empty(&async_cow->extents)) | |
609 | return 0; | |
610 | ||
3e04e7f1 | 611 | again: |
d397712b | 612 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
613 | async_extent = list_entry(async_cow->extents.next, |
614 | struct async_extent, list); | |
615 | list_del(&async_extent->list); | |
c8b97818 | 616 | |
771ed689 CM |
617 | io_tree = &BTRFS_I(inode)->io_tree; |
618 | ||
f5a84ee3 | 619 | retry: |
771ed689 CM |
620 | /* did the compression code fall back to uncompressed IO? */ |
621 | if (!async_extent->pages) { | |
622 | int page_started = 0; | |
623 | unsigned long nr_written = 0; | |
624 | ||
625 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 626 | async_extent->start + |
d0082371 | 627 | async_extent->ram_size - 1); |
771ed689 CM |
628 | |
629 | /* allocate blocks */ | |
f5a84ee3 JB |
630 | ret = cow_file_range(inode, async_cow->locked_page, |
631 | async_extent->start, | |
632 | async_extent->start + | |
633 | async_extent->ram_size - 1, | |
634 | &page_started, &nr_written, 0); | |
771ed689 | 635 | |
79787eaa JM |
636 | /* JDM XXX */ |
637 | ||
771ed689 CM |
638 | /* |
639 | * if page_started, cow_file_range inserted an | |
640 | * inline extent and took care of all the unlocking | |
641 | * and IO for us. Otherwise, we need to submit | |
642 | * all those pages down to the drive. | |
643 | */ | |
f5a84ee3 | 644 | if (!page_started && !ret) |
771ed689 CM |
645 | extent_write_locked_range(io_tree, |
646 | inode, async_extent->start, | |
d397712b | 647 | async_extent->start + |
771ed689 CM |
648 | async_extent->ram_size - 1, |
649 | btrfs_get_extent, | |
650 | WB_SYNC_ALL); | |
3e04e7f1 JB |
651 | else if (ret) |
652 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
653 | kfree(async_extent); |
654 | cond_resched(); | |
655 | continue; | |
656 | } | |
657 | ||
658 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 659 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 660 | |
7a7eaa40 | 661 | trans = btrfs_join_transaction(root); |
79787eaa JM |
662 | if (IS_ERR(trans)) { |
663 | ret = PTR_ERR(trans); | |
664 | } else { | |
665 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
666 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
667 | async_extent->compressed_size, |
668 | async_extent->compressed_size, | |
81c9ad23 | 669 | 0, alloc_hint, &ins, 1); |
962197ba | 670 | if (ret && ret != -ENOSPC) |
79787eaa JM |
671 | btrfs_abort_transaction(trans, root, ret); |
672 | btrfs_end_transaction(trans, root); | |
673 | } | |
c2167754 | 674 | |
f5a84ee3 JB |
675 | if (ret) { |
676 | int i; | |
3e04e7f1 | 677 | |
f5a84ee3 JB |
678 | for (i = 0; i < async_extent->nr_pages; i++) { |
679 | WARN_ON(async_extent->pages[i]->mapping); | |
680 | page_cache_release(async_extent->pages[i]); | |
681 | } | |
682 | kfree(async_extent->pages); | |
683 | async_extent->nr_pages = 0; | |
684 | async_extent->pages = NULL; | |
3e04e7f1 | 685 | |
79787eaa JM |
686 | if (ret == -ENOSPC) |
687 | goto retry; | |
3e04e7f1 | 688 | goto out_free; |
f5a84ee3 JB |
689 | } |
690 | ||
c2167754 YZ |
691 | /* |
692 | * here we're doing allocation and writeback of the | |
693 | * compressed pages | |
694 | */ | |
695 | btrfs_drop_extent_cache(inode, async_extent->start, | |
696 | async_extent->start + | |
697 | async_extent->ram_size - 1, 0); | |
698 | ||
172ddd60 | 699 | em = alloc_extent_map(); |
3e04e7f1 JB |
700 | if (!em) |
701 | goto out_free_reserve; | |
771ed689 CM |
702 | em->start = async_extent->start; |
703 | em->len = async_extent->ram_size; | |
445a6944 | 704 | em->orig_start = em->start; |
2ab28f32 JB |
705 | em->mod_start = em->start; |
706 | em->mod_len = em->len; | |
c8b97818 | 707 | |
771ed689 CM |
708 | em->block_start = ins.objectid; |
709 | em->block_len = ins.offset; | |
b4939680 | 710 | em->orig_block_len = ins.offset; |
771ed689 | 711 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 712 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
713 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
714 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 715 | em->generation = -1; |
771ed689 | 716 | |
d397712b | 717 | while (1) { |
890871be | 718 | write_lock(&em_tree->lock); |
771ed689 | 719 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
720 | if (!ret) |
721 | list_move(&em->list, | |
722 | &em_tree->modified_extents); | |
890871be | 723 | write_unlock(&em_tree->lock); |
771ed689 CM |
724 | if (ret != -EEXIST) { |
725 | free_extent_map(em); | |
726 | break; | |
727 | } | |
728 | btrfs_drop_extent_cache(inode, async_extent->start, | |
729 | async_extent->start + | |
730 | async_extent->ram_size - 1, 0); | |
731 | } | |
732 | ||
3e04e7f1 JB |
733 | if (ret) |
734 | goto out_free_reserve; | |
735 | ||
261507a0 LZ |
736 | ret = btrfs_add_ordered_extent_compress(inode, |
737 | async_extent->start, | |
738 | ins.objectid, | |
739 | async_extent->ram_size, | |
740 | ins.offset, | |
741 | BTRFS_ORDERED_COMPRESSED, | |
742 | async_extent->compress_type); | |
3e04e7f1 JB |
743 | if (ret) |
744 | goto out_free_reserve; | |
771ed689 | 745 | |
771ed689 CM |
746 | /* |
747 | * clear dirty, set writeback and unlock the pages. | |
748 | */ | |
749 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
750 | &BTRFS_I(inode)->io_tree, |
751 | async_extent->start, | |
752 | async_extent->start + | |
753 | async_extent->ram_size - 1, | |
754 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
755 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 756 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 757 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
758 | |
759 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
760 | async_extent->start, |
761 | async_extent->ram_size, | |
762 | ins.objectid, | |
763 | ins.offset, async_extent->pages, | |
764 | async_extent->nr_pages); | |
771ed689 CM |
765 | alloc_hint = ins.objectid + ins.offset; |
766 | kfree(async_extent); | |
3e04e7f1 JB |
767 | if (ret) |
768 | goto out; | |
771ed689 CM |
769 | cond_resched(); |
770 | } | |
79787eaa JM |
771 | ret = 0; |
772 | out: | |
773 | return ret; | |
3e04e7f1 JB |
774 | out_free_reserve: |
775 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 776 | out_free: |
3e04e7f1 JB |
777 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
778 | async_extent->start, | |
779 | async_extent->start + | |
780 | async_extent->ram_size - 1, | |
781 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
782 | EXTENT_CLEAR_UNLOCK | | |
783 | EXTENT_CLEAR_DELALLOC | | |
784 | EXTENT_CLEAR_DIRTY | | |
785 | EXTENT_SET_WRITEBACK | | |
786 | EXTENT_END_WRITEBACK); | |
79787eaa | 787 | kfree(async_extent); |
3e04e7f1 | 788 | goto again; |
771ed689 CM |
789 | } |
790 | ||
4b46fce2 JB |
791 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
792 | u64 num_bytes) | |
793 | { | |
794 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
795 | struct extent_map *em; | |
796 | u64 alloc_hint = 0; | |
797 | ||
798 | read_lock(&em_tree->lock); | |
799 | em = search_extent_mapping(em_tree, start, num_bytes); | |
800 | if (em) { | |
801 | /* | |
802 | * if block start isn't an actual block number then find the | |
803 | * first block in this inode and use that as a hint. If that | |
804 | * block is also bogus then just don't worry about it. | |
805 | */ | |
806 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
807 | free_extent_map(em); | |
808 | em = search_extent_mapping(em_tree, 0, 0); | |
809 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
810 | alloc_hint = em->block_start; | |
811 | if (em) | |
812 | free_extent_map(em); | |
813 | } else { | |
814 | alloc_hint = em->block_start; | |
815 | free_extent_map(em); | |
816 | } | |
817 | } | |
818 | read_unlock(&em_tree->lock); | |
819 | ||
820 | return alloc_hint; | |
821 | } | |
822 | ||
771ed689 CM |
823 | /* |
824 | * when extent_io.c finds a delayed allocation range in the file, | |
825 | * the call backs end up in this code. The basic idea is to | |
826 | * allocate extents on disk for the range, and create ordered data structs | |
827 | * in ram to track those extents. | |
828 | * | |
829 | * locked_page is the page that writepage had locked already. We use | |
830 | * it to make sure we don't do extra locks or unlocks. | |
831 | * | |
832 | * *page_started is set to one if we unlock locked_page and do everything | |
833 | * required to start IO on it. It may be clean and already done with | |
834 | * IO when we return. | |
835 | */ | |
b7d5b0a8 MX |
836 | static noinline int __cow_file_range(struct btrfs_trans_handle *trans, |
837 | struct inode *inode, | |
838 | struct btrfs_root *root, | |
839 | struct page *locked_page, | |
840 | u64 start, u64 end, int *page_started, | |
841 | unsigned long *nr_written, | |
842 | int unlock) | |
771ed689 | 843 | { |
771ed689 CM |
844 | u64 alloc_hint = 0; |
845 | u64 num_bytes; | |
846 | unsigned long ram_size; | |
847 | u64 disk_num_bytes; | |
848 | u64 cur_alloc_size; | |
849 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
850 | struct btrfs_key ins; |
851 | struct extent_map *em; | |
852 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
853 | int ret = 0; | |
854 | ||
83eea1f1 | 855 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 856 | |
771ed689 CM |
857 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
858 | num_bytes = max(blocksize, num_bytes); | |
859 | disk_num_bytes = num_bytes; | |
771ed689 | 860 | |
4cb5300b | 861 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
862 | if (num_bytes < 64 * 1024 && |
863 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
864 | btrfs_add_inode_defrag(trans, inode); |
865 | ||
771ed689 CM |
866 | if (start == 0) { |
867 | /* lets try to make an inline extent */ | |
868 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 869 | start, end, 0, 0, NULL); |
771ed689 CM |
870 | if (ret == 0) { |
871 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
872 | &BTRFS_I(inode)->io_tree, |
873 | start, end, NULL, | |
874 | EXTENT_CLEAR_UNLOCK_PAGE | | |
875 | EXTENT_CLEAR_UNLOCK | | |
876 | EXTENT_CLEAR_DELALLOC | | |
877 | EXTENT_CLEAR_DIRTY | | |
878 | EXTENT_SET_WRITEBACK | | |
879 | EXTENT_END_WRITEBACK); | |
c2167754 | 880 | |
771ed689 CM |
881 | *nr_written = *nr_written + |
882 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
883 | *page_started = 1; | |
771ed689 | 884 | goto out; |
79787eaa JM |
885 | } else if (ret < 0) { |
886 | btrfs_abort_transaction(trans, root, ret); | |
887 | goto out_unlock; | |
771ed689 CM |
888 | } |
889 | } | |
890 | ||
891 | BUG_ON(disk_num_bytes > | |
6c41761f | 892 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 893 | |
4b46fce2 | 894 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
895 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
896 | ||
d397712b | 897 | while (disk_num_bytes > 0) { |
a791e35e CM |
898 | unsigned long op; |
899 | ||
287a0ab9 | 900 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 901 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 902 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 903 | &ins, 1); |
79787eaa JM |
904 | if (ret < 0) { |
905 | btrfs_abort_transaction(trans, root, ret); | |
906 | goto out_unlock; | |
907 | } | |
d397712b | 908 | |
172ddd60 | 909 | em = alloc_extent_map(); |
79787eaa | 910 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 911 | em->start = start; |
445a6944 | 912 | em->orig_start = em->start; |
771ed689 CM |
913 | ram_size = ins.offset; |
914 | em->len = ins.offset; | |
2ab28f32 JB |
915 | em->mod_start = em->start; |
916 | em->mod_len = em->len; | |
c8b97818 | 917 | |
e6dcd2dc | 918 | em->block_start = ins.objectid; |
c8b97818 | 919 | em->block_len = ins.offset; |
b4939680 | 920 | em->orig_block_len = ins.offset; |
e6dcd2dc | 921 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 922 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 923 | em->generation = -1; |
c8b97818 | 924 | |
d397712b | 925 | while (1) { |
890871be | 926 | write_lock(&em_tree->lock); |
e6dcd2dc | 927 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
928 | if (!ret) |
929 | list_move(&em->list, | |
930 | &em_tree->modified_extents); | |
890871be | 931 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
932 | if (ret != -EEXIST) { |
933 | free_extent_map(em); | |
934 | break; | |
935 | } | |
936 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 937 | start + ram_size - 1, 0); |
e6dcd2dc CM |
938 | } |
939 | ||
98d20f67 | 940 | cur_alloc_size = ins.offset; |
e6dcd2dc | 941 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 942 | ram_size, cur_alloc_size, 0); |
79787eaa | 943 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 944 | |
17d217fe YZ |
945 | if (root->root_key.objectid == |
946 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
947 | ret = btrfs_reloc_clone_csums(inode, start, | |
948 | cur_alloc_size); | |
79787eaa JM |
949 | if (ret) { |
950 | btrfs_abort_transaction(trans, root, ret); | |
951 | goto out_unlock; | |
952 | } | |
17d217fe YZ |
953 | } |
954 | ||
d397712b | 955 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 956 | break; |
d397712b | 957 | |
c8b97818 CM |
958 | /* we're not doing compressed IO, don't unlock the first |
959 | * page (which the caller expects to stay locked), don't | |
960 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
961 | * |
962 | * Do set the Private2 bit so we know this page was properly | |
963 | * setup for writepage | |
c8b97818 | 964 | */ |
a791e35e CM |
965 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
966 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
967 | EXTENT_SET_PRIVATE2; | |
968 | ||
c8b97818 CM |
969 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
970 | start, start + ram_size - 1, | |
a791e35e | 971 | locked_page, op); |
c8b97818 | 972 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
973 | num_bytes -= cur_alloc_size; |
974 | alloc_hint = ins.objectid + ins.offset; | |
975 | start += cur_alloc_size; | |
b888db2b | 976 | } |
79787eaa | 977 | out: |
be20aa9d | 978 | return ret; |
b7d5b0a8 | 979 | |
79787eaa JM |
980 | out_unlock: |
981 | extent_clear_unlock_delalloc(inode, | |
982 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 983 | start, end, locked_page, |
79787eaa JM |
984 | EXTENT_CLEAR_UNLOCK_PAGE | |
985 | EXTENT_CLEAR_UNLOCK | | |
986 | EXTENT_CLEAR_DELALLOC | | |
987 | EXTENT_CLEAR_DIRTY | | |
988 | EXTENT_SET_WRITEBACK | | |
989 | EXTENT_END_WRITEBACK); | |
990 | ||
991 | goto out; | |
771ed689 | 992 | } |
c8b97818 | 993 | |
b7d5b0a8 MX |
994 | static noinline int cow_file_range(struct inode *inode, |
995 | struct page *locked_page, | |
996 | u64 start, u64 end, int *page_started, | |
997 | unsigned long *nr_written, | |
998 | int unlock) | |
999 | { | |
1000 | struct btrfs_trans_handle *trans; | |
1001 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1002 | int ret; | |
1003 | ||
1004 | trans = btrfs_join_transaction(root); | |
1005 | if (IS_ERR(trans)) { | |
1006 | extent_clear_unlock_delalloc(inode, | |
1007 | &BTRFS_I(inode)->io_tree, | |
1008 | start, end, locked_page, | |
1009 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1010 | EXTENT_CLEAR_UNLOCK | | |
1011 | EXTENT_CLEAR_DELALLOC | | |
1012 | EXTENT_CLEAR_DIRTY | | |
1013 | EXTENT_SET_WRITEBACK | | |
1014 | EXTENT_END_WRITEBACK); | |
1015 | return PTR_ERR(trans); | |
1016 | } | |
1017 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
1018 | ||
1019 | ret = __cow_file_range(trans, inode, root, locked_page, start, end, | |
1020 | page_started, nr_written, unlock); | |
1021 | ||
1022 | btrfs_end_transaction(trans, root); | |
1023 | ||
1024 | return ret; | |
1025 | } | |
1026 | ||
771ed689 CM |
1027 | /* |
1028 | * work queue call back to started compression on a file and pages | |
1029 | */ | |
1030 | static noinline void async_cow_start(struct btrfs_work *work) | |
1031 | { | |
1032 | struct async_cow *async_cow; | |
1033 | int num_added = 0; | |
1034 | async_cow = container_of(work, struct async_cow, work); | |
1035 | ||
1036 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1037 | async_cow->start, async_cow->end, async_cow, | |
1038 | &num_added); | |
8180ef88 | 1039 | if (num_added == 0) { |
cb77fcd8 | 1040 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1041 | async_cow->inode = NULL; |
8180ef88 | 1042 | } |
771ed689 CM |
1043 | } |
1044 | ||
1045 | /* | |
1046 | * work queue call back to submit previously compressed pages | |
1047 | */ | |
1048 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1049 | { | |
1050 | struct async_cow *async_cow; | |
1051 | struct btrfs_root *root; | |
1052 | unsigned long nr_pages; | |
1053 | ||
1054 | async_cow = container_of(work, struct async_cow, work); | |
1055 | ||
1056 | root = async_cow->root; | |
1057 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1058 | PAGE_CACHE_SHIFT; | |
1059 | ||
66657b31 | 1060 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1061 | 5 * 1024 * 1024 && |
771ed689 CM |
1062 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1063 | wake_up(&root->fs_info->async_submit_wait); | |
1064 | ||
d397712b | 1065 | if (async_cow->inode) |
771ed689 | 1066 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1067 | } |
c8b97818 | 1068 | |
771ed689 CM |
1069 | static noinline void async_cow_free(struct btrfs_work *work) |
1070 | { | |
1071 | struct async_cow *async_cow; | |
1072 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1073 | if (async_cow->inode) |
cb77fcd8 | 1074 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1075 | kfree(async_cow); |
1076 | } | |
1077 | ||
1078 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1079 | u64 start, u64 end, int *page_started, | |
1080 | unsigned long *nr_written) | |
1081 | { | |
1082 | struct async_cow *async_cow; | |
1083 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1084 | unsigned long nr_pages; | |
1085 | u64 cur_end; | |
287082b0 | 1086 | int limit = 10 * 1024 * 1024; |
771ed689 | 1087 | |
a3429ab7 CM |
1088 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1089 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1090 | while (start < end) { |
771ed689 | 1091 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1092 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1093 | async_cow->inode = igrab(inode); |
771ed689 CM |
1094 | async_cow->root = root; |
1095 | async_cow->locked_page = locked_page; | |
1096 | async_cow->start = start; | |
1097 | ||
6cbff00f | 1098 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1099 | cur_end = end; |
1100 | else | |
1101 | cur_end = min(end, start + 512 * 1024 - 1); | |
1102 | ||
1103 | async_cow->end = cur_end; | |
1104 | INIT_LIST_HEAD(&async_cow->extents); | |
1105 | ||
1106 | async_cow->work.func = async_cow_start; | |
1107 | async_cow->work.ordered_func = async_cow_submit; | |
1108 | async_cow->work.ordered_free = async_cow_free; | |
1109 | async_cow->work.flags = 0; | |
1110 | ||
771ed689 CM |
1111 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1112 | PAGE_CACHE_SHIFT; | |
1113 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1114 | ||
1115 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1116 | &async_cow->work); | |
1117 | ||
1118 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1119 | wait_event(root->fs_info->async_submit_wait, | |
1120 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1121 | limit)); | |
1122 | } | |
1123 | ||
d397712b | 1124 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1125 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1126 | wait_event(root->fs_info->async_submit_wait, | |
1127 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1128 | 0)); | |
1129 | } | |
1130 | ||
1131 | *nr_written += nr_pages; | |
1132 | start = cur_end + 1; | |
1133 | } | |
1134 | *page_started = 1; | |
1135 | return 0; | |
be20aa9d CM |
1136 | } |
1137 | ||
d397712b | 1138 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1139 | u64 bytenr, u64 num_bytes) |
1140 | { | |
1141 | int ret; | |
1142 | struct btrfs_ordered_sum *sums; | |
1143 | LIST_HEAD(list); | |
1144 | ||
07d400a6 | 1145 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1146 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1147 | if (ret == 0 && list_empty(&list)) |
1148 | return 0; | |
1149 | ||
1150 | while (!list_empty(&list)) { | |
1151 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1152 | list_del(&sums->list); | |
1153 | kfree(sums); | |
1154 | } | |
1155 | return 1; | |
1156 | } | |
1157 | ||
d352ac68 CM |
1158 | /* |
1159 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1160 | * of the extents that exist in the file, and COWs the file as required. | |
1161 | * | |
1162 | * If no cow copies or snapshots exist, we write directly to the existing | |
1163 | * blocks on disk | |
1164 | */ | |
7f366cfe CM |
1165 | static noinline int run_delalloc_nocow(struct inode *inode, |
1166 | struct page *locked_page, | |
771ed689 CM |
1167 | u64 start, u64 end, int *page_started, int force, |
1168 | unsigned long *nr_written) | |
be20aa9d | 1169 | { |
be20aa9d | 1170 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1171 | struct btrfs_trans_handle *trans; |
be20aa9d | 1172 | struct extent_buffer *leaf; |
be20aa9d | 1173 | struct btrfs_path *path; |
80ff3856 | 1174 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1175 | struct btrfs_key found_key; |
80ff3856 YZ |
1176 | u64 cow_start; |
1177 | u64 cur_offset; | |
1178 | u64 extent_end; | |
5d4f98a2 | 1179 | u64 extent_offset; |
80ff3856 YZ |
1180 | u64 disk_bytenr; |
1181 | u64 num_bytes; | |
b4939680 | 1182 | u64 disk_num_bytes; |
80ff3856 | 1183 | int extent_type; |
79787eaa | 1184 | int ret, err; |
d899e052 | 1185 | int type; |
80ff3856 YZ |
1186 | int nocow; |
1187 | int check_prev = 1; | |
82d5902d | 1188 | bool nolock; |
33345d01 | 1189 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1190 | |
1191 | path = btrfs_alloc_path(); | |
17ca04af JB |
1192 | if (!path) { |
1193 | extent_clear_unlock_delalloc(inode, | |
1194 | &BTRFS_I(inode)->io_tree, | |
1195 | start, end, locked_page, | |
1196 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1197 | EXTENT_CLEAR_UNLOCK | | |
1198 | EXTENT_CLEAR_DELALLOC | | |
1199 | EXTENT_CLEAR_DIRTY | | |
1200 | EXTENT_SET_WRITEBACK | | |
1201 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1202 | return -ENOMEM; |
17ca04af | 1203 | } |
82d5902d | 1204 | |
83eea1f1 | 1205 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1206 | |
1207 | if (nolock) | |
7a7eaa40 | 1208 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1209 | else |
7a7eaa40 | 1210 | trans = btrfs_join_transaction(root); |
ff5714cc | 1211 | |
79787eaa | 1212 | if (IS_ERR(trans)) { |
17ca04af JB |
1213 | extent_clear_unlock_delalloc(inode, |
1214 | &BTRFS_I(inode)->io_tree, | |
1215 | start, end, locked_page, | |
1216 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1217 | EXTENT_CLEAR_UNLOCK | | |
1218 | EXTENT_CLEAR_DELALLOC | | |
1219 | EXTENT_CLEAR_DIRTY | | |
1220 | EXTENT_SET_WRITEBACK | | |
1221 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1222 | btrfs_free_path(path); |
1223 | return PTR_ERR(trans); | |
1224 | } | |
1225 | ||
74b21075 | 1226 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1227 | |
80ff3856 YZ |
1228 | cow_start = (u64)-1; |
1229 | cur_offset = start; | |
1230 | while (1) { | |
33345d01 | 1231 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1232 | cur_offset, 0); |
79787eaa JM |
1233 | if (ret < 0) { |
1234 | btrfs_abort_transaction(trans, root, ret); | |
1235 | goto error; | |
1236 | } | |
80ff3856 YZ |
1237 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1238 | leaf = path->nodes[0]; | |
1239 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1240 | path->slots[0] - 1); | |
33345d01 | 1241 | if (found_key.objectid == ino && |
80ff3856 YZ |
1242 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1243 | path->slots[0]--; | |
1244 | } | |
1245 | check_prev = 0; | |
1246 | next_slot: | |
1247 | leaf = path->nodes[0]; | |
1248 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1249 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1250 | if (ret < 0) { |
1251 | btrfs_abort_transaction(trans, root, ret); | |
1252 | goto error; | |
1253 | } | |
80ff3856 YZ |
1254 | if (ret > 0) |
1255 | break; | |
1256 | leaf = path->nodes[0]; | |
1257 | } | |
be20aa9d | 1258 | |
80ff3856 YZ |
1259 | nocow = 0; |
1260 | disk_bytenr = 0; | |
17d217fe | 1261 | num_bytes = 0; |
80ff3856 YZ |
1262 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1263 | ||
33345d01 | 1264 | if (found_key.objectid > ino || |
80ff3856 YZ |
1265 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1266 | found_key.offset > end) | |
1267 | break; | |
1268 | ||
1269 | if (found_key.offset > cur_offset) { | |
1270 | extent_end = found_key.offset; | |
e9061e21 | 1271 | extent_type = 0; |
80ff3856 YZ |
1272 | goto out_check; |
1273 | } | |
1274 | ||
1275 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1276 | struct btrfs_file_extent_item); | |
1277 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1278 | ||
d899e052 YZ |
1279 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1280 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1281 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1282 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1283 | extent_end = found_key.offset + |
1284 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1285 | disk_num_bytes = |
1286 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1287 | if (extent_end <= start) { |
1288 | path->slots[0]++; | |
1289 | goto next_slot; | |
1290 | } | |
17d217fe YZ |
1291 | if (disk_bytenr == 0) |
1292 | goto out_check; | |
80ff3856 YZ |
1293 | if (btrfs_file_extent_compression(leaf, fi) || |
1294 | btrfs_file_extent_encryption(leaf, fi) || | |
1295 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1296 | goto out_check; | |
d899e052 YZ |
1297 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1298 | goto out_check; | |
d2fb3437 | 1299 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1300 | goto out_check; |
33345d01 | 1301 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1302 | found_key.offset - |
1303 | extent_offset, disk_bytenr)) | |
17d217fe | 1304 | goto out_check; |
5d4f98a2 | 1305 | disk_bytenr += extent_offset; |
17d217fe YZ |
1306 | disk_bytenr += cur_offset - found_key.offset; |
1307 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1308 | /* | |
1309 | * force cow if csum exists in the range. | |
1310 | * this ensure that csum for a given extent are | |
1311 | * either valid or do not exist. | |
1312 | */ | |
1313 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1314 | goto out_check; | |
80ff3856 YZ |
1315 | nocow = 1; |
1316 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1317 | extent_end = found_key.offset + | |
1318 | btrfs_file_extent_inline_len(leaf, fi); | |
1319 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1320 | } else { | |
1321 | BUG_ON(1); | |
1322 | } | |
1323 | out_check: | |
1324 | if (extent_end <= start) { | |
1325 | path->slots[0]++; | |
1326 | goto next_slot; | |
1327 | } | |
1328 | if (!nocow) { | |
1329 | if (cow_start == (u64)-1) | |
1330 | cow_start = cur_offset; | |
1331 | cur_offset = extent_end; | |
1332 | if (cur_offset > end) | |
1333 | break; | |
1334 | path->slots[0]++; | |
1335 | goto next_slot; | |
7ea394f1 YZ |
1336 | } |
1337 | ||
b3b4aa74 | 1338 | btrfs_release_path(path); |
80ff3856 | 1339 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1340 | ret = __cow_file_range(trans, inode, root, locked_page, |
1341 | cow_start, found_key.offset - 1, | |
1342 | page_started, nr_written, 1); | |
79787eaa JM |
1343 | if (ret) { |
1344 | btrfs_abort_transaction(trans, root, ret); | |
1345 | goto error; | |
1346 | } | |
80ff3856 | 1347 | cow_start = (u64)-1; |
7ea394f1 | 1348 | } |
80ff3856 | 1349 | |
d899e052 YZ |
1350 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1351 | struct extent_map *em; | |
1352 | struct extent_map_tree *em_tree; | |
1353 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1354 | em = alloc_extent_map(); |
79787eaa | 1355 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1356 | em->start = cur_offset; |
70c8a91c | 1357 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1358 | em->len = num_bytes; |
1359 | em->block_len = num_bytes; | |
1360 | em->block_start = disk_bytenr; | |
b4939680 | 1361 | em->orig_block_len = disk_num_bytes; |
d899e052 | 1362 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1363 | em->mod_start = em->start; |
1364 | em->mod_len = em->len; | |
d899e052 | 1365 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1366 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1367 | em->generation = -1; |
d899e052 | 1368 | while (1) { |
890871be | 1369 | write_lock(&em_tree->lock); |
d899e052 | 1370 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
1371 | if (!ret) |
1372 | list_move(&em->list, | |
1373 | &em_tree->modified_extents); | |
890871be | 1374 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1375 | if (ret != -EEXIST) { |
1376 | free_extent_map(em); | |
1377 | break; | |
1378 | } | |
1379 | btrfs_drop_extent_cache(inode, em->start, | |
1380 | em->start + em->len - 1, 0); | |
1381 | } | |
1382 | type = BTRFS_ORDERED_PREALLOC; | |
1383 | } else { | |
1384 | type = BTRFS_ORDERED_NOCOW; | |
1385 | } | |
80ff3856 YZ |
1386 | |
1387 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1388 | num_bytes, num_bytes, type); |
79787eaa | 1389 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1390 | |
efa56464 YZ |
1391 | if (root->root_key.objectid == |
1392 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1393 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1394 | num_bytes); | |
79787eaa JM |
1395 | if (ret) { |
1396 | btrfs_abort_transaction(trans, root, ret); | |
1397 | goto error; | |
1398 | } | |
efa56464 YZ |
1399 | } |
1400 | ||
d899e052 | 1401 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1402 | cur_offset, cur_offset + num_bytes - 1, |
1403 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1404 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1405 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1406 | cur_offset = extent_end; |
1407 | if (cur_offset > end) | |
1408 | break; | |
be20aa9d | 1409 | } |
b3b4aa74 | 1410 | btrfs_release_path(path); |
80ff3856 | 1411 | |
17ca04af | 1412 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1413 | cow_start = cur_offset; |
17ca04af JB |
1414 | cur_offset = end; |
1415 | } | |
1416 | ||
80ff3856 | 1417 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1418 | ret = __cow_file_range(trans, inode, root, locked_page, |
1419 | cow_start, end, | |
1420 | page_started, nr_written, 1); | |
79787eaa JM |
1421 | if (ret) { |
1422 | btrfs_abort_transaction(trans, root, ret); | |
1423 | goto error; | |
1424 | } | |
80ff3856 YZ |
1425 | } |
1426 | ||
79787eaa | 1427 | error: |
a698d075 | 1428 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1429 | if (!ret) |
1430 | ret = err; | |
1431 | ||
17ca04af JB |
1432 | if (ret && cur_offset < end) |
1433 | extent_clear_unlock_delalloc(inode, | |
1434 | &BTRFS_I(inode)->io_tree, | |
1435 | cur_offset, end, locked_page, | |
1436 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1437 | EXTENT_CLEAR_UNLOCK | | |
1438 | EXTENT_CLEAR_DELALLOC | | |
1439 | EXTENT_CLEAR_DIRTY | | |
1440 | EXTENT_SET_WRITEBACK | | |
1441 | EXTENT_END_WRITEBACK); | |
1442 | ||
7ea394f1 | 1443 | btrfs_free_path(path); |
79787eaa | 1444 | return ret; |
be20aa9d CM |
1445 | } |
1446 | ||
d352ac68 CM |
1447 | /* |
1448 | * extent_io.c call back to do delayed allocation processing | |
1449 | */ | |
c8b97818 | 1450 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1451 | u64 start, u64 end, int *page_started, |
1452 | unsigned long *nr_written) | |
be20aa9d | 1453 | { |
be20aa9d | 1454 | int ret; |
7f366cfe | 1455 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1456 | |
7ddf5a42 | 1457 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1458 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1459 | page_started, 1, nr_written); |
7ddf5a42 | 1460 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1461 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1462 | page_started, 0, nr_written); |
7ddf5a42 JB |
1463 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1464 | !(BTRFS_I(inode)->force_compress) && | |
1465 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1466 | ret = cow_file_range(inode, locked_page, start, end, |
1467 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1468 | } else { |
1469 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1470 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1471 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1472 | page_started, nr_written); |
7ddf5a42 | 1473 | } |
b888db2b CM |
1474 | return ret; |
1475 | } | |
1476 | ||
1bf85046 JM |
1477 | static void btrfs_split_extent_hook(struct inode *inode, |
1478 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1479 | { |
0ca1f7ce | 1480 | /* not delalloc, ignore it */ |
9ed74f2d | 1481 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1482 | return; |
9ed74f2d | 1483 | |
9e0baf60 JB |
1484 | spin_lock(&BTRFS_I(inode)->lock); |
1485 | BTRFS_I(inode)->outstanding_extents++; | |
1486 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1487 | } |
1488 | ||
1489 | /* | |
1490 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1491 | * extents so we can keep track of new extents that are just merged onto old | |
1492 | * extents, such as when we are doing sequential writes, so we can properly | |
1493 | * account for the metadata space we'll need. | |
1494 | */ | |
1bf85046 JM |
1495 | static void btrfs_merge_extent_hook(struct inode *inode, |
1496 | struct extent_state *new, | |
1497 | struct extent_state *other) | |
9ed74f2d | 1498 | { |
9ed74f2d JB |
1499 | /* not delalloc, ignore it */ |
1500 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1501 | return; |
9ed74f2d | 1502 | |
9e0baf60 JB |
1503 | spin_lock(&BTRFS_I(inode)->lock); |
1504 | BTRFS_I(inode)->outstanding_extents--; | |
1505 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1506 | } |
1507 | ||
d352ac68 CM |
1508 | /* |
1509 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1510 | * bytes in this file, and to maintain the list of inodes that | |
1511 | * have pending delalloc work to be done. | |
1512 | */ | |
1bf85046 JM |
1513 | static void btrfs_set_bit_hook(struct inode *inode, |
1514 | struct extent_state *state, int *bits) | |
291d673e | 1515 | { |
9ed74f2d | 1516 | |
75eff68e CM |
1517 | /* |
1518 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1519 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1520 | * bit, which is only set or cleared with irqs on |
1521 | */ | |
0ca1f7ce | 1522 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1523 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1524 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1525 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1526 | |
9e0baf60 | 1527 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1528 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1529 | } else { |
1530 | spin_lock(&BTRFS_I(inode)->lock); | |
1531 | BTRFS_I(inode)->outstanding_extents++; | |
1532 | spin_unlock(&BTRFS_I(inode)->lock); | |
1533 | } | |
287a0ab9 | 1534 | |
963d678b MX |
1535 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1536 | root->fs_info->delalloc_batch); | |
df0af1a5 MX |
1537 | spin_lock(&BTRFS_I(inode)->lock); |
1538 | BTRFS_I(inode)->delalloc_bytes += len; | |
1539 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1540 | &BTRFS_I(inode)->runtime_flags)) { | |
1541 | spin_lock(&root->fs_info->delalloc_lock); | |
1542 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1543 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1544 | &root->fs_info->delalloc_inodes); | |
1545 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1546 | &BTRFS_I(inode)->runtime_flags); | |
1547 | } | |
1548 | spin_unlock(&root->fs_info->delalloc_lock); | |
ea8c2819 | 1549 | } |
df0af1a5 | 1550 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1551 | } |
291d673e CM |
1552 | } |
1553 | ||
d352ac68 CM |
1554 | /* |
1555 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1556 | */ | |
1bf85046 JM |
1557 | static void btrfs_clear_bit_hook(struct inode *inode, |
1558 | struct extent_state *state, int *bits) | |
291d673e | 1559 | { |
75eff68e CM |
1560 | /* |
1561 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1562 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1563 | * bit, which is only set or cleared with irqs on |
1564 | */ | |
0ca1f7ce | 1565 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1566 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1567 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1568 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1569 | |
9e0baf60 | 1570 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1571 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1572 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1573 | spin_lock(&BTRFS_I(inode)->lock); | |
1574 | BTRFS_I(inode)->outstanding_extents--; | |
1575 | spin_unlock(&BTRFS_I(inode)->lock); | |
1576 | } | |
0ca1f7ce YZ |
1577 | |
1578 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1579 | btrfs_delalloc_release_metadata(inode, len); | |
1580 | ||
0cb59c99 JB |
1581 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1582 | && do_list) | |
0ca1f7ce | 1583 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1584 | |
963d678b MX |
1585 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1586 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1587 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1588 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1589 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 MX |
1590 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
1591 | &BTRFS_I(inode)->runtime_flags)) { | |
1592 | spin_lock(&root->fs_info->delalloc_lock); | |
1593 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1594 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1595 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1596 | &BTRFS_I(inode)->runtime_flags); | |
1597 | } | |
1598 | spin_unlock(&root->fs_info->delalloc_lock); | |
ea8c2819 | 1599 | } |
df0af1a5 | 1600 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1601 | } |
291d673e CM |
1602 | } |
1603 | ||
d352ac68 CM |
1604 | /* |
1605 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1606 | * we don't create bios that span stripes or chunks | |
1607 | */ | |
239b14b3 | 1608 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1609 | size_t size, struct bio *bio, |
1610 | unsigned long bio_flags) | |
239b14b3 CM |
1611 | { |
1612 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1613 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1614 | u64 length = 0; |
1615 | u64 map_length; | |
239b14b3 CM |
1616 | int ret; |
1617 | ||
771ed689 CM |
1618 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1619 | return 0; | |
1620 | ||
f2d8d74d | 1621 | length = bio->bi_size; |
239b14b3 | 1622 | map_length = length; |
3ec706c8 | 1623 | ret = btrfs_map_block(root->fs_info, READ, logical, |
f188591e | 1624 | &map_length, NULL, 0); |
3ec706c8 | 1625 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1626 | BUG_ON(ret < 0); |
d397712b | 1627 | if (map_length < length + size) |
239b14b3 | 1628 | return 1; |
3444a972 | 1629 | return 0; |
239b14b3 CM |
1630 | } |
1631 | ||
d352ac68 CM |
1632 | /* |
1633 | * in order to insert checksums into the metadata in large chunks, | |
1634 | * we wait until bio submission time. All the pages in the bio are | |
1635 | * checksummed and sums are attached onto the ordered extent record. | |
1636 | * | |
1637 | * At IO completion time the cums attached on the ordered extent record | |
1638 | * are inserted into the btree | |
1639 | */ | |
d397712b CM |
1640 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1641 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1642 | unsigned long bio_flags, |
1643 | u64 bio_offset) | |
065631f6 | 1644 | { |
065631f6 | 1645 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1646 | int ret = 0; |
e015640f | 1647 | |
d20f7043 | 1648 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1649 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1650 | return 0; |
1651 | } | |
e015640f | 1652 | |
4a69a410 CM |
1653 | /* |
1654 | * in order to insert checksums into the metadata in large chunks, | |
1655 | * we wait until bio submission time. All the pages in the bio are | |
1656 | * checksummed and sums are attached onto the ordered extent record. | |
1657 | * | |
1658 | * At IO completion time the cums attached on the ordered extent record | |
1659 | * are inserted into the btree | |
1660 | */ | |
b2950863 | 1661 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1662 | int mirror_num, unsigned long bio_flags, |
1663 | u64 bio_offset) | |
4a69a410 CM |
1664 | { |
1665 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1666 | int ret; |
1667 | ||
1668 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1669 | if (ret) | |
1670 | bio_endio(bio, ret); | |
1671 | return ret; | |
44b8bd7e CM |
1672 | } |
1673 | ||
d352ac68 | 1674 | /* |
cad321ad CM |
1675 | * extent_io.c submission hook. This does the right thing for csum calculation |
1676 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1677 | */ |
b2950863 | 1678 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1679 | int mirror_num, unsigned long bio_flags, |
1680 | u64 bio_offset) | |
44b8bd7e CM |
1681 | { |
1682 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1683 | int ret = 0; | |
19b9bdb0 | 1684 | int skip_sum; |
0417341e | 1685 | int metadata = 0; |
b812ce28 | 1686 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1687 | |
6cbff00f | 1688 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1689 | |
83eea1f1 | 1690 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1691 | metadata = 2; |
1692 | ||
7b6d91da | 1693 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1694 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1695 | if (ret) | |
61891923 | 1696 | goto out; |
5fd02043 | 1697 | |
d20f7043 | 1698 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1699 | ret = btrfs_submit_compressed_read(inode, bio, |
1700 | mirror_num, | |
1701 | bio_flags); | |
1702 | goto out; | |
c2db1073 TI |
1703 | } else if (!skip_sum) { |
1704 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1705 | if (ret) | |
61891923 | 1706 | goto out; |
c2db1073 | 1707 | } |
4d1b5fb4 | 1708 | goto mapit; |
b812ce28 | 1709 | } else if (async && !skip_sum) { |
17d217fe YZ |
1710 | /* csum items have already been cloned */ |
1711 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1712 | goto mapit; | |
19b9bdb0 | 1713 | /* we're doing a write, do the async checksumming */ |
61891923 | 1714 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1715 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1716 | bio_flags, bio_offset, |
1717 | __btrfs_submit_bio_start, | |
4a69a410 | 1718 | __btrfs_submit_bio_done); |
61891923 | 1719 | goto out; |
b812ce28 JB |
1720 | } else if (!skip_sum) { |
1721 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1722 | if (ret) | |
1723 | goto out; | |
19b9bdb0 CM |
1724 | } |
1725 | ||
0b86a832 | 1726 | mapit: |
61891923 SB |
1727 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1728 | ||
1729 | out: | |
1730 | if (ret < 0) | |
1731 | bio_endio(bio, ret); | |
1732 | return ret; | |
065631f6 | 1733 | } |
6885f308 | 1734 | |
d352ac68 CM |
1735 | /* |
1736 | * given a list of ordered sums record them in the inode. This happens | |
1737 | * at IO completion time based on sums calculated at bio submission time. | |
1738 | */ | |
ba1da2f4 | 1739 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1740 | struct inode *inode, u64 file_offset, |
1741 | struct list_head *list) | |
1742 | { | |
e6dcd2dc CM |
1743 | struct btrfs_ordered_sum *sum; |
1744 | ||
c6e30871 | 1745 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1746 | btrfs_csum_file_blocks(trans, |
1747 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1748 | } |
1749 | return 0; | |
1750 | } | |
1751 | ||
2ac55d41 JB |
1752 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1753 | struct extent_state **cached_state) | |
ea8c2819 | 1754 | { |
6c1500f2 | 1755 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1756 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1757 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1758 | } |
1759 | ||
d352ac68 | 1760 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1761 | struct btrfs_writepage_fixup { |
1762 | struct page *page; | |
1763 | struct btrfs_work work; | |
1764 | }; | |
1765 | ||
b2950863 | 1766 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1767 | { |
1768 | struct btrfs_writepage_fixup *fixup; | |
1769 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1770 | struct extent_state *cached_state = NULL; |
247e743c CM |
1771 | struct page *page; |
1772 | struct inode *inode; | |
1773 | u64 page_start; | |
1774 | u64 page_end; | |
87826df0 | 1775 | int ret; |
247e743c CM |
1776 | |
1777 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1778 | page = fixup->page; | |
4a096752 | 1779 | again: |
247e743c CM |
1780 | lock_page(page); |
1781 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1782 | ClearPageChecked(page); | |
1783 | goto out_page; | |
1784 | } | |
1785 | ||
1786 | inode = page->mapping->host; | |
1787 | page_start = page_offset(page); | |
1788 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1789 | ||
2ac55d41 | 1790 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1791 | &cached_state); |
4a096752 CM |
1792 | |
1793 | /* already ordered? We're done */ | |
8b62b72b | 1794 | if (PagePrivate2(page)) |
247e743c | 1795 | goto out; |
4a096752 CM |
1796 | |
1797 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1798 | if (ordered) { | |
2ac55d41 JB |
1799 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1800 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1801 | unlock_page(page); |
1802 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1803 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1804 | goto again; |
1805 | } | |
247e743c | 1806 | |
87826df0 JM |
1807 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1808 | if (ret) { | |
1809 | mapping_set_error(page->mapping, ret); | |
1810 | end_extent_writepage(page, ret, page_start, page_end); | |
1811 | ClearPageChecked(page); | |
1812 | goto out; | |
1813 | } | |
1814 | ||
2ac55d41 | 1815 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1816 | ClearPageChecked(page); |
87826df0 | 1817 | set_page_dirty(page); |
247e743c | 1818 | out: |
2ac55d41 JB |
1819 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1820 | &cached_state, GFP_NOFS); | |
247e743c CM |
1821 | out_page: |
1822 | unlock_page(page); | |
1823 | page_cache_release(page); | |
b897abec | 1824 | kfree(fixup); |
247e743c CM |
1825 | } |
1826 | ||
1827 | /* | |
1828 | * There are a few paths in the higher layers of the kernel that directly | |
1829 | * set the page dirty bit without asking the filesystem if it is a | |
1830 | * good idea. This causes problems because we want to make sure COW | |
1831 | * properly happens and the data=ordered rules are followed. | |
1832 | * | |
c8b97818 | 1833 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1834 | * hasn't been properly setup for IO. We kick off an async process |
1835 | * to fix it up. The async helper will wait for ordered extents, set | |
1836 | * the delalloc bit and make it safe to write the page. | |
1837 | */ | |
b2950863 | 1838 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1839 | { |
1840 | struct inode *inode = page->mapping->host; | |
1841 | struct btrfs_writepage_fixup *fixup; | |
1842 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1843 | |
8b62b72b CM |
1844 | /* this page is properly in the ordered list */ |
1845 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1846 | return 0; |
1847 | ||
1848 | if (PageChecked(page)) | |
1849 | return -EAGAIN; | |
1850 | ||
1851 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1852 | if (!fixup) | |
1853 | return -EAGAIN; | |
f421950f | 1854 | |
247e743c CM |
1855 | SetPageChecked(page); |
1856 | page_cache_get(page); | |
1857 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1858 | fixup->page = page; | |
1859 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1860 | return -EBUSY; |
247e743c CM |
1861 | } |
1862 | ||
d899e052 YZ |
1863 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1864 | struct inode *inode, u64 file_pos, | |
1865 | u64 disk_bytenr, u64 disk_num_bytes, | |
1866 | u64 num_bytes, u64 ram_bytes, | |
1867 | u8 compression, u8 encryption, | |
1868 | u16 other_encoding, int extent_type) | |
1869 | { | |
1870 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1871 | struct btrfs_file_extent_item *fi; | |
1872 | struct btrfs_path *path; | |
1873 | struct extent_buffer *leaf; | |
1874 | struct btrfs_key ins; | |
d899e052 YZ |
1875 | int ret; |
1876 | ||
1877 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1878 | if (!path) |
1879 | return -ENOMEM; | |
d899e052 | 1880 | |
b9473439 | 1881 | path->leave_spinning = 1; |
a1ed835e CM |
1882 | |
1883 | /* | |
1884 | * we may be replacing one extent in the tree with another. | |
1885 | * The new extent is pinned in the extent map, and we don't want | |
1886 | * to drop it from the cache until it is completely in the btree. | |
1887 | * | |
1888 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1889 | * the caller is expected to unpin it and allow it to be merged | |
1890 | * with the others. | |
1891 | */ | |
5dc562c5 | 1892 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1893 | file_pos + num_bytes, 0); |
79787eaa JM |
1894 | if (ret) |
1895 | goto out; | |
d899e052 | 1896 | |
33345d01 | 1897 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1898 | ins.offset = file_pos; |
1899 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1900 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1901 | if (ret) |
1902 | goto out; | |
d899e052 YZ |
1903 | leaf = path->nodes[0]; |
1904 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1905 | struct btrfs_file_extent_item); | |
1906 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1907 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1908 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1909 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1910 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1911 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1912 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1913 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1914 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1915 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1916 | |
d899e052 | 1917 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1918 | btrfs_release_path(path); |
d899e052 YZ |
1919 | |
1920 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1921 | |
1922 | ins.objectid = disk_bytenr; | |
1923 | ins.offset = disk_num_bytes; | |
1924 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1925 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1926 | root->root_key.objectid, | |
33345d01 | 1927 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1928 | out: |
d899e052 | 1929 | btrfs_free_path(path); |
b9473439 | 1930 | |
79787eaa | 1931 | return ret; |
d899e052 YZ |
1932 | } |
1933 | ||
5d13a98f CM |
1934 | /* |
1935 | * helper function for btrfs_finish_ordered_io, this | |
1936 | * just reads in some of the csum leaves to prime them into ram | |
1937 | * before we start the transaction. It limits the amount of btree | |
1938 | * reads required while inside the transaction. | |
1939 | */ | |
d352ac68 CM |
1940 | /* as ordered data IO finishes, this gets called so we can finish |
1941 | * an ordered extent if the range of bytes in the file it covers are | |
1942 | * fully written. | |
1943 | */ | |
5fd02043 | 1944 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1945 | { |
5fd02043 | 1946 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1947 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1948 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1949 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1950 | struct extent_state *cached_state = NULL; |
261507a0 | 1951 | int compress_type = 0; |
e6dcd2dc | 1952 | int ret; |
82d5902d | 1953 | bool nolock; |
e6dcd2dc | 1954 | |
83eea1f1 | 1955 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 1956 | |
5fd02043 JB |
1957 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1958 | ret = -EIO; | |
1959 | goto out; | |
1960 | } | |
1961 | ||
c2167754 | 1962 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1963 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
1964 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1965 | if (nolock) | |
1966 | trans = btrfs_join_transaction_nolock(root); | |
1967 | else | |
1968 | trans = btrfs_join_transaction(root); | |
1969 | if (IS_ERR(trans)) { | |
1970 | ret = PTR_ERR(trans); | |
1971 | trans = NULL; | |
1972 | goto out; | |
c2167754 | 1973 | } |
6c760c07 JB |
1974 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
1975 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
1976 | if (ret) /* -ENOMEM or corruption */ | |
1977 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1978 | goto out; |
1979 | } | |
e6dcd2dc | 1980 | |
2ac55d41 JB |
1981 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1982 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1983 | 0, &cached_state); |
e6dcd2dc | 1984 | |
0cb59c99 | 1985 | if (nolock) |
7a7eaa40 | 1986 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1987 | else |
7a7eaa40 | 1988 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1989 | if (IS_ERR(trans)) { |
1990 | ret = PTR_ERR(trans); | |
1991 | trans = NULL; | |
1992 | goto out_unlock; | |
1993 | } | |
0ca1f7ce | 1994 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1995 | |
c8b97818 | 1996 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1997 | compress_type = ordered_extent->compress_type; |
d899e052 | 1998 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1999 | BUG_ON(compress_type); |
920bbbfb | 2000 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2001 | ordered_extent->file_offset, |
2002 | ordered_extent->file_offset + | |
2003 | ordered_extent->len); | |
d899e052 | 2004 | } else { |
0af3d00b | 2005 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2006 | ret = insert_reserved_file_extent(trans, inode, |
2007 | ordered_extent->file_offset, | |
2008 | ordered_extent->start, | |
2009 | ordered_extent->disk_len, | |
2010 | ordered_extent->len, | |
2011 | ordered_extent->len, | |
261507a0 | 2012 | compress_type, 0, 0, |
d899e052 | 2013 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 2014 | } |
5dc562c5 JB |
2015 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2016 | ordered_extent->file_offset, ordered_extent->len, | |
2017 | trans->transid); | |
79787eaa JM |
2018 | if (ret < 0) { |
2019 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2020 | goto out_unlock; |
79787eaa | 2021 | } |
2ac55d41 | 2022 | |
e6dcd2dc CM |
2023 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2024 | &ordered_extent->list); | |
2025 | ||
6c760c07 JB |
2026 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2027 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2028 | if (ret) { /* -ENOMEM or corruption */ | |
2029 | btrfs_abort_transaction(trans, root, ret); | |
2030 | goto out_unlock; | |
1ef30be1 JB |
2031 | } |
2032 | ret = 0; | |
5fd02043 JB |
2033 | out_unlock: |
2034 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2035 | ordered_extent->file_offset + | |
2036 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2037 | out: |
5b0e95bf | 2038 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2039 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2040 | if (trans) |
2041 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2042 | |
0bec9ef5 | 2043 | if (ret) { |
5fd02043 JB |
2044 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, |
2045 | ordered_extent->file_offset + | |
2046 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
2047 | ||
0bec9ef5 JB |
2048 | /* |
2049 | * If the ordered extent had an IOERR or something else went | |
2050 | * wrong we need to return the space for this ordered extent | |
2051 | * back to the allocator. | |
2052 | */ | |
2053 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2054 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) | |
2055 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2056 | ordered_extent->disk_len); | |
2057 | } | |
2058 | ||
2059 | ||
5fd02043 | 2060 | /* |
8bad3c02 LB |
2061 | * This needs to be done to make sure anybody waiting knows we are done |
2062 | * updating everything for this ordered extent. | |
5fd02043 JB |
2063 | */ |
2064 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2065 | ||
e6dcd2dc CM |
2066 | /* once for us */ |
2067 | btrfs_put_ordered_extent(ordered_extent); | |
2068 | /* once for the tree */ | |
2069 | btrfs_put_ordered_extent(ordered_extent); | |
2070 | ||
5fd02043 JB |
2071 | return ret; |
2072 | } | |
2073 | ||
2074 | static void finish_ordered_fn(struct btrfs_work *work) | |
2075 | { | |
2076 | struct btrfs_ordered_extent *ordered_extent; | |
2077 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2078 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2079 | } |
2080 | ||
b2950863 | 2081 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2082 | struct extent_state *state, int uptodate) |
2083 | { | |
5fd02043 JB |
2084 | struct inode *inode = page->mapping->host; |
2085 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2086 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2087 | struct btrfs_workers *workers; | |
2088 | ||
1abe9b8a | 2089 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2090 | ||
8b62b72b | 2091 | ClearPagePrivate2(page); |
5fd02043 JB |
2092 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2093 | end - start + 1, uptodate)) | |
2094 | return 0; | |
2095 | ||
2096 | ordered_extent->work.func = finish_ordered_fn; | |
2097 | ordered_extent->work.flags = 0; | |
2098 | ||
83eea1f1 | 2099 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2100 | workers = &root->fs_info->endio_freespace_worker; |
2101 | else | |
2102 | workers = &root->fs_info->endio_write_workers; | |
2103 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2104 | ||
2105 | return 0; | |
211f90e6 CM |
2106 | } |
2107 | ||
d352ac68 CM |
2108 | /* |
2109 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2110 | * if there's a match, we allow the bio to finish. If not, the code in |
2111 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2112 | */ |
b2950863 | 2113 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2114 | struct extent_state *state, int mirror) |
07157aac | 2115 | { |
4eee4fa4 | 2116 | size_t offset = start - page_offset(page); |
07157aac | 2117 | struct inode *inode = page->mapping->host; |
d1310b2e | 2118 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2119 | char *kaddr; |
aadfeb6e | 2120 | u64 private = ~(u32)0; |
07157aac | 2121 | int ret; |
ff79f819 CM |
2122 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2123 | u32 csum = ~(u32)0; | |
d1310b2e | 2124 | |
d20f7043 CM |
2125 | if (PageChecked(page)) { |
2126 | ClearPageChecked(page); | |
2127 | goto good; | |
2128 | } | |
6cbff00f CH |
2129 | |
2130 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2131 | goto good; |
17d217fe YZ |
2132 | |
2133 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2134 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2135 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2136 | GFP_NOFS); | |
b6cda9bc | 2137 | return 0; |
17d217fe | 2138 | } |
d20f7043 | 2139 | |
c2e639f0 | 2140 | if (state && state->start == start) { |
70dec807 CM |
2141 | private = state->private; |
2142 | ret = 0; | |
2143 | } else { | |
2144 | ret = get_state_private(io_tree, start, &private); | |
2145 | } | |
7ac687d9 | 2146 | kaddr = kmap_atomic(page); |
d397712b | 2147 | if (ret) |
07157aac | 2148 | goto zeroit; |
d397712b | 2149 | |
ff79f819 CM |
2150 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2151 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2152 | if (csum != private) |
07157aac | 2153 | goto zeroit; |
d397712b | 2154 | |
7ac687d9 | 2155 | kunmap_atomic(kaddr); |
d20f7043 | 2156 | good: |
07157aac CM |
2157 | return 0; |
2158 | ||
2159 | zeroit: | |
945d8962 | 2160 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2161 | "private %llu\n", |
2162 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2163 | (unsigned long long)start, csum, |
2164 | (unsigned long long)private); | |
db94535d CM |
2165 | memset(kaddr + offset, 1, end - start + 1); |
2166 | flush_dcache_page(page); | |
7ac687d9 | 2167 | kunmap_atomic(kaddr); |
3b951516 CM |
2168 | if (private == 0) |
2169 | return 0; | |
7e38326f | 2170 | return -EIO; |
07157aac | 2171 | } |
b888db2b | 2172 | |
24bbcf04 YZ |
2173 | struct delayed_iput { |
2174 | struct list_head list; | |
2175 | struct inode *inode; | |
2176 | }; | |
2177 | ||
79787eaa JM |
2178 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2179 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2180 | void btrfs_add_delayed_iput(struct inode *inode) |
2181 | { | |
2182 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2183 | struct delayed_iput *delayed; | |
2184 | ||
2185 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2186 | return; | |
2187 | ||
2188 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2189 | delayed->inode = inode; | |
2190 | ||
2191 | spin_lock(&fs_info->delayed_iput_lock); | |
2192 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2193 | spin_unlock(&fs_info->delayed_iput_lock); | |
2194 | } | |
2195 | ||
2196 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2197 | { | |
2198 | LIST_HEAD(list); | |
2199 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2200 | struct delayed_iput *delayed; | |
2201 | int empty; | |
2202 | ||
2203 | spin_lock(&fs_info->delayed_iput_lock); | |
2204 | empty = list_empty(&fs_info->delayed_iputs); | |
2205 | spin_unlock(&fs_info->delayed_iput_lock); | |
2206 | if (empty) | |
2207 | return; | |
2208 | ||
24bbcf04 YZ |
2209 | spin_lock(&fs_info->delayed_iput_lock); |
2210 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2211 | spin_unlock(&fs_info->delayed_iput_lock); | |
2212 | ||
2213 | while (!list_empty(&list)) { | |
2214 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2215 | list_del(&delayed->list); | |
2216 | iput(delayed->inode); | |
2217 | kfree(delayed); | |
2218 | } | |
24bbcf04 YZ |
2219 | } |
2220 | ||
d68fc57b | 2221 | /* |
42b2aa86 | 2222 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2223 | * files in the subvolume, it removes orphan item and frees block_rsv |
2224 | * structure. | |
2225 | */ | |
2226 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2227 | struct btrfs_root *root) | |
2228 | { | |
90290e19 | 2229 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2230 | int ret; |
2231 | ||
8a35d95f | 2232 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2233 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2234 | return; | |
2235 | ||
90290e19 | 2236 | spin_lock(&root->orphan_lock); |
8a35d95f | 2237 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2238 | spin_unlock(&root->orphan_lock); |
2239 | return; | |
2240 | } | |
2241 | ||
2242 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2243 | spin_unlock(&root->orphan_lock); | |
2244 | return; | |
2245 | } | |
2246 | ||
2247 | block_rsv = root->orphan_block_rsv; | |
2248 | root->orphan_block_rsv = NULL; | |
2249 | spin_unlock(&root->orphan_lock); | |
2250 | ||
d68fc57b YZ |
2251 | if (root->orphan_item_inserted && |
2252 | btrfs_root_refs(&root->root_item) > 0) { | |
2253 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2254 | root->root_key.objectid); | |
2255 | BUG_ON(ret); | |
2256 | root->orphan_item_inserted = 0; | |
2257 | } | |
2258 | ||
90290e19 JB |
2259 | if (block_rsv) { |
2260 | WARN_ON(block_rsv->size > 0); | |
2261 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2262 | } |
2263 | } | |
2264 | ||
7b128766 JB |
2265 | /* |
2266 | * This creates an orphan entry for the given inode in case something goes | |
2267 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2268 | * |
2269 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2270 | * this function. | |
7b128766 JB |
2271 | */ |
2272 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2273 | { | |
2274 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2275 | struct btrfs_block_rsv *block_rsv = NULL; |
2276 | int reserve = 0; | |
2277 | int insert = 0; | |
2278 | int ret; | |
7b128766 | 2279 | |
d68fc57b | 2280 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2281 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2282 | if (!block_rsv) |
2283 | return -ENOMEM; | |
d68fc57b | 2284 | } |
7b128766 | 2285 | |
d68fc57b YZ |
2286 | spin_lock(&root->orphan_lock); |
2287 | if (!root->orphan_block_rsv) { | |
2288 | root->orphan_block_rsv = block_rsv; | |
2289 | } else if (block_rsv) { | |
2290 | btrfs_free_block_rsv(root, block_rsv); | |
2291 | block_rsv = NULL; | |
7b128766 | 2292 | } |
7b128766 | 2293 | |
8a35d95f JB |
2294 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2295 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2296 | #if 0 |
2297 | /* | |
2298 | * For proper ENOSPC handling, we should do orphan | |
2299 | * cleanup when mounting. But this introduces backward | |
2300 | * compatibility issue. | |
2301 | */ | |
2302 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2303 | insert = 2; | |
2304 | else | |
2305 | insert = 1; | |
2306 | #endif | |
2307 | insert = 1; | |
321f0e70 | 2308 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2309 | } |
2310 | ||
72ac3c0d JB |
2311 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2312 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2313 | reserve = 1; |
d68fc57b | 2314 | spin_unlock(&root->orphan_lock); |
7b128766 | 2315 | |
d68fc57b YZ |
2316 | /* grab metadata reservation from transaction handle */ |
2317 | if (reserve) { | |
2318 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2319 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2320 | } |
7b128766 | 2321 | |
d68fc57b YZ |
2322 | /* insert an orphan item to track this unlinked/truncated file */ |
2323 | if (insert >= 1) { | |
33345d01 | 2324 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2325 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2326 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2327 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2328 | btrfs_abort_transaction(trans, root, ret); |
2329 | return ret; | |
2330 | } | |
2331 | ret = 0; | |
d68fc57b YZ |
2332 | } |
2333 | ||
2334 | /* insert an orphan item to track subvolume contains orphan files */ | |
2335 | if (insert >= 2) { | |
2336 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2337 | root->root_key.objectid); | |
79787eaa JM |
2338 | if (ret && ret != -EEXIST) { |
2339 | btrfs_abort_transaction(trans, root, ret); | |
2340 | return ret; | |
2341 | } | |
d68fc57b YZ |
2342 | } |
2343 | return 0; | |
7b128766 JB |
2344 | } |
2345 | ||
2346 | /* | |
2347 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2348 | * item for this particular inode. | |
2349 | */ | |
2350 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2351 | { | |
2352 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2353 | int delete_item = 0; |
2354 | int release_rsv = 0; | |
7b128766 JB |
2355 | int ret = 0; |
2356 | ||
d68fc57b | 2357 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2358 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2359 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2360 | delete_item = 1; |
7b128766 | 2361 | |
72ac3c0d JB |
2362 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2363 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2364 | release_rsv = 1; |
d68fc57b | 2365 | spin_unlock(&root->orphan_lock); |
7b128766 | 2366 | |
d68fc57b | 2367 | if (trans && delete_item) { |
33345d01 | 2368 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2369 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2370 | } |
7b128766 | 2371 | |
8a35d95f | 2372 | if (release_rsv) { |
d68fc57b | 2373 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2374 | atomic_dec(&root->orphan_inodes); |
2375 | } | |
7b128766 | 2376 | |
d68fc57b | 2377 | return 0; |
7b128766 JB |
2378 | } |
2379 | ||
2380 | /* | |
2381 | * this cleans up any orphans that may be left on the list from the last use | |
2382 | * of this root. | |
2383 | */ | |
66b4ffd1 | 2384 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2385 | { |
2386 | struct btrfs_path *path; | |
2387 | struct extent_buffer *leaf; | |
7b128766 JB |
2388 | struct btrfs_key key, found_key; |
2389 | struct btrfs_trans_handle *trans; | |
2390 | struct inode *inode; | |
8f6d7f4f | 2391 | u64 last_objectid = 0; |
7b128766 JB |
2392 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2393 | ||
d68fc57b | 2394 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2395 | return 0; |
c71bf099 YZ |
2396 | |
2397 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2398 | if (!path) { |
2399 | ret = -ENOMEM; | |
2400 | goto out; | |
2401 | } | |
7b128766 JB |
2402 | path->reada = -1; |
2403 | ||
2404 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2405 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2406 | key.offset = (u64)-1; | |
2407 | ||
7b128766 JB |
2408 | while (1) { |
2409 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2410 | if (ret < 0) |
2411 | goto out; | |
7b128766 JB |
2412 | |
2413 | /* | |
2414 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2415 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2416 | * find the key and see if we have stuff that matches |
2417 | */ | |
2418 | if (ret > 0) { | |
66b4ffd1 | 2419 | ret = 0; |
7b128766 JB |
2420 | if (path->slots[0] == 0) |
2421 | break; | |
2422 | path->slots[0]--; | |
2423 | } | |
2424 | ||
2425 | /* pull out the item */ | |
2426 | leaf = path->nodes[0]; | |
7b128766 JB |
2427 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2428 | ||
2429 | /* make sure the item matches what we want */ | |
2430 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2431 | break; | |
2432 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2433 | break; | |
2434 | ||
2435 | /* release the path since we're done with it */ | |
b3b4aa74 | 2436 | btrfs_release_path(path); |
7b128766 JB |
2437 | |
2438 | /* | |
2439 | * this is where we are basically btrfs_lookup, without the | |
2440 | * crossing root thing. we store the inode number in the | |
2441 | * offset of the orphan item. | |
2442 | */ | |
8f6d7f4f JB |
2443 | |
2444 | if (found_key.offset == last_objectid) { | |
2445 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2446 | "stopping orphan cleanup\n"); | |
2447 | ret = -EINVAL; | |
2448 | goto out; | |
2449 | } | |
2450 | ||
2451 | last_objectid = found_key.offset; | |
2452 | ||
5d4f98a2 YZ |
2453 | found_key.objectid = found_key.offset; |
2454 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2455 | found_key.offset = 0; | |
73f73415 | 2456 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2457 | ret = PTR_RET(inode); |
2458 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2459 | goto out; |
7b128766 | 2460 | |
f8e9e0b0 AJ |
2461 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2462 | struct btrfs_root *dead_root; | |
2463 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2464 | int is_dead_root = 0; | |
2465 | ||
2466 | /* | |
2467 | * this is an orphan in the tree root. Currently these | |
2468 | * could come from 2 sources: | |
2469 | * a) a snapshot deletion in progress | |
2470 | * b) a free space cache inode | |
2471 | * We need to distinguish those two, as the snapshot | |
2472 | * orphan must not get deleted. | |
2473 | * find_dead_roots already ran before us, so if this | |
2474 | * is a snapshot deletion, we should find the root | |
2475 | * in the dead_roots list | |
2476 | */ | |
2477 | spin_lock(&fs_info->trans_lock); | |
2478 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2479 | root_list) { | |
2480 | if (dead_root->root_key.objectid == | |
2481 | found_key.objectid) { | |
2482 | is_dead_root = 1; | |
2483 | break; | |
2484 | } | |
2485 | } | |
2486 | spin_unlock(&fs_info->trans_lock); | |
2487 | if (is_dead_root) { | |
2488 | /* prevent this orphan from being found again */ | |
2489 | key.offset = found_key.objectid - 1; | |
2490 | continue; | |
2491 | } | |
2492 | } | |
7b128766 | 2493 | /* |
a8c9e576 JB |
2494 | * Inode is already gone but the orphan item is still there, |
2495 | * kill the orphan item. | |
7b128766 | 2496 | */ |
a8c9e576 JB |
2497 | if (ret == -ESTALE) { |
2498 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2499 | if (IS_ERR(trans)) { |
2500 | ret = PTR_ERR(trans); | |
2501 | goto out; | |
2502 | } | |
8a35d95f JB |
2503 | printk(KERN_ERR "auto deleting %Lu\n", |
2504 | found_key.objectid); | |
a8c9e576 JB |
2505 | ret = btrfs_del_orphan_item(trans, root, |
2506 | found_key.objectid); | |
79787eaa | 2507 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2508 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2509 | continue; |
2510 | } | |
2511 | ||
a8c9e576 JB |
2512 | /* |
2513 | * add this inode to the orphan list so btrfs_orphan_del does | |
2514 | * the proper thing when we hit it | |
2515 | */ | |
8a35d95f JB |
2516 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2517 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 2518 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 2519 | |
7b128766 JB |
2520 | /* if we have links, this was a truncate, lets do that */ |
2521 | if (inode->i_nlink) { | |
a41ad394 JB |
2522 | if (!S_ISREG(inode->i_mode)) { |
2523 | WARN_ON(1); | |
2524 | iput(inode); | |
2525 | continue; | |
2526 | } | |
7b128766 | 2527 | nr_truncate++; |
f3fe820c JB |
2528 | |
2529 | /* 1 for the orphan item deletion. */ | |
2530 | trans = btrfs_start_transaction(root, 1); | |
2531 | if (IS_ERR(trans)) { | |
2532 | ret = PTR_ERR(trans); | |
2533 | goto out; | |
2534 | } | |
2535 | ret = btrfs_orphan_add(trans, inode); | |
2536 | btrfs_end_transaction(trans, root); | |
2537 | if (ret) | |
2538 | goto out; | |
2539 | ||
66b4ffd1 | 2540 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
2541 | if (ret) |
2542 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
2543 | } else { |
2544 | nr_unlink++; | |
2545 | } | |
2546 | ||
2547 | /* this will do delete_inode and everything for us */ | |
2548 | iput(inode); | |
66b4ffd1 JB |
2549 | if (ret) |
2550 | goto out; | |
7b128766 | 2551 | } |
3254c876 MX |
2552 | /* release the path since we're done with it */ |
2553 | btrfs_release_path(path); | |
2554 | ||
d68fc57b YZ |
2555 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2556 | ||
2557 | if (root->orphan_block_rsv) | |
2558 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2559 | (u64)-1); | |
2560 | ||
2561 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2562 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2563 | if (!IS_ERR(trans)) |
2564 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2565 | } |
7b128766 JB |
2566 | |
2567 | if (nr_unlink) | |
2568 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2569 | if (nr_truncate) | |
2570 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2571 | |
2572 | out: | |
2573 | if (ret) | |
2574 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2575 | btrfs_free_path(path); | |
2576 | return ret; | |
7b128766 JB |
2577 | } |
2578 | ||
46a53cca CM |
2579 | /* |
2580 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2581 | * don't find any xattrs, we know there can't be any acls. | |
2582 | * | |
2583 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2584 | */ | |
2585 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2586 | int slot, u64 objectid) | |
2587 | { | |
2588 | u32 nritems = btrfs_header_nritems(leaf); | |
2589 | struct btrfs_key found_key; | |
2590 | int scanned = 0; | |
2591 | ||
2592 | slot++; | |
2593 | while (slot < nritems) { | |
2594 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2595 | ||
2596 | /* we found a different objectid, there must not be acls */ | |
2597 | if (found_key.objectid != objectid) | |
2598 | return 0; | |
2599 | ||
2600 | /* we found an xattr, assume we've got an acl */ | |
2601 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2602 | return 1; | |
2603 | ||
2604 | /* | |
2605 | * we found a key greater than an xattr key, there can't | |
2606 | * be any acls later on | |
2607 | */ | |
2608 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2609 | return 0; | |
2610 | ||
2611 | slot++; | |
2612 | scanned++; | |
2613 | ||
2614 | /* | |
2615 | * it goes inode, inode backrefs, xattrs, extents, | |
2616 | * so if there are a ton of hard links to an inode there can | |
2617 | * be a lot of backrefs. Don't waste time searching too hard, | |
2618 | * this is just an optimization | |
2619 | */ | |
2620 | if (scanned >= 8) | |
2621 | break; | |
2622 | } | |
2623 | /* we hit the end of the leaf before we found an xattr or | |
2624 | * something larger than an xattr. We have to assume the inode | |
2625 | * has acls | |
2626 | */ | |
2627 | return 1; | |
2628 | } | |
2629 | ||
d352ac68 CM |
2630 | /* |
2631 | * read an inode from the btree into the in-memory inode | |
2632 | */ | |
5d4f98a2 | 2633 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2634 | { |
2635 | struct btrfs_path *path; | |
5f39d397 | 2636 | struct extent_buffer *leaf; |
39279cc3 | 2637 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2638 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2639 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2640 | struct btrfs_key location; | |
46a53cca | 2641 | int maybe_acls; |
618e21d5 | 2642 | u32 rdev; |
39279cc3 | 2643 | int ret; |
2f7e33d4 MX |
2644 | bool filled = false; |
2645 | ||
2646 | ret = btrfs_fill_inode(inode, &rdev); | |
2647 | if (!ret) | |
2648 | filled = true; | |
39279cc3 CM |
2649 | |
2650 | path = btrfs_alloc_path(); | |
1748f843 MF |
2651 | if (!path) |
2652 | goto make_bad; | |
2653 | ||
d90c7321 | 2654 | path->leave_spinning = 1; |
39279cc3 | 2655 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2656 | |
39279cc3 | 2657 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2658 | if (ret) |
39279cc3 | 2659 | goto make_bad; |
39279cc3 | 2660 | |
5f39d397 | 2661 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2662 | |
2663 | if (filled) | |
2664 | goto cache_acl; | |
2665 | ||
5f39d397 CM |
2666 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2667 | struct btrfs_inode_item); | |
5f39d397 | 2668 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2669 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
2670 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
2671 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 2672 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2673 | |
2674 | tspec = btrfs_inode_atime(inode_item); | |
2675 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2676 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2677 | ||
2678 | tspec = btrfs_inode_mtime(inode_item); | |
2679 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2680 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2681 | ||
2682 | tspec = btrfs_inode_ctime(inode_item); | |
2683 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2684 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2685 | ||
a76a3cd4 | 2686 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2687 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
2688 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
2689 | ||
2690 | /* | |
2691 | * If we were modified in the current generation and evicted from memory | |
2692 | * and then re-read we need to do a full sync since we don't have any | |
2693 | * idea about which extents were modified before we were evicted from | |
2694 | * cache. | |
2695 | */ | |
2696 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
2697 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2698 | &BTRFS_I(inode)->runtime_flags); | |
2699 | ||
0c4d2d95 | 2700 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2701 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2702 | inode->i_rdev = 0; |
5f39d397 CM |
2703 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2704 | ||
aec7477b | 2705 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2706 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2707 | cache_acl: |
46a53cca CM |
2708 | /* |
2709 | * try to precache a NULL acl entry for files that don't have | |
2710 | * any xattrs or acls | |
2711 | */ | |
33345d01 LZ |
2712 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2713 | btrfs_ino(inode)); | |
72c04902 AV |
2714 | if (!maybe_acls) |
2715 | cache_no_acl(inode); | |
46a53cca | 2716 | |
39279cc3 | 2717 | btrfs_free_path(path); |
39279cc3 | 2718 | |
39279cc3 | 2719 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2720 | case S_IFREG: |
2721 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2722 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2723 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2724 | inode->i_fop = &btrfs_file_operations; |
2725 | inode->i_op = &btrfs_file_inode_operations; | |
2726 | break; | |
2727 | case S_IFDIR: | |
2728 | inode->i_fop = &btrfs_dir_file_operations; | |
2729 | if (root == root->fs_info->tree_root) | |
2730 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2731 | else | |
2732 | inode->i_op = &btrfs_dir_inode_operations; | |
2733 | break; | |
2734 | case S_IFLNK: | |
2735 | inode->i_op = &btrfs_symlink_inode_operations; | |
2736 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2737 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2738 | break; |
618e21d5 | 2739 | default: |
0279b4cd | 2740 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2741 | init_special_inode(inode, inode->i_mode, rdev); |
2742 | break; | |
39279cc3 | 2743 | } |
6cbff00f CH |
2744 | |
2745 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2746 | return; |
2747 | ||
2748 | make_bad: | |
39279cc3 | 2749 | btrfs_free_path(path); |
39279cc3 CM |
2750 | make_bad_inode(inode); |
2751 | } | |
2752 | ||
d352ac68 CM |
2753 | /* |
2754 | * given a leaf and an inode, copy the inode fields into the leaf | |
2755 | */ | |
e02119d5 CM |
2756 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2757 | struct extent_buffer *leaf, | |
5f39d397 | 2758 | struct btrfs_inode_item *item, |
39279cc3 CM |
2759 | struct inode *inode) |
2760 | { | |
51fab693 LB |
2761 | struct btrfs_map_token token; |
2762 | ||
2763 | btrfs_init_map_token(&token); | |
2764 | ||
2765 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
2766 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
2767 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
2768 | &token); | |
2769 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
2770 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
2771 | ||
2772 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), | |
2773 | inode->i_atime.tv_sec, &token); | |
2774 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2775 | inode->i_atime.tv_nsec, &token); | |
2776 | ||
2777 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2778 | inode->i_mtime.tv_sec, &token); | |
2779 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2780 | inode->i_mtime.tv_nsec, &token); | |
2781 | ||
2782 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2783 | inode->i_ctime.tv_sec, &token); | |
2784 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2785 | inode->i_ctime.tv_nsec, &token); | |
2786 | ||
2787 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
2788 | &token); | |
2789 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
2790 | &token); | |
2791 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
2792 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
2793 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
2794 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
2795 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
2796 | } |
2797 | ||
d352ac68 CM |
2798 | /* |
2799 | * copy everything in the in-memory inode into the btree. | |
2800 | */ | |
2115133f | 2801 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2802 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2803 | { |
2804 | struct btrfs_inode_item *inode_item; | |
2805 | struct btrfs_path *path; | |
5f39d397 | 2806 | struct extent_buffer *leaf; |
39279cc3 CM |
2807 | int ret; |
2808 | ||
2809 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2810 | if (!path) |
2811 | return -ENOMEM; | |
2812 | ||
b9473439 | 2813 | path->leave_spinning = 1; |
16cdcec7 MX |
2814 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2815 | 1); | |
39279cc3 CM |
2816 | if (ret) { |
2817 | if (ret > 0) | |
2818 | ret = -ENOENT; | |
2819 | goto failed; | |
2820 | } | |
2821 | ||
b4ce94de | 2822 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2823 | leaf = path->nodes[0]; |
2824 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2825 | struct btrfs_inode_item); |
39279cc3 | 2826 | |
e02119d5 | 2827 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2828 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2829 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2830 | ret = 0; |
2831 | failed: | |
39279cc3 CM |
2832 | btrfs_free_path(path); |
2833 | return ret; | |
2834 | } | |
2835 | ||
2115133f CM |
2836 | /* |
2837 | * copy everything in the in-memory inode into the btree. | |
2838 | */ | |
2839 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2840 | struct btrfs_root *root, struct inode *inode) | |
2841 | { | |
2842 | int ret; | |
2843 | ||
2844 | /* | |
2845 | * If the inode is a free space inode, we can deadlock during commit | |
2846 | * if we put it into the delayed code. | |
2847 | * | |
2848 | * The data relocation inode should also be directly updated | |
2849 | * without delay | |
2850 | */ | |
83eea1f1 | 2851 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 2852 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
2853 | btrfs_update_root_times(trans, root); |
2854 | ||
2115133f CM |
2855 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2856 | if (!ret) | |
2857 | btrfs_set_inode_last_trans(trans, inode); | |
2858 | return ret; | |
2859 | } | |
2860 | ||
2861 | return btrfs_update_inode_item(trans, root, inode); | |
2862 | } | |
2863 | ||
be6aef60 JB |
2864 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
2865 | struct btrfs_root *root, | |
2866 | struct inode *inode) | |
2115133f CM |
2867 | { |
2868 | int ret; | |
2869 | ||
2870 | ret = btrfs_update_inode(trans, root, inode); | |
2871 | if (ret == -ENOSPC) | |
2872 | return btrfs_update_inode_item(trans, root, inode); | |
2873 | return ret; | |
2874 | } | |
2875 | ||
d352ac68 CM |
2876 | /* |
2877 | * unlink helper that gets used here in inode.c and in the tree logging | |
2878 | * recovery code. It remove a link in a directory with a given name, and | |
2879 | * also drops the back refs in the inode to the directory | |
2880 | */ | |
92986796 AV |
2881 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2882 | struct btrfs_root *root, | |
2883 | struct inode *dir, struct inode *inode, | |
2884 | const char *name, int name_len) | |
39279cc3 CM |
2885 | { |
2886 | struct btrfs_path *path; | |
39279cc3 | 2887 | int ret = 0; |
5f39d397 | 2888 | struct extent_buffer *leaf; |
39279cc3 | 2889 | struct btrfs_dir_item *di; |
5f39d397 | 2890 | struct btrfs_key key; |
aec7477b | 2891 | u64 index; |
33345d01 LZ |
2892 | u64 ino = btrfs_ino(inode); |
2893 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2894 | |
2895 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2896 | if (!path) { |
2897 | ret = -ENOMEM; | |
554233a6 | 2898 | goto out; |
54aa1f4d CM |
2899 | } |
2900 | ||
b9473439 | 2901 | path->leave_spinning = 1; |
33345d01 | 2902 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2903 | name, name_len, -1); |
2904 | if (IS_ERR(di)) { | |
2905 | ret = PTR_ERR(di); | |
2906 | goto err; | |
2907 | } | |
2908 | if (!di) { | |
2909 | ret = -ENOENT; | |
2910 | goto err; | |
2911 | } | |
5f39d397 CM |
2912 | leaf = path->nodes[0]; |
2913 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2914 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2915 | if (ret) |
2916 | goto err; | |
b3b4aa74 | 2917 | btrfs_release_path(path); |
39279cc3 | 2918 | |
33345d01 LZ |
2919 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2920 | dir_ino, &index); | |
aec7477b | 2921 | if (ret) { |
d397712b | 2922 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2923 | "inode %llu parent %llu\n", name_len, name, |
2924 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2925 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2926 | goto err; |
2927 | } | |
2928 | ||
16cdcec7 | 2929 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2930 | if (ret) { |
2931 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2932 | goto err; |
79787eaa | 2933 | } |
39279cc3 | 2934 | |
e02119d5 | 2935 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2936 | inode, dir_ino); |
79787eaa JM |
2937 | if (ret != 0 && ret != -ENOENT) { |
2938 | btrfs_abort_transaction(trans, root, ret); | |
2939 | goto err; | |
2940 | } | |
e02119d5 CM |
2941 | |
2942 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2943 | dir, index); | |
6418c961 CM |
2944 | if (ret == -ENOENT) |
2945 | ret = 0; | |
39279cc3 CM |
2946 | err: |
2947 | btrfs_free_path(path); | |
e02119d5 CM |
2948 | if (ret) |
2949 | goto out; | |
2950 | ||
2951 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2952 | inode_inc_iversion(inode); |
2953 | inode_inc_iversion(dir); | |
e02119d5 | 2954 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 2955 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 2956 | out: |
39279cc3 CM |
2957 | return ret; |
2958 | } | |
2959 | ||
92986796 AV |
2960 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2961 | struct btrfs_root *root, | |
2962 | struct inode *dir, struct inode *inode, | |
2963 | const char *name, int name_len) | |
2964 | { | |
2965 | int ret; | |
2966 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2967 | if (!ret) { | |
2968 | btrfs_drop_nlink(inode); | |
2969 | ret = btrfs_update_inode(trans, root, inode); | |
2970 | } | |
2971 | return ret; | |
2972 | } | |
2973 | ||
2974 | ||
a22285a6 YZ |
2975 | /* helper to check if there is any shared block in the path */ |
2976 | static int check_path_shared(struct btrfs_root *root, | |
2977 | struct btrfs_path *path) | |
39279cc3 | 2978 | { |
a22285a6 YZ |
2979 | struct extent_buffer *eb; |
2980 | int level; | |
0e4dcbef | 2981 | u64 refs = 1; |
5df6a9f6 | 2982 | |
a22285a6 | 2983 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2984 | int ret; |
2985 | ||
a22285a6 YZ |
2986 | if (!path->nodes[level]) |
2987 | break; | |
2988 | eb = path->nodes[level]; | |
2989 | if (!btrfs_block_can_be_shared(root, eb)) | |
2990 | continue; | |
2991 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2992 | &refs, NULL); | |
2993 | if (refs > 1) | |
2994 | return 1; | |
5df6a9f6 | 2995 | } |
dedefd72 | 2996 | return 0; |
39279cc3 CM |
2997 | } |
2998 | ||
a22285a6 YZ |
2999 | /* |
3000 | * helper to start transaction for unlink and rmdir. | |
3001 | * | |
3002 | * unlink and rmdir are special in btrfs, they do not always free space. | |
3003 | * so in enospc case, we should make sure they will free space before | |
3004 | * allowing them to use the global metadata reservation. | |
3005 | */ | |
3006 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
3007 | struct dentry *dentry) | |
4df27c4d | 3008 | { |
39279cc3 | 3009 | struct btrfs_trans_handle *trans; |
a22285a6 | 3010 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 3011 | struct btrfs_path *path; |
4df27c4d | 3012 | struct btrfs_dir_item *di; |
7b128766 | 3013 | struct inode *inode = dentry->d_inode; |
4df27c4d | 3014 | u64 index; |
a22285a6 YZ |
3015 | int check_link = 1; |
3016 | int err = -ENOSPC; | |
4df27c4d | 3017 | int ret; |
33345d01 LZ |
3018 | u64 ino = btrfs_ino(inode); |
3019 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 3020 | |
e70bea5f JB |
3021 | /* |
3022 | * 1 for the possible orphan item | |
3023 | * 1 for the dir item | |
3024 | * 1 for the dir index | |
3025 | * 1 for the inode ref | |
3026 | * 1 for the inode ref in the tree log | |
3027 | * 2 for the dir entries in the log | |
3028 | * 1 for the inode | |
3029 | */ | |
3030 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
3031 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3032 | return trans; | |
4df27c4d | 3033 | |
33345d01 | 3034 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 3035 | return ERR_PTR(-ENOSPC); |
4df27c4d | 3036 | |
a22285a6 YZ |
3037 | /* check if there is someone else holds reference */ |
3038 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
3039 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 3040 | |
a22285a6 YZ |
3041 | if (atomic_read(&inode->i_count) > 2) |
3042 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 3043 | |
a22285a6 YZ |
3044 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
3045 | return ERR_PTR(-ENOSPC); | |
3046 | ||
3047 | path = btrfs_alloc_path(); | |
3048 | if (!path) { | |
3049 | root->fs_info->enospc_unlink = 0; | |
3050 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
3051 | } |
3052 | ||
3880a1b4 JB |
3053 | /* 1 for the orphan item */ |
3054 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 3055 | if (IS_ERR(trans)) { |
a22285a6 YZ |
3056 | btrfs_free_path(path); |
3057 | root->fs_info->enospc_unlink = 0; | |
3058 | return trans; | |
3059 | } | |
4df27c4d | 3060 | |
a22285a6 YZ |
3061 | path->skip_locking = 1; |
3062 | path->search_commit_root = 1; | |
4df27c4d | 3063 | |
a22285a6 YZ |
3064 | ret = btrfs_lookup_inode(trans, root, path, |
3065 | &BTRFS_I(dir)->location, 0); | |
3066 | if (ret < 0) { | |
3067 | err = ret; | |
3068 | goto out; | |
3069 | } | |
3070 | if (ret == 0) { | |
3071 | if (check_path_shared(root, path)) | |
3072 | goto out; | |
3073 | } else { | |
3074 | check_link = 0; | |
5df6a9f6 | 3075 | } |
b3b4aa74 | 3076 | btrfs_release_path(path); |
a22285a6 YZ |
3077 | |
3078 | ret = btrfs_lookup_inode(trans, root, path, | |
3079 | &BTRFS_I(inode)->location, 0); | |
3080 | if (ret < 0) { | |
3081 | err = ret; | |
3082 | goto out; | |
3083 | } | |
3084 | if (ret == 0) { | |
3085 | if (check_path_shared(root, path)) | |
3086 | goto out; | |
3087 | } else { | |
3088 | check_link = 0; | |
3089 | } | |
b3b4aa74 | 3090 | btrfs_release_path(path); |
a22285a6 YZ |
3091 | |
3092 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
3093 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 3094 | ino, (u64)-1, 0); |
a22285a6 YZ |
3095 | if (ret < 0) { |
3096 | err = ret; | |
3097 | goto out; | |
3098 | } | |
79787eaa | 3099 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
3100 | if (check_path_shared(root, path)) |
3101 | goto out; | |
b3b4aa74 | 3102 | btrfs_release_path(path); |
a22285a6 YZ |
3103 | } |
3104 | ||
3105 | if (!check_link) { | |
3106 | err = 0; | |
3107 | goto out; | |
3108 | } | |
3109 | ||
33345d01 | 3110 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
3111 | dentry->d_name.name, dentry->d_name.len, 0); |
3112 | if (IS_ERR(di)) { | |
3113 | err = PTR_ERR(di); | |
3114 | goto out; | |
3115 | } | |
3116 | if (di) { | |
3117 | if (check_path_shared(root, path)) | |
3118 | goto out; | |
3119 | } else { | |
3120 | err = 0; | |
3121 | goto out; | |
3122 | } | |
b3b4aa74 | 3123 | btrfs_release_path(path); |
a22285a6 | 3124 | |
f186373f MF |
3125 | ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name, |
3126 | dentry->d_name.len, ino, dir_ino, 0, | |
3127 | &index); | |
3128 | if (ret) { | |
3129 | err = ret; | |
a22285a6 YZ |
3130 | goto out; |
3131 | } | |
f186373f | 3132 | |
a22285a6 YZ |
3133 | if (check_path_shared(root, path)) |
3134 | goto out; | |
f186373f | 3135 | |
b3b4aa74 | 3136 | btrfs_release_path(path); |
a22285a6 | 3137 | |
16cdcec7 MX |
3138 | /* |
3139 | * This is a commit root search, if we can lookup inode item and other | |
3140 | * relative items in the commit root, it means the transaction of | |
3141 | * dir/file creation has been committed, and the dir index item that we | |
3142 | * delay to insert has also been inserted into the commit root. So | |
3143 | * we needn't worry about the delayed insertion of the dir index item | |
3144 | * here. | |
3145 | */ | |
33345d01 | 3146 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
3147 | dentry->d_name.name, dentry->d_name.len, 0); |
3148 | if (IS_ERR(di)) { | |
3149 | err = PTR_ERR(di); | |
3150 | goto out; | |
3151 | } | |
3152 | BUG_ON(ret == -ENOENT); | |
3153 | if (check_path_shared(root, path)) | |
3154 | goto out; | |
3155 | ||
3156 | err = 0; | |
3157 | out: | |
3158 | btrfs_free_path(path); | |
3880a1b4 JB |
3159 | /* Migrate the orphan reservation over */ |
3160 | if (!err) | |
3161 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3162 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3163 | trans->bytes_reserved); |
3880a1b4 | 3164 | |
a22285a6 YZ |
3165 | if (err) { |
3166 | btrfs_end_transaction(trans, root); | |
3167 | root->fs_info->enospc_unlink = 0; | |
3168 | return ERR_PTR(err); | |
3169 | } | |
3170 | ||
3171 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3172 | return trans; | |
3173 | } | |
3174 | ||
3175 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3176 | struct btrfs_root *root) | |
3177 | { | |
66d8f3dd | 3178 | if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) { |
5a77d76c JB |
3179 | btrfs_block_rsv_release(root, trans->block_rsv, |
3180 | trans->bytes_reserved); | |
3181 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3182 | BUG_ON(!root->fs_info->enospc_unlink); |
3183 | root->fs_info->enospc_unlink = 0; | |
3184 | } | |
7ad85bb7 | 3185 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3186 | } |
3187 | ||
3188 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3189 | { | |
3190 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3191 | struct btrfs_trans_handle *trans; | |
3192 | struct inode *inode = dentry->d_inode; | |
3193 | int ret; | |
a22285a6 YZ |
3194 | |
3195 | trans = __unlink_start_trans(dir, dentry); | |
3196 | if (IS_ERR(trans)) | |
3197 | return PTR_ERR(trans); | |
5f39d397 | 3198 | |
12fcfd22 CM |
3199 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3200 | ||
e02119d5 CM |
3201 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3202 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3203 | if (ret) |
3204 | goto out; | |
7b128766 | 3205 | |
a22285a6 | 3206 | if (inode->i_nlink == 0) { |
7b128766 | 3207 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3208 | if (ret) |
3209 | goto out; | |
a22285a6 | 3210 | } |
7b128766 | 3211 | |
b532402e | 3212 | out: |
a22285a6 | 3213 | __unlink_end_trans(trans, root); |
b53d3f5d | 3214 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3215 | return ret; |
3216 | } | |
3217 | ||
4df27c4d YZ |
3218 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3219 | struct btrfs_root *root, | |
3220 | struct inode *dir, u64 objectid, | |
3221 | const char *name, int name_len) | |
3222 | { | |
3223 | struct btrfs_path *path; | |
3224 | struct extent_buffer *leaf; | |
3225 | struct btrfs_dir_item *di; | |
3226 | struct btrfs_key key; | |
3227 | u64 index; | |
3228 | int ret; | |
33345d01 | 3229 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3230 | |
3231 | path = btrfs_alloc_path(); | |
3232 | if (!path) | |
3233 | return -ENOMEM; | |
3234 | ||
33345d01 | 3235 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3236 | name, name_len, -1); |
79787eaa JM |
3237 | if (IS_ERR_OR_NULL(di)) { |
3238 | if (!di) | |
3239 | ret = -ENOENT; | |
3240 | else | |
3241 | ret = PTR_ERR(di); | |
3242 | goto out; | |
3243 | } | |
4df27c4d YZ |
3244 | |
3245 | leaf = path->nodes[0]; | |
3246 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3247 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3248 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3249 | if (ret) { |
3250 | btrfs_abort_transaction(trans, root, ret); | |
3251 | goto out; | |
3252 | } | |
b3b4aa74 | 3253 | btrfs_release_path(path); |
4df27c4d YZ |
3254 | |
3255 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3256 | objectid, root->root_key.objectid, | |
33345d01 | 3257 | dir_ino, &index, name, name_len); |
4df27c4d | 3258 | if (ret < 0) { |
79787eaa JM |
3259 | if (ret != -ENOENT) { |
3260 | btrfs_abort_transaction(trans, root, ret); | |
3261 | goto out; | |
3262 | } | |
33345d01 | 3263 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3264 | name, name_len); |
79787eaa JM |
3265 | if (IS_ERR_OR_NULL(di)) { |
3266 | if (!di) | |
3267 | ret = -ENOENT; | |
3268 | else | |
3269 | ret = PTR_ERR(di); | |
3270 | btrfs_abort_transaction(trans, root, ret); | |
3271 | goto out; | |
3272 | } | |
4df27c4d YZ |
3273 | |
3274 | leaf = path->nodes[0]; | |
3275 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3276 | btrfs_release_path(path); |
4df27c4d YZ |
3277 | index = key.offset; |
3278 | } | |
945d8962 | 3279 | btrfs_release_path(path); |
4df27c4d | 3280 | |
16cdcec7 | 3281 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3282 | if (ret) { |
3283 | btrfs_abort_transaction(trans, root, ret); | |
3284 | goto out; | |
3285 | } | |
4df27c4d YZ |
3286 | |
3287 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3288 | inode_inc_iversion(dir); |
4df27c4d | 3289 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3290 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3291 | if (ret) |
3292 | btrfs_abort_transaction(trans, root, ret); | |
3293 | out: | |
71d7aed0 | 3294 | btrfs_free_path(path); |
79787eaa | 3295 | return ret; |
4df27c4d YZ |
3296 | } |
3297 | ||
39279cc3 CM |
3298 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3299 | { | |
3300 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3301 | int err = 0; |
39279cc3 | 3302 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3303 | struct btrfs_trans_handle *trans; |
39279cc3 | 3304 | |
b3ae244e | 3305 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3306 | return -ENOTEMPTY; |
b3ae244e DS |
3307 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3308 | return -EPERM; | |
134d4512 | 3309 | |
a22285a6 YZ |
3310 | trans = __unlink_start_trans(dir, dentry); |
3311 | if (IS_ERR(trans)) | |
5df6a9f6 | 3312 | return PTR_ERR(trans); |
5df6a9f6 | 3313 | |
33345d01 | 3314 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3315 | err = btrfs_unlink_subvol(trans, root, dir, |
3316 | BTRFS_I(inode)->location.objectid, | |
3317 | dentry->d_name.name, | |
3318 | dentry->d_name.len); | |
3319 | goto out; | |
3320 | } | |
3321 | ||
7b128766 JB |
3322 | err = btrfs_orphan_add(trans, inode); |
3323 | if (err) | |
4df27c4d | 3324 | goto out; |
7b128766 | 3325 | |
39279cc3 | 3326 | /* now the directory is empty */ |
e02119d5 CM |
3327 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3328 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3329 | if (!err) |
dbe674a9 | 3330 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3331 | out: |
a22285a6 | 3332 | __unlink_end_trans(trans, root); |
b53d3f5d | 3333 | btrfs_btree_balance_dirty(root); |
3954401f | 3334 | |
39279cc3 CM |
3335 | return err; |
3336 | } | |
3337 | ||
39279cc3 CM |
3338 | /* |
3339 | * this can truncate away extent items, csum items and directory items. | |
3340 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3341 | * any higher than new_size |
39279cc3 CM |
3342 | * |
3343 | * csum items that cross the new i_size are truncated to the new size | |
3344 | * as well. | |
7b128766 JB |
3345 | * |
3346 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3347 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3348 | */ |
8082510e YZ |
3349 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3350 | struct btrfs_root *root, | |
3351 | struct inode *inode, | |
3352 | u64 new_size, u32 min_type) | |
39279cc3 | 3353 | { |
39279cc3 | 3354 | struct btrfs_path *path; |
5f39d397 | 3355 | struct extent_buffer *leaf; |
39279cc3 | 3356 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3357 | struct btrfs_key key; |
3358 | struct btrfs_key found_key; | |
39279cc3 | 3359 | u64 extent_start = 0; |
db94535d | 3360 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3361 | u64 extent_offset = 0; |
39279cc3 | 3362 | u64 item_end = 0; |
8082510e YZ |
3363 | u64 mask = root->sectorsize - 1; |
3364 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3365 | int found_extent; |
3366 | int del_item; | |
85e21bac CM |
3367 | int pending_del_nr = 0; |
3368 | int pending_del_slot = 0; | |
179e29e4 | 3369 | int extent_type = -1; |
8082510e YZ |
3370 | int ret; |
3371 | int err = 0; | |
33345d01 | 3372 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3373 | |
3374 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3375 | |
0eb0e19c MF |
3376 | path = btrfs_alloc_path(); |
3377 | if (!path) | |
3378 | return -ENOMEM; | |
3379 | path->reada = -1; | |
3380 | ||
5dc562c5 JB |
3381 | /* |
3382 | * We want to drop from the next block forward in case this new size is | |
3383 | * not block aligned since we will be keeping the last block of the | |
3384 | * extent just the way it is. | |
3385 | */ | |
0af3d00b | 3386 | if (root->ref_cows || root == root->fs_info->tree_root) |
5dc562c5 | 3387 | btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0); |
8082510e | 3388 | |
16cdcec7 MX |
3389 | /* |
3390 | * This function is also used to drop the items in the log tree before | |
3391 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3392 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3393 | * items. | |
3394 | */ | |
3395 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3396 | btrfs_kill_delayed_inode_items(inode); | |
3397 | ||
33345d01 | 3398 | key.objectid = ino; |
39279cc3 | 3399 | key.offset = (u64)-1; |
5f39d397 CM |
3400 | key.type = (u8)-1; |
3401 | ||
85e21bac | 3402 | search_again: |
b9473439 | 3403 | path->leave_spinning = 1; |
85e21bac | 3404 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3405 | if (ret < 0) { |
3406 | err = ret; | |
3407 | goto out; | |
3408 | } | |
d397712b | 3409 | |
85e21bac | 3410 | if (ret > 0) { |
e02119d5 CM |
3411 | /* there are no items in the tree for us to truncate, we're |
3412 | * done | |
3413 | */ | |
8082510e YZ |
3414 | if (path->slots[0] == 0) |
3415 | goto out; | |
85e21bac CM |
3416 | path->slots[0]--; |
3417 | } | |
3418 | ||
d397712b | 3419 | while (1) { |
39279cc3 | 3420 | fi = NULL; |
5f39d397 CM |
3421 | leaf = path->nodes[0]; |
3422 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3423 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3424 | |
33345d01 | 3425 | if (found_key.objectid != ino) |
39279cc3 | 3426 | break; |
5f39d397 | 3427 | |
85e21bac | 3428 | if (found_type < min_type) |
39279cc3 CM |
3429 | break; |
3430 | ||
5f39d397 | 3431 | item_end = found_key.offset; |
39279cc3 | 3432 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3433 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3434 | struct btrfs_file_extent_item); |
179e29e4 CM |
3435 | extent_type = btrfs_file_extent_type(leaf, fi); |
3436 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3437 | item_end += |
db94535d | 3438 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3439 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3440 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3441 | fi); |
39279cc3 | 3442 | } |
008630c1 | 3443 | item_end--; |
39279cc3 | 3444 | } |
8082510e YZ |
3445 | if (found_type > min_type) { |
3446 | del_item = 1; | |
3447 | } else { | |
3448 | if (item_end < new_size) | |
b888db2b | 3449 | break; |
8082510e YZ |
3450 | if (found_key.offset >= new_size) |
3451 | del_item = 1; | |
3452 | else | |
3453 | del_item = 0; | |
39279cc3 | 3454 | } |
39279cc3 | 3455 | found_extent = 0; |
39279cc3 | 3456 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3457 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3458 | goto delete; | |
3459 | ||
3460 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3461 | u64 num_dec; |
db94535d | 3462 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3463 | if (!del_item) { |
db94535d CM |
3464 | u64 orig_num_bytes = |
3465 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3466 | extent_num_bytes = new_size - |
5f39d397 | 3467 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3468 | extent_num_bytes = extent_num_bytes & |
3469 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3470 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3471 | extent_num_bytes); | |
3472 | num_dec = (orig_num_bytes - | |
9069218d | 3473 | extent_num_bytes); |
e02119d5 | 3474 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3475 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3476 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3477 | } else { |
db94535d CM |
3478 | extent_num_bytes = |
3479 | btrfs_file_extent_disk_num_bytes(leaf, | |
3480 | fi); | |
5d4f98a2 YZ |
3481 | extent_offset = found_key.offset - |
3482 | btrfs_file_extent_offset(leaf, fi); | |
3483 | ||
39279cc3 | 3484 | /* FIXME blocksize != 4096 */ |
9069218d | 3485 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3486 | if (extent_start != 0) { |
3487 | found_extent = 1; | |
e02119d5 | 3488 | if (root->ref_cows) |
a76a3cd4 | 3489 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3490 | } |
39279cc3 | 3491 | } |
9069218d | 3492 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3493 | /* |
3494 | * we can't truncate inline items that have had | |
3495 | * special encodings | |
3496 | */ | |
3497 | if (!del_item && | |
3498 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3499 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3500 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3501 | u32 size = new_size - found_key.offset; |
3502 | ||
3503 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3504 | inode_sub_bytes(inode, item_end + 1 - |
3505 | new_size); | |
e02119d5 CM |
3506 | } |
3507 | size = | |
3508 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3509 | btrfs_truncate_item(trans, root, path, |
3510 | size, 1); | |
e02119d5 | 3511 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3512 | inode_sub_bytes(inode, item_end + 1 - |
3513 | found_key.offset); | |
9069218d | 3514 | } |
39279cc3 | 3515 | } |
179e29e4 | 3516 | delete: |
39279cc3 | 3517 | if (del_item) { |
85e21bac CM |
3518 | if (!pending_del_nr) { |
3519 | /* no pending yet, add ourselves */ | |
3520 | pending_del_slot = path->slots[0]; | |
3521 | pending_del_nr = 1; | |
3522 | } else if (pending_del_nr && | |
3523 | path->slots[0] + 1 == pending_del_slot) { | |
3524 | /* hop on the pending chunk */ | |
3525 | pending_del_nr++; | |
3526 | pending_del_slot = path->slots[0]; | |
3527 | } else { | |
d397712b | 3528 | BUG(); |
85e21bac | 3529 | } |
39279cc3 CM |
3530 | } else { |
3531 | break; | |
3532 | } | |
0af3d00b JB |
3533 | if (found_extent && (root->ref_cows || |
3534 | root == root->fs_info->tree_root)) { | |
b9473439 | 3535 | btrfs_set_path_blocking(path); |
39279cc3 | 3536 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3537 | extent_num_bytes, 0, |
3538 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3539 | ino, extent_offset, 0); |
39279cc3 CM |
3540 | BUG_ON(ret); |
3541 | } | |
85e21bac | 3542 | |
8082510e YZ |
3543 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3544 | break; | |
3545 | ||
3546 | if (path->slots[0] == 0 || | |
3547 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
3548 | if (pending_del_nr) { |
3549 | ret = btrfs_del_items(trans, root, path, | |
3550 | pending_del_slot, | |
3551 | pending_del_nr); | |
79787eaa JM |
3552 | if (ret) { |
3553 | btrfs_abort_transaction(trans, | |
3554 | root, ret); | |
3555 | goto error; | |
3556 | } | |
8082510e YZ |
3557 | pending_del_nr = 0; |
3558 | } | |
b3b4aa74 | 3559 | btrfs_release_path(path); |
85e21bac | 3560 | goto search_again; |
8082510e YZ |
3561 | } else { |
3562 | path->slots[0]--; | |
85e21bac | 3563 | } |
39279cc3 | 3564 | } |
8082510e | 3565 | out: |
85e21bac CM |
3566 | if (pending_del_nr) { |
3567 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3568 | pending_del_nr); | |
79787eaa JM |
3569 | if (ret) |
3570 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3571 | } |
79787eaa | 3572 | error: |
39279cc3 | 3573 | btrfs_free_path(path); |
8082510e | 3574 | return err; |
39279cc3 CM |
3575 | } |
3576 | ||
3577 | /* | |
2aaa6655 JB |
3578 | * btrfs_truncate_page - read, zero a chunk and write a page |
3579 | * @inode - inode that we're zeroing | |
3580 | * @from - the offset to start zeroing | |
3581 | * @len - the length to zero, 0 to zero the entire range respective to the | |
3582 | * offset | |
3583 | * @front - zero up to the offset instead of from the offset on | |
3584 | * | |
3585 | * This will find the page for the "from" offset and cow the page and zero the | |
3586 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 3587 | */ |
2aaa6655 JB |
3588 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
3589 | int front) | |
39279cc3 | 3590 | { |
2aaa6655 | 3591 | struct address_space *mapping = inode->i_mapping; |
db94535d | 3592 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3593 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3594 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3595 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3596 | char *kaddr; |
db94535d | 3597 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3598 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3599 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3600 | struct page *page; | |
3b16a4e3 | 3601 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3602 | int ret = 0; |
a52d9a80 | 3603 | u64 page_start; |
e6dcd2dc | 3604 | u64 page_end; |
39279cc3 | 3605 | |
2aaa6655 JB |
3606 | if ((offset & (blocksize - 1)) == 0 && |
3607 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 3608 | goto out; |
0ca1f7ce | 3609 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3610 | if (ret) |
3611 | goto out; | |
39279cc3 | 3612 | |
211c17f5 | 3613 | again: |
3b16a4e3 | 3614 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3615 | if (!page) { |
0ca1f7ce | 3616 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 3617 | ret = -ENOMEM; |
39279cc3 | 3618 | goto out; |
5d5e103a | 3619 | } |
e6dcd2dc CM |
3620 | |
3621 | page_start = page_offset(page); | |
3622 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3623 | ||
39279cc3 | 3624 | if (!PageUptodate(page)) { |
9ebefb18 | 3625 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3626 | lock_page(page); |
211c17f5 CM |
3627 | if (page->mapping != mapping) { |
3628 | unlock_page(page); | |
3629 | page_cache_release(page); | |
3630 | goto again; | |
3631 | } | |
39279cc3 CM |
3632 | if (!PageUptodate(page)) { |
3633 | ret = -EIO; | |
89642229 | 3634 | goto out_unlock; |
39279cc3 CM |
3635 | } |
3636 | } | |
211c17f5 | 3637 | wait_on_page_writeback(page); |
e6dcd2dc | 3638 | |
d0082371 | 3639 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3640 | set_page_extent_mapped(page); |
3641 | ||
3642 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3643 | if (ordered) { | |
2ac55d41 JB |
3644 | unlock_extent_cached(io_tree, page_start, page_end, |
3645 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3646 | unlock_page(page); |
3647 | page_cache_release(page); | |
eb84ae03 | 3648 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3649 | btrfs_put_ordered_extent(ordered); |
3650 | goto again; | |
3651 | } | |
3652 | ||
2ac55d41 | 3653 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
3654 | EXTENT_DIRTY | EXTENT_DELALLOC | |
3655 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 3656 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3657 | |
2ac55d41 JB |
3658 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3659 | &cached_state); | |
9ed74f2d | 3660 | if (ret) { |
2ac55d41 JB |
3661 | unlock_extent_cached(io_tree, page_start, page_end, |
3662 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3663 | goto out_unlock; |
3664 | } | |
3665 | ||
e6dcd2dc | 3666 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
3667 | if (!len) |
3668 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 3669 | kaddr = kmap(page); |
2aaa6655 JB |
3670 | if (front) |
3671 | memset(kaddr, 0, offset); | |
3672 | else | |
3673 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
3674 | flush_dcache_page(page); |
3675 | kunmap(page); | |
3676 | } | |
247e743c | 3677 | ClearPageChecked(page); |
e6dcd2dc | 3678 | set_page_dirty(page); |
2ac55d41 JB |
3679 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3680 | GFP_NOFS); | |
39279cc3 | 3681 | |
89642229 | 3682 | out_unlock: |
5d5e103a | 3683 | if (ret) |
0ca1f7ce | 3684 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3685 | unlock_page(page); |
3686 | page_cache_release(page); | |
3687 | out: | |
3688 | return ret; | |
3689 | } | |
3690 | ||
695a0d0d JB |
3691 | /* |
3692 | * This function puts in dummy file extents for the area we're creating a hole | |
3693 | * for. So if we are truncating this file to a larger size we need to insert | |
3694 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3695 | * the range between oldsize and size | |
3696 | */ | |
a41ad394 | 3697 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3698 | { |
9036c102 YZ |
3699 | struct btrfs_trans_handle *trans; |
3700 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3701 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3702 | struct extent_map *em = NULL; |
2ac55d41 | 3703 | struct extent_state *cached_state = NULL; |
5dc562c5 | 3704 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9036c102 | 3705 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3706 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3707 | u64 block_end = (size + mask) & ~mask; |
3708 | u64 last_byte; | |
3709 | u64 cur_offset; | |
3710 | u64 hole_size; | |
9ed74f2d | 3711 | int err = 0; |
39279cc3 | 3712 | |
9036c102 YZ |
3713 | if (size <= hole_start) |
3714 | return 0; | |
3715 | ||
9036c102 YZ |
3716 | while (1) { |
3717 | struct btrfs_ordered_extent *ordered; | |
3718 | btrfs_wait_ordered_range(inode, hole_start, | |
3719 | block_end - hole_start); | |
2ac55d41 | 3720 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3721 | &cached_state); |
9036c102 YZ |
3722 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3723 | if (!ordered) | |
3724 | break; | |
2ac55d41 JB |
3725 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3726 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3727 | btrfs_put_ordered_extent(ordered); |
3728 | } | |
39279cc3 | 3729 | |
9036c102 YZ |
3730 | cur_offset = hole_start; |
3731 | while (1) { | |
3732 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3733 | block_end - cur_offset, 0); | |
79787eaa JM |
3734 | if (IS_ERR(em)) { |
3735 | err = PTR_ERR(em); | |
f2767956 | 3736 | em = NULL; |
79787eaa JM |
3737 | break; |
3738 | } | |
9036c102 YZ |
3739 | last_byte = min(extent_map_end(em), block_end); |
3740 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3741 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 3742 | struct extent_map *hole_em; |
9036c102 | 3743 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3744 | |
3642320e | 3745 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3746 | if (IS_ERR(trans)) { |
3747 | err = PTR_ERR(trans); | |
9ed74f2d | 3748 | break; |
a22285a6 | 3749 | } |
8082510e | 3750 | |
5dc562c5 JB |
3751 | err = btrfs_drop_extents(trans, root, inode, |
3752 | cur_offset, | |
2671485d | 3753 | cur_offset + hole_size, 1); |
5b397377 | 3754 | if (err) { |
79787eaa | 3755 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3756 | btrfs_end_transaction(trans, root); |
3893e33b | 3757 | break; |
5b397377 | 3758 | } |
8082510e | 3759 | |
9036c102 | 3760 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3761 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3762 | 0, hole_size, 0, hole_size, |
3763 | 0, 0, 0); | |
5b397377 | 3764 | if (err) { |
79787eaa | 3765 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3766 | btrfs_end_transaction(trans, root); |
3893e33b | 3767 | break; |
5b397377 | 3768 | } |
8082510e | 3769 | |
5dc562c5 JB |
3770 | btrfs_drop_extent_cache(inode, cur_offset, |
3771 | cur_offset + hole_size - 1, 0); | |
3772 | hole_em = alloc_extent_map(); | |
3773 | if (!hole_em) { | |
3774 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3775 | &BTRFS_I(inode)->runtime_flags); | |
3776 | goto next; | |
3777 | } | |
3778 | hole_em->start = cur_offset; | |
3779 | hole_em->len = hole_size; | |
3780 | hole_em->orig_start = cur_offset; | |
8082510e | 3781 | |
5dc562c5 JB |
3782 | hole_em->block_start = EXTENT_MAP_HOLE; |
3783 | hole_em->block_len = 0; | |
b4939680 | 3784 | hole_em->orig_block_len = 0; |
5dc562c5 JB |
3785 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
3786 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
3787 | hole_em->generation = trans->transid; | |
8082510e | 3788 | |
5dc562c5 JB |
3789 | while (1) { |
3790 | write_lock(&em_tree->lock); | |
3791 | err = add_extent_mapping(em_tree, hole_em); | |
3792 | if (!err) | |
3793 | list_move(&hole_em->list, | |
3794 | &em_tree->modified_extents); | |
3795 | write_unlock(&em_tree->lock); | |
3796 | if (err != -EEXIST) | |
3797 | break; | |
3798 | btrfs_drop_extent_cache(inode, cur_offset, | |
3799 | cur_offset + | |
3800 | hole_size - 1, 0); | |
3801 | } | |
3802 | free_extent_map(hole_em); | |
3803 | next: | |
3642320e | 3804 | btrfs_update_inode(trans, root, inode); |
8082510e | 3805 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3806 | } |
3807 | free_extent_map(em); | |
a22285a6 | 3808 | em = NULL; |
9036c102 | 3809 | cur_offset = last_byte; |
8082510e | 3810 | if (cur_offset >= block_end) |
9036c102 YZ |
3811 | break; |
3812 | } | |
1832a6d5 | 3813 | |
a22285a6 | 3814 | free_extent_map(em); |
2ac55d41 JB |
3815 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3816 | GFP_NOFS); | |
9036c102 YZ |
3817 | return err; |
3818 | } | |
39279cc3 | 3819 | |
3972f260 | 3820 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 3821 | { |
f4a2f4c5 MX |
3822 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3823 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3824 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
3825 | loff_t newsize = attr->ia_size; |
3826 | int mask = attr->ia_valid; | |
8082510e YZ |
3827 | int ret; |
3828 | ||
a41ad394 | 3829 | if (newsize == oldsize) |
8082510e YZ |
3830 | return 0; |
3831 | ||
3972f260 ES |
3832 | /* |
3833 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
3834 | * special case where we need to update the times despite not having | |
3835 | * these flags set. For all other operations the VFS set these flags | |
3836 | * explicitly if it wants a timestamp update. | |
3837 | */ | |
3838 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
3839 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
3840 | ||
a41ad394 | 3841 | if (newsize > oldsize) { |
a41ad394 JB |
3842 | truncate_pagecache(inode, oldsize, newsize); |
3843 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3844 | if (ret) |
8082510e | 3845 | return ret; |
8082510e | 3846 | |
f4a2f4c5 MX |
3847 | trans = btrfs_start_transaction(root, 1); |
3848 | if (IS_ERR(trans)) | |
3849 | return PTR_ERR(trans); | |
3850 | ||
3851 | i_size_write(inode, newsize); | |
3852 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3853 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3854 | btrfs_end_transaction(trans, root); |
a41ad394 | 3855 | } else { |
8082510e | 3856 | |
a41ad394 JB |
3857 | /* |
3858 | * We're truncating a file that used to have good data down to | |
3859 | * zero. Make sure it gets into the ordered flush list so that | |
3860 | * any new writes get down to disk quickly. | |
3861 | */ | |
3862 | if (newsize == 0) | |
72ac3c0d JB |
3863 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3864 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3865 | |
f3fe820c JB |
3866 | /* |
3867 | * 1 for the orphan item we're going to add | |
3868 | * 1 for the orphan item deletion. | |
3869 | */ | |
3870 | trans = btrfs_start_transaction(root, 2); | |
3871 | if (IS_ERR(trans)) | |
3872 | return PTR_ERR(trans); | |
3873 | ||
3874 | /* | |
3875 | * We need to do this in case we fail at _any_ point during the | |
3876 | * actual truncate. Once we do the truncate_setsize we could | |
3877 | * invalidate pages which forces any outstanding ordered io to | |
3878 | * be instantly completed which will give us extents that need | |
3879 | * to be truncated. If we fail to get an orphan inode down we | |
3880 | * could have left over extents that were never meant to live, | |
3881 | * so we need to garuntee from this point on that everything | |
3882 | * will be consistent. | |
3883 | */ | |
3884 | ret = btrfs_orphan_add(trans, inode); | |
3885 | btrfs_end_transaction(trans, root); | |
3886 | if (ret) | |
3887 | return ret; | |
3888 | ||
a41ad394 JB |
3889 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3890 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
3891 | |
3892 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
3893 | btrfs_inode_block_unlocked_dio(inode); | |
3894 | inode_dio_wait(inode); | |
3895 | btrfs_inode_resume_unlocked_dio(inode); | |
3896 | ||
a41ad394 | 3897 | ret = btrfs_truncate(inode); |
f3fe820c JB |
3898 | if (ret && inode->i_nlink) |
3899 | btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
3900 | } |
3901 | ||
a41ad394 | 3902 | return ret; |
8082510e YZ |
3903 | } |
3904 | ||
9036c102 YZ |
3905 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3906 | { | |
3907 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3908 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3909 | int err; |
39279cc3 | 3910 | |
b83cc969 LZ |
3911 | if (btrfs_root_readonly(root)) |
3912 | return -EROFS; | |
3913 | ||
9036c102 YZ |
3914 | err = inode_change_ok(inode, attr); |
3915 | if (err) | |
3916 | return err; | |
2bf5a725 | 3917 | |
5a3f23d5 | 3918 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 3919 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
3920 | if (err) |
3921 | return err; | |
39279cc3 | 3922 | } |
9036c102 | 3923 | |
1025774c CH |
3924 | if (attr->ia_valid) { |
3925 | setattr_copy(inode, attr); | |
0c4d2d95 | 3926 | inode_inc_iversion(inode); |
22c44fe6 | 3927 | err = btrfs_dirty_inode(inode); |
1025774c | 3928 | |
22c44fe6 | 3929 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3930 | err = btrfs_acl_chmod(inode); |
3931 | } | |
33268eaf | 3932 | |
39279cc3 CM |
3933 | return err; |
3934 | } | |
61295eb8 | 3935 | |
bd555975 | 3936 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3937 | { |
3938 | struct btrfs_trans_handle *trans; | |
3939 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3940 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3941 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
3942 | int ret; |
3943 | ||
1abe9b8a | 3944 | trace_btrfs_inode_evict(inode); |
3945 | ||
39279cc3 | 3946 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3947 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 3948 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
3949 | goto no_delete; |
3950 | ||
39279cc3 | 3951 | if (is_bad_inode(inode)) { |
7b128766 | 3952 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3953 | goto no_delete; |
3954 | } | |
bd555975 | 3955 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3956 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3957 | |
c71bf099 | 3958 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 3959 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 3960 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
3961 | goto no_delete; |
3962 | } | |
3963 | ||
76dda93c YZ |
3964 | if (inode->i_nlink > 0) { |
3965 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3966 | goto no_delete; | |
3967 | } | |
3968 | ||
0e8c36a9 MX |
3969 | ret = btrfs_commit_inode_delayed_inode(inode); |
3970 | if (ret) { | |
3971 | btrfs_orphan_del(NULL, inode); | |
3972 | goto no_delete; | |
3973 | } | |
3974 | ||
66d8f3dd | 3975 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
3976 | if (!rsv) { |
3977 | btrfs_orphan_del(NULL, inode); | |
3978 | goto no_delete; | |
3979 | } | |
4a338542 | 3980 | rsv->size = min_size; |
ca7e70f5 | 3981 | rsv->failfast = 1; |
726c35fa | 3982 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3983 | |
dbe674a9 | 3984 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3985 | |
4289a667 | 3986 | /* |
8407aa46 MX |
3987 | * This is a bit simpler than btrfs_truncate since we've already |
3988 | * reserved our space for our orphan item in the unlink, so we just | |
3989 | * need to reserve some slack space in case we add bytes and update | |
3990 | * inode item when doing the truncate. | |
4289a667 | 3991 | */ |
8082510e | 3992 | while (1) { |
08e007d2 MX |
3993 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
3994 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
3995 | |
3996 | /* | |
3997 | * Try and steal from the global reserve since we will | |
3998 | * likely not use this space anyway, we want to try as | |
3999 | * hard as possible to get this to work. | |
4000 | */ | |
4001 | if (ret) | |
4002 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4003 | |
d68fc57b | 4004 | if (ret) { |
4289a667 | 4005 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 4006 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
4007 | btrfs_orphan_del(NULL, inode); |
4008 | btrfs_free_block_rsv(root, rsv); | |
4009 | goto no_delete; | |
d68fc57b | 4010 | } |
7b128766 | 4011 | |
0e8c36a9 | 4012 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4013 | if (IS_ERR(trans)) { |
4014 | btrfs_orphan_del(NULL, inode); | |
4015 | btrfs_free_block_rsv(root, rsv); | |
4016 | goto no_delete; | |
d68fc57b | 4017 | } |
7b128766 | 4018 | |
4289a667 JB |
4019 | trans->block_rsv = rsv; |
4020 | ||
d68fc57b | 4021 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4022 | if (ret != -ENOSPC) |
8082510e | 4023 | break; |
85e21bac | 4024 | |
8407aa46 | 4025 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4026 | btrfs_end_transaction(trans, root); |
4027 | trans = NULL; | |
b53d3f5d | 4028 | btrfs_btree_balance_dirty(root); |
8082510e | 4029 | } |
5f39d397 | 4030 | |
4289a667 JB |
4031 | btrfs_free_block_rsv(root, rsv); |
4032 | ||
8082510e | 4033 | if (ret == 0) { |
4289a667 | 4034 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
4035 | ret = btrfs_orphan_del(trans, inode); |
4036 | BUG_ON(ret); | |
4037 | } | |
54aa1f4d | 4038 | |
4289a667 | 4039 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4040 | if (!(root == root->fs_info->tree_root || |
4041 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4042 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4043 | |
54aa1f4d | 4044 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4045 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4046 | no_delete: |
dbd5768f | 4047 | clear_inode(inode); |
8082510e | 4048 | return; |
39279cc3 CM |
4049 | } |
4050 | ||
4051 | /* | |
4052 | * this returns the key found in the dir entry in the location pointer. | |
4053 | * If no dir entries were found, location->objectid is 0. | |
4054 | */ | |
4055 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4056 | struct btrfs_key *location) | |
4057 | { | |
4058 | const char *name = dentry->d_name.name; | |
4059 | int namelen = dentry->d_name.len; | |
4060 | struct btrfs_dir_item *di; | |
4061 | struct btrfs_path *path; | |
4062 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4063 | int ret = 0; |
39279cc3 CM |
4064 | |
4065 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4066 | if (!path) |
4067 | return -ENOMEM; | |
3954401f | 4068 | |
33345d01 | 4069 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4070 | namelen, 0); |
0d9f7f3e Y |
4071 | if (IS_ERR(di)) |
4072 | ret = PTR_ERR(di); | |
d397712b | 4073 | |
c704005d | 4074 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4075 | goto out_err; |
d397712b | 4076 | |
5f39d397 | 4077 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4078 | out: |
39279cc3 CM |
4079 | btrfs_free_path(path); |
4080 | return ret; | |
3954401f CM |
4081 | out_err: |
4082 | location->objectid = 0; | |
4083 | goto out; | |
39279cc3 CM |
4084 | } |
4085 | ||
4086 | /* | |
4087 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4088 | * needs to be changed to reflect the root directory of the tree root. This | |
4089 | * is kind of like crossing a mount point. | |
4090 | */ | |
4091 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4092 | struct inode *dir, |
4093 | struct dentry *dentry, | |
4094 | struct btrfs_key *location, | |
4095 | struct btrfs_root **sub_root) | |
39279cc3 | 4096 | { |
4df27c4d YZ |
4097 | struct btrfs_path *path; |
4098 | struct btrfs_root *new_root; | |
4099 | struct btrfs_root_ref *ref; | |
4100 | struct extent_buffer *leaf; | |
4101 | int ret; | |
4102 | int err = 0; | |
39279cc3 | 4103 | |
4df27c4d YZ |
4104 | path = btrfs_alloc_path(); |
4105 | if (!path) { | |
4106 | err = -ENOMEM; | |
4107 | goto out; | |
4108 | } | |
39279cc3 | 4109 | |
4df27c4d YZ |
4110 | err = -ENOENT; |
4111 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
4112 | BTRFS_I(dir)->root->root_key.objectid, | |
4113 | location->objectid); | |
4114 | if (ret) { | |
4115 | if (ret < 0) | |
4116 | err = ret; | |
4117 | goto out; | |
4118 | } | |
39279cc3 | 4119 | |
4df27c4d YZ |
4120 | leaf = path->nodes[0]; |
4121 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4122 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4123 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4124 | goto out; | |
39279cc3 | 4125 | |
4df27c4d YZ |
4126 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4127 | (unsigned long)(ref + 1), | |
4128 | dentry->d_name.len); | |
4129 | if (ret) | |
4130 | goto out; | |
4131 | ||
b3b4aa74 | 4132 | btrfs_release_path(path); |
4df27c4d YZ |
4133 | |
4134 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4135 | if (IS_ERR(new_root)) { | |
4136 | err = PTR_ERR(new_root); | |
4137 | goto out; | |
4138 | } | |
4139 | ||
4140 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
4141 | err = -ENOENT; | |
4142 | goto out; | |
4143 | } | |
4144 | ||
4145 | *sub_root = new_root; | |
4146 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4147 | location->type = BTRFS_INODE_ITEM_KEY; | |
4148 | location->offset = 0; | |
4149 | err = 0; | |
4150 | out: | |
4151 | btrfs_free_path(path); | |
4152 | return err; | |
39279cc3 CM |
4153 | } |
4154 | ||
5d4f98a2 YZ |
4155 | static void inode_tree_add(struct inode *inode) |
4156 | { | |
4157 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4158 | struct btrfs_inode *entry; | |
03e860bd NP |
4159 | struct rb_node **p; |
4160 | struct rb_node *parent; | |
33345d01 | 4161 | u64 ino = btrfs_ino(inode); |
03e860bd NP |
4162 | again: |
4163 | p = &root->inode_tree.rb_node; | |
4164 | parent = NULL; | |
5d4f98a2 | 4165 | |
1d3382cb | 4166 | if (inode_unhashed(inode)) |
76dda93c YZ |
4167 | return; |
4168 | ||
5d4f98a2 YZ |
4169 | spin_lock(&root->inode_lock); |
4170 | while (*p) { | |
4171 | parent = *p; | |
4172 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4173 | ||
33345d01 | 4174 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4175 | p = &parent->rb_left; |
33345d01 | 4176 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4177 | p = &parent->rb_right; |
5d4f98a2 YZ |
4178 | else { |
4179 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4180 | (I_WILL_FREE | I_FREEING))); |
03e860bd NP |
4181 | rb_erase(parent, &root->inode_tree); |
4182 | RB_CLEAR_NODE(parent); | |
4183 | spin_unlock(&root->inode_lock); | |
4184 | goto again; | |
5d4f98a2 YZ |
4185 | } |
4186 | } | |
4187 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4188 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4189 | spin_unlock(&root->inode_lock); | |
4190 | } | |
4191 | ||
4192 | static void inode_tree_del(struct inode *inode) | |
4193 | { | |
4194 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4195 | int empty = 0; |
5d4f98a2 | 4196 | |
03e860bd | 4197 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4198 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4199 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4200 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4201 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4202 | } |
03e860bd | 4203 | spin_unlock(&root->inode_lock); |
76dda93c | 4204 | |
0af3d00b JB |
4205 | /* |
4206 | * Free space cache has inodes in the tree root, but the tree root has a | |
4207 | * root_refs of 0, so this could end up dropping the tree root as a | |
4208 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4209 | * make sure we don't drop it. | |
4210 | */ | |
4211 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4212 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4213 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4214 | spin_lock(&root->inode_lock); | |
4215 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4216 | spin_unlock(&root->inode_lock); | |
4217 | if (empty) | |
4218 | btrfs_add_dead_root(root); | |
4219 | } | |
4220 | } | |
4221 | ||
143bede5 | 4222 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4223 | { |
4224 | struct rb_node *node; | |
4225 | struct rb_node *prev; | |
4226 | struct btrfs_inode *entry; | |
4227 | struct inode *inode; | |
4228 | u64 objectid = 0; | |
4229 | ||
4230 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4231 | ||
4232 | spin_lock(&root->inode_lock); | |
4233 | again: | |
4234 | node = root->inode_tree.rb_node; | |
4235 | prev = NULL; | |
4236 | while (node) { | |
4237 | prev = node; | |
4238 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4239 | ||
33345d01 | 4240 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4241 | node = node->rb_left; |
33345d01 | 4242 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4243 | node = node->rb_right; |
4244 | else | |
4245 | break; | |
4246 | } | |
4247 | if (!node) { | |
4248 | while (prev) { | |
4249 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4250 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4251 | node = prev; |
4252 | break; | |
4253 | } | |
4254 | prev = rb_next(prev); | |
4255 | } | |
4256 | } | |
4257 | while (node) { | |
4258 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4259 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4260 | inode = igrab(&entry->vfs_inode); |
4261 | if (inode) { | |
4262 | spin_unlock(&root->inode_lock); | |
4263 | if (atomic_read(&inode->i_count) > 1) | |
4264 | d_prune_aliases(inode); | |
4265 | /* | |
45321ac5 | 4266 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4267 | * the inode cache when its usage count |
4268 | * hits zero. | |
4269 | */ | |
4270 | iput(inode); | |
4271 | cond_resched(); | |
4272 | spin_lock(&root->inode_lock); | |
4273 | goto again; | |
4274 | } | |
4275 | ||
4276 | if (cond_resched_lock(&root->inode_lock)) | |
4277 | goto again; | |
4278 | ||
4279 | node = rb_next(node); | |
4280 | } | |
4281 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4282 | } |
4283 | ||
e02119d5 CM |
4284 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4285 | { | |
4286 | struct btrfs_iget_args *args = p; | |
4287 | inode->i_ino = args->ino; | |
e02119d5 | 4288 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4289 | return 0; |
4290 | } | |
4291 | ||
4292 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4293 | { | |
4294 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4295 | return args->ino == btrfs_ino(inode) && |
d397712b | 4296 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4297 | } |
4298 | ||
5d4f98a2 YZ |
4299 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4300 | u64 objectid, | |
4301 | struct btrfs_root *root) | |
39279cc3 CM |
4302 | { |
4303 | struct inode *inode; | |
4304 | struct btrfs_iget_args args; | |
4305 | args.ino = objectid; | |
4306 | args.root = root; | |
4307 | ||
4308 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4309 | btrfs_init_locked_inode, | |
4310 | (void *)&args); | |
4311 | return inode; | |
4312 | } | |
4313 | ||
1a54ef8c BR |
4314 | /* Get an inode object given its location and corresponding root. |
4315 | * Returns in *is_new if the inode was read from disk | |
4316 | */ | |
4317 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4318 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4319 | { |
4320 | struct inode *inode; | |
4321 | ||
4322 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4323 | if (!inode) | |
5d4f98a2 | 4324 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4325 | |
4326 | if (inode->i_state & I_NEW) { | |
4327 | BTRFS_I(inode)->root = root; | |
4328 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4329 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4330 | if (!is_bad_inode(inode)) { |
4331 | inode_tree_add(inode); | |
4332 | unlock_new_inode(inode); | |
4333 | if (new) | |
4334 | *new = 1; | |
4335 | } else { | |
e0b6d65b ST |
4336 | unlock_new_inode(inode); |
4337 | iput(inode); | |
4338 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4339 | } |
4340 | } | |
4341 | ||
1a54ef8c BR |
4342 | return inode; |
4343 | } | |
4344 | ||
4df27c4d YZ |
4345 | static struct inode *new_simple_dir(struct super_block *s, |
4346 | struct btrfs_key *key, | |
4347 | struct btrfs_root *root) | |
4348 | { | |
4349 | struct inode *inode = new_inode(s); | |
4350 | ||
4351 | if (!inode) | |
4352 | return ERR_PTR(-ENOMEM); | |
4353 | ||
4df27c4d YZ |
4354 | BTRFS_I(inode)->root = root; |
4355 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4356 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4357 | |
4358 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4359 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4360 | inode->i_fop = &simple_dir_operations; |
4361 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4362 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4363 | ||
4364 | return inode; | |
4365 | } | |
4366 | ||
3de4586c | 4367 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4368 | { |
d397712b | 4369 | struct inode *inode; |
4df27c4d | 4370 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4371 | struct btrfs_root *sub_root = root; |
4372 | struct btrfs_key location; | |
76dda93c | 4373 | int index; |
b4aff1f8 | 4374 | int ret = 0; |
39279cc3 CM |
4375 | |
4376 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4377 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4378 | |
b4aff1f8 JB |
4379 | if (unlikely(d_need_lookup(dentry))) { |
4380 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4381 | kfree(dentry->d_fsdata); | |
4382 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4383 | /* This thing is hashed, drop it for now */ |
4384 | d_drop(dentry); | |
b4aff1f8 JB |
4385 | } else { |
4386 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4387 | } | |
5f39d397 | 4388 | |
39279cc3 CM |
4389 | if (ret < 0) |
4390 | return ERR_PTR(ret); | |
5f39d397 | 4391 | |
4df27c4d YZ |
4392 | if (location.objectid == 0) |
4393 | return NULL; | |
4394 | ||
4395 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4396 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4397 | return inode; |
4398 | } | |
4399 | ||
4400 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4401 | ||
76dda93c | 4402 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4403 | ret = fixup_tree_root_location(root, dir, dentry, |
4404 | &location, &sub_root); | |
4405 | if (ret < 0) { | |
4406 | if (ret != -ENOENT) | |
4407 | inode = ERR_PTR(ret); | |
4408 | else | |
4409 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4410 | } else { | |
73f73415 | 4411 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4412 | } |
76dda93c YZ |
4413 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4414 | ||
34d19bad | 4415 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4416 | down_read(&root->fs_info->cleanup_work_sem); |
4417 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4418 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4419 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4420 | if (ret) |
4421 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4422 | } |
4423 | ||
3de4586c CM |
4424 | return inode; |
4425 | } | |
4426 | ||
fe15ce44 | 4427 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4428 | { |
4429 | struct btrfs_root *root; | |
848cce0d | 4430 | struct inode *inode = dentry->d_inode; |
76dda93c | 4431 | |
848cce0d LZ |
4432 | if (!inode && !IS_ROOT(dentry)) |
4433 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4434 | |
848cce0d LZ |
4435 | if (inode) { |
4436 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4437 | if (btrfs_root_refs(&root->root_item) == 0) |
4438 | return 1; | |
848cce0d LZ |
4439 | |
4440 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4441 | return 1; | |
efefb143 | 4442 | } |
76dda93c YZ |
4443 | return 0; |
4444 | } | |
4445 | ||
b4aff1f8 JB |
4446 | static void btrfs_dentry_release(struct dentry *dentry) |
4447 | { | |
4448 | if (dentry->d_fsdata) | |
4449 | kfree(dentry->d_fsdata); | |
4450 | } | |
4451 | ||
3de4586c | 4452 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4453 | unsigned int flags) |
3de4586c | 4454 | { |
a66e7cc6 JB |
4455 | struct dentry *ret; |
4456 | ||
4457 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4458 | if (unlikely(d_need_lookup(dentry))) { | |
4459 | spin_lock(&dentry->d_lock); | |
4460 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4461 | spin_unlock(&dentry->d_lock); | |
4462 | } | |
4463 | return ret; | |
39279cc3 CM |
4464 | } |
4465 | ||
16cdcec7 | 4466 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4467 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4468 | }; | |
4469 | ||
cbdf5a24 DW |
4470 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4471 | filldir_t filldir) | |
39279cc3 | 4472 | { |
6da6abae | 4473 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4474 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4475 | struct btrfs_item *item; | |
4476 | struct btrfs_dir_item *di; | |
4477 | struct btrfs_key key; | |
5f39d397 | 4478 | struct btrfs_key found_key; |
39279cc3 | 4479 | struct btrfs_path *path; |
16cdcec7 MX |
4480 | struct list_head ins_list; |
4481 | struct list_head del_list; | |
39279cc3 | 4482 | int ret; |
5f39d397 | 4483 | struct extent_buffer *leaf; |
39279cc3 | 4484 | int slot; |
39279cc3 CM |
4485 | unsigned char d_type; |
4486 | int over = 0; | |
4487 | u32 di_cur; | |
4488 | u32 di_total; | |
4489 | u32 di_len; | |
4490 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4491 | char tmp_name[32]; |
4492 | char *name_ptr; | |
4493 | int name_len; | |
16cdcec7 | 4494 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4495 | |
4496 | /* FIXME, use a real flag for deciding about the key type */ | |
4497 | if (root->fs_info->tree_root == root) | |
4498 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4499 | |
3954401f CM |
4500 | /* special case for "." */ |
4501 | if (filp->f_pos == 0) { | |
3765fefa HS |
4502 | over = filldir(dirent, ".", 1, |
4503 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4504 | if (over) |
4505 | return 0; | |
4506 | filp->f_pos = 1; | |
4507 | } | |
3954401f CM |
4508 | /* special case for .., just use the back ref */ |
4509 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4510 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4511 | over = filldir(dirent, "..", 2, |
3765fefa | 4512 | filp->f_pos, pino, DT_DIR); |
3954401f | 4513 | if (over) |
49593bfa | 4514 | return 0; |
3954401f CM |
4515 | filp->f_pos = 2; |
4516 | } | |
49593bfa | 4517 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4518 | if (!path) |
4519 | return -ENOMEM; | |
ff5714cc | 4520 | |
026fd317 | 4521 | path->reada = 1; |
49593bfa | 4522 | |
16cdcec7 MX |
4523 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4524 | INIT_LIST_HEAD(&ins_list); | |
4525 | INIT_LIST_HEAD(&del_list); | |
4526 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4527 | } | |
4528 | ||
39279cc3 CM |
4529 | btrfs_set_key_type(&key, key_type); |
4530 | key.offset = filp->f_pos; | |
33345d01 | 4531 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4532 | |
39279cc3 CM |
4533 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4534 | if (ret < 0) | |
4535 | goto err; | |
49593bfa DW |
4536 | |
4537 | while (1) { | |
5f39d397 | 4538 | leaf = path->nodes[0]; |
39279cc3 | 4539 | slot = path->slots[0]; |
b9e03af0 LZ |
4540 | if (slot >= btrfs_header_nritems(leaf)) { |
4541 | ret = btrfs_next_leaf(root, path); | |
4542 | if (ret < 0) | |
4543 | goto err; | |
4544 | else if (ret > 0) | |
4545 | break; | |
4546 | continue; | |
39279cc3 | 4547 | } |
3de4586c | 4548 | |
5f39d397 CM |
4549 | item = btrfs_item_nr(leaf, slot); |
4550 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4551 | ||
4552 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4553 | break; |
5f39d397 | 4554 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4555 | break; |
5f39d397 | 4556 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4557 | goto next; |
16cdcec7 MX |
4558 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4559 | btrfs_should_delete_dir_index(&del_list, | |
4560 | found_key.offset)) | |
4561 | goto next; | |
5f39d397 CM |
4562 | |
4563 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4564 | is_curr = 1; |
49593bfa | 4565 | |
39279cc3 CM |
4566 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4567 | di_cur = 0; | |
5f39d397 | 4568 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4569 | |
4570 | while (di_cur < di_total) { | |
5f39d397 CM |
4571 | struct btrfs_key location; |
4572 | ||
22a94d44 JB |
4573 | if (verify_dir_item(root, leaf, di)) |
4574 | break; | |
4575 | ||
5f39d397 | 4576 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4577 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4578 | name_ptr = tmp_name; |
4579 | } else { | |
4580 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4581 | if (!name_ptr) { |
4582 | ret = -ENOMEM; | |
4583 | goto err; | |
4584 | } | |
5f39d397 CM |
4585 | } |
4586 | read_extent_buffer(leaf, name_ptr, | |
4587 | (unsigned long)(di + 1), name_len); | |
4588 | ||
4589 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4590 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4591 | |
fede766f | 4592 | |
3de4586c | 4593 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4594 | * skip it. |
4595 | * | |
4596 | * In contrast to old kernels, we insert the snapshot's | |
4597 | * dir item and dir index after it has been created, so | |
4598 | * we won't find a reference to our own snapshot. We | |
4599 | * still keep the following code for backward | |
4600 | * compatibility. | |
3de4586c CM |
4601 | */ |
4602 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4603 | location.objectid == root->root_key.objectid) { | |
4604 | over = 0; | |
4605 | goto skip; | |
4606 | } | |
5f39d397 | 4607 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4608 | found_key.offset, location.objectid, |
39279cc3 | 4609 | d_type); |
5f39d397 | 4610 | |
3de4586c | 4611 | skip: |
5f39d397 CM |
4612 | if (name_ptr != tmp_name) |
4613 | kfree(name_ptr); | |
4614 | ||
39279cc3 CM |
4615 | if (over) |
4616 | goto nopos; | |
5103e947 | 4617 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4618 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4619 | di_cur += di_len; |
4620 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4621 | } | |
b9e03af0 LZ |
4622 | next: |
4623 | path->slots[0]++; | |
39279cc3 | 4624 | } |
49593bfa | 4625 | |
16cdcec7 MX |
4626 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4627 | if (is_curr) | |
4628 | filp->f_pos++; | |
4629 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4630 | &ins_list); | |
4631 | if (ret) | |
4632 | goto nopos; | |
4633 | } | |
4634 | ||
49593bfa | 4635 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4636 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4637 | /* |
4638 | * 32-bit glibc will use getdents64, but then strtol - | |
4639 | * so the last number we can serve is this. | |
4640 | */ | |
4641 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4642 | else |
4643 | filp->f_pos++; | |
39279cc3 CM |
4644 | nopos: |
4645 | ret = 0; | |
4646 | err: | |
16cdcec7 MX |
4647 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4648 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4649 | btrfs_free_path(path); |
39279cc3 CM |
4650 | return ret; |
4651 | } | |
4652 | ||
a9185b41 | 4653 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4654 | { |
4655 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4656 | struct btrfs_trans_handle *trans; | |
4657 | int ret = 0; | |
0af3d00b | 4658 | bool nolock = false; |
39279cc3 | 4659 | |
72ac3c0d | 4660 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4661 | return 0; |
4662 | ||
83eea1f1 | 4663 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 4664 | nolock = true; |
0af3d00b | 4665 | |
a9185b41 | 4666 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4667 | if (nolock) |
7a7eaa40 | 4668 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4669 | else |
7a7eaa40 | 4670 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4671 | if (IS_ERR(trans)) |
4672 | return PTR_ERR(trans); | |
a698d075 | 4673 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
4674 | } |
4675 | return ret; | |
4676 | } | |
4677 | ||
4678 | /* | |
54aa1f4d | 4679 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4680 | * inode changes. But, it is most likely to find the inode in cache. |
4681 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4682 | * to keep or drop this code. | |
4683 | */ | |
22c44fe6 | 4684 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4685 | { |
4686 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4687 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4688 | int ret; |
4689 | ||
72ac3c0d | 4690 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4691 | return 0; |
39279cc3 | 4692 | |
7a7eaa40 | 4693 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4694 | if (IS_ERR(trans)) |
4695 | return PTR_ERR(trans); | |
8929ecfa YZ |
4696 | |
4697 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4698 | if (ret && ret == -ENOSPC) { |
4699 | /* whoops, lets try again with the full transaction */ | |
4700 | btrfs_end_transaction(trans, root); | |
4701 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4702 | if (IS_ERR(trans)) |
4703 | return PTR_ERR(trans); | |
8929ecfa | 4704 | |
94b60442 | 4705 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4706 | } |
39279cc3 | 4707 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4708 | if (BTRFS_I(inode)->delayed_node) |
4709 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4710 | |
4711 | return ret; | |
4712 | } | |
4713 | ||
4714 | /* | |
4715 | * This is a copy of file_update_time. We need this so we can return error on | |
4716 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4717 | */ | |
e41f941a JB |
4718 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
4719 | int flags) | |
22c44fe6 | 4720 | { |
2bc55652 AB |
4721 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4722 | ||
4723 | if (btrfs_root_readonly(root)) | |
4724 | return -EROFS; | |
4725 | ||
e41f941a | 4726 | if (flags & S_VERSION) |
22c44fe6 | 4727 | inode_inc_iversion(inode); |
e41f941a JB |
4728 | if (flags & S_CTIME) |
4729 | inode->i_ctime = *now; | |
4730 | if (flags & S_MTIME) | |
4731 | inode->i_mtime = *now; | |
4732 | if (flags & S_ATIME) | |
4733 | inode->i_atime = *now; | |
4734 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
4735 | } |
4736 | ||
d352ac68 CM |
4737 | /* |
4738 | * find the highest existing sequence number in a directory | |
4739 | * and then set the in-memory index_cnt variable to reflect | |
4740 | * free sequence numbers | |
4741 | */ | |
aec7477b JB |
4742 | static int btrfs_set_inode_index_count(struct inode *inode) |
4743 | { | |
4744 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4745 | struct btrfs_key key, found_key; | |
4746 | struct btrfs_path *path; | |
4747 | struct extent_buffer *leaf; | |
4748 | int ret; | |
4749 | ||
33345d01 | 4750 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4751 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4752 | key.offset = (u64)-1; | |
4753 | ||
4754 | path = btrfs_alloc_path(); | |
4755 | if (!path) | |
4756 | return -ENOMEM; | |
4757 | ||
4758 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4759 | if (ret < 0) | |
4760 | goto out; | |
4761 | /* FIXME: we should be able to handle this */ | |
4762 | if (ret == 0) | |
4763 | goto out; | |
4764 | ret = 0; | |
4765 | ||
4766 | /* | |
4767 | * MAGIC NUMBER EXPLANATION: | |
4768 | * since we search a directory based on f_pos we have to start at 2 | |
4769 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4770 | * else has to start at 2 | |
4771 | */ | |
4772 | if (path->slots[0] == 0) { | |
4773 | BTRFS_I(inode)->index_cnt = 2; | |
4774 | goto out; | |
4775 | } | |
4776 | ||
4777 | path->slots[0]--; | |
4778 | ||
4779 | leaf = path->nodes[0]; | |
4780 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4781 | ||
33345d01 | 4782 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4783 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4784 | BTRFS_I(inode)->index_cnt = 2; | |
4785 | goto out; | |
4786 | } | |
4787 | ||
4788 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4789 | out: | |
4790 | btrfs_free_path(path); | |
4791 | return ret; | |
4792 | } | |
4793 | ||
d352ac68 CM |
4794 | /* |
4795 | * helper to find a free sequence number in a given directory. This current | |
4796 | * code is very simple, later versions will do smarter things in the btree | |
4797 | */ | |
3de4586c | 4798 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4799 | { |
4800 | int ret = 0; | |
4801 | ||
4802 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4803 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4804 | if (ret) { | |
4805 | ret = btrfs_set_inode_index_count(dir); | |
4806 | if (ret) | |
4807 | return ret; | |
4808 | } | |
aec7477b JB |
4809 | } |
4810 | ||
00e4e6b3 | 4811 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4812 | BTRFS_I(dir)->index_cnt++; |
4813 | ||
4814 | return ret; | |
4815 | } | |
4816 | ||
39279cc3 CM |
4817 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4818 | struct btrfs_root *root, | |
aec7477b | 4819 | struct inode *dir, |
9c58309d | 4820 | const char *name, int name_len, |
175a4eb7 AV |
4821 | u64 ref_objectid, u64 objectid, |
4822 | umode_t mode, u64 *index) | |
39279cc3 CM |
4823 | { |
4824 | struct inode *inode; | |
5f39d397 | 4825 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4826 | struct btrfs_key *location; |
5f39d397 | 4827 | struct btrfs_path *path; |
9c58309d CM |
4828 | struct btrfs_inode_ref *ref; |
4829 | struct btrfs_key key[2]; | |
4830 | u32 sizes[2]; | |
4831 | unsigned long ptr; | |
39279cc3 CM |
4832 | int ret; |
4833 | int owner; | |
4834 | ||
5f39d397 | 4835 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4836 | if (!path) |
4837 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4838 | |
39279cc3 | 4839 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4840 | if (!inode) { |
4841 | btrfs_free_path(path); | |
39279cc3 | 4842 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4843 | } |
39279cc3 | 4844 | |
581bb050 LZ |
4845 | /* |
4846 | * we have to initialize this early, so we can reclaim the inode | |
4847 | * number if we fail afterwards in this function. | |
4848 | */ | |
4849 | inode->i_ino = objectid; | |
4850 | ||
aec7477b | 4851 | if (dir) { |
1abe9b8a | 4852 | trace_btrfs_inode_request(dir); |
4853 | ||
3de4586c | 4854 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4855 | if (ret) { |
8fb27640 | 4856 | btrfs_free_path(path); |
09771430 | 4857 | iput(inode); |
aec7477b | 4858 | return ERR_PTR(ret); |
09771430 | 4859 | } |
aec7477b JB |
4860 | } |
4861 | /* | |
4862 | * index_cnt is ignored for everything but a dir, | |
4863 | * btrfs_get_inode_index_count has an explanation for the magic | |
4864 | * number | |
4865 | */ | |
4866 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4867 | BTRFS_I(inode)->root = root; |
e02119d5 | 4868 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4869 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 4870 | |
5dc562c5 JB |
4871 | /* |
4872 | * We could have gotten an inode number from somebody who was fsynced | |
4873 | * and then removed in this same transaction, so let's just set full | |
4874 | * sync since it will be a full sync anyway and this will blow away the | |
4875 | * old info in the log. | |
4876 | */ | |
4877 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
4878 | ||
569254b0 | 4879 | if (S_ISDIR(mode)) |
39279cc3 CM |
4880 | owner = 0; |
4881 | else | |
4882 | owner = 1; | |
9c58309d CM |
4883 | |
4884 | key[0].objectid = objectid; | |
4885 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4886 | key[0].offset = 0; | |
4887 | ||
f186373f MF |
4888 | /* |
4889 | * Start new inodes with an inode_ref. This is slightly more | |
4890 | * efficient for small numbers of hard links since they will | |
4891 | * be packed into one item. Extended refs will kick in if we | |
4892 | * add more hard links than can fit in the ref item. | |
4893 | */ | |
9c58309d CM |
4894 | key[1].objectid = objectid; |
4895 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4896 | key[1].offset = ref_objectid; | |
4897 | ||
4898 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4899 | sizes[1] = name_len + sizeof(*ref); | |
4900 | ||
b9473439 | 4901 | path->leave_spinning = 1; |
9c58309d CM |
4902 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4903 | if (ret != 0) | |
5f39d397 CM |
4904 | goto fail; |
4905 | ||
ecc11fab | 4906 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4907 | inode_set_bytes(inode, 0); |
39279cc3 | 4908 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4909 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4910 | struct btrfs_inode_item); | |
293f7e07 LZ |
4911 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
4912 | sizeof(*inode_item)); | |
e02119d5 | 4913 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4914 | |
4915 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4916 | struct btrfs_inode_ref); | |
4917 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4918 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4919 | ptr = (unsigned long)(ref + 1); |
4920 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4921 | ||
5f39d397 CM |
4922 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4923 | btrfs_free_path(path); | |
4924 | ||
39279cc3 CM |
4925 | location = &BTRFS_I(inode)->location; |
4926 | location->objectid = objectid; | |
39279cc3 CM |
4927 | location->offset = 0; |
4928 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4929 | ||
6cbff00f CH |
4930 | btrfs_inherit_iflags(inode, dir); |
4931 | ||
569254b0 | 4932 | if (S_ISREG(mode)) { |
94272164 CM |
4933 | if (btrfs_test_opt(root, NODATASUM)) |
4934 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 4935 | if (btrfs_test_opt(root, NODATACOW)) |
94272164 CM |
4936 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4937 | } | |
4938 | ||
39279cc3 | 4939 | insert_inode_hash(inode); |
5d4f98a2 | 4940 | inode_tree_add(inode); |
1abe9b8a | 4941 | |
4942 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4943 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4944 | |
8ea05e3a AB |
4945 | btrfs_update_root_times(trans, root); |
4946 | ||
39279cc3 | 4947 | return inode; |
5f39d397 | 4948 | fail: |
aec7477b JB |
4949 | if (dir) |
4950 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4951 | btrfs_free_path(path); |
09771430 | 4952 | iput(inode); |
5f39d397 | 4953 | return ERR_PTR(ret); |
39279cc3 CM |
4954 | } |
4955 | ||
4956 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4957 | { | |
4958 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4959 | } | |
4960 | ||
d352ac68 CM |
4961 | /* |
4962 | * utility function to add 'inode' into 'parent_inode' with | |
4963 | * a give name and a given sequence number. | |
4964 | * if 'add_backref' is true, also insert a backref from the | |
4965 | * inode to the parent directory. | |
4966 | */ | |
e02119d5 CM |
4967 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4968 | struct inode *parent_inode, struct inode *inode, | |
4969 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4970 | { |
4df27c4d | 4971 | int ret = 0; |
39279cc3 | 4972 | struct btrfs_key key; |
e02119d5 | 4973 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4974 | u64 ino = btrfs_ino(inode); |
4975 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4976 | |
33345d01 | 4977 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4978 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4979 | } else { | |
33345d01 | 4980 | key.objectid = ino; |
4df27c4d YZ |
4981 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4982 | key.offset = 0; | |
4983 | } | |
4984 | ||
33345d01 | 4985 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4986 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4987 | key.objectid, root->root_key.objectid, | |
33345d01 | 4988 | parent_ino, index, name, name_len); |
4df27c4d | 4989 | } else if (add_backref) { |
33345d01 LZ |
4990 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4991 | parent_ino, index); | |
4df27c4d | 4992 | } |
39279cc3 | 4993 | |
79787eaa JM |
4994 | /* Nothing to clean up yet */ |
4995 | if (ret) | |
4996 | return ret; | |
4df27c4d | 4997 | |
79787eaa JM |
4998 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4999 | parent_inode, &key, | |
5000 | btrfs_inode_type(inode), index); | |
9c52057c | 5001 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5002 | goto fail_dir_item; |
5003 | else if (ret) { | |
5004 | btrfs_abort_transaction(trans, root, ret); | |
5005 | return ret; | |
39279cc3 | 5006 | } |
79787eaa JM |
5007 | |
5008 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5009 | name_len * 2); | |
0c4d2d95 | 5010 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5011 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5012 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5013 | if (ret) | |
5014 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5015 | return ret; |
fe66a05a CM |
5016 | |
5017 | fail_dir_item: | |
5018 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5019 | u64 local_index; | |
5020 | int err; | |
5021 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5022 | key.objectid, root->root_key.objectid, | |
5023 | parent_ino, &local_index, name, name_len); | |
5024 | ||
5025 | } else if (add_backref) { | |
5026 | u64 local_index; | |
5027 | int err; | |
5028 | ||
5029 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5030 | ino, parent_ino, &local_index); | |
5031 | } | |
5032 | return ret; | |
39279cc3 CM |
5033 | } |
5034 | ||
5035 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5036 | struct inode *dir, struct dentry *dentry, |
5037 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5038 | { |
a1b075d2 JB |
5039 | int err = btrfs_add_link(trans, dir, inode, |
5040 | dentry->d_name.name, dentry->d_name.len, | |
5041 | backref, index); | |
39279cc3 CM |
5042 | if (err > 0) |
5043 | err = -EEXIST; | |
5044 | return err; | |
5045 | } | |
5046 | ||
618e21d5 | 5047 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5048 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5049 | { |
5050 | struct btrfs_trans_handle *trans; | |
5051 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5052 | struct inode *inode = NULL; |
618e21d5 JB |
5053 | int err; |
5054 | int drop_inode = 0; | |
5055 | u64 objectid; | |
00e4e6b3 | 5056 | u64 index = 0; |
618e21d5 JB |
5057 | |
5058 | if (!new_valid_dev(rdev)) | |
5059 | return -EINVAL; | |
5060 | ||
9ed74f2d JB |
5061 | /* |
5062 | * 2 for inode item and ref | |
5063 | * 2 for dir items | |
5064 | * 1 for xattr if selinux is on | |
5065 | */ | |
a22285a6 YZ |
5066 | trans = btrfs_start_transaction(root, 5); |
5067 | if (IS_ERR(trans)) | |
5068 | return PTR_ERR(trans); | |
1832a6d5 | 5069 | |
581bb050 LZ |
5070 | err = btrfs_find_free_ino(root, &objectid); |
5071 | if (err) | |
5072 | goto out_unlock; | |
5073 | ||
aec7477b | 5074 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5075 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5076 | mode, &index); |
7cf96da3 TI |
5077 | if (IS_ERR(inode)) { |
5078 | err = PTR_ERR(inode); | |
618e21d5 | 5079 | goto out_unlock; |
7cf96da3 | 5080 | } |
618e21d5 | 5081 | |
2a7dba39 | 5082 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5083 | if (err) { |
5084 | drop_inode = 1; | |
5085 | goto out_unlock; | |
5086 | } | |
5087 | ||
ad19db71 CS |
5088 | /* |
5089 | * If the active LSM wants to access the inode during | |
5090 | * d_instantiate it needs these. Smack checks to see | |
5091 | * if the filesystem supports xattrs by looking at the | |
5092 | * ops vector. | |
5093 | */ | |
5094 | ||
5095 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5096 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5097 | if (err) |
5098 | drop_inode = 1; | |
5099 | else { | |
618e21d5 | 5100 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5101 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5102 | d_instantiate(dentry, inode); |
618e21d5 | 5103 | } |
618e21d5 | 5104 | out_unlock: |
7ad85bb7 | 5105 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5106 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5107 | if (drop_inode) { |
5108 | inode_dec_link_count(inode); | |
5109 | iput(inode); | |
5110 | } | |
618e21d5 JB |
5111 | return err; |
5112 | } | |
5113 | ||
39279cc3 | 5114 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5115 | umode_t mode, bool excl) |
39279cc3 CM |
5116 | { |
5117 | struct btrfs_trans_handle *trans; | |
5118 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5119 | struct inode *inode = NULL; |
43baa579 | 5120 | int drop_inode_on_err = 0; |
a22285a6 | 5121 | int err; |
39279cc3 | 5122 | u64 objectid; |
00e4e6b3 | 5123 | u64 index = 0; |
39279cc3 | 5124 | |
9ed74f2d JB |
5125 | /* |
5126 | * 2 for inode item and ref | |
5127 | * 2 for dir items | |
5128 | * 1 for xattr if selinux is on | |
5129 | */ | |
a22285a6 YZ |
5130 | trans = btrfs_start_transaction(root, 5); |
5131 | if (IS_ERR(trans)) | |
5132 | return PTR_ERR(trans); | |
9ed74f2d | 5133 | |
581bb050 LZ |
5134 | err = btrfs_find_free_ino(root, &objectid); |
5135 | if (err) | |
5136 | goto out_unlock; | |
5137 | ||
aec7477b | 5138 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5139 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5140 | mode, &index); |
7cf96da3 TI |
5141 | if (IS_ERR(inode)) { |
5142 | err = PTR_ERR(inode); | |
39279cc3 | 5143 | goto out_unlock; |
7cf96da3 | 5144 | } |
43baa579 | 5145 | drop_inode_on_err = 1; |
39279cc3 | 5146 | |
2a7dba39 | 5147 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5148 | if (err) |
33268eaf | 5149 | goto out_unlock; |
33268eaf | 5150 | |
9185aa58 FB |
5151 | err = btrfs_update_inode(trans, root, inode); |
5152 | if (err) | |
5153 | goto out_unlock; | |
5154 | ||
ad19db71 CS |
5155 | /* |
5156 | * If the active LSM wants to access the inode during | |
5157 | * d_instantiate it needs these. Smack checks to see | |
5158 | * if the filesystem supports xattrs by looking at the | |
5159 | * ops vector. | |
5160 | */ | |
5161 | inode->i_fop = &btrfs_file_operations; | |
5162 | inode->i_op = &btrfs_file_inode_operations; | |
5163 | ||
a1b075d2 | 5164 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5165 | if (err) |
43baa579 FB |
5166 | goto out_unlock; |
5167 | ||
5168 | inode->i_mapping->a_ops = &btrfs_aops; | |
5169 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5170 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5171 | d_instantiate(dentry, inode); | |
5172 | ||
39279cc3 | 5173 | out_unlock: |
7ad85bb7 | 5174 | btrfs_end_transaction(trans, root); |
43baa579 | 5175 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5176 | inode_dec_link_count(inode); |
5177 | iput(inode); | |
5178 | } | |
b53d3f5d | 5179 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5180 | return err; |
5181 | } | |
5182 | ||
5183 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5184 | struct dentry *dentry) | |
5185 | { | |
5186 | struct btrfs_trans_handle *trans; | |
5187 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5188 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5189 | u64 index; |
39279cc3 CM |
5190 | int err; |
5191 | int drop_inode = 0; | |
5192 | ||
4a8be425 TH |
5193 | /* do not allow sys_link's with other subvols of the same device */ |
5194 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5195 | return -EXDEV; |
4a8be425 | 5196 | |
f186373f | 5197 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5198 | return -EMLINK; |
4a8be425 | 5199 | |
3de4586c | 5200 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5201 | if (err) |
5202 | goto fail; | |
5203 | ||
a22285a6 | 5204 | /* |
7e6b6465 | 5205 | * 2 items for inode and inode ref |
a22285a6 | 5206 | * 2 items for dir items |
7e6b6465 | 5207 | * 1 item for parent inode |
a22285a6 | 5208 | */ |
7e6b6465 | 5209 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5210 | if (IS_ERR(trans)) { |
5211 | err = PTR_ERR(trans); | |
5212 | goto fail; | |
5213 | } | |
5f39d397 | 5214 | |
3153495d | 5215 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5216 | inode_inc_iversion(inode); |
3153495d | 5217 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5218 | ihold(inode); |
e9976151 | 5219 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5220 | |
a1b075d2 | 5221 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5222 | |
a5719521 | 5223 | if (err) { |
54aa1f4d | 5224 | drop_inode = 1; |
a5719521 | 5225 | } else { |
10d9f309 | 5226 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5227 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5228 | if (err) |
5229 | goto fail; | |
08c422c2 | 5230 | d_instantiate(dentry, inode); |
6a912213 | 5231 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5232 | } |
39279cc3 | 5233 | |
7ad85bb7 | 5234 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5235 | fail: |
39279cc3 CM |
5236 | if (drop_inode) { |
5237 | inode_dec_link_count(inode); | |
5238 | iput(inode); | |
5239 | } | |
b53d3f5d | 5240 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5241 | return err; |
5242 | } | |
5243 | ||
18bb1db3 | 5244 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5245 | { |
b9d86667 | 5246 | struct inode *inode = NULL; |
39279cc3 CM |
5247 | struct btrfs_trans_handle *trans; |
5248 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5249 | int err = 0; | |
5250 | int drop_on_err = 0; | |
b9d86667 | 5251 | u64 objectid = 0; |
00e4e6b3 | 5252 | u64 index = 0; |
39279cc3 | 5253 | |
9ed74f2d JB |
5254 | /* |
5255 | * 2 items for inode and ref | |
5256 | * 2 items for dir items | |
5257 | * 1 for xattr if selinux is on | |
5258 | */ | |
a22285a6 YZ |
5259 | trans = btrfs_start_transaction(root, 5); |
5260 | if (IS_ERR(trans)) | |
5261 | return PTR_ERR(trans); | |
39279cc3 | 5262 | |
581bb050 LZ |
5263 | err = btrfs_find_free_ino(root, &objectid); |
5264 | if (err) | |
5265 | goto out_fail; | |
5266 | ||
aec7477b | 5267 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5268 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5269 | S_IFDIR | mode, &index); |
39279cc3 CM |
5270 | if (IS_ERR(inode)) { |
5271 | err = PTR_ERR(inode); | |
5272 | goto out_fail; | |
5273 | } | |
5f39d397 | 5274 | |
39279cc3 | 5275 | drop_on_err = 1; |
33268eaf | 5276 | |
2a7dba39 | 5277 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5278 | if (err) |
5279 | goto out_fail; | |
5280 | ||
39279cc3 CM |
5281 | inode->i_op = &btrfs_dir_inode_operations; |
5282 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5283 | |
dbe674a9 | 5284 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5285 | err = btrfs_update_inode(trans, root, inode); |
5286 | if (err) | |
5287 | goto out_fail; | |
5f39d397 | 5288 | |
a1b075d2 JB |
5289 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5290 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5291 | if (err) |
5292 | goto out_fail; | |
5f39d397 | 5293 | |
39279cc3 CM |
5294 | d_instantiate(dentry, inode); |
5295 | drop_on_err = 0; | |
39279cc3 CM |
5296 | |
5297 | out_fail: | |
7ad85bb7 | 5298 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5299 | if (drop_on_err) |
5300 | iput(inode); | |
b53d3f5d | 5301 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5302 | return err; |
5303 | } | |
5304 | ||
d352ac68 CM |
5305 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5306 | * and an extent that you want to insert, deal with overlap and insert | |
5307 | * the new extent into the tree. | |
5308 | */ | |
3b951516 CM |
5309 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5310 | struct extent_map *existing, | |
e6dcd2dc CM |
5311 | struct extent_map *em, |
5312 | u64 map_start, u64 map_len) | |
3b951516 CM |
5313 | { |
5314 | u64 start_diff; | |
3b951516 | 5315 | |
e6dcd2dc CM |
5316 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5317 | start_diff = map_start - em->start; | |
5318 | em->start = map_start; | |
5319 | em->len = map_len; | |
c8b97818 CM |
5320 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5321 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5322 | em->block_start += start_diff; |
c8b97818 CM |
5323 | em->block_len -= start_diff; |
5324 | } | |
e6dcd2dc | 5325 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5326 | } |
5327 | ||
c8b97818 CM |
5328 | static noinline int uncompress_inline(struct btrfs_path *path, |
5329 | struct inode *inode, struct page *page, | |
5330 | size_t pg_offset, u64 extent_offset, | |
5331 | struct btrfs_file_extent_item *item) | |
5332 | { | |
5333 | int ret; | |
5334 | struct extent_buffer *leaf = path->nodes[0]; | |
5335 | char *tmp; | |
5336 | size_t max_size; | |
5337 | unsigned long inline_size; | |
5338 | unsigned long ptr; | |
261507a0 | 5339 | int compress_type; |
c8b97818 CM |
5340 | |
5341 | WARN_ON(pg_offset != 0); | |
261507a0 | 5342 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5343 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5344 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5345 | btrfs_item_nr(leaf, path->slots[0])); | |
5346 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5347 | if (!tmp) |
5348 | return -ENOMEM; | |
c8b97818 CM |
5349 | ptr = btrfs_file_extent_inline_start(item); |
5350 | ||
5351 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5352 | ||
5b050f04 | 5353 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5354 | ret = btrfs_decompress(compress_type, tmp, page, |
5355 | extent_offset, inline_size, max_size); | |
c8b97818 | 5356 | if (ret) { |
7ac687d9 | 5357 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5358 | unsigned long copy_size = min_t(u64, |
5359 | PAGE_CACHE_SIZE - pg_offset, | |
5360 | max_size - extent_offset); | |
5361 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5362 | kunmap_atomic(kaddr); |
c8b97818 CM |
5363 | } |
5364 | kfree(tmp); | |
5365 | return 0; | |
5366 | } | |
5367 | ||
d352ac68 CM |
5368 | /* |
5369 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5370 | * the ugly parts come from merging extents from the disk with the in-ram |
5371 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5372 | * where the in-ram extents might be locked pending data=ordered completion. |
5373 | * | |
5374 | * This also copies inline extents directly into the page. | |
5375 | */ | |
d397712b | 5376 | |
a52d9a80 | 5377 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5378 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5379 | int create) |
5380 | { | |
5381 | int ret; | |
5382 | int err = 0; | |
db94535d | 5383 | u64 bytenr; |
a52d9a80 CM |
5384 | u64 extent_start = 0; |
5385 | u64 extent_end = 0; | |
33345d01 | 5386 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5387 | u32 found_type; |
f421950f | 5388 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5389 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5390 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5391 | struct extent_buffer *leaf; |
5392 | struct btrfs_key found_key; | |
a52d9a80 CM |
5393 | struct extent_map *em = NULL; |
5394 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5395 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5396 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5397 | int compress_type; |
a52d9a80 | 5398 | |
a52d9a80 | 5399 | again: |
890871be | 5400 | read_lock(&em_tree->lock); |
d1310b2e | 5401 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5402 | if (em) |
5403 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5404 | read_unlock(&em_tree->lock); |
d1310b2e | 5405 | |
a52d9a80 | 5406 | if (em) { |
e1c4b745 CM |
5407 | if (em->start > start || em->start + em->len <= start) |
5408 | free_extent_map(em); | |
5409 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5410 | free_extent_map(em); |
5411 | else | |
5412 | goto out; | |
a52d9a80 | 5413 | } |
172ddd60 | 5414 | em = alloc_extent_map(); |
a52d9a80 | 5415 | if (!em) { |
d1310b2e CM |
5416 | err = -ENOMEM; |
5417 | goto out; | |
a52d9a80 | 5418 | } |
e6dcd2dc | 5419 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5420 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5421 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5422 | em->len = (u64)-1; |
c8b97818 | 5423 | em->block_len = (u64)-1; |
f421950f CM |
5424 | |
5425 | if (!path) { | |
5426 | path = btrfs_alloc_path(); | |
026fd317 JB |
5427 | if (!path) { |
5428 | err = -ENOMEM; | |
5429 | goto out; | |
5430 | } | |
5431 | /* | |
5432 | * Chances are we'll be called again, so go ahead and do | |
5433 | * readahead | |
5434 | */ | |
5435 | path->reada = 1; | |
f421950f CM |
5436 | } |
5437 | ||
179e29e4 CM |
5438 | ret = btrfs_lookup_file_extent(trans, root, path, |
5439 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5440 | if (ret < 0) { |
5441 | err = ret; | |
5442 | goto out; | |
5443 | } | |
5444 | ||
5445 | if (ret != 0) { | |
5446 | if (path->slots[0] == 0) | |
5447 | goto not_found; | |
5448 | path->slots[0]--; | |
5449 | } | |
5450 | ||
5f39d397 CM |
5451 | leaf = path->nodes[0]; |
5452 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5453 | struct btrfs_file_extent_item); |
a52d9a80 | 5454 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5455 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5456 | found_type = btrfs_key_type(&found_key); | |
5457 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5458 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5459 | goto not_found; | |
5460 | } | |
5461 | ||
5f39d397 CM |
5462 | found_type = btrfs_file_extent_type(leaf, item); |
5463 | extent_start = found_key.offset; | |
261507a0 | 5464 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5465 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5466 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5467 | extent_end = extent_start + |
db94535d | 5468 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5469 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5470 | size_t size; | |
5471 | size = btrfs_file_extent_inline_len(leaf, item); | |
5472 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5473 | ~((u64)root->sectorsize - 1); | |
5474 | } | |
5475 | ||
5476 | if (start >= extent_end) { | |
5477 | path->slots[0]++; | |
5478 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5479 | ret = btrfs_next_leaf(root, path); | |
5480 | if (ret < 0) { | |
5481 | err = ret; | |
5482 | goto out; | |
a52d9a80 | 5483 | } |
9036c102 YZ |
5484 | if (ret > 0) |
5485 | goto not_found; | |
5486 | leaf = path->nodes[0]; | |
a52d9a80 | 5487 | } |
9036c102 YZ |
5488 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5489 | if (found_key.objectid != objectid || | |
5490 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5491 | goto not_found; | |
5492 | if (start + len <= found_key.offset) | |
5493 | goto not_found; | |
5494 | em->start = start; | |
70c8a91c | 5495 | em->orig_start = start; |
9036c102 YZ |
5496 | em->len = found_key.offset - start; |
5497 | goto not_found_em; | |
5498 | } | |
5499 | ||
d899e052 YZ |
5500 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5501 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5502 | em->start = extent_start; |
5503 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5504 | em->orig_start = extent_start - |
5505 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
5506 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
5507 | item); | |
db94535d CM |
5508 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5509 | if (bytenr == 0) { | |
5f39d397 | 5510 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5511 | goto insert; |
5512 | } | |
261507a0 | 5513 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5514 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5515 | em->compress_type = compress_type; |
c8b97818 | 5516 | em->block_start = bytenr; |
b4939680 | 5517 | em->block_len = em->orig_block_len; |
c8b97818 CM |
5518 | } else { |
5519 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5520 | em->block_start = bytenr; | |
5521 | em->block_len = em->len; | |
d899e052 YZ |
5522 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5523 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5524 | } |
a52d9a80 CM |
5525 | goto insert; |
5526 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5527 | unsigned long ptr; |
a52d9a80 | 5528 | char *map; |
3326d1b0 CM |
5529 | size_t size; |
5530 | size_t extent_offset; | |
5531 | size_t copy_size; | |
a52d9a80 | 5532 | |
689f9346 | 5533 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5534 | if (!page || create) { |
689f9346 | 5535 | em->start = extent_start; |
9036c102 | 5536 | em->len = extent_end - extent_start; |
689f9346 Y |
5537 | goto out; |
5538 | } | |
5f39d397 | 5539 | |
9036c102 YZ |
5540 | size = btrfs_file_extent_inline_len(leaf, item); |
5541 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5542 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5543 | size - extent_offset); |
3326d1b0 | 5544 | em->start = extent_start + extent_offset; |
70dec807 CM |
5545 | em->len = (copy_size + root->sectorsize - 1) & |
5546 | ~((u64)root->sectorsize - 1); | |
b4939680 | 5547 | em->orig_block_len = em->len; |
70c8a91c | 5548 | em->orig_start = em->start; |
261507a0 | 5549 | if (compress_type) { |
c8b97818 | 5550 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5551 | em->compress_type = compress_type; |
5552 | } | |
689f9346 | 5553 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5554 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5555 | if (btrfs_file_extent_compression(leaf, item) != |
5556 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5557 | ret = uncompress_inline(path, inode, page, |
5558 | pg_offset, | |
5559 | extent_offset, item); | |
79787eaa | 5560 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5561 | } else { |
5562 | map = kmap(page); | |
5563 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5564 | copy_size); | |
93c82d57 CM |
5565 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5566 | memset(map + pg_offset + copy_size, 0, | |
5567 | PAGE_CACHE_SIZE - pg_offset - | |
5568 | copy_size); | |
5569 | } | |
c8b97818 CM |
5570 | kunmap(page); |
5571 | } | |
179e29e4 CM |
5572 | flush_dcache_page(page); |
5573 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5574 | BUG(); |
179e29e4 CM |
5575 | if (!trans) { |
5576 | kunmap(page); | |
5577 | free_extent_map(em); | |
5578 | em = NULL; | |
ff5714cc | 5579 | |
b3b4aa74 | 5580 | btrfs_release_path(path); |
7a7eaa40 | 5581 | trans = btrfs_join_transaction(root); |
ff5714cc | 5582 | |
3612b495 TI |
5583 | if (IS_ERR(trans)) |
5584 | return ERR_CAST(trans); | |
179e29e4 CM |
5585 | goto again; |
5586 | } | |
c8b97818 | 5587 | map = kmap(page); |
70dec807 | 5588 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5589 | copy_size); |
c8b97818 | 5590 | kunmap(page); |
179e29e4 | 5591 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5592 | } |
d1310b2e | 5593 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5594 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5595 | goto insert; |
5596 | } else { | |
31b1a2bd | 5597 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5598 | } |
5599 | not_found: | |
5600 | em->start = start; | |
70c8a91c | 5601 | em->orig_start = start; |
d1310b2e | 5602 | em->len = len; |
a52d9a80 | 5603 | not_found_em: |
5f39d397 | 5604 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5605 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5606 | insert: |
b3b4aa74 | 5607 | btrfs_release_path(path); |
d1310b2e | 5608 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5609 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5610 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5611 | (unsigned long long)em->len, | |
5612 | (unsigned long long)start, | |
5613 | (unsigned long long)len); | |
a52d9a80 CM |
5614 | err = -EIO; |
5615 | goto out; | |
5616 | } | |
d1310b2e CM |
5617 | |
5618 | err = 0; | |
890871be | 5619 | write_lock(&em_tree->lock); |
a52d9a80 | 5620 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5621 | /* it is possible that someone inserted the extent into the tree |
5622 | * while we had the lock dropped. It is also possible that | |
5623 | * an overlapping map exists in the tree | |
5624 | */ | |
a52d9a80 | 5625 | if (ret == -EEXIST) { |
3b951516 | 5626 | struct extent_map *existing; |
e6dcd2dc CM |
5627 | |
5628 | ret = 0; | |
5629 | ||
3b951516 | 5630 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5631 | if (existing && (existing->start > start || |
5632 | existing->start + existing->len <= start)) { | |
5633 | free_extent_map(existing); | |
5634 | existing = NULL; | |
5635 | } | |
3b951516 CM |
5636 | if (!existing) { |
5637 | existing = lookup_extent_mapping(em_tree, em->start, | |
5638 | em->len); | |
5639 | if (existing) { | |
5640 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5641 | em, start, |
5642 | root->sectorsize); | |
3b951516 CM |
5643 | free_extent_map(existing); |
5644 | if (err) { | |
5645 | free_extent_map(em); | |
5646 | em = NULL; | |
5647 | } | |
5648 | } else { | |
5649 | err = -EIO; | |
3b951516 CM |
5650 | free_extent_map(em); |
5651 | em = NULL; | |
5652 | } | |
5653 | } else { | |
5654 | free_extent_map(em); | |
5655 | em = existing; | |
e6dcd2dc | 5656 | err = 0; |
a52d9a80 | 5657 | } |
a52d9a80 | 5658 | } |
890871be | 5659 | write_unlock(&em_tree->lock); |
a52d9a80 | 5660 | out: |
1abe9b8a | 5661 | |
f0bd95ea TI |
5662 | if (em) |
5663 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 5664 | |
f421950f CM |
5665 | if (path) |
5666 | btrfs_free_path(path); | |
a52d9a80 CM |
5667 | if (trans) { |
5668 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5669 | if (!err) |
a52d9a80 CM |
5670 | err = ret; |
5671 | } | |
a52d9a80 CM |
5672 | if (err) { |
5673 | free_extent_map(em); | |
a52d9a80 CM |
5674 | return ERR_PTR(err); |
5675 | } | |
79787eaa | 5676 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5677 | return em; |
5678 | } | |
5679 | ||
ec29ed5b CM |
5680 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5681 | size_t pg_offset, u64 start, u64 len, | |
5682 | int create) | |
5683 | { | |
5684 | struct extent_map *em; | |
5685 | struct extent_map *hole_em = NULL; | |
5686 | u64 range_start = start; | |
5687 | u64 end; | |
5688 | u64 found; | |
5689 | u64 found_end; | |
5690 | int err = 0; | |
5691 | ||
5692 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5693 | if (IS_ERR(em)) | |
5694 | return em; | |
5695 | if (em) { | |
5696 | /* | |
f9e4fb53 LB |
5697 | * if our em maps to |
5698 | * - a hole or | |
5699 | * - a pre-alloc extent, | |
5700 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 5701 | */ |
f9e4fb53 LB |
5702 | if (em->block_start != EXTENT_MAP_HOLE && |
5703 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
5704 | return em; |
5705 | else | |
5706 | hole_em = em; | |
5707 | } | |
5708 | ||
5709 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5710 | end = start + len; | |
5711 | if (end < start) | |
5712 | end = (u64)-1; | |
5713 | else | |
5714 | end -= 1; | |
5715 | ||
5716 | em = NULL; | |
5717 | ||
5718 | /* ok, we didn't find anything, lets look for delalloc */ | |
5719 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5720 | end, len, EXTENT_DELALLOC, 1); | |
5721 | found_end = range_start + found; | |
5722 | if (found_end < range_start) | |
5723 | found_end = (u64)-1; | |
5724 | ||
5725 | /* | |
5726 | * we didn't find anything useful, return | |
5727 | * the original results from get_extent() | |
5728 | */ | |
5729 | if (range_start > end || found_end <= start) { | |
5730 | em = hole_em; | |
5731 | hole_em = NULL; | |
5732 | goto out; | |
5733 | } | |
5734 | ||
5735 | /* adjust the range_start to make sure it doesn't | |
5736 | * go backwards from the start they passed in | |
5737 | */ | |
5738 | range_start = max(start,range_start); | |
5739 | found = found_end - range_start; | |
5740 | ||
5741 | if (found > 0) { | |
5742 | u64 hole_start = start; | |
5743 | u64 hole_len = len; | |
5744 | ||
172ddd60 | 5745 | em = alloc_extent_map(); |
ec29ed5b CM |
5746 | if (!em) { |
5747 | err = -ENOMEM; | |
5748 | goto out; | |
5749 | } | |
5750 | /* | |
5751 | * when btrfs_get_extent can't find anything it | |
5752 | * returns one huge hole | |
5753 | * | |
5754 | * make sure what it found really fits our range, and | |
5755 | * adjust to make sure it is based on the start from | |
5756 | * the caller | |
5757 | */ | |
5758 | if (hole_em) { | |
5759 | u64 calc_end = extent_map_end(hole_em); | |
5760 | ||
5761 | if (calc_end <= start || (hole_em->start > end)) { | |
5762 | free_extent_map(hole_em); | |
5763 | hole_em = NULL; | |
5764 | } else { | |
5765 | hole_start = max(hole_em->start, start); | |
5766 | hole_len = calc_end - hole_start; | |
5767 | } | |
5768 | } | |
5769 | em->bdev = NULL; | |
5770 | if (hole_em && range_start > hole_start) { | |
5771 | /* our hole starts before our delalloc, so we | |
5772 | * have to return just the parts of the hole | |
5773 | * that go until the delalloc starts | |
5774 | */ | |
5775 | em->len = min(hole_len, | |
5776 | range_start - hole_start); | |
5777 | em->start = hole_start; | |
5778 | em->orig_start = hole_start; | |
5779 | /* | |
5780 | * don't adjust block start at all, | |
5781 | * it is fixed at EXTENT_MAP_HOLE | |
5782 | */ | |
5783 | em->block_start = hole_em->block_start; | |
5784 | em->block_len = hole_len; | |
f9e4fb53 LB |
5785 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
5786 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
5787 | } else { |
5788 | em->start = range_start; | |
5789 | em->len = found; | |
5790 | em->orig_start = range_start; | |
5791 | em->block_start = EXTENT_MAP_DELALLOC; | |
5792 | em->block_len = found; | |
5793 | } | |
5794 | } else if (hole_em) { | |
5795 | return hole_em; | |
5796 | } | |
5797 | out: | |
5798 | ||
5799 | free_extent_map(hole_em); | |
5800 | if (err) { | |
5801 | free_extent_map(em); | |
5802 | return ERR_PTR(err); | |
5803 | } | |
5804 | return em; | |
5805 | } | |
5806 | ||
4b46fce2 JB |
5807 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
5808 | u64 start, u64 len) | |
5809 | { | |
5810 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5811 | struct btrfs_trans_handle *trans; | |
70c8a91c | 5812 | struct extent_map *em; |
4b46fce2 JB |
5813 | struct btrfs_key ins; |
5814 | u64 alloc_hint; | |
5815 | int ret; | |
4b46fce2 | 5816 | |
7a7eaa40 | 5817 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5818 | if (IS_ERR(trans)) |
5819 | return ERR_CAST(trans); | |
4b46fce2 JB |
5820 | |
5821 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5822 | ||
5823 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5824 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5825 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5826 | if (ret) { |
5827 | em = ERR_PTR(ret); | |
5828 | goto out; | |
5829 | } | |
5830 | ||
70c8a91c JB |
5831 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
5832 | ins.offset, ins.offset, 0); | |
5833 | if (IS_ERR(em)) | |
5834 | goto out; | |
4b46fce2 JB |
5835 | |
5836 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5837 | ins.offset, ins.offset, 0); | |
5838 | if (ret) { | |
5839 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5840 | em = ERR_PTR(ret); | |
5841 | } | |
5842 | out: | |
5843 | btrfs_end_transaction(trans, root); | |
5844 | return em; | |
5845 | } | |
5846 | ||
46bfbb5c CM |
5847 | /* |
5848 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5849 | * block must be cow'd | |
5850 | */ | |
5851 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5852 | struct inode *inode, u64 offset, u64 len) | |
5853 | { | |
5854 | struct btrfs_path *path; | |
5855 | int ret; | |
5856 | struct extent_buffer *leaf; | |
5857 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5858 | struct btrfs_file_extent_item *fi; | |
5859 | struct btrfs_key key; | |
5860 | u64 disk_bytenr; | |
5861 | u64 backref_offset; | |
5862 | u64 extent_end; | |
5863 | u64 num_bytes; | |
5864 | int slot; | |
5865 | int found_type; | |
5866 | ||
5867 | path = btrfs_alloc_path(); | |
5868 | if (!path) | |
5869 | return -ENOMEM; | |
5870 | ||
33345d01 | 5871 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5872 | offset, 0); |
5873 | if (ret < 0) | |
5874 | goto out; | |
5875 | ||
5876 | slot = path->slots[0]; | |
5877 | if (ret == 1) { | |
5878 | if (slot == 0) { | |
5879 | /* can't find the item, must cow */ | |
5880 | ret = 0; | |
5881 | goto out; | |
5882 | } | |
5883 | slot--; | |
5884 | } | |
5885 | ret = 0; | |
5886 | leaf = path->nodes[0]; | |
5887 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5888 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5889 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5890 | /* not our file or wrong item type, must cow */ | |
5891 | goto out; | |
5892 | } | |
5893 | ||
5894 | if (key.offset > offset) { | |
5895 | /* Wrong offset, must cow */ | |
5896 | goto out; | |
5897 | } | |
5898 | ||
5899 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5900 | found_type = btrfs_file_extent_type(leaf, fi); | |
5901 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5902 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5903 | /* not a regular extent, must cow */ | |
5904 | goto out; | |
5905 | } | |
5906 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5907 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5908 | ||
5909 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5910 | if (extent_end < offset + len) { | |
5911 | /* extent doesn't include our full range, must cow */ | |
5912 | goto out; | |
5913 | } | |
5914 | ||
5915 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5916 | goto out; | |
5917 | ||
5918 | /* | |
5919 | * look for other files referencing this extent, if we | |
5920 | * find any we must cow | |
5921 | */ | |
33345d01 | 5922 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5923 | key.offset - backref_offset, disk_bytenr)) |
5924 | goto out; | |
5925 | ||
5926 | /* | |
5927 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5928 | * in this extent we are about to write. If there | |
5929 | * are any csums in that range we have to cow in order | |
5930 | * to keep the csums correct | |
5931 | */ | |
5932 | disk_bytenr += backref_offset; | |
5933 | disk_bytenr += offset - key.offset; | |
5934 | num_bytes = min(offset + len, extent_end) - offset; | |
5935 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5936 | goto out; | |
5937 | /* | |
5938 | * all of the above have passed, it is safe to overwrite this extent | |
5939 | * without cow | |
5940 | */ | |
5941 | ret = 1; | |
5942 | out: | |
5943 | btrfs_free_path(path); | |
5944 | return ret; | |
5945 | } | |
5946 | ||
eb838e73 JB |
5947 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
5948 | struct extent_state **cached_state, int writing) | |
5949 | { | |
5950 | struct btrfs_ordered_extent *ordered; | |
5951 | int ret = 0; | |
5952 | ||
5953 | while (1) { | |
5954 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5955 | 0, cached_state); | |
5956 | /* | |
5957 | * We're concerned with the entire range that we're going to be | |
5958 | * doing DIO to, so we need to make sure theres no ordered | |
5959 | * extents in this range. | |
5960 | */ | |
5961 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
5962 | lockend - lockstart + 1); | |
5963 | ||
5964 | /* | |
5965 | * We need to make sure there are no buffered pages in this | |
5966 | * range either, we could have raced between the invalidate in | |
5967 | * generic_file_direct_write and locking the extent. The | |
5968 | * invalidate needs to happen so that reads after a write do not | |
5969 | * get stale data. | |
5970 | */ | |
5971 | if (!ordered && (!writing || | |
5972 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
5973 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
5974 | *cached_state))) | |
5975 | break; | |
5976 | ||
5977 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5978 | cached_state, GFP_NOFS); | |
5979 | ||
5980 | if (ordered) { | |
5981 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5982 | btrfs_put_ordered_extent(ordered); | |
5983 | } else { | |
5984 | /* Screw you mmap */ | |
5985 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
5986 | lockstart, | |
5987 | lockend); | |
5988 | if (ret) | |
5989 | break; | |
5990 | ||
5991 | /* | |
5992 | * If we found a page that couldn't be invalidated just | |
5993 | * fall back to buffered. | |
5994 | */ | |
5995 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5996 | lockstart >> PAGE_CACHE_SHIFT, | |
5997 | lockend >> PAGE_CACHE_SHIFT); | |
5998 | if (ret) | |
5999 | break; | |
6000 | } | |
6001 | ||
6002 | cond_resched(); | |
6003 | } | |
6004 | ||
6005 | return ret; | |
6006 | } | |
6007 | ||
69ffb543 JB |
6008 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
6009 | u64 len, u64 orig_start, | |
6010 | u64 block_start, u64 block_len, | |
b4939680 | 6011 | u64 orig_block_len, int type) |
69ffb543 JB |
6012 | { |
6013 | struct extent_map_tree *em_tree; | |
6014 | struct extent_map *em; | |
6015 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6016 | int ret; | |
6017 | ||
6018 | em_tree = &BTRFS_I(inode)->extent_tree; | |
6019 | em = alloc_extent_map(); | |
6020 | if (!em) | |
6021 | return ERR_PTR(-ENOMEM); | |
6022 | ||
6023 | em->start = start; | |
6024 | em->orig_start = orig_start; | |
2ab28f32 JB |
6025 | em->mod_start = start; |
6026 | em->mod_len = len; | |
69ffb543 JB |
6027 | em->len = len; |
6028 | em->block_len = block_len; | |
6029 | em->block_start = block_start; | |
6030 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6031 | em->orig_block_len = orig_block_len; |
70c8a91c | 6032 | em->generation = -1; |
69ffb543 JB |
6033 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6034 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6035 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6036 | |
6037 | do { | |
6038 | btrfs_drop_extent_cache(inode, em->start, | |
6039 | em->start + em->len - 1, 0); | |
6040 | write_lock(&em_tree->lock); | |
6041 | ret = add_extent_mapping(em_tree, em); | |
70c8a91c JB |
6042 | if (!ret) |
6043 | list_move(&em->list, | |
6044 | &em_tree->modified_extents); | |
69ffb543 JB |
6045 | write_unlock(&em_tree->lock); |
6046 | } while (ret == -EEXIST); | |
6047 | ||
6048 | if (ret) { | |
6049 | free_extent_map(em); | |
6050 | return ERR_PTR(ret); | |
6051 | } | |
6052 | ||
6053 | return em; | |
6054 | } | |
6055 | ||
6056 | ||
4b46fce2 JB |
6057 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6058 | struct buffer_head *bh_result, int create) | |
6059 | { | |
6060 | struct extent_map *em; | |
6061 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6062 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6063 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6064 | u64 lockstart, lockend; |
4b46fce2 | 6065 | u64 len = bh_result->b_size; |
46bfbb5c | 6066 | struct btrfs_trans_handle *trans; |
eb838e73 | 6067 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 6068 | int ret = 0; |
eb838e73 | 6069 | |
eb838e73 | 6070 | if (create) { |
0934856d MX |
6071 | spin_lock(&BTRFS_I(inode)->lock); |
6072 | BTRFS_I(inode)->outstanding_extents++; | |
6073 | spin_unlock(&BTRFS_I(inode)->lock); | |
eb838e73 | 6074 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
0934856d | 6075 | } else |
c329861d | 6076 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 6077 | |
c329861d JB |
6078 | lockstart = start; |
6079 | lockend = start + len - 1; | |
6080 | ||
eb838e73 JB |
6081 | /* |
6082 | * If this errors out it's because we couldn't invalidate pagecache for | |
6083 | * this range and we need to fallback to buffered. | |
6084 | */ | |
6085 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6086 | return -ENOTBLK; | |
6087 | ||
4b46fce2 | 6088 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
6089 | if (IS_ERR(em)) { |
6090 | ret = PTR_ERR(em); | |
6091 | goto unlock_err; | |
6092 | } | |
4b46fce2 JB |
6093 | |
6094 | /* | |
6095 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6096 | * io. INLINE is special, and we could probably kludge it in here, but | |
6097 | * it's still buffered so for safety lets just fall back to the generic | |
6098 | * buffered path. | |
6099 | * | |
6100 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6101 | * decompress it, so there will be buffering required no matter what we | |
6102 | * do, so go ahead and fallback to buffered. | |
6103 | * | |
6104 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6105 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6106 | * the generic code. | |
6107 | */ | |
6108 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6109 | em->block_start == EXTENT_MAP_INLINE) { | |
6110 | free_extent_map(em); | |
eb838e73 JB |
6111 | ret = -ENOTBLK; |
6112 | goto unlock_err; | |
4b46fce2 JB |
6113 | } |
6114 | ||
6115 | /* Just a good old fashioned hole, return */ | |
6116 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6117 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6118 | free_extent_map(em); | |
eb838e73 | 6119 | goto unlock_err; |
4b46fce2 JB |
6120 | } |
6121 | ||
6122 | /* | |
6123 | * We don't allocate a new extent in the following cases | |
6124 | * | |
6125 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6126 | * existing extent. | |
6127 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6128 | * just use the extent. | |
6129 | * | |
6130 | */ | |
46bfbb5c | 6131 | if (!create) { |
eb838e73 JB |
6132 | len = min(len, em->len - (start - em->start)); |
6133 | lockstart = start + len; | |
6134 | goto unlock; | |
46bfbb5c | 6135 | } |
4b46fce2 JB |
6136 | |
6137 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6138 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6139 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6140 | int type; |
6141 | int ret; | |
46bfbb5c | 6142 | u64 block_start; |
4b46fce2 JB |
6143 | |
6144 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6145 | type = BTRFS_ORDERED_PREALLOC; | |
6146 | else | |
6147 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6148 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6149 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6150 | |
6151 | /* | |
6152 | * we're not going to log anything, but we do need | |
6153 | * to make sure the current transaction stays open | |
6154 | * while we look for nocow cross refs | |
6155 | */ | |
7a7eaa40 | 6156 | trans = btrfs_join_transaction(root); |
3612b495 | 6157 | if (IS_ERR(trans)) |
46bfbb5c CM |
6158 | goto must_cow; |
6159 | ||
6160 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
70c8a91c | 6161 | u64 orig_start = em->orig_start; |
b4939680 | 6162 | u64 orig_block_len = em->orig_block_len; |
69ffb543 JB |
6163 | |
6164 | if (type == BTRFS_ORDERED_PREALLOC) { | |
6165 | free_extent_map(em); | |
6166 | em = create_pinned_em(inode, start, len, | |
6167 | orig_start, | |
b4939680 JB |
6168 | block_start, len, |
6169 | orig_block_len, type); | |
69ffb543 JB |
6170 | if (IS_ERR(em)) { |
6171 | btrfs_end_transaction(trans, root); | |
6172 | goto unlock_err; | |
6173 | } | |
6174 | } | |
6175 | ||
46bfbb5c CM |
6176 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6177 | block_start, len, len, type); | |
6178 | btrfs_end_transaction(trans, root); | |
6179 | if (ret) { | |
6180 | free_extent_map(em); | |
eb838e73 | 6181 | goto unlock_err; |
46bfbb5c CM |
6182 | } |
6183 | goto unlock; | |
4b46fce2 | 6184 | } |
46bfbb5c | 6185 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6186 | } |
46bfbb5c CM |
6187 | must_cow: |
6188 | /* | |
6189 | * this will cow the extent, reset the len in case we changed | |
6190 | * it above | |
6191 | */ | |
6192 | len = bh_result->b_size; | |
70c8a91c JB |
6193 | free_extent_map(em); |
6194 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6195 | if (IS_ERR(em)) { |
6196 | ret = PTR_ERR(em); | |
6197 | goto unlock_err; | |
6198 | } | |
46bfbb5c CM |
6199 | len = min(len, em->len - (start - em->start)); |
6200 | unlock: | |
4b46fce2 JB |
6201 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6202 | inode->i_blkbits; | |
46bfbb5c | 6203 | bh_result->b_size = len; |
4b46fce2 JB |
6204 | bh_result->b_bdev = em->bdev; |
6205 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6206 | if (create) { |
6207 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6208 | set_buffer_new(bh_result); | |
6209 | ||
6210 | /* | |
6211 | * Need to update the i_size under the extent lock so buffered | |
6212 | * readers will get the updated i_size when we unlock. | |
6213 | */ | |
6214 | if (start + len > i_size_read(inode)) | |
6215 | i_size_write(inode, start + len); | |
0934856d MX |
6216 | |
6217 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6218 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
6219 | &cached_state, GFP_NOFS); | |
6220 | BUG_ON(ret); | |
c3473e83 | 6221 | } |
4b46fce2 | 6222 | |
eb838e73 JB |
6223 | /* |
6224 | * In the case of write we need to clear and unlock the entire range, | |
6225 | * in the case of read we need to unlock only the end area that we | |
6226 | * aren't using if there is any left over space. | |
6227 | */ | |
24c03fa5 | 6228 | if (lockstart < lockend) { |
0934856d MX |
6229 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6230 | lockend, unlock_bits, 1, 0, | |
6231 | &cached_state, GFP_NOFS); | |
24c03fa5 | 6232 | } else { |
eb838e73 | 6233 | free_extent_state(cached_state); |
24c03fa5 | 6234 | } |
eb838e73 | 6235 | |
4b46fce2 JB |
6236 | free_extent_map(em); |
6237 | ||
6238 | return 0; | |
eb838e73 JB |
6239 | |
6240 | unlock_err: | |
eb838e73 JB |
6241 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6242 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6243 | return ret; | |
4b46fce2 JB |
6244 | } |
6245 | ||
6246 | struct btrfs_dio_private { | |
6247 | struct inode *inode; | |
6248 | u64 logical_offset; | |
6249 | u64 disk_bytenr; | |
6250 | u64 bytes; | |
4b46fce2 | 6251 | void *private; |
e65e1535 MX |
6252 | |
6253 | /* number of bios pending for this dio */ | |
6254 | atomic_t pending_bios; | |
6255 | ||
6256 | /* IO errors */ | |
6257 | int errors; | |
6258 | ||
6259 | struct bio *orig_bio; | |
4b46fce2 JB |
6260 | }; |
6261 | ||
6262 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6263 | { | |
e65e1535 | 6264 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6265 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6266 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6267 | struct inode *inode = dip->inode; |
6268 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6269 | u64 start; | |
4b46fce2 JB |
6270 | |
6271 | start = dip->logical_offset; | |
6272 | do { | |
6273 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6274 | struct page *page = bvec->bv_page; | |
6275 | char *kaddr; | |
6276 | u32 csum = ~(u32)0; | |
c329861d | 6277 | u64 private = ~(u32)0; |
4b46fce2 JB |
6278 | unsigned long flags; |
6279 | ||
c329861d JB |
6280 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6281 | start, &private)) | |
6282 | goto failed; | |
4b46fce2 | 6283 | local_irq_save(flags); |
7ac687d9 | 6284 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
6285 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
6286 | csum, bvec->bv_len); | |
6287 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6288 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6289 | local_irq_restore(flags); |
6290 | ||
6291 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6292 | if (csum != private) { |
6293 | failed: | |
33345d01 | 6294 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 6295 | " %llu csum %u private %u\n", |
33345d01 LZ |
6296 | (unsigned long long)btrfs_ino(inode), |
6297 | (unsigned long long)start, | |
c329861d | 6298 | csum, (unsigned)private); |
4b46fce2 JB |
6299 | err = -EIO; |
6300 | } | |
6301 | } | |
6302 | ||
6303 | start += bvec->bv_len; | |
4b46fce2 JB |
6304 | bvec++; |
6305 | } while (bvec <= bvec_end); | |
6306 | ||
6307 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6308 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
6309 | bio->bi_private = dip->private; |
6310 | ||
4b46fce2 | 6311 | kfree(dip); |
c0da7aa1 JB |
6312 | |
6313 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6314 | if (err) | |
6315 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6316 | dio_end_io(bio, err); |
6317 | } | |
6318 | ||
6319 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6320 | { | |
6321 | struct btrfs_dio_private *dip = bio->bi_private; | |
6322 | struct inode *inode = dip->inode; | |
6323 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6324 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6325 | u64 ordered_offset = dip->logical_offset; |
6326 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
6327 | int ret; |
6328 | ||
6329 | if (err) | |
6330 | goto out_done; | |
163cf09c CM |
6331 | again: |
6332 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6333 | &ordered_offset, | |
5fd02043 | 6334 | ordered_bytes, !err); |
4b46fce2 | 6335 | if (!ret) |
163cf09c | 6336 | goto out_test; |
4b46fce2 | 6337 | |
5fd02043 JB |
6338 | ordered->work.func = finish_ordered_fn; |
6339 | ordered->work.flags = 0; | |
6340 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6341 | &ordered->work); | |
163cf09c CM |
6342 | out_test: |
6343 | /* | |
6344 | * our bio might span multiple ordered extents. If we haven't | |
6345 | * completed the accounting for the whole dio, go back and try again | |
6346 | */ | |
6347 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6348 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6349 | ordered_offset; | |
5fd02043 | 6350 | ordered = NULL; |
163cf09c CM |
6351 | goto again; |
6352 | } | |
4b46fce2 JB |
6353 | out_done: |
6354 | bio->bi_private = dip->private; | |
6355 | ||
4b46fce2 | 6356 | kfree(dip); |
c0da7aa1 JB |
6357 | |
6358 | /* If we had an error make sure to clear the uptodate flag */ | |
6359 | if (err) | |
6360 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6361 | dio_end_io(bio, err); |
6362 | } | |
6363 | ||
eaf25d93 CM |
6364 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6365 | struct bio *bio, int mirror_num, | |
6366 | unsigned long bio_flags, u64 offset) | |
6367 | { | |
6368 | int ret; | |
6369 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6370 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6371 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6372 | return 0; |
6373 | } | |
6374 | ||
e65e1535 MX |
6375 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6376 | { | |
6377 | struct btrfs_dio_private *dip = bio->bi_private; | |
6378 | ||
6379 | if (err) { | |
33345d01 | 6380 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6381 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6382 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6383 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6384 | dip->errors = 1; |
6385 | ||
6386 | /* | |
6387 | * before atomic variable goto zero, we must make sure | |
6388 | * dip->errors is perceived to be set. | |
6389 | */ | |
6390 | smp_mb__before_atomic_dec(); | |
6391 | } | |
6392 | ||
6393 | /* if there are more bios still pending for this dio, just exit */ | |
6394 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6395 | goto out; | |
6396 | ||
6397 | if (dip->errors) | |
6398 | bio_io_error(dip->orig_bio); | |
6399 | else { | |
6400 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6401 | bio_endio(dip->orig_bio, 0); | |
6402 | } | |
6403 | out: | |
6404 | bio_put(bio); | |
6405 | } | |
6406 | ||
6407 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6408 | u64 first_sector, gfp_t gfp_flags) | |
6409 | { | |
6410 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6411 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6412 | } | |
6413 | ||
6414 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6415 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6416 | int async_submit) |
e65e1535 MX |
6417 | { |
6418 | int write = rw & REQ_WRITE; | |
6419 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6420 | int ret; | |
6421 | ||
b812ce28 JB |
6422 | if (async_submit) |
6423 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
6424 | ||
e65e1535 | 6425 | bio_get(bio); |
5fd02043 JB |
6426 | |
6427 | if (!write) { | |
6428 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6429 | if (ret) | |
6430 | goto err; | |
6431 | } | |
e65e1535 | 6432 | |
1ae39938 JB |
6433 | if (skip_sum) |
6434 | goto map; | |
6435 | ||
6436 | if (write && async_submit) { | |
e65e1535 MX |
6437 | ret = btrfs_wq_submit_bio(root->fs_info, |
6438 | inode, rw, bio, 0, 0, | |
6439 | file_offset, | |
6440 | __btrfs_submit_bio_start_direct_io, | |
6441 | __btrfs_submit_bio_done); | |
6442 | goto err; | |
1ae39938 JB |
6443 | } else if (write) { |
6444 | /* | |
6445 | * If we aren't doing async submit, calculate the csum of the | |
6446 | * bio now. | |
6447 | */ | |
6448 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6449 | if (ret) | |
6450 | goto err; | |
c2db1073 | 6451 | } else if (!skip_sum) { |
c329861d | 6452 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6453 | if (ret) |
6454 | goto err; | |
6455 | } | |
e65e1535 | 6456 | |
1ae39938 JB |
6457 | map: |
6458 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6459 | err: |
6460 | bio_put(bio); | |
6461 | return ret; | |
6462 | } | |
6463 | ||
6464 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6465 | int skip_sum) | |
6466 | { | |
6467 | struct inode *inode = dip->inode; | |
6468 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
6469 | struct bio *bio; |
6470 | struct bio *orig_bio = dip->orig_bio; | |
6471 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6472 | u64 start_sector = orig_bio->bi_sector; | |
6473 | u64 file_offset = dip->logical_offset; | |
6474 | u64 submit_len = 0; | |
6475 | u64 map_length; | |
6476 | int nr_pages = 0; | |
e65e1535 | 6477 | int ret = 0; |
1ae39938 | 6478 | int async_submit = 0; |
e65e1535 | 6479 | |
e65e1535 | 6480 | map_length = orig_bio->bi_size; |
3ec706c8 | 6481 | ret = btrfs_map_block(root->fs_info, READ, start_sector << 9, |
e65e1535 MX |
6482 | &map_length, NULL, 0); |
6483 | if (ret) { | |
64728bbb | 6484 | bio_put(orig_bio); |
e65e1535 MX |
6485 | return -EIO; |
6486 | } | |
6487 | ||
02f57c7a JB |
6488 | if (map_length >= orig_bio->bi_size) { |
6489 | bio = orig_bio; | |
6490 | goto submit; | |
6491 | } | |
6492 | ||
1ae39938 | 6493 | async_submit = 1; |
02f57c7a JB |
6494 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6495 | if (!bio) | |
6496 | return -ENOMEM; | |
6497 | bio->bi_private = dip; | |
6498 | bio->bi_end_io = btrfs_end_dio_bio; | |
6499 | atomic_inc(&dip->pending_bios); | |
6500 | ||
e65e1535 MX |
6501 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6502 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6503 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6504 | bvec->bv_offset) < bvec->bv_len)) { | |
6505 | /* | |
6506 | * inc the count before we submit the bio so | |
6507 | * we know the end IO handler won't happen before | |
6508 | * we inc the count. Otherwise, the dip might get freed | |
6509 | * before we're done setting it up | |
6510 | */ | |
6511 | atomic_inc(&dip->pending_bios); | |
6512 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6513 | file_offset, skip_sum, | |
c329861d | 6514 | async_submit); |
e65e1535 MX |
6515 | if (ret) { |
6516 | bio_put(bio); | |
6517 | atomic_dec(&dip->pending_bios); | |
6518 | goto out_err; | |
6519 | } | |
6520 | ||
e65e1535 MX |
6521 | start_sector += submit_len >> 9; |
6522 | file_offset += submit_len; | |
6523 | ||
6524 | submit_len = 0; | |
6525 | nr_pages = 0; | |
6526 | ||
6527 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6528 | start_sector, GFP_NOFS); | |
6529 | if (!bio) | |
6530 | goto out_err; | |
6531 | bio->bi_private = dip; | |
6532 | bio->bi_end_io = btrfs_end_dio_bio; | |
6533 | ||
6534 | map_length = orig_bio->bi_size; | |
3ec706c8 SB |
6535 | ret = btrfs_map_block(root->fs_info, READ, |
6536 | start_sector << 9, | |
e65e1535 MX |
6537 | &map_length, NULL, 0); |
6538 | if (ret) { | |
6539 | bio_put(bio); | |
6540 | goto out_err; | |
6541 | } | |
6542 | } else { | |
6543 | submit_len += bvec->bv_len; | |
6544 | nr_pages ++; | |
6545 | bvec++; | |
6546 | } | |
6547 | } | |
6548 | ||
02f57c7a | 6549 | submit: |
e65e1535 | 6550 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 6551 | async_submit); |
e65e1535 MX |
6552 | if (!ret) |
6553 | return 0; | |
6554 | ||
6555 | bio_put(bio); | |
6556 | out_err: | |
6557 | dip->errors = 1; | |
6558 | /* | |
6559 | * before atomic variable goto zero, we must | |
6560 | * make sure dip->errors is perceived to be set. | |
6561 | */ | |
6562 | smp_mb__before_atomic_dec(); | |
6563 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6564 | bio_io_error(dip->orig_bio); | |
6565 | ||
6566 | /* bio_end_io() will handle error, so we needn't return it */ | |
6567 | return 0; | |
6568 | } | |
6569 | ||
4b46fce2 JB |
6570 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6571 | loff_t file_offset) | |
6572 | { | |
6573 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6574 | struct btrfs_dio_private *dip; | |
6575 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6576 | int skip_sum; |
7b6d91da | 6577 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6578 | int ret = 0; |
6579 | ||
6580 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6581 | ||
6582 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6583 | if (!dip) { | |
6584 | ret = -ENOMEM; | |
6585 | goto free_ordered; | |
6586 | } | |
4b46fce2 JB |
6587 | |
6588 | dip->private = bio->bi_private; | |
6589 | dip->inode = inode; | |
6590 | dip->logical_offset = file_offset; | |
6591 | ||
4b46fce2 JB |
6592 | dip->bytes = 0; |
6593 | do { | |
6594 | dip->bytes += bvec->bv_len; | |
6595 | bvec++; | |
6596 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6597 | ||
46bfbb5c | 6598 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6599 | bio->bi_private = dip; |
e65e1535 MX |
6600 | dip->errors = 0; |
6601 | dip->orig_bio = bio; | |
6602 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6603 | |
6604 | if (write) | |
6605 | bio->bi_end_io = btrfs_endio_direct_write; | |
6606 | else | |
6607 | bio->bi_end_io = btrfs_endio_direct_read; | |
6608 | ||
e65e1535 MX |
6609 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6610 | if (!ret) | |
eaf25d93 | 6611 | return; |
4b46fce2 JB |
6612 | free_ordered: |
6613 | /* | |
6614 | * If this is a write, we need to clean up the reserved space and kill | |
6615 | * the ordered extent. | |
6616 | */ | |
6617 | if (write) { | |
6618 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6619 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6620 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6621 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6622 | btrfs_free_reserved_extent(root, ordered->start, | |
6623 | ordered->disk_len); | |
6624 | btrfs_put_ordered_extent(ordered); | |
6625 | btrfs_put_ordered_extent(ordered); | |
6626 | } | |
6627 | bio_endio(bio, ret); | |
6628 | } | |
6629 | ||
5a5f79b5 CM |
6630 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6631 | const struct iovec *iov, loff_t offset, | |
6632 | unsigned long nr_segs) | |
6633 | { | |
6634 | int seg; | |
a1b75f7d | 6635 | int i; |
5a5f79b5 CM |
6636 | size_t size; |
6637 | unsigned long addr; | |
6638 | unsigned blocksize_mask = root->sectorsize - 1; | |
6639 | ssize_t retval = -EINVAL; | |
6640 | loff_t end = offset; | |
6641 | ||
6642 | if (offset & blocksize_mask) | |
6643 | goto out; | |
6644 | ||
6645 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6646 | for (seg = 0; seg < nr_segs; seg++) { | |
6647 | addr = (unsigned long)iov[seg].iov_base; | |
6648 | size = iov[seg].iov_len; | |
6649 | end += size; | |
a1b75f7d | 6650 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6651 | goto out; |
a1b75f7d JB |
6652 | |
6653 | /* If this is a write we don't need to check anymore */ | |
6654 | if (rw & WRITE) | |
6655 | continue; | |
6656 | ||
6657 | /* | |
6658 | * Check to make sure we don't have duplicate iov_base's in this | |
6659 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6660 | * when reading back. | |
6661 | */ | |
6662 | for (i = seg + 1; i < nr_segs; i++) { | |
6663 | if (iov[seg].iov_base == iov[i].iov_base) | |
6664 | goto out; | |
6665 | } | |
5a5f79b5 CM |
6666 | } |
6667 | retval = 0; | |
6668 | out: | |
6669 | return retval; | |
6670 | } | |
eb838e73 | 6671 | |
16432985 CM |
6672 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6673 | const struct iovec *iov, loff_t offset, | |
6674 | unsigned long nr_segs) | |
6675 | { | |
4b46fce2 JB |
6676 | struct file *file = iocb->ki_filp; |
6677 | struct inode *inode = file->f_mapping->host; | |
0934856d | 6678 | size_t count = 0; |
2e60a51e | 6679 | int flags = 0; |
38851cc1 MX |
6680 | bool wakeup = true; |
6681 | bool relock = false; | |
0934856d | 6682 | ssize_t ret; |
4b46fce2 | 6683 | |
5a5f79b5 | 6684 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 6685 | offset, nr_segs)) |
5a5f79b5 | 6686 | return 0; |
3f7c579c | 6687 | |
38851cc1 MX |
6688 | atomic_inc(&inode->i_dio_count); |
6689 | smp_mb__after_atomic_inc(); | |
6690 | ||
0934856d MX |
6691 | if (rw & WRITE) { |
6692 | count = iov_length(iov, nr_segs); | |
38851cc1 MX |
6693 | /* |
6694 | * If the write DIO is beyond the EOF, we need update | |
6695 | * the isize, but it is protected by i_mutex. So we can | |
6696 | * not unlock the i_mutex at this case. | |
6697 | */ | |
6698 | if (offset + count <= inode->i_size) { | |
6699 | mutex_unlock(&inode->i_mutex); | |
6700 | relock = true; | |
6701 | } | |
0934856d MX |
6702 | ret = btrfs_delalloc_reserve_space(inode, count); |
6703 | if (ret) | |
38851cc1 MX |
6704 | goto out; |
6705 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
6706 | &BTRFS_I(inode)->runtime_flags))) { | |
6707 | inode_dio_done(inode); | |
6708 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
6709 | wakeup = false; | |
0934856d MX |
6710 | } |
6711 | ||
6712 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
6713 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6714 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
2e60a51e | 6715 | btrfs_submit_direct, flags); |
0934856d MX |
6716 | if (rw & WRITE) { |
6717 | if (ret < 0 && ret != -EIOCBQUEUED) | |
6718 | btrfs_delalloc_release_space(inode, count); | |
6719 | else if (ret > 0 && (size_t)ret < count) { | |
6720 | spin_lock(&BTRFS_I(inode)->lock); | |
6721 | BTRFS_I(inode)->outstanding_extents++; | |
6722 | spin_unlock(&BTRFS_I(inode)->lock); | |
6723 | btrfs_delalloc_release_space(inode, | |
6724 | count - (size_t)ret); | |
6725 | } | |
6726 | btrfs_delalloc_release_metadata(inode, 0); | |
6727 | } | |
38851cc1 | 6728 | out: |
2e60a51e MX |
6729 | if (wakeup) |
6730 | inode_dio_done(inode); | |
38851cc1 MX |
6731 | if (relock) |
6732 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
6733 | |
6734 | return ret; | |
16432985 CM |
6735 | } |
6736 | ||
05dadc09 TI |
6737 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
6738 | ||
1506fcc8 YS |
6739 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6740 | __u64 start, __u64 len) | |
6741 | { | |
05dadc09 TI |
6742 | int ret; |
6743 | ||
6744 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
6745 | if (ret) | |
6746 | return ret; | |
6747 | ||
ec29ed5b | 6748 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6749 | } |
6750 | ||
a52d9a80 | 6751 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6752 | { |
d1310b2e CM |
6753 | struct extent_io_tree *tree; |
6754 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6755 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6756 | } |
1832a6d5 | 6757 | |
a52d9a80 | 6758 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6759 | { |
d1310b2e | 6760 | struct extent_io_tree *tree; |
b888db2b CM |
6761 | |
6762 | ||
6763 | if (current->flags & PF_MEMALLOC) { | |
6764 | redirty_page_for_writepage(wbc, page); | |
6765 | unlock_page(page); | |
6766 | return 0; | |
6767 | } | |
d1310b2e | 6768 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6769 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6770 | } |
6771 | ||
f421950f CM |
6772 | int btrfs_writepages(struct address_space *mapping, |
6773 | struct writeback_control *wbc) | |
b293f02e | 6774 | { |
d1310b2e | 6775 | struct extent_io_tree *tree; |
771ed689 | 6776 | |
d1310b2e | 6777 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6778 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6779 | } | |
6780 | ||
3ab2fb5a CM |
6781 | static int |
6782 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6783 | struct list_head *pages, unsigned nr_pages) | |
6784 | { | |
d1310b2e CM |
6785 | struct extent_io_tree *tree; |
6786 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6787 | return extent_readpages(tree, mapping, pages, nr_pages, |
6788 | btrfs_get_extent); | |
6789 | } | |
e6dcd2dc | 6790 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6791 | { |
d1310b2e CM |
6792 | struct extent_io_tree *tree; |
6793 | struct extent_map_tree *map; | |
a52d9a80 | 6794 | int ret; |
8c2383c3 | 6795 | |
d1310b2e CM |
6796 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6797 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6798 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6799 | if (ret == 1) { |
6800 | ClearPagePrivate(page); | |
6801 | set_page_private(page, 0); | |
6802 | page_cache_release(page); | |
39279cc3 | 6803 | } |
a52d9a80 | 6804 | return ret; |
39279cc3 CM |
6805 | } |
6806 | ||
e6dcd2dc CM |
6807 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6808 | { | |
98509cfc CM |
6809 | if (PageWriteback(page) || PageDirty(page)) |
6810 | return 0; | |
b335b003 | 6811 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6812 | } |
6813 | ||
a52d9a80 | 6814 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6815 | { |
5fd02043 | 6816 | struct inode *inode = page->mapping->host; |
d1310b2e | 6817 | struct extent_io_tree *tree; |
e6dcd2dc | 6818 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6819 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6820 | u64 page_start = page_offset(page); |
6821 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6822 | |
8b62b72b CM |
6823 | /* |
6824 | * we have the page locked, so new writeback can't start, | |
6825 | * and the dirty bit won't be cleared while we are here. | |
6826 | * | |
6827 | * Wait for IO on this page so that we can safely clear | |
6828 | * the PagePrivate2 bit and do ordered accounting | |
6829 | */ | |
e6dcd2dc | 6830 | wait_on_page_writeback(page); |
8b62b72b | 6831 | |
5fd02043 | 6832 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6833 | if (offset) { |
6834 | btrfs_releasepage(page, GFP_NOFS); | |
6835 | return; | |
6836 | } | |
d0082371 | 6837 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
4eee4fa4 | 6838 | ordered = btrfs_lookup_ordered_extent(inode, page_offset(page)); |
e6dcd2dc | 6839 | if (ordered) { |
eb84ae03 CM |
6840 | /* |
6841 | * IO on this page will never be started, so we need | |
6842 | * to account for any ordered extents now | |
6843 | */ | |
e6dcd2dc CM |
6844 | clear_extent_bit(tree, page_start, page_end, |
6845 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
6846 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
6847 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
6848 | /* |
6849 | * whoever cleared the private bit is responsible | |
6850 | * for the finish_ordered_io | |
6851 | */ | |
5fd02043 JB |
6852 | if (TestClearPagePrivate2(page) && |
6853 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6854 | PAGE_CACHE_SIZE, 1)) { | |
6855 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6856 | } |
e6dcd2dc | 6857 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6858 | cached_state = NULL; |
d0082371 | 6859 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6860 | } |
6861 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6862 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
6863 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
6864 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
6865 | __btrfs_releasepage(page, GFP_NOFS); |
6866 | ||
4a096752 | 6867 | ClearPageChecked(page); |
9ad6b7bc | 6868 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6869 | ClearPagePrivate(page); |
6870 | set_page_private(page, 0); | |
6871 | page_cache_release(page); | |
6872 | } | |
39279cc3 CM |
6873 | } |
6874 | ||
9ebefb18 CM |
6875 | /* |
6876 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6877 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6878 | * be careful to check for EOF conditions here. We set the page up correctly | |
6879 | * for a written page which means we get ENOSPC checking when writing into | |
6880 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6881 | * support these features. | |
6882 | * | |
6883 | * We are not allowed to take the i_mutex here so we have to play games to | |
6884 | * protect against truncate races as the page could now be beyond EOF. Because | |
6885 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6886 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6887 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6888 | * unlock the page. | |
6889 | */ | |
c2ec175c | 6890 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6891 | { |
c2ec175c | 6892 | struct page *page = vmf->page; |
6da6abae | 6893 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6894 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6895 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6896 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6897 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6898 | char *kaddr; |
6899 | unsigned long zero_start; | |
9ebefb18 | 6900 | loff_t size; |
1832a6d5 | 6901 | int ret; |
9998eb70 | 6902 | int reserved = 0; |
a52d9a80 | 6903 | u64 page_start; |
e6dcd2dc | 6904 | u64 page_end; |
9ebefb18 | 6905 | |
b2b5ef5c | 6906 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 6907 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6908 | if (!ret) { |
e41f941a | 6909 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
6910 | reserved = 1; |
6911 | } | |
56a76f82 NP |
6912 | if (ret) { |
6913 | if (ret == -ENOMEM) | |
6914 | ret = VM_FAULT_OOM; | |
6915 | else /* -ENOSPC, -EIO, etc */ | |
6916 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6917 | if (reserved) |
6918 | goto out; | |
6919 | goto out_noreserve; | |
56a76f82 | 6920 | } |
1832a6d5 | 6921 | |
56a76f82 | 6922 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6923 | again: |
9ebefb18 | 6924 | lock_page(page); |
9ebefb18 | 6925 | size = i_size_read(inode); |
e6dcd2dc CM |
6926 | page_start = page_offset(page); |
6927 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6928 | |
9ebefb18 | 6929 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6930 | (page_start >= size)) { |
9ebefb18 CM |
6931 | /* page got truncated out from underneath us */ |
6932 | goto out_unlock; | |
6933 | } | |
e6dcd2dc CM |
6934 | wait_on_page_writeback(page); |
6935 | ||
d0082371 | 6936 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6937 | set_page_extent_mapped(page); |
6938 | ||
eb84ae03 CM |
6939 | /* |
6940 | * we can't set the delalloc bits if there are pending ordered | |
6941 | * extents. Drop our locks and wait for them to finish | |
6942 | */ | |
e6dcd2dc CM |
6943 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6944 | if (ordered) { | |
2ac55d41 JB |
6945 | unlock_extent_cached(io_tree, page_start, page_end, |
6946 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6947 | unlock_page(page); |
eb84ae03 | 6948 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6949 | btrfs_put_ordered_extent(ordered); |
6950 | goto again; | |
6951 | } | |
6952 | ||
fbf19087 JB |
6953 | /* |
6954 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6955 | * if it was already dirty, so for space accounting reasons we need to | |
6956 | * clear any delalloc bits for the range we are fixing to save. There | |
6957 | * is probably a better way to do this, but for now keep consistent with | |
6958 | * prepare_pages in the normal write path. | |
6959 | */ | |
2ac55d41 | 6960 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
6961 | EXTENT_DIRTY | EXTENT_DELALLOC | |
6962 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 6963 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6964 | |
2ac55d41 JB |
6965 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6966 | &cached_state); | |
9ed74f2d | 6967 | if (ret) { |
2ac55d41 JB |
6968 | unlock_extent_cached(io_tree, page_start, page_end, |
6969 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6970 | ret = VM_FAULT_SIGBUS; |
6971 | goto out_unlock; | |
6972 | } | |
e6dcd2dc | 6973 | ret = 0; |
9ebefb18 CM |
6974 | |
6975 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6976 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6977 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6978 | else |
e6dcd2dc | 6979 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6980 | |
e6dcd2dc CM |
6981 | if (zero_start != PAGE_CACHE_SIZE) { |
6982 | kaddr = kmap(page); | |
6983 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6984 | flush_dcache_page(page); | |
6985 | kunmap(page); | |
6986 | } | |
247e743c | 6987 | ClearPageChecked(page); |
e6dcd2dc | 6988 | set_page_dirty(page); |
50a9b214 | 6989 | SetPageUptodate(page); |
5a3f23d5 | 6990 | |
257c62e1 CM |
6991 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6992 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 6993 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 6994 | |
2ac55d41 | 6995 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6996 | |
6997 | out_unlock: | |
b2b5ef5c JK |
6998 | if (!ret) { |
6999 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 7000 | return VM_FAULT_LOCKED; |
b2b5ef5c | 7001 | } |
9ebefb18 | 7002 | unlock_page(page); |
1832a6d5 | 7003 | out: |
ec39e180 | 7004 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7005 | out_noreserve: |
b2b5ef5c | 7006 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
7007 | return ret; |
7008 | } | |
7009 | ||
a41ad394 | 7010 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
7011 | { |
7012 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 7013 | struct btrfs_block_rsv *rsv; |
39279cc3 | 7014 | int ret; |
3893e33b | 7015 | int err = 0; |
39279cc3 | 7016 | struct btrfs_trans_handle *trans; |
dbe674a9 | 7017 | u64 mask = root->sectorsize - 1; |
07127184 | 7018 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 7019 | |
2aaa6655 | 7020 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
5d5e103a | 7021 | if (ret) |
a41ad394 | 7022 | return ret; |
8082510e | 7023 | |
4a096752 | 7024 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 7025 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 7026 | |
fcb80c2a JB |
7027 | /* |
7028 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
7029 | * 3 things going on here | |
7030 | * | |
7031 | * 1) We need to reserve space for our orphan item and the space to | |
7032 | * delete our orphan item. Lord knows we don't want to have a dangling | |
7033 | * orphan item because we didn't reserve space to remove it. | |
7034 | * | |
7035 | * 2) We need to reserve space to update our inode. | |
7036 | * | |
7037 | * 3) We need to have something to cache all the space that is going to | |
7038 | * be free'd up by the truncate operation, but also have some slack | |
7039 | * space reserved in case it uses space during the truncate (thank you | |
7040 | * very much snapshotting). | |
7041 | * | |
7042 | * And we need these to all be seperate. The fact is we can use alot of | |
7043 | * space doing the truncate, and we have no earthly idea how much space | |
7044 | * we will use, so we need the truncate reservation to be seperate so it | |
7045 | * doesn't end up using space reserved for updating the inode or | |
7046 | * removing the orphan item. We also need to be able to stop the | |
7047 | * transaction and start a new one, which means we need to be able to | |
7048 | * update the inode several times, and we have no idea of knowing how | |
7049 | * many times that will be, so we can't just reserve 1 item for the | |
7050 | * entirety of the opration, so that has to be done seperately as well. | |
7051 | * Then there is the orphan item, which does indeed need to be held on | |
7052 | * to for the whole operation, and we need nobody to touch this reserved | |
7053 | * space except the orphan code. | |
7054 | * | |
7055 | * So that leaves us with | |
7056 | * | |
7057 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7058 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7059 | * transaction reservation. | |
7060 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7061 | * updating the inode. | |
7062 | */ | |
66d8f3dd | 7063 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7064 | if (!rsv) |
7065 | return -ENOMEM; | |
4a338542 | 7066 | rsv->size = min_size; |
ca7e70f5 | 7067 | rsv->failfast = 1; |
f0cd846e | 7068 | |
907cbceb | 7069 | /* |
07127184 | 7070 | * 1 for the truncate slack space |
907cbceb JB |
7071 | * 1 for updating the inode. |
7072 | */ | |
f3fe820c | 7073 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7074 | if (IS_ERR(trans)) { |
7075 | err = PTR_ERR(trans); | |
7076 | goto out; | |
7077 | } | |
f0cd846e | 7078 | |
907cbceb JB |
7079 | /* Migrate the slack space for the truncate to our reserve */ |
7080 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7081 | min_size); | |
fcb80c2a | 7082 | BUG_ON(ret); |
f0cd846e | 7083 | |
5a3f23d5 CM |
7084 | /* |
7085 | * setattr is responsible for setting the ordered_data_close flag, | |
7086 | * but that is only tested during the last file release. That | |
7087 | * could happen well after the next commit, leaving a great big | |
7088 | * window where new writes may get lost if someone chooses to write | |
7089 | * to this file after truncating to zero | |
7090 | * | |
7091 | * The inode doesn't have any dirty data here, and so if we commit | |
7092 | * this is a noop. If someone immediately starts writing to the inode | |
7093 | * it is very likely we'll catch some of their writes in this | |
7094 | * transaction, and the commit will find this file on the ordered | |
7095 | * data list with good things to send down. | |
7096 | * | |
7097 | * This is a best effort solution, there is still a window where | |
7098 | * using truncate to replace the contents of the file will | |
7099 | * end up with a zero length file after a crash. | |
7100 | */ | |
72ac3c0d JB |
7101 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7102 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7103 | btrfs_add_ordered_operation(trans, root, inode); |
7104 | ||
5dc562c5 JB |
7105 | /* |
7106 | * So if we truncate and then write and fsync we normally would just | |
7107 | * write the extents that changed, which is a problem if we need to | |
7108 | * first truncate that entire inode. So set this flag so we write out | |
7109 | * all of the extents in the inode to the sync log so we're completely | |
7110 | * safe. | |
7111 | */ | |
7112 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7113 | trans->block_rsv = rsv; |
907cbceb | 7114 | |
8082510e YZ |
7115 | while (1) { |
7116 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7117 | inode->i_size, | |
7118 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7119 | if (ret != -ENOSPC) { |
3893e33b | 7120 | err = ret; |
8082510e | 7121 | break; |
3893e33b | 7122 | } |
39279cc3 | 7123 | |
fcb80c2a | 7124 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7125 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7126 | if (ret) { |
7127 | err = ret; | |
7128 | break; | |
7129 | } | |
ca7e70f5 | 7130 | |
8082510e | 7131 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7132 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7133 | |
7134 | trans = btrfs_start_transaction(root, 2); | |
7135 | if (IS_ERR(trans)) { | |
7136 | ret = err = PTR_ERR(trans); | |
7137 | trans = NULL; | |
7138 | break; | |
7139 | } | |
7140 | ||
7141 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7142 | rsv, min_size); | |
7143 | BUG_ON(ret); /* shouldn't happen */ | |
7144 | trans->block_rsv = rsv; | |
8082510e YZ |
7145 | } |
7146 | ||
7147 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7148 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7149 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7150 | if (ret) |
7151 | err = ret; | |
8082510e YZ |
7152 | } |
7153 | ||
917c16b2 CM |
7154 | if (trans) { |
7155 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7156 | ret = btrfs_update_inode(trans, root, inode); | |
7157 | if (ret && !err) | |
7158 | err = ret; | |
7b128766 | 7159 | |
7ad85bb7 | 7160 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7161 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7162 | } |
fcb80c2a JB |
7163 | |
7164 | out: | |
7165 | btrfs_free_block_rsv(root, rsv); | |
7166 | ||
3893e33b JB |
7167 | if (ret && !err) |
7168 | err = ret; | |
a41ad394 | 7169 | |
3893e33b | 7170 | return err; |
39279cc3 CM |
7171 | } |
7172 | ||
d352ac68 CM |
7173 | /* |
7174 | * create a new subvolume directory/inode (helper for the ioctl). | |
7175 | */ | |
d2fb3437 | 7176 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7177 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7178 | { |
39279cc3 | 7179 | struct inode *inode; |
76dda93c | 7180 | int err; |
00e4e6b3 | 7181 | u64 index = 0; |
39279cc3 | 7182 | |
12fc9d09 FA |
7183 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7184 | new_dirid, new_dirid, | |
7185 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7186 | &index); | |
54aa1f4d | 7187 | if (IS_ERR(inode)) |
f46b5a66 | 7188 | return PTR_ERR(inode); |
39279cc3 CM |
7189 | inode->i_op = &btrfs_dir_inode_operations; |
7190 | inode->i_fop = &btrfs_dir_file_operations; | |
7191 | ||
bfe86848 | 7192 | set_nlink(inode, 1); |
dbe674a9 | 7193 | btrfs_i_size_write(inode, 0); |
3b96362c | 7194 | |
76dda93c | 7195 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7196 | |
76dda93c | 7197 | iput(inode); |
ce598979 | 7198 | return err; |
39279cc3 CM |
7199 | } |
7200 | ||
39279cc3 CM |
7201 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7202 | { | |
7203 | struct btrfs_inode *ei; | |
2ead6ae7 | 7204 | struct inode *inode; |
39279cc3 CM |
7205 | |
7206 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7207 | if (!ei) | |
7208 | return NULL; | |
2ead6ae7 YZ |
7209 | |
7210 | ei->root = NULL; | |
2ead6ae7 | 7211 | ei->generation = 0; |
15ee9bc7 | 7212 | ei->last_trans = 0; |
257c62e1 | 7213 | ei->last_sub_trans = 0; |
e02119d5 | 7214 | ei->logged_trans = 0; |
2ead6ae7 | 7215 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7216 | ei->disk_i_size = 0; |
7217 | ei->flags = 0; | |
7709cde3 | 7218 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7219 | ei->index_cnt = (u64)-1; |
7220 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7221 | ei->last_log_commit = 0; |
2ead6ae7 | 7222 | |
9e0baf60 JB |
7223 | spin_lock_init(&ei->lock); |
7224 | ei->outstanding_extents = 0; | |
7225 | ei->reserved_extents = 0; | |
2ead6ae7 | 7226 | |
72ac3c0d | 7227 | ei->runtime_flags = 0; |
261507a0 | 7228 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7229 | |
16cdcec7 MX |
7230 | ei->delayed_node = NULL; |
7231 | ||
2ead6ae7 | 7232 | inode = &ei->vfs_inode; |
a8067e02 | 7233 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7234 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7235 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7236 | ei->io_tree.track_uptodate = 1; |
7237 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7238 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7239 | mutex_init(&ei->log_mutex); |
f248679e | 7240 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7241 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7242 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7243 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7244 | RB_CLEAR_NODE(&ei->rb_node); |
7245 | ||
7246 | return inode; | |
39279cc3 CM |
7247 | } |
7248 | ||
fa0d7e3d NP |
7249 | static void btrfs_i_callback(struct rcu_head *head) |
7250 | { | |
7251 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7252 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7253 | } | |
7254 | ||
39279cc3 CM |
7255 | void btrfs_destroy_inode(struct inode *inode) |
7256 | { | |
e6dcd2dc | 7257 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7258 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7259 | ||
b3d9b7a3 | 7260 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7261 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7262 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7263 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7264 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7265 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7266 | |
a6dbd429 JB |
7267 | /* |
7268 | * This can happen where we create an inode, but somebody else also | |
7269 | * created the same inode and we need to destroy the one we already | |
7270 | * created. | |
7271 | */ | |
7272 | if (!root) | |
7273 | goto free; | |
7274 | ||
5a3f23d5 CM |
7275 | /* |
7276 | * Make sure we're properly removed from the ordered operation | |
7277 | * lists. | |
7278 | */ | |
7279 | smp_mb(); | |
7280 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
7281 | spin_lock(&root->fs_info->ordered_extent_lock); | |
7282 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
7283 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
7284 | } | |
7285 | ||
8a35d95f JB |
7286 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7287 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
7288 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
7289 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7290 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7291 | } |
7b128766 | 7292 | |
d397712b | 7293 | while (1) { |
e6dcd2dc CM |
7294 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7295 | if (!ordered) | |
7296 | break; | |
7297 | else { | |
d397712b CM |
7298 | printk(KERN_ERR "btrfs found ordered " |
7299 | "extent %llu %llu on inode cleanup\n", | |
7300 | (unsigned long long)ordered->file_offset, | |
7301 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7302 | btrfs_remove_ordered_extent(inode, ordered); |
7303 | btrfs_put_ordered_extent(ordered); | |
7304 | btrfs_put_ordered_extent(ordered); | |
7305 | } | |
7306 | } | |
5d4f98a2 | 7307 | inode_tree_del(inode); |
5b21f2ed | 7308 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7309 | free: |
16cdcec7 | 7310 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 7311 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7312 | } |
7313 | ||
45321ac5 | 7314 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7315 | { |
7316 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7317 | |
0af3d00b | 7318 | if (btrfs_root_refs(&root->root_item) == 0 && |
83eea1f1 | 7319 | !btrfs_is_free_space_inode(inode)) |
45321ac5 | 7320 | return 1; |
76dda93c | 7321 | else |
45321ac5 | 7322 | return generic_drop_inode(inode); |
76dda93c YZ |
7323 | } |
7324 | ||
0ee0fda0 | 7325 | static void init_once(void *foo) |
39279cc3 CM |
7326 | { |
7327 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7328 | ||
7329 | inode_init_once(&ei->vfs_inode); | |
7330 | } | |
7331 | ||
7332 | void btrfs_destroy_cachep(void) | |
7333 | { | |
8c0a8537 KS |
7334 | /* |
7335 | * Make sure all delayed rcu free inodes are flushed before we | |
7336 | * destroy cache. | |
7337 | */ | |
7338 | rcu_barrier(); | |
39279cc3 CM |
7339 | if (btrfs_inode_cachep) |
7340 | kmem_cache_destroy(btrfs_inode_cachep); | |
7341 | if (btrfs_trans_handle_cachep) | |
7342 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7343 | if (btrfs_transaction_cachep) | |
7344 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7345 | if (btrfs_path_cachep) |
7346 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7347 | if (btrfs_free_space_cachep) |
7348 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7349 | if (btrfs_delalloc_work_cachep) |
7350 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7351 | } |
7352 | ||
7353 | int btrfs_init_cachep(void) | |
7354 | { | |
837e1972 | 7355 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7356 | sizeof(struct btrfs_inode), 0, |
7357 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7358 | if (!btrfs_inode_cachep) |
7359 | goto fail; | |
9601e3f6 | 7360 | |
837e1972 | 7361 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7362 | sizeof(struct btrfs_trans_handle), 0, |
7363 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7364 | if (!btrfs_trans_handle_cachep) |
7365 | goto fail; | |
9601e3f6 | 7366 | |
837e1972 | 7367 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7368 | sizeof(struct btrfs_transaction), 0, |
7369 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7370 | if (!btrfs_transaction_cachep) |
7371 | goto fail; | |
9601e3f6 | 7372 | |
837e1972 | 7373 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7374 | sizeof(struct btrfs_path), 0, |
7375 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7376 | if (!btrfs_path_cachep) |
7377 | goto fail; | |
9601e3f6 | 7378 | |
837e1972 | 7379 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7380 | sizeof(struct btrfs_free_space), 0, |
7381 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7382 | if (!btrfs_free_space_cachep) | |
7383 | goto fail; | |
7384 | ||
8ccf6f19 MX |
7385 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7386 | sizeof(struct btrfs_delalloc_work), 0, | |
7387 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7388 | NULL); | |
7389 | if (!btrfs_delalloc_work_cachep) | |
7390 | goto fail; | |
7391 | ||
39279cc3 CM |
7392 | return 0; |
7393 | fail: | |
7394 | btrfs_destroy_cachep(); | |
7395 | return -ENOMEM; | |
7396 | } | |
7397 | ||
7398 | static int btrfs_getattr(struct vfsmount *mnt, | |
7399 | struct dentry *dentry, struct kstat *stat) | |
7400 | { | |
df0af1a5 | 7401 | u64 delalloc_bytes; |
39279cc3 | 7402 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
7403 | u32 blocksize = inode->i_sb->s_blocksize; |
7404 | ||
39279cc3 | 7405 | generic_fillattr(inode, stat); |
0ee5dc67 | 7406 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7407 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
7408 | |
7409 | spin_lock(&BTRFS_I(inode)->lock); | |
7410 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
7411 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 7412 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 7413 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
7414 | return 0; |
7415 | } | |
7416 | ||
75e7cb7f LB |
7417 | /* |
7418 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7419 | * directory. | |
7420 | */ | |
7421 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7422 | { | |
7423 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7424 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7425 | ||
7426 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7427 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7428 | else | |
7429 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7430 | ||
bc178237 | 7431 | if (b_dir->flags & BTRFS_INODE_COMPRESS) { |
75e7cb7f | 7432 | b_inode->flags |= BTRFS_INODE_COMPRESS; |
bc178237 LB |
7433 | b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
7434 | } else { | |
7435 | b_inode->flags &= ~(BTRFS_INODE_COMPRESS | | |
7436 | BTRFS_INODE_NOCOMPRESS); | |
7437 | } | |
75e7cb7f LB |
7438 | } |
7439 | ||
d397712b CM |
7440 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7441 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7442 | { |
7443 | struct btrfs_trans_handle *trans; | |
7444 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7445 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7446 | struct inode *new_inode = new_dentry->d_inode; |
7447 | struct inode *old_inode = old_dentry->d_inode; | |
7448 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7449 | u64 index = 0; |
4df27c4d | 7450 | u64 root_objectid; |
39279cc3 | 7451 | int ret; |
33345d01 | 7452 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7453 | |
33345d01 | 7454 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7455 | return -EPERM; |
7456 | ||
4df27c4d | 7457 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7458 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7459 | return -EXDEV; |
7460 | ||
33345d01 LZ |
7461 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7462 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7463 | return -ENOTEMPTY; |
5f39d397 | 7464 | |
4df27c4d YZ |
7465 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7466 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7467 | return -ENOTEMPTY; | |
9c52057c CM |
7468 | |
7469 | ||
7470 | /* check for collisions, even if the name isn't there */ | |
7471 | ret = btrfs_check_dir_item_collision(root, new_dir->i_ino, | |
7472 | new_dentry->d_name.name, | |
7473 | new_dentry->d_name.len); | |
7474 | ||
7475 | if (ret) { | |
7476 | if (ret == -EEXIST) { | |
7477 | /* we shouldn't get | |
7478 | * eexist without a new_inode */ | |
7479 | if (!new_inode) { | |
7480 | WARN_ON(1); | |
7481 | return ret; | |
7482 | } | |
7483 | } else { | |
7484 | /* maybe -EOVERFLOW */ | |
7485 | return ret; | |
7486 | } | |
7487 | } | |
7488 | ret = 0; | |
7489 | ||
5a3f23d5 CM |
7490 | /* |
7491 | * we're using rename to replace one file with another. | |
7492 | * and the replacement file is large. Start IO on it now so | |
7493 | * we don't add too much work to the end of the transaction | |
7494 | */ | |
4baf8c92 | 7495 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7496 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7497 | filemap_flush(old_inode->i_mapping); | |
7498 | ||
76dda93c | 7499 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7500 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7501 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7502 | /* |
7503 | * We want to reserve the absolute worst case amount of items. So if | |
7504 | * both inodes are subvols and we need to unlink them then that would | |
7505 | * require 4 item modifications, but if they are both normal inodes it | |
7506 | * would require 5 item modifications, so we'll assume their normal | |
7507 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7508 | * should cover the worst case number of items we'll modify. | |
7509 | */ | |
7510 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7511 | if (IS_ERR(trans)) { |
7512 | ret = PTR_ERR(trans); | |
7513 | goto out_notrans; | |
7514 | } | |
76dda93c | 7515 | |
4df27c4d YZ |
7516 | if (dest != root) |
7517 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7518 | |
a5719521 YZ |
7519 | ret = btrfs_set_inode_index(new_dir, &index); |
7520 | if (ret) | |
7521 | goto out_fail; | |
5a3f23d5 | 7522 | |
33345d01 | 7523 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7524 | /* force full log commit if subvolume involved. */ |
7525 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7526 | } else { | |
a5719521 YZ |
7527 | ret = btrfs_insert_inode_ref(trans, dest, |
7528 | new_dentry->d_name.name, | |
7529 | new_dentry->d_name.len, | |
33345d01 LZ |
7530 | old_ino, |
7531 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7532 | if (ret) |
7533 | goto out_fail; | |
4df27c4d YZ |
7534 | /* |
7535 | * this is an ugly little race, but the rename is required | |
7536 | * to make sure that if we crash, the inode is either at the | |
7537 | * old name or the new one. pinning the log transaction lets | |
7538 | * us make sure we don't allow a log commit to come in after | |
7539 | * we unlink the name but before we add the new name back in. | |
7540 | */ | |
7541 | btrfs_pin_log_trans(root); | |
7542 | } | |
5a3f23d5 CM |
7543 | /* |
7544 | * make sure the inode gets flushed if it is replacing | |
7545 | * something. | |
7546 | */ | |
33345d01 | 7547 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7548 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7549 | |
0c4d2d95 JB |
7550 | inode_inc_iversion(old_dir); |
7551 | inode_inc_iversion(new_dir); | |
7552 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7553 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7554 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7555 | old_inode->i_ctime = ctime; | |
5f39d397 | 7556 | |
12fcfd22 CM |
7557 | if (old_dentry->d_parent != new_dentry->d_parent) |
7558 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7559 | ||
33345d01 | 7560 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7561 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7562 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7563 | old_dentry->d_name.name, | |
7564 | old_dentry->d_name.len); | |
7565 | } else { | |
92986796 AV |
7566 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7567 | old_dentry->d_inode, | |
7568 | old_dentry->d_name.name, | |
7569 | old_dentry->d_name.len); | |
7570 | if (!ret) | |
7571 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7572 | } |
79787eaa JM |
7573 | if (ret) { |
7574 | btrfs_abort_transaction(trans, root, ret); | |
7575 | goto out_fail; | |
7576 | } | |
39279cc3 CM |
7577 | |
7578 | if (new_inode) { | |
0c4d2d95 | 7579 | inode_inc_iversion(new_inode); |
39279cc3 | 7580 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7581 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7582 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7583 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7584 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7585 | root_objectid, | |
7586 | new_dentry->d_name.name, | |
7587 | new_dentry->d_name.len); | |
7588 | BUG_ON(new_inode->i_nlink == 0); | |
7589 | } else { | |
7590 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7591 | new_dentry->d_inode, | |
7592 | new_dentry->d_name.name, | |
7593 | new_dentry->d_name.len); | |
7594 | } | |
79787eaa | 7595 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7596 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7597 | BUG_ON(ret); |
7b128766 | 7598 | } |
79787eaa JM |
7599 | if (ret) { |
7600 | btrfs_abort_transaction(trans, root, ret); | |
7601 | goto out_fail; | |
7602 | } | |
39279cc3 | 7603 | } |
aec7477b | 7604 | |
75e7cb7f LB |
7605 | fixup_inode_flags(new_dir, old_inode); |
7606 | ||
4df27c4d YZ |
7607 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7608 | new_dentry->d_name.name, | |
a5719521 | 7609 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7610 | if (ret) { |
7611 | btrfs_abort_transaction(trans, root, ret); | |
7612 | goto out_fail; | |
7613 | } | |
39279cc3 | 7614 | |
33345d01 | 7615 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7616 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7617 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7618 | btrfs_end_log_trans(root); |
7619 | } | |
39279cc3 | 7620 | out_fail: |
7ad85bb7 | 7621 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7622 | out_notrans: |
33345d01 | 7623 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7624 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7625 | |
39279cc3 CM |
7626 | return ret; |
7627 | } | |
7628 | ||
8ccf6f19 MX |
7629 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
7630 | { | |
7631 | struct btrfs_delalloc_work *delalloc_work; | |
7632 | ||
7633 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
7634 | work); | |
7635 | if (delalloc_work->wait) | |
7636 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
7637 | else | |
7638 | filemap_flush(delalloc_work->inode->i_mapping); | |
7639 | ||
7640 | if (delalloc_work->delay_iput) | |
7641 | btrfs_add_delayed_iput(delalloc_work->inode); | |
7642 | else | |
7643 | iput(delalloc_work->inode); | |
7644 | complete(&delalloc_work->completion); | |
7645 | } | |
7646 | ||
7647 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
7648 | int wait, int delay_iput) | |
7649 | { | |
7650 | struct btrfs_delalloc_work *work; | |
7651 | ||
7652 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
7653 | if (!work) | |
7654 | return NULL; | |
7655 | ||
7656 | init_completion(&work->completion); | |
7657 | INIT_LIST_HEAD(&work->list); | |
7658 | work->inode = inode; | |
7659 | work->wait = wait; | |
7660 | work->delay_iput = delay_iput; | |
7661 | work->work.func = btrfs_run_delalloc_work; | |
7662 | ||
7663 | return work; | |
7664 | } | |
7665 | ||
7666 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
7667 | { | |
7668 | wait_for_completion(&work->completion); | |
7669 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
7670 | } | |
7671 | ||
d352ac68 CM |
7672 | /* |
7673 | * some fairly slow code that needs optimization. This walks the list | |
7674 | * of all the inodes with pending delalloc and forces them to disk. | |
7675 | */ | |
24bbcf04 | 7676 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 7677 | { |
ea8c2819 | 7678 | struct btrfs_inode *binode; |
5b21f2ed | 7679 | struct inode *inode; |
8ccf6f19 MX |
7680 | struct btrfs_delalloc_work *work, *next; |
7681 | struct list_head works; | |
1eafa6c7 | 7682 | struct list_head splice; |
8ccf6f19 | 7683 | int ret = 0; |
ea8c2819 | 7684 | |
c146afad YZ |
7685 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7686 | return -EROFS; | |
7687 | ||
8ccf6f19 | 7688 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 7689 | INIT_LIST_HEAD(&splice); |
63607cc8 | 7690 | |
75eff68e | 7691 | spin_lock(&root->fs_info->delalloc_lock); |
1eafa6c7 MX |
7692 | list_splice_init(&root->fs_info->delalloc_inodes, &splice); |
7693 | while (!list_empty(&splice)) { | |
7694 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 7695 | delalloc_inodes); |
1eafa6c7 MX |
7696 | |
7697 | list_del_init(&binode->delalloc_inodes); | |
7698 | ||
5b21f2ed | 7699 | inode = igrab(&binode->vfs_inode); |
df0af1a5 MX |
7700 | if (!inode) { |
7701 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
7702 | &binode->runtime_flags); | |
1eafa6c7 | 7703 | continue; |
df0af1a5 | 7704 | } |
1eafa6c7 MX |
7705 | |
7706 | list_add_tail(&binode->delalloc_inodes, | |
7707 | &root->fs_info->delalloc_inodes); | |
75eff68e | 7708 | spin_unlock(&root->fs_info->delalloc_lock); |
1eafa6c7 MX |
7709 | |
7710 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
7711 | if (unlikely(!work)) { | |
7712 | ret = -ENOMEM; | |
7713 | goto out; | |
5b21f2ed | 7714 | } |
1eafa6c7 MX |
7715 | list_add_tail(&work->list, &works); |
7716 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
7717 | &work->work); | |
7718 | ||
5b21f2ed | 7719 | cond_resched(); |
75eff68e | 7720 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7721 | } |
75eff68e | 7722 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d | 7723 | |
1eafa6c7 MX |
7724 | list_for_each_entry_safe(work, next, &works, list) { |
7725 | list_del_init(&work->list); | |
7726 | btrfs_wait_and_free_delalloc_work(work); | |
7727 | } | |
7728 | ||
8c8bee1d CM |
7729 | /* the filemap_flush will queue IO into the worker threads, but |
7730 | * we have to make sure the IO is actually started and that | |
7731 | * ordered extents get created before we return | |
7732 | */ | |
7733 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7734 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7735 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7736 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7737 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7738 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7739 | } |
7740 | atomic_dec(&root->fs_info->async_submit_draining); | |
1eafa6c7 | 7741 | return 0; |
8ccf6f19 MX |
7742 | out: |
7743 | list_for_each_entry_safe(work, next, &works, list) { | |
7744 | list_del_init(&work->list); | |
7745 | btrfs_wait_and_free_delalloc_work(work); | |
7746 | } | |
1eafa6c7 MX |
7747 | |
7748 | if (!list_empty_careful(&splice)) { | |
7749 | spin_lock(&root->fs_info->delalloc_lock); | |
7750 | list_splice_tail(&splice, &root->fs_info->delalloc_inodes); | |
7751 | spin_unlock(&root->fs_info->delalloc_lock); | |
7752 | } | |
8ccf6f19 | 7753 | return ret; |
ea8c2819 CM |
7754 | } |
7755 | ||
39279cc3 CM |
7756 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7757 | const char *symname) | |
7758 | { | |
7759 | struct btrfs_trans_handle *trans; | |
7760 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7761 | struct btrfs_path *path; | |
7762 | struct btrfs_key key; | |
1832a6d5 | 7763 | struct inode *inode = NULL; |
39279cc3 CM |
7764 | int err; |
7765 | int drop_inode = 0; | |
7766 | u64 objectid; | |
00e4e6b3 | 7767 | u64 index = 0 ; |
39279cc3 CM |
7768 | int name_len; |
7769 | int datasize; | |
5f39d397 | 7770 | unsigned long ptr; |
39279cc3 | 7771 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7772 | struct extent_buffer *leaf; |
39279cc3 CM |
7773 | |
7774 | name_len = strlen(symname) + 1; | |
7775 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7776 | return -ENAMETOOLONG; | |
1832a6d5 | 7777 | |
9ed74f2d JB |
7778 | /* |
7779 | * 2 items for inode item and ref | |
7780 | * 2 items for dir items | |
7781 | * 1 item for xattr if selinux is on | |
7782 | */ | |
a22285a6 YZ |
7783 | trans = btrfs_start_transaction(root, 5); |
7784 | if (IS_ERR(trans)) | |
7785 | return PTR_ERR(trans); | |
1832a6d5 | 7786 | |
581bb050 LZ |
7787 | err = btrfs_find_free_ino(root, &objectid); |
7788 | if (err) | |
7789 | goto out_unlock; | |
7790 | ||
aec7477b | 7791 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7792 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7793 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7794 | if (IS_ERR(inode)) { |
7795 | err = PTR_ERR(inode); | |
39279cc3 | 7796 | goto out_unlock; |
7cf96da3 | 7797 | } |
39279cc3 | 7798 | |
2a7dba39 | 7799 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7800 | if (err) { |
7801 | drop_inode = 1; | |
7802 | goto out_unlock; | |
7803 | } | |
7804 | ||
ad19db71 CS |
7805 | /* |
7806 | * If the active LSM wants to access the inode during | |
7807 | * d_instantiate it needs these. Smack checks to see | |
7808 | * if the filesystem supports xattrs by looking at the | |
7809 | * ops vector. | |
7810 | */ | |
7811 | inode->i_fop = &btrfs_file_operations; | |
7812 | inode->i_op = &btrfs_file_inode_operations; | |
7813 | ||
a1b075d2 | 7814 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7815 | if (err) |
7816 | drop_inode = 1; | |
7817 | else { | |
7818 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7819 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7820 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7821 | } |
39279cc3 CM |
7822 | if (drop_inode) |
7823 | goto out_unlock; | |
7824 | ||
7825 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7826 | if (!path) { |
7827 | err = -ENOMEM; | |
7828 | drop_inode = 1; | |
7829 | goto out_unlock; | |
7830 | } | |
33345d01 | 7831 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7832 | key.offset = 0; |
39279cc3 CM |
7833 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7834 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7835 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7836 | datasize); | |
54aa1f4d CM |
7837 | if (err) { |
7838 | drop_inode = 1; | |
b0839166 | 7839 | btrfs_free_path(path); |
54aa1f4d CM |
7840 | goto out_unlock; |
7841 | } | |
5f39d397 CM |
7842 | leaf = path->nodes[0]; |
7843 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7844 | struct btrfs_file_extent_item); | |
7845 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7846 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7847 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7848 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7849 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7850 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7851 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7852 | ||
39279cc3 | 7853 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7854 | write_extent_buffer(leaf, symname, ptr, name_len); |
7855 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7856 | btrfs_free_path(path); |
5f39d397 | 7857 | |
39279cc3 CM |
7858 | inode->i_op = &btrfs_symlink_inode_operations; |
7859 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7860 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7861 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7862 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7863 | err = btrfs_update_inode(trans, root, inode); |
7864 | if (err) | |
7865 | drop_inode = 1; | |
39279cc3 CM |
7866 | |
7867 | out_unlock: | |
08c422c2 AV |
7868 | if (!err) |
7869 | d_instantiate(dentry, inode); | |
7ad85bb7 | 7870 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7871 | if (drop_inode) { |
7872 | inode_dec_link_count(inode); | |
7873 | iput(inode); | |
7874 | } | |
b53d3f5d | 7875 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
7876 | return err; |
7877 | } | |
16432985 | 7878 | |
0af3d00b JB |
7879 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7880 | u64 start, u64 num_bytes, u64 min_size, | |
7881 | loff_t actual_len, u64 *alloc_hint, | |
7882 | struct btrfs_trans_handle *trans) | |
d899e052 | 7883 | { |
5dc562c5 JB |
7884 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7885 | struct extent_map *em; | |
d899e052 YZ |
7886 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7887 | struct btrfs_key ins; | |
d899e052 | 7888 | u64 cur_offset = start; |
55a61d1d | 7889 | u64 i_size; |
d899e052 | 7890 | int ret = 0; |
0af3d00b | 7891 | bool own_trans = true; |
d899e052 | 7892 | |
0af3d00b JB |
7893 | if (trans) |
7894 | own_trans = false; | |
d899e052 | 7895 | while (num_bytes > 0) { |
0af3d00b JB |
7896 | if (own_trans) { |
7897 | trans = btrfs_start_transaction(root, 3); | |
7898 | if (IS_ERR(trans)) { | |
7899 | ret = PTR_ERR(trans); | |
7900 | break; | |
7901 | } | |
5a303d5d YZ |
7902 | } |
7903 | ||
efa56464 | 7904 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7905 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7906 | if (ret) { |
0af3d00b JB |
7907 | if (own_trans) |
7908 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7909 | break; |
d899e052 | 7910 | } |
5a303d5d | 7911 | |
d899e052 YZ |
7912 | ret = insert_reserved_file_extent(trans, inode, |
7913 | cur_offset, ins.objectid, | |
7914 | ins.offset, ins.offset, | |
920bbbfb | 7915 | ins.offset, 0, 0, 0, |
d899e052 | 7916 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7917 | if (ret) { |
7918 | btrfs_abort_transaction(trans, root, ret); | |
7919 | if (own_trans) | |
7920 | btrfs_end_transaction(trans, root); | |
7921 | break; | |
7922 | } | |
a1ed835e CM |
7923 | btrfs_drop_extent_cache(inode, cur_offset, |
7924 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7925 | |
5dc562c5 JB |
7926 | em = alloc_extent_map(); |
7927 | if (!em) { | |
7928 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
7929 | &BTRFS_I(inode)->runtime_flags); | |
7930 | goto next; | |
7931 | } | |
7932 | ||
7933 | em->start = cur_offset; | |
7934 | em->orig_start = cur_offset; | |
7935 | em->len = ins.offset; | |
7936 | em->block_start = ins.objectid; | |
7937 | em->block_len = ins.offset; | |
b4939680 | 7938 | em->orig_block_len = ins.offset; |
5dc562c5 JB |
7939 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7940 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
7941 | em->generation = trans->transid; | |
7942 | ||
7943 | while (1) { | |
7944 | write_lock(&em_tree->lock); | |
7945 | ret = add_extent_mapping(em_tree, em); | |
7946 | if (!ret) | |
7947 | list_move(&em->list, | |
7948 | &em_tree->modified_extents); | |
7949 | write_unlock(&em_tree->lock); | |
7950 | if (ret != -EEXIST) | |
7951 | break; | |
7952 | btrfs_drop_extent_cache(inode, cur_offset, | |
7953 | cur_offset + ins.offset - 1, | |
7954 | 0); | |
7955 | } | |
7956 | free_extent_map(em); | |
7957 | next: | |
d899e052 YZ |
7958 | num_bytes -= ins.offset; |
7959 | cur_offset += ins.offset; | |
efa56464 | 7960 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7961 | |
0c4d2d95 | 7962 | inode_inc_iversion(inode); |
d899e052 | 7963 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7964 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7965 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7966 | (actual_len > inode->i_size) && |
7967 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7968 | if (cur_offset > actual_len) |
55a61d1d | 7969 | i_size = actual_len; |
d1ea6a61 | 7970 | else |
55a61d1d JB |
7971 | i_size = cur_offset; |
7972 | i_size_write(inode, i_size); | |
7973 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7974 | } |
7975 | ||
d899e052 | 7976 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7977 | |
7978 | if (ret) { | |
7979 | btrfs_abort_transaction(trans, root, ret); | |
7980 | if (own_trans) | |
7981 | btrfs_end_transaction(trans, root); | |
7982 | break; | |
7983 | } | |
d899e052 | 7984 | |
0af3d00b JB |
7985 | if (own_trans) |
7986 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7987 | } |
d899e052 YZ |
7988 | return ret; |
7989 | } | |
7990 | ||
0af3d00b JB |
7991 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7992 | u64 start, u64 num_bytes, u64 min_size, | |
7993 | loff_t actual_len, u64 *alloc_hint) | |
7994 | { | |
7995 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7996 | min_size, actual_len, alloc_hint, | |
7997 | NULL); | |
7998 | } | |
7999 | ||
8000 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
8001 | struct btrfs_trans_handle *trans, int mode, | |
8002 | u64 start, u64 num_bytes, u64 min_size, | |
8003 | loff_t actual_len, u64 *alloc_hint) | |
8004 | { | |
8005 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8006 | min_size, actual_len, alloc_hint, trans); | |
8007 | } | |
8008 | ||
e6dcd2dc CM |
8009 | static int btrfs_set_page_dirty(struct page *page) |
8010 | { | |
e6dcd2dc CM |
8011 | return __set_page_dirty_nobuffers(page); |
8012 | } | |
8013 | ||
10556cb2 | 8014 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 8015 | { |
b83cc969 | 8016 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 8017 | umode_t mode = inode->i_mode; |
b83cc969 | 8018 | |
cb6db4e5 JM |
8019 | if (mask & MAY_WRITE && |
8020 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
8021 | if (btrfs_root_readonly(root)) | |
8022 | return -EROFS; | |
8023 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
8024 | return -EACCES; | |
8025 | } | |
2830ba7f | 8026 | return generic_permission(inode, mask); |
fdebe2bd | 8027 | } |
39279cc3 | 8028 | |
6e1d5dcc | 8029 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 8030 | .getattr = btrfs_getattr, |
39279cc3 CM |
8031 | .lookup = btrfs_lookup, |
8032 | .create = btrfs_create, | |
8033 | .unlink = btrfs_unlink, | |
8034 | .link = btrfs_link, | |
8035 | .mkdir = btrfs_mkdir, | |
8036 | .rmdir = btrfs_rmdir, | |
8037 | .rename = btrfs_rename, | |
8038 | .symlink = btrfs_symlink, | |
8039 | .setattr = btrfs_setattr, | |
618e21d5 | 8040 | .mknod = btrfs_mknod, |
95819c05 CH |
8041 | .setxattr = btrfs_setxattr, |
8042 | .getxattr = btrfs_getxattr, | |
5103e947 | 8043 | .listxattr = btrfs_listxattr, |
95819c05 | 8044 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8045 | .permission = btrfs_permission, |
4e34e719 | 8046 | .get_acl = btrfs_get_acl, |
39279cc3 | 8047 | }; |
6e1d5dcc | 8048 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8049 | .lookup = btrfs_lookup, |
fdebe2bd | 8050 | .permission = btrfs_permission, |
4e34e719 | 8051 | .get_acl = btrfs_get_acl, |
39279cc3 | 8052 | }; |
76dda93c | 8053 | |
828c0950 | 8054 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8055 | .llseek = generic_file_llseek, |
8056 | .read = generic_read_dir, | |
cbdf5a24 | 8057 | .readdir = btrfs_real_readdir, |
34287aa3 | 8058 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8059 | #ifdef CONFIG_COMPAT |
34287aa3 | 8060 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8061 | #endif |
6bf13c0c | 8062 | .release = btrfs_release_file, |
e02119d5 | 8063 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8064 | }; |
8065 | ||
d1310b2e | 8066 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8067 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8068 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8069 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8070 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8071 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8072 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8073 | .set_bit_hook = btrfs_set_bit_hook, |
8074 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8075 | .merge_extent_hook = btrfs_merge_extent_hook, |
8076 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8077 | }; |
8078 | ||
35054394 CM |
8079 | /* |
8080 | * btrfs doesn't support the bmap operation because swapfiles | |
8081 | * use bmap to make a mapping of extents in the file. They assume | |
8082 | * these extents won't change over the life of the file and they | |
8083 | * use the bmap result to do IO directly to the drive. | |
8084 | * | |
8085 | * the btrfs bmap call would return logical addresses that aren't | |
8086 | * suitable for IO and they also will change frequently as COW | |
8087 | * operations happen. So, swapfile + btrfs == corruption. | |
8088 | * | |
8089 | * For now we're avoiding this by dropping bmap. | |
8090 | */ | |
7f09410b | 8091 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8092 | .readpage = btrfs_readpage, |
8093 | .writepage = btrfs_writepage, | |
b293f02e | 8094 | .writepages = btrfs_writepages, |
3ab2fb5a | 8095 | .readpages = btrfs_readpages, |
16432985 | 8096 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8097 | .invalidatepage = btrfs_invalidatepage, |
8098 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8099 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8100 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8101 | }; |
8102 | ||
7f09410b | 8103 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8104 | .readpage = btrfs_readpage, |
8105 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8106 | .invalidatepage = btrfs_invalidatepage, |
8107 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8108 | }; |
8109 | ||
6e1d5dcc | 8110 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8111 | .getattr = btrfs_getattr, |
8112 | .setattr = btrfs_setattr, | |
95819c05 CH |
8113 | .setxattr = btrfs_setxattr, |
8114 | .getxattr = btrfs_getxattr, | |
5103e947 | 8115 | .listxattr = btrfs_listxattr, |
95819c05 | 8116 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8117 | .permission = btrfs_permission, |
1506fcc8 | 8118 | .fiemap = btrfs_fiemap, |
4e34e719 | 8119 | .get_acl = btrfs_get_acl, |
e41f941a | 8120 | .update_time = btrfs_update_time, |
39279cc3 | 8121 | }; |
6e1d5dcc | 8122 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8123 | .getattr = btrfs_getattr, |
8124 | .setattr = btrfs_setattr, | |
fdebe2bd | 8125 | .permission = btrfs_permission, |
95819c05 CH |
8126 | .setxattr = btrfs_setxattr, |
8127 | .getxattr = btrfs_getxattr, | |
33268eaf | 8128 | .listxattr = btrfs_listxattr, |
95819c05 | 8129 | .removexattr = btrfs_removexattr, |
4e34e719 | 8130 | .get_acl = btrfs_get_acl, |
e41f941a | 8131 | .update_time = btrfs_update_time, |
618e21d5 | 8132 | }; |
6e1d5dcc | 8133 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8134 | .readlink = generic_readlink, |
8135 | .follow_link = page_follow_link_light, | |
8136 | .put_link = page_put_link, | |
f209561a | 8137 | .getattr = btrfs_getattr, |
22c44fe6 | 8138 | .setattr = btrfs_setattr, |
fdebe2bd | 8139 | .permission = btrfs_permission, |
0279b4cd JO |
8140 | .setxattr = btrfs_setxattr, |
8141 | .getxattr = btrfs_getxattr, | |
8142 | .listxattr = btrfs_listxattr, | |
8143 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8144 | .get_acl = btrfs_get_acl, |
e41f941a | 8145 | .update_time = btrfs_update_time, |
39279cc3 | 8146 | }; |
76dda93c | 8147 | |
82d339d9 | 8148 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8149 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8150 | .d_release = btrfs_dentry_release, |
76dda93c | 8151 | }; |