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