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> |
4b4e25f2 | 42 | #include "compat.h" |
39279cc3 CM |
43 | #include "ctree.h" |
44 | #include "disk-io.h" | |
45 | #include "transaction.h" | |
46 | #include "btrfs_inode.h" | |
47 | #include "ioctl.h" | |
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; | |
74 | struct kmem_cache *btrfs_trans_handle_cachep; | |
75 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 76 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 77 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
78 | |
79 | #define S_SHIFT 12 | |
80 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
81 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
82 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
83 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
84 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
85 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
86 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
87 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
88 | }; | |
89 | ||
a41ad394 JB |
90 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
91 | static int btrfs_truncate(struct inode *inode); | |
5fd02043 | 92 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
93 | static noinline int cow_file_range(struct inode *inode, |
94 | struct page *locked_page, | |
95 | u64 start, u64 end, int *page_started, | |
96 | unsigned long *nr_written, int unlock); | |
2115133f CM |
97 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
98 | struct btrfs_root *root, struct inode *inode); | |
7b128766 | 99 | |
f34f57a3 | 100 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
101 | struct inode *inode, struct inode *dir, |
102 | const struct qstr *qstr) | |
0279b4cd JO |
103 | { |
104 | int err; | |
105 | ||
f34f57a3 | 106 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 107 | if (!err) |
2a7dba39 | 108 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
109 | return err; |
110 | } | |
111 | ||
c8b97818 CM |
112 | /* |
113 | * this does all the hard work for inserting an inline extent into | |
114 | * the btree. The caller should have done a btrfs_drop_extents so that | |
115 | * no overlapping inline items exist in the btree | |
116 | */ | |
d397712b | 117 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
118 | struct btrfs_root *root, struct inode *inode, |
119 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 120 | int compress_type, |
c8b97818 CM |
121 | struct page **compressed_pages) |
122 | { | |
123 | struct btrfs_key key; | |
124 | struct btrfs_path *path; | |
125 | struct extent_buffer *leaf; | |
126 | struct page *page = NULL; | |
127 | char *kaddr; | |
128 | unsigned long ptr; | |
129 | struct btrfs_file_extent_item *ei; | |
130 | int err = 0; | |
131 | int ret; | |
132 | size_t cur_size = size; | |
133 | size_t datasize; | |
134 | unsigned long offset; | |
c8b97818 | 135 | |
fe3f566c | 136 | if (compressed_size && compressed_pages) |
c8b97818 | 137 | cur_size = compressed_size; |
c8b97818 | 138 | |
d397712b CM |
139 | path = btrfs_alloc_path(); |
140 | if (!path) | |
c8b97818 CM |
141 | return -ENOMEM; |
142 | ||
b9473439 | 143 | path->leave_spinning = 1; |
c8b97818 | 144 | |
33345d01 | 145 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
146 | key.offset = start; |
147 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
148 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
149 | ||
150 | inode_add_bytes(inode, size); | |
151 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
152 | datasize); | |
c8b97818 CM |
153 | if (ret) { |
154 | err = ret; | |
c8b97818 CM |
155 | goto fail; |
156 | } | |
157 | leaf = path->nodes[0]; | |
158 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
159 | struct btrfs_file_extent_item); | |
160 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
161 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
162 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
163 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
164 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
165 | ptr = btrfs_file_extent_inline_start(ei); | |
166 | ||
261507a0 | 167 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
168 | struct page *cpage; |
169 | int i = 0; | |
d397712b | 170 | while (compressed_size > 0) { |
c8b97818 | 171 | cpage = compressed_pages[i]; |
5b050f04 | 172 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
173 | PAGE_CACHE_SIZE); |
174 | ||
7ac687d9 | 175 | kaddr = kmap_atomic(cpage); |
c8b97818 | 176 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 177 | kunmap_atomic(kaddr); |
c8b97818 CM |
178 | |
179 | i++; | |
180 | ptr += cur_size; | |
181 | compressed_size -= cur_size; | |
182 | } | |
183 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 184 | compress_type); |
c8b97818 CM |
185 | } else { |
186 | page = find_get_page(inode->i_mapping, | |
187 | start >> PAGE_CACHE_SHIFT); | |
188 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 189 | kaddr = kmap_atomic(page); |
c8b97818 CM |
190 | offset = start & (PAGE_CACHE_SIZE - 1); |
191 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 192 | kunmap_atomic(kaddr); |
c8b97818 CM |
193 | page_cache_release(page); |
194 | } | |
195 | btrfs_mark_buffer_dirty(leaf); | |
196 | btrfs_free_path(path); | |
197 | ||
c2167754 YZ |
198 | /* |
199 | * we're an inline extent, so nobody can | |
200 | * extend the file past i_size without locking | |
201 | * a page we already have locked. | |
202 | * | |
203 | * We must do any isize and inode updates | |
204 | * before we unlock the pages. Otherwise we | |
205 | * could end up racing with unlink. | |
206 | */ | |
c8b97818 | 207 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 208 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 209 | |
79787eaa | 210 | return ret; |
c8b97818 CM |
211 | fail: |
212 | btrfs_free_path(path); | |
213 | return err; | |
214 | } | |
215 | ||
216 | ||
217 | /* | |
218 | * conditionally insert an inline extent into the file. This | |
219 | * does the checks required to make sure the data is small enough | |
220 | * to fit as an inline extent. | |
221 | */ | |
7f366cfe | 222 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
223 | struct btrfs_root *root, |
224 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 225 | size_t compressed_size, int compress_type, |
c8b97818 CM |
226 | struct page **compressed_pages) |
227 | { | |
228 | u64 isize = i_size_read(inode); | |
229 | u64 actual_end = min(end + 1, isize); | |
230 | u64 inline_len = actual_end - start; | |
231 | u64 aligned_end = (end + root->sectorsize - 1) & | |
232 | ~((u64)root->sectorsize - 1); | |
233 | u64 hint_byte; | |
234 | u64 data_len = inline_len; | |
235 | int ret; | |
236 | ||
237 | if (compressed_size) | |
238 | data_len = compressed_size; | |
239 | ||
240 | if (start > 0 || | |
70b99e69 | 241 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
242 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
243 | (!compressed_size && | |
244 | (actual_end & (root->sectorsize - 1)) == 0) || | |
245 | end + 1 < isize || | |
246 | data_len > root->fs_info->max_inline) { | |
247 | return 1; | |
248 | } | |
249 | ||
920bbbfb | 250 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 251 | &hint_byte, 1); |
79787eaa JM |
252 | if (ret) |
253 | return ret; | |
c8b97818 CM |
254 | |
255 | if (isize > actual_end) | |
256 | inline_len = min_t(u64, isize, actual_end); | |
257 | ret = insert_inline_extent(trans, root, inode, start, | |
258 | inline_len, compressed_size, | |
fe3f566c | 259 | compress_type, compressed_pages); |
2adcac1a | 260 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
261 | btrfs_abort_transaction(trans, root, ret); |
262 | return ret; | |
2adcac1a JB |
263 | } else if (ret == -ENOSPC) { |
264 | return 1; | |
79787eaa | 265 | } |
2adcac1a | 266 | |
0ca1f7ce | 267 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 268 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
269 | return 0; |
270 | } | |
271 | ||
771ed689 CM |
272 | struct async_extent { |
273 | u64 start; | |
274 | u64 ram_size; | |
275 | u64 compressed_size; | |
276 | struct page **pages; | |
277 | unsigned long nr_pages; | |
261507a0 | 278 | int compress_type; |
771ed689 CM |
279 | struct list_head list; |
280 | }; | |
281 | ||
282 | struct async_cow { | |
283 | struct inode *inode; | |
284 | struct btrfs_root *root; | |
285 | struct page *locked_page; | |
286 | u64 start; | |
287 | u64 end; | |
288 | struct list_head extents; | |
289 | struct btrfs_work work; | |
290 | }; | |
291 | ||
292 | static noinline int add_async_extent(struct async_cow *cow, | |
293 | u64 start, u64 ram_size, | |
294 | u64 compressed_size, | |
295 | struct page **pages, | |
261507a0 LZ |
296 | unsigned long nr_pages, |
297 | int compress_type) | |
771ed689 CM |
298 | { |
299 | struct async_extent *async_extent; | |
300 | ||
301 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 302 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
303 | async_extent->start = start; |
304 | async_extent->ram_size = ram_size; | |
305 | async_extent->compressed_size = compressed_size; | |
306 | async_extent->pages = pages; | |
307 | async_extent->nr_pages = nr_pages; | |
261507a0 | 308 | async_extent->compress_type = compress_type; |
771ed689 CM |
309 | list_add_tail(&async_extent->list, &cow->extents); |
310 | return 0; | |
311 | } | |
312 | ||
d352ac68 | 313 | /* |
771ed689 CM |
314 | * we create compressed extents in two phases. The first |
315 | * phase compresses a range of pages that have already been | |
316 | * locked (both pages and state bits are locked). | |
c8b97818 | 317 | * |
771ed689 CM |
318 | * This is done inside an ordered work queue, and the compression |
319 | * is spread across many cpus. The actual IO submission is step | |
320 | * two, and the ordered work queue takes care of making sure that | |
321 | * happens in the same order things were put onto the queue by | |
322 | * writepages and friends. | |
c8b97818 | 323 | * |
771ed689 CM |
324 | * If this code finds it can't get good compression, it puts an |
325 | * entry onto the work queue to write the uncompressed bytes. This | |
326 | * makes sure that both compressed inodes and uncompressed inodes | |
327 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 328 | */ |
771ed689 CM |
329 | static noinline int compress_file_range(struct inode *inode, |
330 | struct page *locked_page, | |
331 | u64 start, u64 end, | |
332 | struct async_cow *async_cow, | |
333 | int *num_added) | |
b888db2b CM |
334 | { |
335 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
336 | struct btrfs_trans_handle *trans; | |
db94535d | 337 | u64 num_bytes; |
db94535d | 338 | u64 blocksize = root->sectorsize; |
c8b97818 | 339 | u64 actual_end; |
42dc7bab | 340 | u64 isize = i_size_read(inode); |
e6dcd2dc | 341 | int ret = 0; |
c8b97818 CM |
342 | struct page **pages = NULL; |
343 | unsigned long nr_pages; | |
344 | unsigned long nr_pages_ret = 0; | |
345 | unsigned long total_compressed = 0; | |
346 | unsigned long total_in = 0; | |
347 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 348 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
349 | int i; |
350 | int will_compress; | |
261507a0 | 351 | int compress_type = root->fs_info->compress_type; |
b888db2b | 352 | |
4cb13e5d LB |
353 | /* if this is a small write inside eof, kick off a defrag */ |
354 | if ((end - start + 1) < 16 * 1024 && | |
355 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
356 | btrfs_add_inode_defrag(NULL, inode); |
357 | ||
42dc7bab | 358 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
359 | again: |
360 | will_compress = 0; | |
361 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
362 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 363 | |
f03d9301 CM |
364 | /* |
365 | * we don't want to send crud past the end of i_size through | |
366 | * compression, that's just a waste of CPU time. So, if the | |
367 | * end of the file is before the start of our current | |
368 | * requested range of bytes, we bail out to the uncompressed | |
369 | * cleanup code that can deal with all of this. | |
370 | * | |
371 | * It isn't really the fastest way to fix things, but this is a | |
372 | * very uncommon corner. | |
373 | */ | |
374 | if (actual_end <= start) | |
375 | goto cleanup_and_bail_uncompressed; | |
376 | ||
c8b97818 CM |
377 | total_compressed = actual_end - start; |
378 | ||
379 | /* we want to make sure that amount of ram required to uncompress | |
380 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
381 | * of a compressed extent to 128k. This is a crucial number |
382 | * because it also controls how easily we can spread reads across | |
383 | * cpus for decompression. | |
384 | * | |
385 | * We also want to make sure the amount of IO required to do | |
386 | * a random read is reasonably small, so we limit the size of | |
387 | * a compressed extent to 128k. | |
c8b97818 CM |
388 | */ |
389 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 390 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 391 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
392 | total_in = 0; |
393 | ret = 0; | |
db94535d | 394 | |
771ed689 CM |
395 | /* |
396 | * we do compression for mount -o compress and when the | |
397 | * inode has not been flagged as nocompress. This flag can | |
398 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 399 | */ |
6cbff00f | 400 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 401 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
402 | (BTRFS_I(inode)->force_compress) || |
403 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 404 | WARN_ON(pages); |
cfbc246e | 405 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
406 | if (!pages) { |
407 | /* just bail out to the uncompressed code */ | |
408 | goto cont; | |
409 | } | |
c8b97818 | 410 | |
261507a0 LZ |
411 | if (BTRFS_I(inode)->force_compress) |
412 | compress_type = BTRFS_I(inode)->force_compress; | |
413 | ||
414 | ret = btrfs_compress_pages(compress_type, | |
415 | inode->i_mapping, start, | |
416 | total_compressed, pages, | |
417 | nr_pages, &nr_pages_ret, | |
418 | &total_in, | |
419 | &total_compressed, | |
420 | max_compressed); | |
c8b97818 CM |
421 | |
422 | if (!ret) { | |
423 | unsigned long offset = total_compressed & | |
424 | (PAGE_CACHE_SIZE - 1); | |
425 | struct page *page = pages[nr_pages_ret - 1]; | |
426 | char *kaddr; | |
427 | ||
428 | /* zero the tail end of the last page, we might be | |
429 | * sending it down to disk | |
430 | */ | |
431 | if (offset) { | |
7ac687d9 | 432 | kaddr = kmap_atomic(page); |
c8b97818 CM |
433 | memset(kaddr + offset, 0, |
434 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 435 | kunmap_atomic(kaddr); |
c8b97818 CM |
436 | } |
437 | will_compress = 1; | |
438 | } | |
439 | } | |
560f7d75 | 440 | cont: |
c8b97818 | 441 | if (start == 0) { |
7a7eaa40 | 442 | trans = btrfs_join_transaction(root); |
79787eaa JM |
443 | if (IS_ERR(trans)) { |
444 | ret = PTR_ERR(trans); | |
445 | trans = NULL; | |
446 | goto cleanup_and_out; | |
447 | } | |
0ca1f7ce | 448 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 449 | |
c8b97818 | 450 | /* lets try to make an inline extent */ |
771ed689 | 451 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 452 | /* we didn't compress the entire range, try |
771ed689 | 453 | * to make an uncompressed inline extent. |
c8b97818 CM |
454 | */ |
455 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 456 | start, end, 0, 0, NULL); |
c8b97818 | 457 | } else { |
771ed689 | 458 | /* try making a compressed inline extent */ |
c8b97818 CM |
459 | ret = cow_file_range_inline(trans, root, inode, |
460 | start, end, | |
fe3f566c LZ |
461 | total_compressed, |
462 | compress_type, pages); | |
c8b97818 | 463 | } |
79787eaa | 464 | if (ret <= 0) { |
771ed689 | 465 | /* |
79787eaa JM |
466 | * inline extent creation worked or returned error, |
467 | * we don't need to create any more async work items. | |
468 | * Unlock and free up our temp pages. | |
771ed689 | 469 | */ |
c8b97818 | 470 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
471 | &BTRFS_I(inode)->io_tree, |
472 | start, end, NULL, | |
473 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 474 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 475 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
476 | |
477 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
478 | goto free_pages_out; |
479 | } | |
c2167754 | 480 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
481 | } |
482 | ||
483 | if (will_compress) { | |
484 | /* | |
485 | * we aren't doing an inline extent round the compressed size | |
486 | * up to a block size boundary so the allocator does sane | |
487 | * things | |
488 | */ | |
489 | total_compressed = (total_compressed + blocksize - 1) & | |
490 | ~(blocksize - 1); | |
491 | ||
492 | /* | |
493 | * one last check to make sure the compression is really a | |
494 | * win, compare the page count read with the blocks on disk | |
495 | */ | |
496 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
497 | ~(PAGE_CACHE_SIZE - 1); | |
498 | if (total_compressed >= total_in) { | |
499 | will_compress = 0; | |
500 | } else { | |
c8b97818 CM |
501 | num_bytes = total_in; |
502 | } | |
503 | } | |
504 | if (!will_compress && pages) { | |
505 | /* | |
506 | * the compression code ran but failed to make things smaller, | |
507 | * free any pages it allocated and our page pointer array | |
508 | */ | |
509 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 510 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
511 | page_cache_release(pages[i]); |
512 | } | |
513 | kfree(pages); | |
514 | pages = NULL; | |
515 | total_compressed = 0; | |
516 | nr_pages_ret = 0; | |
517 | ||
518 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
519 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
520 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 521 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 522 | } |
c8b97818 | 523 | } |
771ed689 CM |
524 | if (will_compress) { |
525 | *num_added += 1; | |
c8b97818 | 526 | |
771ed689 CM |
527 | /* the async work queues will take care of doing actual |
528 | * allocation on disk for these compressed pages, | |
529 | * and will submit them to the elevator. | |
530 | */ | |
531 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
532 | total_compressed, pages, nr_pages_ret, |
533 | compress_type); | |
179e29e4 | 534 | |
24ae6365 | 535 | if (start + num_bytes < end) { |
771ed689 CM |
536 | start += num_bytes; |
537 | pages = NULL; | |
538 | cond_resched(); | |
539 | goto again; | |
540 | } | |
541 | } else { | |
f03d9301 | 542 | cleanup_and_bail_uncompressed: |
771ed689 CM |
543 | /* |
544 | * No compression, but we still need to write the pages in | |
545 | * the file we've been given so far. redirty the locked | |
546 | * page if it corresponds to our extent and set things up | |
547 | * for the async work queue to run cow_file_range to do | |
548 | * the normal delalloc dance | |
549 | */ | |
550 | if (page_offset(locked_page) >= start && | |
551 | page_offset(locked_page) <= end) { | |
552 | __set_page_dirty_nobuffers(locked_page); | |
553 | /* unlocked later on in the async handlers */ | |
554 | } | |
261507a0 LZ |
555 | add_async_extent(async_cow, start, end - start + 1, |
556 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
557 | *num_added += 1; |
558 | } | |
3b951516 | 559 | |
771ed689 | 560 | out: |
79787eaa | 561 | return ret; |
771ed689 CM |
562 | |
563 | free_pages_out: | |
564 | for (i = 0; i < nr_pages_ret; i++) { | |
565 | WARN_ON(pages[i]->mapping); | |
566 | page_cache_release(pages[i]); | |
567 | } | |
d397712b | 568 | kfree(pages); |
771ed689 CM |
569 | |
570 | goto out; | |
79787eaa JM |
571 | |
572 | cleanup_and_out: | |
573 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
574 | start, end, NULL, | |
575 | EXTENT_CLEAR_UNLOCK_PAGE | | |
576 | EXTENT_CLEAR_DIRTY | | |
577 | EXTENT_CLEAR_DELALLOC | | |
578 | EXTENT_SET_WRITEBACK | | |
579 | EXTENT_END_WRITEBACK); | |
580 | if (!trans || IS_ERR(trans)) | |
581 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
582 | else | |
583 | btrfs_abort_transaction(trans, root, ret); | |
584 | goto free_pages_out; | |
771ed689 CM |
585 | } |
586 | ||
587 | /* | |
588 | * phase two of compressed writeback. This is the ordered portion | |
589 | * of the code, which only gets called in the order the work was | |
590 | * queued. We walk all the async extents created by compress_file_range | |
591 | * and send them down to the disk. | |
592 | */ | |
593 | static noinline int submit_compressed_extents(struct inode *inode, | |
594 | struct async_cow *async_cow) | |
595 | { | |
596 | struct async_extent *async_extent; | |
597 | u64 alloc_hint = 0; | |
598 | struct btrfs_trans_handle *trans; | |
599 | struct btrfs_key ins; | |
600 | struct extent_map *em; | |
601 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
602 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
603 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 604 | int ret = 0; |
771ed689 CM |
605 | |
606 | if (list_empty(&async_cow->extents)) | |
607 | return 0; | |
608 | ||
771ed689 | 609 | |
d397712b | 610 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
611 | async_extent = list_entry(async_cow->extents.next, |
612 | struct async_extent, list); | |
613 | list_del(&async_extent->list); | |
c8b97818 | 614 | |
771ed689 CM |
615 | io_tree = &BTRFS_I(inode)->io_tree; |
616 | ||
f5a84ee3 | 617 | retry: |
771ed689 CM |
618 | /* did the compression code fall back to uncompressed IO? */ |
619 | if (!async_extent->pages) { | |
620 | int page_started = 0; | |
621 | unsigned long nr_written = 0; | |
622 | ||
623 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 624 | async_extent->start + |
d0082371 | 625 | async_extent->ram_size - 1); |
771ed689 CM |
626 | |
627 | /* allocate blocks */ | |
f5a84ee3 JB |
628 | ret = cow_file_range(inode, async_cow->locked_page, |
629 | async_extent->start, | |
630 | async_extent->start + | |
631 | async_extent->ram_size - 1, | |
632 | &page_started, &nr_written, 0); | |
771ed689 | 633 | |
79787eaa JM |
634 | /* JDM XXX */ |
635 | ||
771ed689 CM |
636 | /* |
637 | * if page_started, cow_file_range inserted an | |
638 | * inline extent and took care of all the unlocking | |
639 | * and IO for us. Otherwise, we need to submit | |
640 | * all those pages down to the drive. | |
641 | */ | |
f5a84ee3 | 642 | if (!page_started && !ret) |
771ed689 CM |
643 | extent_write_locked_range(io_tree, |
644 | inode, async_extent->start, | |
d397712b | 645 | async_extent->start + |
771ed689 CM |
646 | async_extent->ram_size - 1, |
647 | btrfs_get_extent, | |
648 | WB_SYNC_ALL); | |
649 | kfree(async_extent); | |
650 | cond_resched(); | |
651 | continue; | |
652 | } | |
653 | ||
654 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 655 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 656 | |
7a7eaa40 | 657 | trans = btrfs_join_transaction(root); |
79787eaa JM |
658 | if (IS_ERR(trans)) { |
659 | ret = PTR_ERR(trans); | |
660 | } else { | |
661 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
662 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
663 | async_extent->compressed_size, |
664 | async_extent->compressed_size, | |
81c9ad23 | 665 | 0, alloc_hint, &ins, 1); |
79787eaa JM |
666 | if (ret) |
667 | btrfs_abort_transaction(trans, root, ret); | |
668 | btrfs_end_transaction(trans, root); | |
669 | } | |
c2167754 | 670 | |
f5a84ee3 JB |
671 | if (ret) { |
672 | int i; | |
673 | for (i = 0; i < async_extent->nr_pages; i++) { | |
674 | WARN_ON(async_extent->pages[i]->mapping); | |
675 | page_cache_release(async_extent->pages[i]); | |
676 | } | |
677 | kfree(async_extent->pages); | |
678 | async_extent->nr_pages = 0; | |
679 | async_extent->pages = NULL; | |
680 | unlock_extent(io_tree, async_extent->start, | |
681 | async_extent->start + | |
d0082371 | 682 | async_extent->ram_size - 1); |
79787eaa JM |
683 | if (ret == -ENOSPC) |
684 | goto retry; | |
685 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
686 | } |
687 | ||
c2167754 YZ |
688 | /* |
689 | * here we're doing allocation and writeback of the | |
690 | * compressed pages | |
691 | */ | |
692 | btrfs_drop_extent_cache(inode, async_extent->start, | |
693 | async_extent->start + | |
694 | async_extent->ram_size - 1, 0); | |
695 | ||
172ddd60 | 696 | em = alloc_extent_map(); |
79787eaa | 697 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
698 | em->start = async_extent->start; |
699 | em->len = async_extent->ram_size; | |
445a6944 | 700 | em->orig_start = em->start; |
c8b97818 | 701 | |
771ed689 CM |
702 | em->block_start = ins.objectid; |
703 | em->block_len = ins.offset; | |
704 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 705 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
706 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
707 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
708 | ||
d397712b | 709 | while (1) { |
890871be | 710 | write_lock(&em_tree->lock); |
771ed689 | 711 | ret = add_extent_mapping(em_tree, em); |
890871be | 712 | write_unlock(&em_tree->lock); |
771ed689 CM |
713 | if (ret != -EEXIST) { |
714 | free_extent_map(em); | |
715 | break; | |
716 | } | |
717 | btrfs_drop_extent_cache(inode, async_extent->start, | |
718 | async_extent->start + | |
719 | async_extent->ram_size - 1, 0); | |
720 | } | |
721 | ||
261507a0 LZ |
722 | ret = btrfs_add_ordered_extent_compress(inode, |
723 | async_extent->start, | |
724 | ins.objectid, | |
725 | async_extent->ram_size, | |
726 | ins.offset, | |
727 | BTRFS_ORDERED_COMPRESSED, | |
728 | async_extent->compress_type); | |
79787eaa | 729 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 730 | |
771ed689 CM |
731 | /* |
732 | * clear dirty, set writeback and unlock the pages. | |
733 | */ | |
734 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
735 | &BTRFS_I(inode)->io_tree, |
736 | async_extent->start, | |
737 | async_extent->start + | |
738 | async_extent->ram_size - 1, | |
739 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
740 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 741 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 742 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
743 | |
744 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
745 | async_extent->start, |
746 | async_extent->ram_size, | |
747 | ins.objectid, | |
748 | ins.offset, async_extent->pages, | |
749 | async_extent->nr_pages); | |
771ed689 | 750 | |
79787eaa | 751 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
752 | alloc_hint = ins.objectid + ins.offset; |
753 | kfree(async_extent); | |
754 | cond_resched(); | |
755 | } | |
79787eaa JM |
756 | ret = 0; |
757 | out: | |
758 | return ret; | |
759 | out_free: | |
760 | kfree(async_extent); | |
761 | goto out; | |
771ed689 CM |
762 | } |
763 | ||
4b46fce2 JB |
764 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
765 | u64 num_bytes) | |
766 | { | |
767 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
768 | struct extent_map *em; | |
769 | u64 alloc_hint = 0; | |
770 | ||
771 | read_lock(&em_tree->lock); | |
772 | em = search_extent_mapping(em_tree, start, num_bytes); | |
773 | if (em) { | |
774 | /* | |
775 | * if block start isn't an actual block number then find the | |
776 | * first block in this inode and use that as a hint. If that | |
777 | * block is also bogus then just don't worry about it. | |
778 | */ | |
779 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
780 | free_extent_map(em); | |
781 | em = search_extent_mapping(em_tree, 0, 0); | |
782 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
783 | alloc_hint = em->block_start; | |
784 | if (em) | |
785 | free_extent_map(em); | |
786 | } else { | |
787 | alloc_hint = em->block_start; | |
788 | free_extent_map(em); | |
789 | } | |
790 | } | |
791 | read_unlock(&em_tree->lock); | |
792 | ||
793 | return alloc_hint; | |
794 | } | |
795 | ||
771ed689 CM |
796 | /* |
797 | * when extent_io.c finds a delayed allocation range in the file, | |
798 | * the call backs end up in this code. The basic idea is to | |
799 | * allocate extents on disk for the range, and create ordered data structs | |
800 | * in ram to track those extents. | |
801 | * | |
802 | * locked_page is the page that writepage had locked already. We use | |
803 | * it to make sure we don't do extra locks or unlocks. | |
804 | * | |
805 | * *page_started is set to one if we unlock locked_page and do everything | |
806 | * required to start IO on it. It may be clean and already done with | |
807 | * IO when we return. | |
808 | */ | |
809 | static noinline int cow_file_range(struct inode *inode, | |
810 | struct page *locked_page, | |
811 | u64 start, u64 end, int *page_started, | |
812 | unsigned long *nr_written, | |
813 | int unlock) | |
814 | { | |
815 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
816 | struct btrfs_trans_handle *trans; | |
817 | u64 alloc_hint = 0; | |
818 | u64 num_bytes; | |
819 | unsigned long ram_size; | |
820 | u64 disk_num_bytes; | |
821 | u64 cur_alloc_size; | |
822 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
823 | struct btrfs_key ins; |
824 | struct extent_map *em; | |
825 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
826 | int ret = 0; | |
827 | ||
2cf8572d | 828 | BUG_ON(btrfs_is_free_space_inode(root, inode)); |
7a7eaa40 | 829 | trans = btrfs_join_transaction(root); |
79787eaa JM |
830 | if (IS_ERR(trans)) { |
831 | extent_clear_unlock_delalloc(inode, | |
832 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 833 | start, end, locked_page, |
79787eaa JM |
834 | EXTENT_CLEAR_UNLOCK_PAGE | |
835 | EXTENT_CLEAR_UNLOCK | | |
836 | EXTENT_CLEAR_DELALLOC | | |
837 | EXTENT_CLEAR_DIRTY | | |
838 | EXTENT_SET_WRITEBACK | | |
839 | EXTENT_END_WRITEBACK); | |
840 | return PTR_ERR(trans); | |
841 | } | |
0ca1f7ce | 842 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 843 | |
771ed689 CM |
844 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
845 | num_bytes = max(blocksize, num_bytes); | |
846 | disk_num_bytes = num_bytes; | |
847 | ret = 0; | |
848 | ||
4cb5300b | 849 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
850 | if (num_bytes < 64 * 1024 && |
851 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
852 | btrfs_add_inode_defrag(trans, inode); |
853 | ||
771ed689 CM |
854 | if (start == 0) { |
855 | /* lets try to make an inline extent */ | |
856 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 857 | start, end, 0, 0, NULL); |
771ed689 CM |
858 | if (ret == 0) { |
859 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
860 | &BTRFS_I(inode)->io_tree, |
861 | start, end, NULL, | |
862 | EXTENT_CLEAR_UNLOCK_PAGE | | |
863 | EXTENT_CLEAR_UNLOCK | | |
864 | EXTENT_CLEAR_DELALLOC | | |
865 | EXTENT_CLEAR_DIRTY | | |
866 | EXTENT_SET_WRITEBACK | | |
867 | EXTENT_END_WRITEBACK); | |
c2167754 | 868 | |
771ed689 CM |
869 | *nr_written = *nr_written + |
870 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
871 | *page_started = 1; | |
771ed689 | 872 | goto out; |
79787eaa JM |
873 | } else if (ret < 0) { |
874 | btrfs_abort_transaction(trans, root, ret); | |
875 | goto out_unlock; | |
771ed689 CM |
876 | } |
877 | } | |
878 | ||
879 | BUG_ON(disk_num_bytes > | |
6c41761f | 880 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 881 | |
4b46fce2 | 882 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
883 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
884 | ||
d397712b | 885 | while (disk_num_bytes > 0) { |
a791e35e CM |
886 | unsigned long op; |
887 | ||
287a0ab9 | 888 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 889 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 890 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 891 | &ins, 1); |
79787eaa JM |
892 | if (ret < 0) { |
893 | btrfs_abort_transaction(trans, root, ret); | |
894 | goto out_unlock; | |
895 | } | |
d397712b | 896 | |
172ddd60 | 897 | em = alloc_extent_map(); |
79787eaa | 898 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 899 | em->start = start; |
445a6944 | 900 | em->orig_start = em->start; |
771ed689 CM |
901 | ram_size = ins.offset; |
902 | em->len = ins.offset; | |
c8b97818 | 903 | |
e6dcd2dc | 904 | em->block_start = ins.objectid; |
c8b97818 | 905 | em->block_len = ins.offset; |
e6dcd2dc | 906 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 907 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 908 | |
d397712b | 909 | while (1) { |
890871be | 910 | write_lock(&em_tree->lock); |
e6dcd2dc | 911 | ret = add_extent_mapping(em_tree, em); |
890871be | 912 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
913 | if (ret != -EEXIST) { |
914 | free_extent_map(em); | |
915 | break; | |
916 | } | |
917 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 918 | start + ram_size - 1, 0); |
e6dcd2dc CM |
919 | } |
920 | ||
98d20f67 | 921 | cur_alloc_size = ins.offset; |
e6dcd2dc | 922 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 923 | ram_size, cur_alloc_size, 0); |
79787eaa | 924 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 925 | |
17d217fe YZ |
926 | if (root->root_key.objectid == |
927 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
928 | ret = btrfs_reloc_clone_csums(inode, start, | |
929 | cur_alloc_size); | |
79787eaa JM |
930 | if (ret) { |
931 | btrfs_abort_transaction(trans, root, ret); | |
932 | goto out_unlock; | |
933 | } | |
17d217fe YZ |
934 | } |
935 | ||
d397712b | 936 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 937 | break; |
d397712b | 938 | |
c8b97818 CM |
939 | /* we're not doing compressed IO, don't unlock the first |
940 | * page (which the caller expects to stay locked), don't | |
941 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
942 | * |
943 | * Do set the Private2 bit so we know this page was properly | |
944 | * setup for writepage | |
c8b97818 | 945 | */ |
a791e35e CM |
946 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
947 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
948 | EXTENT_SET_PRIVATE2; | |
949 | ||
c8b97818 CM |
950 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
951 | start, start + ram_size - 1, | |
a791e35e | 952 | locked_page, op); |
c8b97818 | 953 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
954 | num_bytes -= cur_alloc_size; |
955 | alloc_hint = ins.objectid + ins.offset; | |
956 | start += cur_alloc_size; | |
b888db2b | 957 | } |
771ed689 | 958 | ret = 0; |
79787eaa | 959 | out: |
b888db2b | 960 | btrfs_end_transaction(trans, root); |
c8b97818 | 961 | |
be20aa9d | 962 | return ret; |
79787eaa JM |
963 | out_unlock: |
964 | extent_clear_unlock_delalloc(inode, | |
965 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 966 | start, end, locked_page, |
79787eaa JM |
967 | EXTENT_CLEAR_UNLOCK_PAGE | |
968 | EXTENT_CLEAR_UNLOCK | | |
969 | EXTENT_CLEAR_DELALLOC | | |
970 | EXTENT_CLEAR_DIRTY | | |
971 | EXTENT_SET_WRITEBACK | | |
972 | EXTENT_END_WRITEBACK); | |
973 | ||
974 | goto out; | |
771ed689 | 975 | } |
c8b97818 | 976 | |
771ed689 CM |
977 | /* |
978 | * work queue call back to started compression on a file and pages | |
979 | */ | |
980 | static noinline void async_cow_start(struct btrfs_work *work) | |
981 | { | |
982 | struct async_cow *async_cow; | |
983 | int num_added = 0; | |
984 | async_cow = container_of(work, struct async_cow, work); | |
985 | ||
986 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
987 | async_cow->start, async_cow->end, async_cow, | |
988 | &num_added); | |
989 | if (num_added == 0) | |
990 | async_cow->inode = NULL; | |
991 | } | |
992 | ||
993 | /* | |
994 | * work queue call back to submit previously compressed pages | |
995 | */ | |
996 | static noinline void async_cow_submit(struct btrfs_work *work) | |
997 | { | |
998 | struct async_cow *async_cow; | |
999 | struct btrfs_root *root; | |
1000 | unsigned long nr_pages; | |
1001 | ||
1002 | async_cow = container_of(work, struct async_cow, work); | |
1003 | ||
1004 | root = async_cow->root; | |
1005 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1006 | PAGE_CACHE_SHIFT; | |
1007 | ||
1008 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
1009 | ||
1010 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1011 | 5 * 1042 * 1024 && | |
1012 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
1013 | wake_up(&root->fs_info->async_submit_wait); | |
1014 | ||
d397712b | 1015 | if (async_cow->inode) |
771ed689 | 1016 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1017 | } |
c8b97818 | 1018 | |
771ed689 CM |
1019 | static noinline void async_cow_free(struct btrfs_work *work) |
1020 | { | |
1021 | struct async_cow *async_cow; | |
1022 | async_cow = container_of(work, struct async_cow, work); | |
1023 | kfree(async_cow); | |
1024 | } | |
1025 | ||
1026 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1027 | u64 start, u64 end, int *page_started, | |
1028 | unsigned long *nr_written) | |
1029 | { | |
1030 | struct async_cow *async_cow; | |
1031 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1032 | unsigned long nr_pages; | |
1033 | u64 cur_end; | |
1034 | int limit = 10 * 1024 * 1042; | |
1035 | ||
a3429ab7 CM |
1036 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1037 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1038 | while (start < end) { |
771ed689 | 1039 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1040 | BUG_ON(!async_cow); /* -ENOMEM */ |
771ed689 CM |
1041 | async_cow->inode = inode; |
1042 | async_cow->root = root; | |
1043 | async_cow->locked_page = locked_page; | |
1044 | async_cow->start = start; | |
1045 | ||
6cbff00f | 1046 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1047 | cur_end = end; |
1048 | else | |
1049 | cur_end = min(end, start + 512 * 1024 - 1); | |
1050 | ||
1051 | async_cow->end = cur_end; | |
1052 | INIT_LIST_HEAD(&async_cow->extents); | |
1053 | ||
1054 | async_cow->work.func = async_cow_start; | |
1055 | async_cow->work.ordered_func = async_cow_submit; | |
1056 | async_cow->work.ordered_free = async_cow_free; | |
1057 | async_cow->work.flags = 0; | |
1058 | ||
771ed689 CM |
1059 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1060 | PAGE_CACHE_SHIFT; | |
1061 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1062 | ||
1063 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1064 | &async_cow->work); | |
1065 | ||
1066 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1067 | wait_event(root->fs_info->async_submit_wait, | |
1068 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1069 | limit)); | |
1070 | } | |
1071 | ||
d397712b | 1072 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1073 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1074 | wait_event(root->fs_info->async_submit_wait, | |
1075 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1076 | 0)); | |
1077 | } | |
1078 | ||
1079 | *nr_written += nr_pages; | |
1080 | start = cur_end + 1; | |
1081 | } | |
1082 | *page_started = 1; | |
1083 | return 0; | |
be20aa9d CM |
1084 | } |
1085 | ||
d397712b | 1086 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1087 | u64 bytenr, u64 num_bytes) |
1088 | { | |
1089 | int ret; | |
1090 | struct btrfs_ordered_sum *sums; | |
1091 | LIST_HEAD(list); | |
1092 | ||
07d400a6 | 1093 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1094 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1095 | if (ret == 0 && list_empty(&list)) |
1096 | return 0; | |
1097 | ||
1098 | while (!list_empty(&list)) { | |
1099 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1100 | list_del(&sums->list); | |
1101 | kfree(sums); | |
1102 | } | |
1103 | return 1; | |
1104 | } | |
1105 | ||
d352ac68 CM |
1106 | /* |
1107 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1108 | * of the extents that exist in the file, and COWs the file as required. | |
1109 | * | |
1110 | * If no cow copies or snapshots exist, we write directly to the existing | |
1111 | * blocks on disk | |
1112 | */ | |
7f366cfe CM |
1113 | static noinline int run_delalloc_nocow(struct inode *inode, |
1114 | struct page *locked_page, | |
771ed689 CM |
1115 | u64 start, u64 end, int *page_started, int force, |
1116 | unsigned long *nr_written) | |
be20aa9d | 1117 | { |
be20aa9d | 1118 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1119 | struct btrfs_trans_handle *trans; |
be20aa9d | 1120 | struct extent_buffer *leaf; |
be20aa9d | 1121 | struct btrfs_path *path; |
80ff3856 | 1122 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1123 | struct btrfs_key found_key; |
80ff3856 YZ |
1124 | u64 cow_start; |
1125 | u64 cur_offset; | |
1126 | u64 extent_end; | |
5d4f98a2 | 1127 | u64 extent_offset; |
80ff3856 YZ |
1128 | u64 disk_bytenr; |
1129 | u64 num_bytes; | |
1130 | int extent_type; | |
79787eaa | 1131 | int ret, err; |
d899e052 | 1132 | int type; |
80ff3856 YZ |
1133 | int nocow; |
1134 | int check_prev = 1; | |
82d5902d | 1135 | bool nolock; |
33345d01 | 1136 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1137 | |
1138 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1139 | if (!path) |
1140 | return -ENOMEM; | |
82d5902d | 1141 | |
2cf8572d | 1142 | nolock = btrfs_is_free_space_inode(root, inode); |
82d5902d LZ |
1143 | |
1144 | if (nolock) | |
7a7eaa40 | 1145 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1146 | else |
7a7eaa40 | 1147 | trans = btrfs_join_transaction(root); |
ff5714cc | 1148 | |
79787eaa JM |
1149 | if (IS_ERR(trans)) { |
1150 | btrfs_free_path(path); | |
1151 | return PTR_ERR(trans); | |
1152 | } | |
1153 | ||
74b21075 | 1154 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1155 | |
80ff3856 YZ |
1156 | cow_start = (u64)-1; |
1157 | cur_offset = start; | |
1158 | while (1) { | |
33345d01 | 1159 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1160 | cur_offset, 0); |
79787eaa JM |
1161 | if (ret < 0) { |
1162 | btrfs_abort_transaction(trans, root, ret); | |
1163 | goto error; | |
1164 | } | |
80ff3856 YZ |
1165 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1166 | leaf = path->nodes[0]; | |
1167 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1168 | path->slots[0] - 1); | |
33345d01 | 1169 | if (found_key.objectid == ino && |
80ff3856 YZ |
1170 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1171 | path->slots[0]--; | |
1172 | } | |
1173 | check_prev = 0; | |
1174 | next_slot: | |
1175 | leaf = path->nodes[0]; | |
1176 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1177 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1178 | if (ret < 0) { |
1179 | btrfs_abort_transaction(trans, root, ret); | |
1180 | goto error; | |
1181 | } | |
80ff3856 YZ |
1182 | if (ret > 0) |
1183 | break; | |
1184 | leaf = path->nodes[0]; | |
1185 | } | |
be20aa9d | 1186 | |
80ff3856 YZ |
1187 | nocow = 0; |
1188 | disk_bytenr = 0; | |
17d217fe | 1189 | num_bytes = 0; |
80ff3856 YZ |
1190 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1191 | ||
33345d01 | 1192 | if (found_key.objectid > ino || |
80ff3856 YZ |
1193 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1194 | found_key.offset > end) | |
1195 | break; | |
1196 | ||
1197 | if (found_key.offset > cur_offset) { | |
1198 | extent_end = found_key.offset; | |
e9061e21 | 1199 | extent_type = 0; |
80ff3856 YZ |
1200 | goto out_check; |
1201 | } | |
1202 | ||
1203 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1204 | struct btrfs_file_extent_item); | |
1205 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1206 | ||
d899e052 YZ |
1207 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1208 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1209 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1210 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1211 | extent_end = found_key.offset + |
1212 | btrfs_file_extent_num_bytes(leaf, fi); | |
1213 | if (extent_end <= start) { | |
1214 | path->slots[0]++; | |
1215 | goto next_slot; | |
1216 | } | |
17d217fe YZ |
1217 | if (disk_bytenr == 0) |
1218 | goto out_check; | |
80ff3856 YZ |
1219 | if (btrfs_file_extent_compression(leaf, fi) || |
1220 | btrfs_file_extent_encryption(leaf, fi) || | |
1221 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1222 | goto out_check; | |
d899e052 YZ |
1223 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1224 | goto out_check; | |
d2fb3437 | 1225 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1226 | goto out_check; |
33345d01 | 1227 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1228 | found_key.offset - |
1229 | extent_offset, disk_bytenr)) | |
17d217fe | 1230 | goto out_check; |
5d4f98a2 | 1231 | disk_bytenr += extent_offset; |
17d217fe YZ |
1232 | disk_bytenr += cur_offset - found_key.offset; |
1233 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1234 | /* | |
1235 | * force cow if csum exists in the range. | |
1236 | * this ensure that csum for a given extent are | |
1237 | * either valid or do not exist. | |
1238 | */ | |
1239 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1240 | goto out_check; | |
80ff3856 YZ |
1241 | nocow = 1; |
1242 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1243 | extent_end = found_key.offset + | |
1244 | btrfs_file_extent_inline_len(leaf, fi); | |
1245 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1246 | } else { | |
1247 | BUG_ON(1); | |
1248 | } | |
1249 | out_check: | |
1250 | if (extent_end <= start) { | |
1251 | path->slots[0]++; | |
1252 | goto next_slot; | |
1253 | } | |
1254 | if (!nocow) { | |
1255 | if (cow_start == (u64)-1) | |
1256 | cow_start = cur_offset; | |
1257 | cur_offset = extent_end; | |
1258 | if (cur_offset > end) | |
1259 | break; | |
1260 | path->slots[0]++; | |
1261 | goto next_slot; | |
7ea394f1 YZ |
1262 | } |
1263 | ||
b3b4aa74 | 1264 | btrfs_release_path(path); |
80ff3856 YZ |
1265 | if (cow_start != (u64)-1) { |
1266 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1267 | found_key.offset - 1, page_started, |
1268 | nr_written, 1); | |
79787eaa JM |
1269 | if (ret) { |
1270 | btrfs_abort_transaction(trans, root, ret); | |
1271 | goto error; | |
1272 | } | |
80ff3856 | 1273 | cow_start = (u64)-1; |
7ea394f1 | 1274 | } |
80ff3856 | 1275 | |
d899e052 YZ |
1276 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1277 | struct extent_map *em; | |
1278 | struct extent_map_tree *em_tree; | |
1279 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1280 | em = alloc_extent_map(); |
79787eaa | 1281 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1282 | em->start = cur_offset; |
445a6944 | 1283 | em->orig_start = em->start; |
d899e052 YZ |
1284 | em->len = num_bytes; |
1285 | em->block_len = num_bytes; | |
1286 | em->block_start = disk_bytenr; | |
1287 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1288 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1289 | while (1) { | |
890871be | 1290 | write_lock(&em_tree->lock); |
d899e052 | 1291 | ret = add_extent_mapping(em_tree, em); |
890871be | 1292 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1293 | if (ret != -EEXIST) { |
1294 | free_extent_map(em); | |
1295 | break; | |
1296 | } | |
1297 | btrfs_drop_extent_cache(inode, em->start, | |
1298 | em->start + em->len - 1, 0); | |
1299 | } | |
1300 | type = BTRFS_ORDERED_PREALLOC; | |
1301 | } else { | |
1302 | type = BTRFS_ORDERED_NOCOW; | |
1303 | } | |
80ff3856 YZ |
1304 | |
1305 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1306 | num_bytes, num_bytes, type); |
79787eaa | 1307 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1308 | |
efa56464 YZ |
1309 | if (root->root_key.objectid == |
1310 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1311 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1312 | num_bytes); | |
79787eaa JM |
1313 | if (ret) { |
1314 | btrfs_abort_transaction(trans, root, ret); | |
1315 | goto error; | |
1316 | } | |
efa56464 YZ |
1317 | } |
1318 | ||
d899e052 | 1319 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1320 | cur_offset, cur_offset + num_bytes - 1, |
1321 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1322 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1323 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1324 | cur_offset = extent_end; |
1325 | if (cur_offset > end) | |
1326 | break; | |
be20aa9d | 1327 | } |
b3b4aa74 | 1328 | btrfs_release_path(path); |
80ff3856 YZ |
1329 | |
1330 | if (cur_offset <= end && cow_start == (u64)-1) | |
1331 | cow_start = cur_offset; | |
1332 | if (cow_start != (u64)-1) { | |
1333 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1334 | page_started, nr_written, 1); |
79787eaa JM |
1335 | if (ret) { |
1336 | btrfs_abort_transaction(trans, root, ret); | |
1337 | goto error; | |
1338 | } | |
80ff3856 YZ |
1339 | } |
1340 | ||
79787eaa | 1341 | error: |
0cb59c99 | 1342 | if (nolock) { |
79787eaa | 1343 | err = btrfs_end_transaction_nolock(trans, root); |
0cb59c99 | 1344 | } else { |
79787eaa | 1345 | err = btrfs_end_transaction(trans, root); |
0cb59c99 | 1346 | } |
79787eaa JM |
1347 | if (!ret) |
1348 | ret = err; | |
1349 | ||
7ea394f1 | 1350 | btrfs_free_path(path); |
79787eaa | 1351 | return ret; |
be20aa9d CM |
1352 | } |
1353 | ||
d352ac68 CM |
1354 | /* |
1355 | * extent_io.c call back to do delayed allocation processing | |
1356 | */ | |
c8b97818 | 1357 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1358 | u64 start, u64 end, int *page_started, |
1359 | unsigned long *nr_written) | |
be20aa9d | 1360 | { |
be20aa9d | 1361 | int ret; |
7f366cfe | 1362 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1363 | |
6cbff00f | 1364 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1365 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1366 | page_started, 1, nr_written); |
6cbff00f | 1367 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1368 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1369 | page_started, 0, nr_written); |
1e701a32 | 1370 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1371 | !(BTRFS_I(inode)->force_compress) && |
1372 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1373 | ret = cow_file_range(inode, locked_page, start, end, |
1374 | page_started, nr_written, 1); | |
be20aa9d | 1375 | else |
771ed689 | 1376 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1377 | page_started, nr_written); |
b888db2b CM |
1378 | return ret; |
1379 | } | |
1380 | ||
1bf85046 JM |
1381 | static void btrfs_split_extent_hook(struct inode *inode, |
1382 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1383 | { |
0ca1f7ce | 1384 | /* not delalloc, ignore it */ |
9ed74f2d | 1385 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1386 | return; |
9ed74f2d | 1387 | |
9e0baf60 JB |
1388 | spin_lock(&BTRFS_I(inode)->lock); |
1389 | BTRFS_I(inode)->outstanding_extents++; | |
1390 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1391 | } |
1392 | ||
1393 | /* | |
1394 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1395 | * extents so we can keep track of new extents that are just merged onto old | |
1396 | * extents, such as when we are doing sequential writes, so we can properly | |
1397 | * account for the metadata space we'll need. | |
1398 | */ | |
1bf85046 JM |
1399 | static void btrfs_merge_extent_hook(struct inode *inode, |
1400 | struct extent_state *new, | |
1401 | struct extent_state *other) | |
9ed74f2d | 1402 | { |
9ed74f2d JB |
1403 | /* not delalloc, ignore it */ |
1404 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1405 | return; |
9ed74f2d | 1406 | |
9e0baf60 JB |
1407 | spin_lock(&BTRFS_I(inode)->lock); |
1408 | BTRFS_I(inode)->outstanding_extents--; | |
1409 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1410 | } |
1411 | ||
d352ac68 CM |
1412 | /* |
1413 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1414 | * bytes in this file, and to maintain the list of inodes that | |
1415 | * have pending delalloc work to be done. | |
1416 | */ | |
1bf85046 JM |
1417 | static void btrfs_set_bit_hook(struct inode *inode, |
1418 | struct extent_state *state, int *bits) | |
291d673e | 1419 | { |
9ed74f2d | 1420 | |
75eff68e CM |
1421 | /* |
1422 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1423 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1424 | * bit, which is only set or cleared with irqs on |
1425 | */ | |
0ca1f7ce | 1426 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1427 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1428 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1429 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
9ed74f2d | 1430 | |
9e0baf60 | 1431 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1432 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1433 | } else { |
1434 | spin_lock(&BTRFS_I(inode)->lock); | |
1435 | BTRFS_I(inode)->outstanding_extents++; | |
1436 | spin_unlock(&BTRFS_I(inode)->lock); | |
1437 | } | |
287a0ab9 | 1438 | |
75eff68e | 1439 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1440 | BTRFS_I(inode)->delalloc_bytes += len; |
1441 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1442 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1443 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1444 | &root->fs_info->delalloc_inodes); | |
1445 | } | |
75eff68e | 1446 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1447 | } |
291d673e CM |
1448 | } |
1449 | ||
d352ac68 CM |
1450 | /* |
1451 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1452 | */ | |
1bf85046 JM |
1453 | static void btrfs_clear_bit_hook(struct inode *inode, |
1454 | struct extent_state *state, int *bits) | |
291d673e | 1455 | { |
75eff68e CM |
1456 | /* |
1457 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1458 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1459 | * bit, which is only set or cleared with irqs on |
1460 | */ | |
0ca1f7ce | 1461 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1462 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1463 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1464 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
bcbfce8a | 1465 | |
9e0baf60 | 1466 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1467 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1468 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1469 | spin_lock(&BTRFS_I(inode)->lock); | |
1470 | BTRFS_I(inode)->outstanding_extents--; | |
1471 | spin_unlock(&BTRFS_I(inode)->lock); | |
1472 | } | |
0ca1f7ce YZ |
1473 | |
1474 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1475 | btrfs_delalloc_release_metadata(inode, len); | |
1476 | ||
0cb59c99 JB |
1477 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1478 | && do_list) | |
0ca1f7ce | 1479 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1480 | |
75eff68e | 1481 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1482 | root->fs_info->delalloc_bytes -= len; |
1483 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1484 | ||
0cb59c99 | 1485 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1486 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1487 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1488 | } | |
75eff68e | 1489 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1490 | } |
291d673e CM |
1491 | } |
1492 | ||
d352ac68 CM |
1493 | /* |
1494 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1495 | * we don't create bios that span stripes or chunks | |
1496 | */ | |
239b14b3 | 1497 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1498 | size_t size, struct bio *bio, |
1499 | unsigned long bio_flags) | |
239b14b3 CM |
1500 | { |
1501 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1502 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1503 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1504 | u64 length = 0; |
1505 | u64 map_length; | |
239b14b3 CM |
1506 | int ret; |
1507 | ||
771ed689 CM |
1508 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1509 | return 0; | |
1510 | ||
f2d8d74d | 1511 | length = bio->bi_size; |
239b14b3 CM |
1512 | map_tree = &root->fs_info->mapping_tree; |
1513 | map_length = length; | |
cea9e445 | 1514 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1515 | &map_length, NULL, 0); |
3444a972 JM |
1516 | /* Will always return 0 or 1 with map_multi == NULL */ |
1517 | BUG_ON(ret < 0); | |
d397712b | 1518 | if (map_length < length + size) |
239b14b3 | 1519 | return 1; |
3444a972 | 1520 | return 0; |
239b14b3 CM |
1521 | } |
1522 | ||
d352ac68 CM |
1523 | /* |
1524 | * in order to insert checksums into the metadata in large chunks, | |
1525 | * we wait until bio submission time. All the pages in the bio are | |
1526 | * checksummed and sums are attached onto the ordered extent record. | |
1527 | * | |
1528 | * At IO completion time the cums attached on the ordered extent record | |
1529 | * are inserted into the btree | |
1530 | */ | |
d397712b CM |
1531 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1532 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1533 | unsigned long bio_flags, |
1534 | u64 bio_offset) | |
065631f6 | 1535 | { |
065631f6 | 1536 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1537 | int ret = 0; |
e015640f | 1538 | |
d20f7043 | 1539 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1540 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1541 | return 0; |
1542 | } | |
e015640f | 1543 | |
4a69a410 CM |
1544 | /* |
1545 | * in order to insert checksums into the metadata in large chunks, | |
1546 | * we wait until bio submission time. All the pages in the bio are | |
1547 | * checksummed and sums are attached onto the ordered extent record. | |
1548 | * | |
1549 | * At IO completion time the cums attached on the ordered extent record | |
1550 | * are inserted into the btree | |
1551 | */ | |
b2950863 | 1552 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1553 | int mirror_num, unsigned long bio_flags, |
1554 | u64 bio_offset) | |
4a69a410 CM |
1555 | { |
1556 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1557 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1558 | } |
1559 | ||
d352ac68 | 1560 | /* |
cad321ad CM |
1561 | * extent_io.c submission hook. This does the right thing for csum calculation |
1562 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1563 | */ |
b2950863 | 1564 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1565 | int mirror_num, unsigned long bio_flags, |
1566 | u64 bio_offset) | |
44b8bd7e CM |
1567 | { |
1568 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1569 | int ret = 0; | |
19b9bdb0 | 1570 | int skip_sum; |
0417341e | 1571 | int metadata = 0; |
44b8bd7e | 1572 | |
6cbff00f | 1573 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1574 | |
2cf8572d | 1575 | if (btrfs_is_free_space_inode(root, inode)) |
0417341e JM |
1576 | metadata = 2; |
1577 | ||
7b6d91da | 1578 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1579 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1580 | if (ret) | |
1581 | return ret; | |
1582 | ||
d20f7043 | 1583 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1584 | return btrfs_submit_compressed_read(inode, bio, |
1585 | mirror_num, bio_flags); | |
c2db1073 TI |
1586 | } else if (!skip_sum) { |
1587 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1588 | if (ret) | |
1589 | return ret; | |
1590 | } | |
4d1b5fb4 | 1591 | goto mapit; |
19b9bdb0 | 1592 | } else if (!skip_sum) { |
17d217fe YZ |
1593 | /* csum items have already been cloned */ |
1594 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1595 | goto mapit; | |
19b9bdb0 CM |
1596 | /* we're doing a write, do the async checksumming */ |
1597 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1598 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1599 | bio_flags, bio_offset, |
1600 | __btrfs_submit_bio_start, | |
4a69a410 | 1601 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1602 | } |
1603 | ||
0b86a832 | 1604 | mapit: |
8b712842 | 1605 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1606 | } |
6885f308 | 1607 | |
d352ac68 CM |
1608 | /* |
1609 | * given a list of ordered sums record them in the inode. This happens | |
1610 | * at IO completion time based on sums calculated at bio submission time. | |
1611 | */ | |
ba1da2f4 | 1612 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1613 | struct inode *inode, u64 file_offset, |
1614 | struct list_head *list) | |
1615 | { | |
e6dcd2dc CM |
1616 | struct btrfs_ordered_sum *sum; |
1617 | ||
c6e30871 | 1618 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1619 | btrfs_csum_file_blocks(trans, |
1620 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1621 | } |
1622 | return 0; | |
1623 | } | |
1624 | ||
2ac55d41 JB |
1625 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1626 | struct extent_state **cached_state) | |
ea8c2819 | 1627 | { |
d397712b | 1628 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1629 | WARN_ON(1); |
ea8c2819 | 1630 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1631 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1632 | } |
1633 | ||
d352ac68 | 1634 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1635 | struct btrfs_writepage_fixup { |
1636 | struct page *page; | |
1637 | struct btrfs_work work; | |
1638 | }; | |
1639 | ||
b2950863 | 1640 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1641 | { |
1642 | struct btrfs_writepage_fixup *fixup; | |
1643 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1644 | struct extent_state *cached_state = NULL; |
247e743c CM |
1645 | struct page *page; |
1646 | struct inode *inode; | |
1647 | u64 page_start; | |
1648 | u64 page_end; | |
87826df0 | 1649 | int ret; |
247e743c CM |
1650 | |
1651 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1652 | page = fixup->page; | |
4a096752 | 1653 | again: |
247e743c CM |
1654 | lock_page(page); |
1655 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1656 | ClearPageChecked(page); | |
1657 | goto out_page; | |
1658 | } | |
1659 | ||
1660 | inode = page->mapping->host; | |
1661 | page_start = page_offset(page); | |
1662 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1663 | ||
2ac55d41 | 1664 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1665 | &cached_state); |
4a096752 CM |
1666 | |
1667 | /* already ordered? We're done */ | |
8b62b72b | 1668 | if (PagePrivate2(page)) |
247e743c | 1669 | goto out; |
4a096752 CM |
1670 | |
1671 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1672 | if (ordered) { | |
2ac55d41 JB |
1673 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1674 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1675 | unlock_page(page); |
1676 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1677 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1678 | goto again; |
1679 | } | |
247e743c | 1680 | |
87826df0 JM |
1681 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1682 | if (ret) { | |
1683 | mapping_set_error(page->mapping, ret); | |
1684 | end_extent_writepage(page, ret, page_start, page_end); | |
1685 | ClearPageChecked(page); | |
1686 | goto out; | |
1687 | } | |
1688 | ||
2ac55d41 | 1689 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1690 | ClearPageChecked(page); |
87826df0 | 1691 | set_page_dirty(page); |
247e743c | 1692 | out: |
2ac55d41 JB |
1693 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1694 | &cached_state, GFP_NOFS); | |
247e743c CM |
1695 | out_page: |
1696 | unlock_page(page); | |
1697 | page_cache_release(page); | |
b897abec | 1698 | kfree(fixup); |
247e743c CM |
1699 | } |
1700 | ||
1701 | /* | |
1702 | * There are a few paths in the higher layers of the kernel that directly | |
1703 | * set the page dirty bit without asking the filesystem if it is a | |
1704 | * good idea. This causes problems because we want to make sure COW | |
1705 | * properly happens and the data=ordered rules are followed. | |
1706 | * | |
c8b97818 | 1707 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1708 | * hasn't been properly setup for IO. We kick off an async process |
1709 | * to fix it up. The async helper will wait for ordered extents, set | |
1710 | * the delalloc bit and make it safe to write the page. | |
1711 | */ | |
b2950863 | 1712 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1713 | { |
1714 | struct inode *inode = page->mapping->host; | |
1715 | struct btrfs_writepage_fixup *fixup; | |
1716 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1717 | |
8b62b72b CM |
1718 | /* this page is properly in the ordered list */ |
1719 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1720 | return 0; |
1721 | ||
1722 | if (PageChecked(page)) | |
1723 | return -EAGAIN; | |
1724 | ||
1725 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1726 | if (!fixup) | |
1727 | return -EAGAIN; | |
f421950f | 1728 | |
247e743c CM |
1729 | SetPageChecked(page); |
1730 | page_cache_get(page); | |
1731 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1732 | fixup->page = page; | |
1733 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1734 | return -EBUSY; |
247e743c CM |
1735 | } |
1736 | ||
d899e052 YZ |
1737 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1738 | struct inode *inode, u64 file_pos, | |
1739 | u64 disk_bytenr, u64 disk_num_bytes, | |
1740 | u64 num_bytes, u64 ram_bytes, | |
1741 | u8 compression, u8 encryption, | |
1742 | u16 other_encoding, int extent_type) | |
1743 | { | |
1744 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1745 | struct btrfs_file_extent_item *fi; | |
1746 | struct btrfs_path *path; | |
1747 | struct extent_buffer *leaf; | |
1748 | struct btrfs_key ins; | |
1749 | u64 hint; | |
1750 | int ret; | |
1751 | ||
1752 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1753 | if (!path) |
1754 | return -ENOMEM; | |
d899e052 | 1755 | |
b9473439 | 1756 | path->leave_spinning = 1; |
a1ed835e CM |
1757 | |
1758 | /* | |
1759 | * we may be replacing one extent in the tree with another. | |
1760 | * The new extent is pinned in the extent map, and we don't want | |
1761 | * to drop it from the cache until it is completely in the btree. | |
1762 | * | |
1763 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1764 | * the caller is expected to unpin it and allow it to be merged | |
1765 | * with the others. | |
1766 | */ | |
920bbbfb YZ |
1767 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1768 | &hint, 0); | |
79787eaa JM |
1769 | if (ret) |
1770 | goto out; | |
d899e052 | 1771 | |
33345d01 | 1772 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1773 | ins.offset = file_pos; |
1774 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1775 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1776 | if (ret) |
1777 | goto out; | |
d899e052 YZ |
1778 | leaf = path->nodes[0]; |
1779 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1780 | struct btrfs_file_extent_item); | |
1781 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1782 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1783 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1784 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1785 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1786 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1787 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1788 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1789 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1790 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1791 | |
1792 | btrfs_unlock_up_safe(path, 1); | |
1793 | btrfs_set_lock_blocking(leaf); | |
1794 | ||
d899e052 YZ |
1795 | btrfs_mark_buffer_dirty(leaf); |
1796 | ||
1797 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1798 | |
1799 | ins.objectid = disk_bytenr; | |
1800 | ins.offset = disk_num_bytes; | |
1801 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1802 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1803 | root->root_key.objectid, | |
33345d01 | 1804 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1805 | out: |
d899e052 | 1806 | btrfs_free_path(path); |
b9473439 | 1807 | |
79787eaa | 1808 | return ret; |
d899e052 YZ |
1809 | } |
1810 | ||
5d13a98f CM |
1811 | /* |
1812 | * helper function for btrfs_finish_ordered_io, this | |
1813 | * just reads in some of the csum leaves to prime them into ram | |
1814 | * before we start the transaction. It limits the amount of btree | |
1815 | * reads required while inside the transaction. | |
1816 | */ | |
d352ac68 CM |
1817 | /* as ordered data IO finishes, this gets called so we can finish |
1818 | * an ordered extent if the range of bytes in the file it covers are | |
1819 | * fully written. | |
1820 | */ | |
5fd02043 | 1821 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1822 | { |
5fd02043 | 1823 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1824 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1825 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1826 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1827 | struct extent_state *cached_state = NULL; |
261507a0 | 1828 | int compress_type = 0; |
e6dcd2dc | 1829 | int ret; |
82d5902d | 1830 | bool nolock; |
e6dcd2dc | 1831 | |
2cf8572d | 1832 | nolock = btrfs_is_free_space_inode(root, inode); |
0cb59c99 | 1833 | |
5fd02043 JB |
1834 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1835 | ret = -EIO; | |
1836 | goto out; | |
1837 | } | |
1838 | ||
c2167754 | 1839 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1840 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
c2167754 YZ |
1841 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1842 | if (!ret) { | |
0cb59c99 | 1843 | if (nolock) |
7a7eaa40 | 1844 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1845 | else |
7a7eaa40 | 1846 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1847 | if (IS_ERR(trans)) |
1848 | return PTR_ERR(trans); | |
0ca1f7ce | 1849 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1850 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1851 | if (ret) /* -ENOMEM or corruption */ |
1852 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1853 | } |
1854 | goto out; | |
1855 | } | |
e6dcd2dc | 1856 | |
2ac55d41 JB |
1857 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1858 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1859 | 0, &cached_state); |
e6dcd2dc | 1860 | |
0cb59c99 | 1861 | if (nolock) |
7a7eaa40 | 1862 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1863 | else |
7a7eaa40 | 1864 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1865 | if (IS_ERR(trans)) { |
1866 | ret = PTR_ERR(trans); | |
1867 | trans = NULL; | |
1868 | goto out_unlock; | |
1869 | } | |
0ca1f7ce | 1870 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1871 | |
c8b97818 | 1872 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1873 | compress_type = ordered_extent->compress_type; |
d899e052 | 1874 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1875 | BUG_ON(compress_type); |
920bbbfb | 1876 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1877 | ordered_extent->file_offset, |
1878 | ordered_extent->file_offset + | |
1879 | ordered_extent->len); | |
d899e052 | 1880 | } else { |
0af3d00b | 1881 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1882 | ret = insert_reserved_file_extent(trans, inode, |
1883 | ordered_extent->file_offset, | |
1884 | ordered_extent->start, | |
1885 | ordered_extent->disk_len, | |
1886 | ordered_extent->len, | |
1887 | ordered_extent->len, | |
261507a0 | 1888 | compress_type, 0, 0, |
d899e052 | 1889 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1890 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1891 | ordered_extent->file_offset, | |
1892 | ordered_extent->len); | |
d899e052 | 1893 | } |
5fd02043 | 1894 | |
79787eaa JM |
1895 | if (ret < 0) { |
1896 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1897 | goto out_unlock; |
79787eaa | 1898 | } |
2ac55d41 | 1899 | |
e6dcd2dc CM |
1900 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1901 | &ordered_extent->list); | |
1902 | ||
1ef30be1 | 1903 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1904 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1905 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1906 | if (ret) { /* -ENOMEM or corruption */ |
1907 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1908 | goto out_unlock; |
79787eaa | 1909 | } |
1ef30be1 JB |
1910 | } |
1911 | ret = 0; | |
5fd02043 JB |
1912 | out_unlock: |
1913 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
1914 | ordered_extent->file_offset + | |
1915 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 1916 | out: |
5b0e95bf | 1917 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1918 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
5b0e95bf JB |
1919 | if (trans) { |
1920 | if (nolock) | |
0cb59c99 | 1921 | btrfs_end_transaction_nolock(trans, root); |
5b0e95bf | 1922 | else |
0cb59c99 JB |
1923 | btrfs_end_transaction(trans, root); |
1924 | } | |
1925 | ||
5fd02043 JB |
1926 | if (ret) |
1927 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, | |
1928 | ordered_extent->file_offset + | |
1929 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
1930 | ||
1931 | /* | |
1932 | * This needs to be dont to make sure anybody waiting knows we are done | |
1933 | * upating everything for this ordered extent. | |
1934 | */ | |
1935 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
1936 | ||
e6dcd2dc CM |
1937 | /* once for us */ |
1938 | btrfs_put_ordered_extent(ordered_extent); | |
1939 | /* once for the tree */ | |
1940 | btrfs_put_ordered_extent(ordered_extent); | |
1941 | ||
5fd02043 JB |
1942 | return ret; |
1943 | } | |
1944 | ||
1945 | static void finish_ordered_fn(struct btrfs_work *work) | |
1946 | { | |
1947 | struct btrfs_ordered_extent *ordered_extent; | |
1948 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
1949 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
1950 | } |
1951 | ||
b2950863 | 1952 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1953 | struct extent_state *state, int uptodate) |
1954 | { | |
5fd02043 JB |
1955 | struct inode *inode = page->mapping->host; |
1956 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1957 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
1958 | struct btrfs_workers *workers; | |
1959 | ||
1abe9b8a | 1960 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1961 | ||
8b62b72b | 1962 | ClearPagePrivate2(page); |
5fd02043 JB |
1963 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1964 | end - start + 1, uptodate)) | |
1965 | return 0; | |
1966 | ||
1967 | ordered_extent->work.func = finish_ordered_fn; | |
1968 | ordered_extent->work.flags = 0; | |
1969 | ||
1970 | if (btrfs_is_free_space_inode(root, inode)) | |
1971 | workers = &root->fs_info->endio_freespace_worker; | |
1972 | else | |
1973 | workers = &root->fs_info->endio_write_workers; | |
1974 | btrfs_queue_worker(workers, &ordered_extent->work); | |
1975 | ||
1976 | return 0; | |
211f90e6 CM |
1977 | } |
1978 | ||
d352ac68 CM |
1979 | /* |
1980 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
1981 | * if there's a match, we allow the bio to finish. If not, the code in |
1982 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 1983 | */ |
b2950863 | 1984 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 1985 | struct extent_state *state, int mirror) |
07157aac | 1986 | { |
35ebb934 | 1987 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1988 | struct inode *inode = page->mapping->host; |
d1310b2e | 1989 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1990 | char *kaddr; |
aadfeb6e | 1991 | u64 private = ~(u32)0; |
07157aac | 1992 | int ret; |
ff79f819 CM |
1993 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1994 | u32 csum = ~(u32)0; | |
d1310b2e | 1995 | |
d20f7043 CM |
1996 | if (PageChecked(page)) { |
1997 | ClearPageChecked(page); | |
1998 | goto good; | |
1999 | } | |
6cbff00f CH |
2000 | |
2001 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2002 | goto good; |
17d217fe YZ |
2003 | |
2004 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2005 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2006 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2007 | GFP_NOFS); | |
b6cda9bc | 2008 | return 0; |
17d217fe | 2009 | } |
d20f7043 | 2010 | |
c2e639f0 | 2011 | if (state && state->start == start) { |
70dec807 CM |
2012 | private = state->private; |
2013 | ret = 0; | |
2014 | } else { | |
2015 | ret = get_state_private(io_tree, start, &private); | |
2016 | } | |
7ac687d9 | 2017 | kaddr = kmap_atomic(page); |
d397712b | 2018 | if (ret) |
07157aac | 2019 | goto zeroit; |
d397712b | 2020 | |
ff79f819 CM |
2021 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2022 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2023 | if (csum != private) |
07157aac | 2024 | goto zeroit; |
d397712b | 2025 | |
7ac687d9 | 2026 | kunmap_atomic(kaddr); |
d20f7043 | 2027 | good: |
07157aac CM |
2028 | return 0; |
2029 | ||
2030 | zeroit: | |
945d8962 | 2031 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2032 | "private %llu\n", |
2033 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2034 | (unsigned long long)start, csum, |
2035 | (unsigned long long)private); | |
db94535d CM |
2036 | memset(kaddr + offset, 1, end - start + 1); |
2037 | flush_dcache_page(page); | |
7ac687d9 | 2038 | kunmap_atomic(kaddr); |
3b951516 CM |
2039 | if (private == 0) |
2040 | return 0; | |
7e38326f | 2041 | return -EIO; |
07157aac | 2042 | } |
b888db2b | 2043 | |
24bbcf04 YZ |
2044 | struct delayed_iput { |
2045 | struct list_head list; | |
2046 | struct inode *inode; | |
2047 | }; | |
2048 | ||
79787eaa JM |
2049 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2050 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2051 | void btrfs_add_delayed_iput(struct inode *inode) |
2052 | { | |
2053 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2054 | struct delayed_iput *delayed; | |
2055 | ||
2056 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2057 | return; | |
2058 | ||
2059 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2060 | delayed->inode = inode; | |
2061 | ||
2062 | spin_lock(&fs_info->delayed_iput_lock); | |
2063 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2064 | spin_unlock(&fs_info->delayed_iput_lock); | |
2065 | } | |
2066 | ||
2067 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2068 | { | |
2069 | LIST_HEAD(list); | |
2070 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2071 | struct delayed_iput *delayed; | |
2072 | int empty; | |
2073 | ||
2074 | spin_lock(&fs_info->delayed_iput_lock); | |
2075 | empty = list_empty(&fs_info->delayed_iputs); | |
2076 | spin_unlock(&fs_info->delayed_iput_lock); | |
2077 | if (empty) | |
2078 | return; | |
2079 | ||
2080 | down_read(&root->fs_info->cleanup_work_sem); | |
2081 | spin_lock(&fs_info->delayed_iput_lock); | |
2082 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2083 | spin_unlock(&fs_info->delayed_iput_lock); | |
2084 | ||
2085 | while (!list_empty(&list)) { | |
2086 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2087 | list_del(&delayed->list); | |
2088 | iput(delayed->inode); | |
2089 | kfree(delayed); | |
2090 | } | |
2091 | up_read(&root->fs_info->cleanup_work_sem); | |
2092 | } | |
2093 | ||
d68fc57b YZ |
2094 | enum btrfs_orphan_cleanup_state { |
2095 | ORPHAN_CLEANUP_STARTED = 1, | |
2096 | ORPHAN_CLEANUP_DONE = 2, | |
2097 | }; | |
2098 | ||
2099 | /* | |
42b2aa86 | 2100 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2101 | * files in the subvolume, it removes orphan item and frees block_rsv |
2102 | * structure. | |
2103 | */ | |
2104 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2105 | struct btrfs_root *root) | |
2106 | { | |
90290e19 | 2107 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2108 | int ret; |
2109 | ||
8a35d95f | 2110 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2111 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2112 | return; | |
2113 | ||
90290e19 | 2114 | spin_lock(&root->orphan_lock); |
8a35d95f | 2115 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2116 | spin_unlock(&root->orphan_lock); |
2117 | return; | |
2118 | } | |
2119 | ||
2120 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2121 | spin_unlock(&root->orphan_lock); | |
2122 | return; | |
2123 | } | |
2124 | ||
2125 | block_rsv = root->orphan_block_rsv; | |
2126 | root->orphan_block_rsv = NULL; | |
2127 | spin_unlock(&root->orphan_lock); | |
2128 | ||
d68fc57b YZ |
2129 | if (root->orphan_item_inserted && |
2130 | btrfs_root_refs(&root->root_item) > 0) { | |
2131 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2132 | root->root_key.objectid); | |
2133 | BUG_ON(ret); | |
2134 | root->orphan_item_inserted = 0; | |
2135 | } | |
2136 | ||
90290e19 JB |
2137 | if (block_rsv) { |
2138 | WARN_ON(block_rsv->size > 0); | |
2139 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2140 | } |
2141 | } | |
2142 | ||
7b128766 JB |
2143 | /* |
2144 | * This creates an orphan entry for the given inode in case something goes | |
2145 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2146 | * |
2147 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2148 | * this function. | |
7b128766 JB |
2149 | */ |
2150 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2151 | { | |
2152 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2153 | struct btrfs_block_rsv *block_rsv = NULL; |
2154 | int reserve = 0; | |
2155 | int insert = 0; | |
2156 | int ret; | |
7b128766 | 2157 | |
d68fc57b YZ |
2158 | if (!root->orphan_block_rsv) { |
2159 | block_rsv = btrfs_alloc_block_rsv(root); | |
b532402e TI |
2160 | if (!block_rsv) |
2161 | return -ENOMEM; | |
d68fc57b | 2162 | } |
7b128766 | 2163 | |
d68fc57b YZ |
2164 | spin_lock(&root->orphan_lock); |
2165 | if (!root->orphan_block_rsv) { | |
2166 | root->orphan_block_rsv = block_rsv; | |
2167 | } else if (block_rsv) { | |
2168 | btrfs_free_block_rsv(root, block_rsv); | |
2169 | block_rsv = NULL; | |
7b128766 | 2170 | } |
7b128766 | 2171 | |
8a35d95f JB |
2172 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2173 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2174 | #if 0 |
2175 | /* | |
2176 | * For proper ENOSPC handling, we should do orphan | |
2177 | * cleanup when mounting. But this introduces backward | |
2178 | * compatibility issue. | |
2179 | */ | |
2180 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2181 | insert = 2; | |
2182 | else | |
2183 | insert = 1; | |
2184 | #endif | |
2185 | insert = 1; | |
8a35d95f | 2186 | atomic_dec(&root->orphan_inodes); |
7b128766 JB |
2187 | } |
2188 | ||
72ac3c0d JB |
2189 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2190 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2191 | reserve = 1; |
d68fc57b | 2192 | spin_unlock(&root->orphan_lock); |
7b128766 | 2193 | |
d68fc57b YZ |
2194 | /* grab metadata reservation from transaction handle */ |
2195 | if (reserve) { | |
2196 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2197 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2198 | } |
7b128766 | 2199 | |
d68fc57b YZ |
2200 | /* insert an orphan item to track this unlinked/truncated file */ |
2201 | if (insert >= 1) { | |
33345d01 | 2202 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2203 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2204 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2205 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2206 | btrfs_abort_transaction(trans, root, ret); |
2207 | return ret; | |
2208 | } | |
2209 | ret = 0; | |
d68fc57b YZ |
2210 | } |
2211 | ||
2212 | /* insert an orphan item to track subvolume contains orphan files */ | |
2213 | if (insert >= 2) { | |
2214 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2215 | root->root_key.objectid); | |
79787eaa JM |
2216 | if (ret && ret != -EEXIST) { |
2217 | btrfs_abort_transaction(trans, root, ret); | |
2218 | return ret; | |
2219 | } | |
d68fc57b YZ |
2220 | } |
2221 | return 0; | |
7b128766 JB |
2222 | } |
2223 | ||
2224 | /* | |
2225 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2226 | * item for this particular inode. | |
2227 | */ | |
2228 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2229 | { | |
2230 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2231 | int delete_item = 0; |
2232 | int release_rsv = 0; | |
7b128766 JB |
2233 | int ret = 0; |
2234 | ||
d68fc57b | 2235 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2236 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2237 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2238 | delete_item = 1; |
7b128766 | 2239 | |
72ac3c0d JB |
2240 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2241 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2242 | release_rsv = 1; |
d68fc57b | 2243 | spin_unlock(&root->orphan_lock); |
7b128766 | 2244 | |
d68fc57b | 2245 | if (trans && delete_item) { |
33345d01 | 2246 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2247 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2248 | } |
7b128766 | 2249 | |
8a35d95f | 2250 | if (release_rsv) { |
d68fc57b | 2251 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2252 | atomic_dec(&root->orphan_inodes); |
2253 | } | |
7b128766 | 2254 | |
d68fc57b | 2255 | return 0; |
7b128766 JB |
2256 | } |
2257 | ||
2258 | /* | |
2259 | * this cleans up any orphans that may be left on the list from the last use | |
2260 | * of this root. | |
2261 | */ | |
66b4ffd1 | 2262 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2263 | { |
2264 | struct btrfs_path *path; | |
2265 | struct extent_buffer *leaf; | |
7b128766 JB |
2266 | struct btrfs_key key, found_key; |
2267 | struct btrfs_trans_handle *trans; | |
2268 | struct inode *inode; | |
8f6d7f4f | 2269 | u64 last_objectid = 0; |
7b128766 JB |
2270 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2271 | ||
d68fc57b | 2272 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2273 | return 0; |
c71bf099 YZ |
2274 | |
2275 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2276 | if (!path) { |
2277 | ret = -ENOMEM; | |
2278 | goto out; | |
2279 | } | |
7b128766 JB |
2280 | path->reada = -1; |
2281 | ||
2282 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2283 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2284 | key.offset = (u64)-1; | |
2285 | ||
7b128766 JB |
2286 | while (1) { |
2287 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2288 | if (ret < 0) |
2289 | goto out; | |
7b128766 JB |
2290 | |
2291 | /* | |
2292 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2293 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2294 | * find the key and see if we have stuff that matches |
2295 | */ | |
2296 | if (ret > 0) { | |
66b4ffd1 | 2297 | ret = 0; |
7b128766 JB |
2298 | if (path->slots[0] == 0) |
2299 | break; | |
2300 | path->slots[0]--; | |
2301 | } | |
2302 | ||
2303 | /* pull out the item */ | |
2304 | leaf = path->nodes[0]; | |
7b128766 JB |
2305 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2306 | ||
2307 | /* make sure the item matches what we want */ | |
2308 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2309 | break; | |
2310 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2311 | break; | |
2312 | ||
2313 | /* release the path since we're done with it */ | |
b3b4aa74 | 2314 | btrfs_release_path(path); |
7b128766 JB |
2315 | |
2316 | /* | |
2317 | * this is where we are basically btrfs_lookup, without the | |
2318 | * crossing root thing. we store the inode number in the | |
2319 | * offset of the orphan item. | |
2320 | */ | |
8f6d7f4f JB |
2321 | |
2322 | if (found_key.offset == last_objectid) { | |
2323 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2324 | "stopping orphan cleanup\n"); | |
2325 | ret = -EINVAL; | |
2326 | goto out; | |
2327 | } | |
2328 | ||
2329 | last_objectid = found_key.offset; | |
2330 | ||
5d4f98a2 YZ |
2331 | found_key.objectid = found_key.offset; |
2332 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2333 | found_key.offset = 0; | |
73f73415 | 2334 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2335 | ret = PTR_RET(inode); |
2336 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2337 | goto out; |
7b128766 | 2338 | |
f8e9e0b0 AJ |
2339 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2340 | struct btrfs_root *dead_root; | |
2341 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2342 | int is_dead_root = 0; | |
2343 | ||
2344 | /* | |
2345 | * this is an orphan in the tree root. Currently these | |
2346 | * could come from 2 sources: | |
2347 | * a) a snapshot deletion in progress | |
2348 | * b) a free space cache inode | |
2349 | * We need to distinguish those two, as the snapshot | |
2350 | * orphan must not get deleted. | |
2351 | * find_dead_roots already ran before us, so if this | |
2352 | * is a snapshot deletion, we should find the root | |
2353 | * in the dead_roots list | |
2354 | */ | |
2355 | spin_lock(&fs_info->trans_lock); | |
2356 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2357 | root_list) { | |
2358 | if (dead_root->root_key.objectid == | |
2359 | found_key.objectid) { | |
2360 | is_dead_root = 1; | |
2361 | break; | |
2362 | } | |
2363 | } | |
2364 | spin_unlock(&fs_info->trans_lock); | |
2365 | if (is_dead_root) { | |
2366 | /* prevent this orphan from being found again */ | |
2367 | key.offset = found_key.objectid - 1; | |
2368 | continue; | |
2369 | } | |
2370 | } | |
7b128766 | 2371 | /* |
a8c9e576 JB |
2372 | * Inode is already gone but the orphan item is still there, |
2373 | * kill the orphan item. | |
7b128766 | 2374 | */ |
a8c9e576 JB |
2375 | if (ret == -ESTALE) { |
2376 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2377 | if (IS_ERR(trans)) { |
2378 | ret = PTR_ERR(trans); | |
2379 | goto out; | |
2380 | } | |
8a35d95f JB |
2381 | printk(KERN_ERR "auto deleting %Lu\n", |
2382 | found_key.objectid); | |
a8c9e576 JB |
2383 | ret = btrfs_del_orphan_item(trans, root, |
2384 | found_key.objectid); | |
79787eaa | 2385 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2386 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2387 | continue; |
2388 | } | |
2389 | ||
a8c9e576 JB |
2390 | /* |
2391 | * add this inode to the orphan list so btrfs_orphan_del does | |
2392 | * the proper thing when we hit it | |
2393 | */ | |
8a35d95f JB |
2394 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2395 | &BTRFS_I(inode)->runtime_flags); | |
a8c9e576 | 2396 | |
7b128766 JB |
2397 | /* if we have links, this was a truncate, lets do that */ |
2398 | if (inode->i_nlink) { | |
a41ad394 JB |
2399 | if (!S_ISREG(inode->i_mode)) { |
2400 | WARN_ON(1); | |
2401 | iput(inode); | |
2402 | continue; | |
2403 | } | |
7b128766 | 2404 | nr_truncate++; |
66b4ffd1 | 2405 | ret = btrfs_truncate(inode); |
7b128766 JB |
2406 | } else { |
2407 | nr_unlink++; | |
2408 | } | |
2409 | ||
2410 | /* this will do delete_inode and everything for us */ | |
2411 | iput(inode); | |
66b4ffd1 JB |
2412 | if (ret) |
2413 | goto out; | |
7b128766 | 2414 | } |
3254c876 MX |
2415 | /* release the path since we're done with it */ |
2416 | btrfs_release_path(path); | |
2417 | ||
d68fc57b YZ |
2418 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2419 | ||
2420 | if (root->orphan_block_rsv) | |
2421 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2422 | (u64)-1); | |
2423 | ||
2424 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2425 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2426 | if (!IS_ERR(trans)) |
2427 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2428 | } |
7b128766 JB |
2429 | |
2430 | if (nr_unlink) | |
2431 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2432 | if (nr_truncate) | |
2433 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2434 | |
2435 | out: | |
2436 | if (ret) | |
2437 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2438 | btrfs_free_path(path); | |
2439 | return ret; | |
7b128766 JB |
2440 | } |
2441 | ||
46a53cca CM |
2442 | /* |
2443 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2444 | * don't find any xattrs, we know there can't be any acls. | |
2445 | * | |
2446 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2447 | */ | |
2448 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2449 | int slot, u64 objectid) | |
2450 | { | |
2451 | u32 nritems = btrfs_header_nritems(leaf); | |
2452 | struct btrfs_key found_key; | |
2453 | int scanned = 0; | |
2454 | ||
2455 | slot++; | |
2456 | while (slot < nritems) { | |
2457 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2458 | ||
2459 | /* we found a different objectid, there must not be acls */ | |
2460 | if (found_key.objectid != objectid) | |
2461 | return 0; | |
2462 | ||
2463 | /* we found an xattr, assume we've got an acl */ | |
2464 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2465 | return 1; | |
2466 | ||
2467 | /* | |
2468 | * we found a key greater than an xattr key, there can't | |
2469 | * be any acls later on | |
2470 | */ | |
2471 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2472 | return 0; | |
2473 | ||
2474 | slot++; | |
2475 | scanned++; | |
2476 | ||
2477 | /* | |
2478 | * it goes inode, inode backrefs, xattrs, extents, | |
2479 | * so if there are a ton of hard links to an inode there can | |
2480 | * be a lot of backrefs. Don't waste time searching too hard, | |
2481 | * this is just an optimization | |
2482 | */ | |
2483 | if (scanned >= 8) | |
2484 | break; | |
2485 | } | |
2486 | /* we hit the end of the leaf before we found an xattr or | |
2487 | * something larger than an xattr. We have to assume the inode | |
2488 | * has acls | |
2489 | */ | |
2490 | return 1; | |
2491 | } | |
2492 | ||
d352ac68 CM |
2493 | /* |
2494 | * read an inode from the btree into the in-memory inode | |
2495 | */ | |
5d4f98a2 | 2496 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2497 | { |
2498 | struct btrfs_path *path; | |
5f39d397 | 2499 | struct extent_buffer *leaf; |
39279cc3 | 2500 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2501 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2502 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2503 | struct btrfs_key location; | |
46a53cca | 2504 | int maybe_acls; |
618e21d5 | 2505 | u32 rdev; |
39279cc3 | 2506 | int ret; |
2f7e33d4 MX |
2507 | bool filled = false; |
2508 | ||
2509 | ret = btrfs_fill_inode(inode, &rdev); | |
2510 | if (!ret) | |
2511 | filled = true; | |
39279cc3 CM |
2512 | |
2513 | path = btrfs_alloc_path(); | |
1748f843 MF |
2514 | if (!path) |
2515 | goto make_bad; | |
2516 | ||
d90c7321 | 2517 | path->leave_spinning = 1; |
39279cc3 | 2518 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2519 | |
39279cc3 | 2520 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2521 | if (ret) |
39279cc3 | 2522 | goto make_bad; |
39279cc3 | 2523 | |
5f39d397 | 2524 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2525 | |
2526 | if (filled) | |
2527 | goto cache_acl; | |
2528 | ||
5f39d397 CM |
2529 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2530 | struct btrfs_inode_item); | |
5f39d397 | 2531 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2532 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
5f39d397 CM |
2533 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); |
2534 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2535 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2536 | |
2537 | tspec = btrfs_inode_atime(inode_item); | |
2538 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2539 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2540 | ||
2541 | tspec = btrfs_inode_mtime(inode_item); | |
2542 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2543 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2544 | ||
2545 | tspec = btrfs_inode_ctime(inode_item); | |
2546 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2547 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2548 | ||
a76a3cd4 | 2549 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2550 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
0c4d2d95 | 2551 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2552 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2553 | inode->i_rdev = 0; |
5f39d397 CM |
2554 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2555 | ||
aec7477b | 2556 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2557 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2558 | cache_acl: |
46a53cca CM |
2559 | /* |
2560 | * try to precache a NULL acl entry for files that don't have | |
2561 | * any xattrs or acls | |
2562 | */ | |
33345d01 LZ |
2563 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2564 | btrfs_ino(inode)); | |
72c04902 AV |
2565 | if (!maybe_acls) |
2566 | cache_no_acl(inode); | |
46a53cca | 2567 | |
39279cc3 | 2568 | btrfs_free_path(path); |
39279cc3 | 2569 | |
39279cc3 | 2570 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2571 | case S_IFREG: |
2572 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2573 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2574 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2575 | inode->i_fop = &btrfs_file_operations; |
2576 | inode->i_op = &btrfs_file_inode_operations; | |
2577 | break; | |
2578 | case S_IFDIR: | |
2579 | inode->i_fop = &btrfs_dir_file_operations; | |
2580 | if (root == root->fs_info->tree_root) | |
2581 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2582 | else | |
2583 | inode->i_op = &btrfs_dir_inode_operations; | |
2584 | break; | |
2585 | case S_IFLNK: | |
2586 | inode->i_op = &btrfs_symlink_inode_operations; | |
2587 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2588 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2589 | break; |
618e21d5 | 2590 | default: |
0279b4cd | 2591 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2592 | init_special_inode(inode, inode->i_mode, rdev); |
2593 | break; | |
39279cc3 | 2594 | } |
6cbff00f CH |
2595 | |
2596 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2597 | return; |
2598 | ||
2599 | make_bad: | |
39279cc3 | 2600 | btrfs_free_path(path); |
39279cc3 CM |
2601 | make_bad_inode(inode); |
2602 | } | |
2603 | ||
d352ac68 CM |
2604 | /* |
2605 | * given a leaf and an inode, copy the inode fields into the leaf | |
2606 | */ | |
e02119d5 CM |
2607 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2608 | struct extent_buffer *leaf, | |
5f39d397 | 2609 | struct btrfs_inode_item *item, |
39279cc3 CM |
2610 | struct inode *inode) |
2611 | { | |
5f39d397 CM |
2612 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2613 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2614 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2615 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2616 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2617 | ||
2618 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2619 | inode->i_atime.tv_sec); | |
2620 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2621 | inode->i_atime.tv_nsec); | |
2622 | ||
2623 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2624 | inode->i_mtime.tv_sec); | |
2625 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2626 | inode->i_mtime.tv_nsec); | |
2627 | ||
2628 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2629 | inode->i_ctime.tv_sec); | |
2630 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2631 | inode->i_ctime.tv_nsec); | |
2632 | ||
a76a3cd4 | 2633 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2634 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
0c4d2d95 | 2635 | btrfs_set_inode_sequence(leaf, item, inode->i_version); |
e02119d5 | 2636 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2637 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2638 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2639 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2640 | } |
2641 | ||
d352ac68 CM |
2642 | /* |
2643 | * copy everything in the in-memory inode into the btree. | |
2644 | */ | |
2115133f | 2645 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2646 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2647 | { |
2648 | struct btrfs_inode_item *inode_item; | |
2649 | struct btrfs_path *path; | |
5f39d397 | 2650 | struct extent_buffer *leaf; |
39279cc3 CM |
2651 | int ret; |
2652 | ||
2653 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2654 | if (!path) |
2655 | return -ENOMEM; | |
2656 | ||
b9473439 | 2657 | path->leave_spinning = 1; |
16cdcec7 MX |
2658 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2659 | 1); | |
39279cc3 CM |
2660 | if (ret) { |
2661 | if (ret > 0) | |
2662 | ret = -ENOENT; | |
2663 | goto failed; | |
2664 | } | |
2665 | ||
b4ce94de | 2666 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2667 | leaf = path->nodes[0]; |
2668 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2669 | struct btrfs_inode_item); |
39279cc3 | 2670 | |
e02119d5 | 2671 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2672 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2673 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2674 | ret = 0; |
2675 | failed: | |
39279cc3 CM |
2676 | btrfs_free_path(path); |
2677 | return ret; | |
2678 | } | |
2679 | ||
2115133f CM |
2680 | /* |
2681 | * copy everything in the in-memory inode into the btree. | |
2682 | */ | |
2683 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2684 | struct btrfs_root *root, struct inode *inode) | |
2685 | { | |
2686 | int ret; | |
2687 | ||
2688 | /* | |
2689 | * If the inode is a free space inode, we can deadlock during commit | |
2690 | * if we put it into the delayed code. | |
2691 | * | |
2692 | * The data relocation inode should also be directly updated | |
2693 | * without delay | |
2694 | */ | |
2695 | if (!btrfs_is_free_space_inode(root, inode) | |
2696 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
2697 | ret = btrfs_delayed_update_inode(trans, root, inode); | |
2698 | if (!ret) | |
2699 | btrfs_set_inode_last_trans(trans, inode); | |
2700 | return ret; | |
2701 | } | |
2702 | ||
2703 | return btrfs_update_inode_item(trans, root, inode); | |
2704 | } | |
2705 | ||
2706 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, | |
2707 | struct btrfs_root *root, struct inode *inode) | |
2708 | { | |
2709 | int ret; | |
2710 | ||
2711 | ret = btrfs_update_inode(trans, root, inode); | |
2712 | if (ret == -ENOSPC) | |
2713 | return btrfs_update_inode_item(trans, root, inode); | |
2714 | return ret; | |
2715 | } | |
2716 | ||
d352ac68 CM |
2717 | /* |
2718 | * unlink helper that gets used here in inode.c and in the tree logging | |
2719 | * recovery code. It remove a link in a directory with a given name, and | |
2720 | * also drops the back refs in the inode to the directory | |
2721 | */ | |
92986796 AV |
2722 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2723 | struct btrfs_root *root, | |
2724 | struct inode *dir, struct inode *inode, | |
2725 | const char *name, int name_len) | |
39279cc3 CM |
2726 | { |
2727 | struct btrfs_path *path; | |
39279cc3 | 2728 | int ret = 0; |
5f39d397 | 2729 | struct extent_buffer *leaf; |
39279cc3 | 2730 | struct btrfs_dir_item *di; |
5f39d397 | 2731 | struct btrfs_key key; |
aec7477b | 2732 | u64 index; |
33345d01 LZ |
2733 | u64 ino = btrfs_ino(inode); |
2734 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2735 | |
2736 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2737 | if (!path) { |
2738 | ret = -ENOMEM; | |
554233a6 | 2739 | goto out; |
54aa1f4d CM |
2740 | } |
2741 | ||
b9473439 | 2742 | path->leave_spinning = 1; |
33345d01 | 2743 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2744 | name, name_len, -1); |
2745 | if (IS_ERR(di)) { | |
2746 | ret = PTR_ERR(di); | |
2747 | goto err; | |
2748 | } | |
2749 | if (!di) { | |
2750 | ret = -ENOENT; | |
2751 | goto err; | |
2752 | } | |
5f39d397 CM |
2753 | leaf = path->nodes[0]; |
2754 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2755 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2756 | if (ret) |
2757 | goto err; | |
b3b4aa74 | 2758 | btrfs_release_path(path); |
39279cc3 | 2759 | |
33345d01 LZ |
2760 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2761 | dir_ino, &index); | |
aec7477b | 2762 | if (ret) { |
d397712b | 2763 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2764 | "inode %llu parent %llu\n", name_len, name, |
2765 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2766 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2767 | goto err; |
2768 | } | |
2769 | ||
16cdcec7 | 2770 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2771 | if (ret) { |
2772 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2773 | goto err; |
79787eaa | 2774 | } |
39279cc3 | 2775 | |
e02119d5 | 2776 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2777 | inode, dir_ino); |
79787eaa JM |
2778 | if (ret != 0 && ret != -ENOENT) { |
2779 | btrfs_abort_transaction(trans, root, ret); | |
2780 | goto err; | |
2781 | } | |
e02119d5 CM |
2782 | |
2783 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2784 | dir, index); | |
6418c961 CM |
2785 | if (ret == -ENOENT) |
2786 | ret = 0; | |
39279cc3 CM |
2787 | err: |
2788 | btrfs_free_path(path); | |
e02119d5 CM |
2789 | if (ret) |
2790 | goto out; | |
2791 | ||
2792 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2793 | inode_inc_iversion(inode); |
2794 | inode_inc_iversion(dir); | |
e02119d5 CM |
2795 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
2796 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2797 | out: |
39279cc3 CM |
2798 | return ret; |
2799 | } | |
2800 | ||
92986796 AV |
2801 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2802 | struct btrfs_root *root, | |
2803 | struct inode *dir, struct inode *inode, | |
2804 | const char *name, int name_len) | |
2805 | { | |
2806 | int ret; | |
2807 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2808 | if (!ret) { | |
2809 | btrfs_drop_nlink(inode); | |
2810 | ret = btrfs_update_inode(trans, root, inode); | |
2811 | } | |
2812 | return ret; | |
2813 | } | |
2814 | ||
2815 | ||
a22285a6 YZ |
2816 | /* helper to check if there is any shared block in the path */ |
2817 | static int check_path_shared(struct btrfs_root *root, | |
2818 | struct btrfs_path *path) | |
39279cc3 | 2819 | { |
a22285a6 YZ |
2820 | struct extent_buffer *eb; |
2821 | int level; | |
0e4dcbef | 2822 | u64 refs = 1; |
5df6a9f6 | 2823 | |
a22285a6 | 2824 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2825 | int ret; |
2826 | ||
a22285a6 YZ |
2827 | if (!path->nodes[level]) |
2828 | break; | |
2829 | eb = path->nodes[level]; | |
2830 | if (!btrfs_block_can_be_shared(root, eb)) | |
2831 | continue; | |
2832 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2833 | &refs, NULL); | |
2834 | if (refs > 1) | |
2835 | return 1; | |
5df6a9f6 | 2836 | } |
dedefd72 | 2837 | return 0; |
39279cc3 CM |
2838 | } |
2839 | ||
a22285a6 YZ |
2840 | /* |
2841 | * helper to start transaction for unlink and rmdir. | |
2842 | * | |
2843 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2844 | * so in enospc case, we should make sure they will free space before | |
2845 | * allowing them to use the global metadata reservation. | |
2846 | */ | |
2847 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2848 | struct dentry *dentry) | |
4df27c4d | 2849 | { |
39279cc3 | 2850 | struct btrfs_trans_handle *trans; |
a22285a6 | 2851 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2852 | struct btrfs_path *path; |
a22285a6 | 2853 | struct btrfs_inode_ref *ref; |
4df27c4d | 2854 | struct btrfs_dir_item *di; |
7b128766 | 2855 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2856 | u64 index; |
a22285a6 YZ |
2857 | int check_link = 1; |
2858 | int err = -ENOSPC; | |
4df27c4d | 2859 | int ret; |
33345d01 LZ |
2860 | u64 ino = btrfs_ino(inode); |
2861 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2862 | |
e70bea5f JB |
2863 | /* |
2864 | * 1 for the possible orphan item | |
2865 | * 1 for the dir item | |
2866 | * 1 for the dir index | |
2867 | * 1 for the inode ref | |
2868 | * 1 for the inode ref in the tree log | |
2869 | * 2 for the dir entries in the log | |
2870 | * 1 for the inode | |
2871 | */ | |
2872 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2873 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2874 | return trans; | |
4df27c4d | 2875 | |
33345d01 | 2876 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2877 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2878 | |
a22285a6 YZ |
2879 | /* check if there is someone else holds reference */ |
2880 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2881 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2882 | |
a22285a6 YZ |
2883 | if (atomic_read(&inode->i_count) > 2) |
2884 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2885 | |
a22285a6 YZ |
2886 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2887 | return ERR_PTR(-ENOSPC); | |
2888 | ||
2889 | path = btrfs_alloc_path(); | |
2890 | if (!path) { | |
2891 | root->fs_info->enospc_unlink = 0; | |
2892 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2893 | } |
2894 | ||
3880a1b4 JB |
2895 | /* 1 for the orphan item */ |
2896 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2897 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2898 | btrfs_free_path(path); |
2899 | root->fs_info->enospc_unlink = 0; | |
2900 | return trans; | |
2901 | } | |
4df27c4d | 2902 | |
a22285a6 YZ |
2903 | path->skip_locking = 1; |
2904 | path->search_commit_root = 1; | |
4df27c4d | 2905 | |
a22285a6 YZ |
2906 | ret = btrfs_lookup_inode(trans, root, path, |
2907 | &BTRFS_I(dir)->location, 0); | |
2908 | if (ret < 0) { | |
2909 | err = ret; | |
2910 | goto out; | |
2911 | } | |
2912 | if (ret == 0) { | |
2913 | if (check_path_shared(root, path)) | |
2914 | goto out; | |
2915 | } else { | |
2916 | check_link = 0; | |
5df6a9f6 | 2917 | } |
b3b4aa74 | 2918 | btrfs_release_path(path); |
a22285a6 YZ |
2919 | |
2920 | ret = btrfs_lookup_inode(trans, root, path, | |
2921 | &BTRFS_I(inode)->location, 0); | |
2922 | if (ret < 0) { | |
2923 | err = ret; | |
2924 | goto out; | |
2925 | } | |
2926 | if (ret == 0) { | |
2927 | if (check_path_shared(root, path)) | |
2928 | goto out; | |
2929 | } else { | |
2930 | check_link = 0; | |
2931 | } | |
b3b4aa74 | 2932 | btrfs_release_path(path); |
a22285a6 YZ |
2933 | |
2934 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2935 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2936 | ino, (u64)-1, 0); |
a22285a6 YZ |
2937 | if (ret < 0) { |
2938 | err = ret; | |
2939 | goto out; | |
2940 | } | |
79787eaa | 2941 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
2942 | if (check_path_shared(root, path)) |
2943 | goto out; | |
b3b4aa74 | 2944 | btrfs_release_path(path); |
a22285a6 YZ |
2945 | } |
2946 | ||
2947 | if (!check_link) { | |
2948 | err = 0; | |
2949 | goto out; | |
2950 | } | |
2951 | ||
33345d01 | 2952 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2953 | dentry->d_name.name, dentry->d_name.len, 0); |
2954 | if (IS_ERR(di)) { | |
2955 | err = PTR_ERR(di); | |
2956 | goto out; | |
2957 | } | |
2958 | if (di) { | |
2959 | if (check_path_shared(root, path)) | |
2960 | goto out; | |
2961 | } else { | |
2962 | err = 0; | |
2963 | goto out; | |
2964 | } | |
b3b4aa74 | 2965 | btrfs_release_path(path); |
a22285a6 YZ |
2966 | |
2967 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2968 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 2969 | ino, dir_ino, 0); |
a22285a6 YZ |
2970 | if (IS_ERR(ref)) { |
2971 | err = PTR_ERR(ref); | |
2972 | goto out; | |
2973 | } | |
79787eaa | 2974 | BUG_ON(!ref); /* Logic error */ |
a22285a6 YZ |
2975 | if (check_path_shared(root, path)) |
2976 | goto out; | |
2977 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 2978 | btrfs_release_path(path); |
a22285a6 | 2979 | |
16cdcec7 MX |
2980 | /* |
2981 | * This is a commit root search, if we can lookup inode item and other | |
2982 | * relative items in the commit root, it means the transaction of | |
2983 | * dir/file creation has been committed, and the dir index item that we | |
2984 | * delay to insert has also been inserted into the commit root. So | |
2985 | * we needn't worry about the delayed insertion of the dir index item | |
2986 | * here. | |
2987 | */ | |
33345d01 | 2988 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
2989 | dentry->d_name.name, dentry->d_name.len, 0); |
2990 | if (IS_ERR(di)) { | |
2991 | err = PTR_ERR(di); | |
2992 | goto out; | |
2993 | } | |
2994 | BUG_ON(ret == -ENOENT); | |
2995 | if (check_path_shared(root, path)) | |
2996 | goto out; | |
2997 | ||
2998 | err = 0; | |
2999 | out: | |
3000 | btrfs_free_path(path); | |
3880a1b4 JB |
3001 | /* Migrate the orphan reservation over */ |
3002 | if (!err) | |
3003 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3004 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3005 | trans->bytes_reserved); |
3880a1b4 | 3006 | |
a22285a6 YZ |
3007 | if (err) { |
3008 | btrfs_end_transaction(trans, root); | |
3009 | root->fs_info->enospc_unlink = 0; | |
3010 | return ERR_PTR(err); | |
3011 | } | |
3012 | ||
3013 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3014 | return trans; | |
3015 | } | |
3016 | ||
3017 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3018 | struct btrfs_root *root) | |
3019 | { | |
3020 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
5a77d76c JB |
3021 | btrfs_block_rsv_release(root, trans->block_rsv, |
3022 | trans->bytes_reserved); | |
3023 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3024 | BUG_ON(!root->fs_info->enospc_unlink); |
3025 | root->fs_info->enospc_unlink = 0; | |
3026 | } | |
7ad85bb7 | 3027 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3028 | } |
3029 | ||
3030 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3031 | { | |
3032 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3033 | struct btrfs_trans_handle *trans; | |
3034 | struct inode *inode = dentry->d_inode; | |
3035 | int ret; | |
3036 | unsigned long nr = 0; | |
3037 | ||
3038 | trans = __unlink_start_trans(dir, dentry); | |
3039 | if (IS_ERR(trans)) | |
3040 | return PTR_ERR(trans); | |
5f39d397 | 3041 | |
12fcfd22 CM |
3042 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3043 | ||
e02119d5 CM |
3044 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3045 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3046 | if (ret) |
3047 | goto out; | |
7b128766 | 3048 | |
a22285a6 | 3049 | if (inode->i_nlink == 0) { |
7b128766 | 3050 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3051 | if (ret) |
3052 | goto out; | |
a22285a6 | 3053 | } |
7b128766 | 3054 | |
b532402e | 3055 | out: |
d3c2fdcf | 3056 | nr = trans->blocks_used; |
a22285a6 | 3057 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3058 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3059 | return ret; |
3060 | } | |
3061 | ||
4df27c4d YZ |
3062 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3063 | struct btrfs_root *root, | |
3064 | struct inode *dir, u64 objectid, | |
3065 | const char *name, int name_len) | |
3066 | { | |
3067 | struct btrfs_path *path; | |
3068 | struct extent_buffer *leaf; | |
3069 | struct btrfs_dir_item *di; | |
3070 | struct btrfs_key key; | |
3071 | u64 index; | |
3072 | int ret; | |
33345d01 | 3073 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3074 | |
3075 | path = btrfs_alloc_path(); | |
3076 | if (!path) | |
3077 | return -ENOMEM; | |
3078 | ||
33345d01 | 3079 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3080 | name, name_len, -1); |
79787eaa JM |
3081 | if (IS_ERR_OR_NULL(di)) { |
3082 | if (!di) | |
3083 | ret = -ENOENT; | |
3084 | else | |
3085 | ret = PTR_ERR(di); | |
3086 | goto out; | |
3087 | } | |
4df27c4d YZ |
3088 | |
3089 | leaf = path->nodes[0]; | |
3090 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3091 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3092 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3093 | if (ret) { |
3094 | btrfs_abort_transaction(trans, root, ret); | |
3095 | goto out; | |
3096 | } | |
b3b4aa74 | 3097 | btrfs_release_path(path); |
4df27c4d YZ |
3098 | |
3099 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3100 | objectid, root->root_key.objectid, | |
33345d01 | 3101 | dir_ino, &index, name, name_len); |
4df27c4d | 3102 | if (ret < 0) { |
79787eaa JM |
3103 | if (ret != -ENOENT) { |
3104 | btrfs_abort_transaction(trans, root, ret); | |
3105 | goto out; | |
3106 | } | |
33345d01 | 3107 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3108 | name, name_len); |
79787eaa JM |
3109 | if (IS_ERR_OR_NULL(di)) { |
3110 | if (!di) | |
3111 | ret = -ENOENT; | |
3112 | else | |
3113 | ret = PTR_ERR(di); | |
3114 | btrfs_abort_transaction(trans, root, ret); | |
3115 | goto out; | |
3116 | } | |
4df27c4d YZ |
3117 | |
3118 | leaf = path->nodes[0]; | |
3119 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3120 | btrfs_release_path(path); |
4df27c4d YZ |
3121 | index = key.offset; |
3122 | } | |
945d8962 | 3123 | btrfs_release_path(path); |
4df27c4d | 3124 | |
16cdcec7 | 3125 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3126 | if (ret) { |
3127 | btrfs_abort_transaction(trans, root, ret); | |
3128 | goto out; | |
3129 | } | |
4df27c4d YZ |
3130 | |
3131 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3132 | inode_inc_iversion(dir); |
4df27c4d YZ |
3133 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
3134 | ret = btrfs_update_inode(trans, root, dir); | |
79787eaa JM |
3135 | if (ret) |
3136 | btrfs_abort_transaction(trans, root, ret); | |
3137 | out: | |
71d7aed0 | 3138 | btrfs_free_path(path); |
79787eaa | 3139 | return ret; |
4df27c4d YZ |
3140 | } |
3141 | ||
39279cc3 CM |
3142 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3143 | { | |
3144 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3145 | int err = 0; |
39279cc3 | 3146 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3147 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3148 | unsigned long nr = 0; |
39279cc3 | 3149 | |
3394e160 | 3150 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 3151 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3152 | return -ENOTEMPTY; |
3153 | ||
a22285a6 YZ |
3154 | trans = __unlink_start_trans(dir, dentry); |
3155 | if (IS_ERR(trans)) | |
5df6a9f6 | 3156 | return PTR_ERR(trans); |
5df6a9f6 | 3157 | |
33345d01 | 3158 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3159 | err = btrfs_unlink_subvol(trans, root, dir, |
3160 | BTRFS_I(inode)->location.objectid, | |
3161 | dentry->d_name.name, | |
3162 | dentry->d_name.len); | |
3163 | goto out; | |
3164 | } | |
3165 | ||
7b128766 JB |
3166 | err = btrfs_orphan_add(trans, inode); |
3167 | if (err) | |
4df27c4d | 3168 | goto out; |
7b128766 | 3169 | |
39279cc3 | 3170 | /* now the directory is empty */ |
e02119d5 CM |
3171 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3172 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3173 | if (!err) |
dbe674a9 | 3174 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3175 | out: |
d3c2fdcf | 3176 | nr = trans->blocks_used; |
a22285a6 | 3177 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3178 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3179 | |
39279cc3 CM |
3180 | return err; |
3181 | } | |
3182 | ||
39279cc3 CM |
3183 | /* |
3184 | * this can truncate away extent items, csum items and directory items. | |
3185 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3186 | * any higher than new_size |
39279cc3 CM |
3187 | * |
3188 | * csum items that cross the new i_size are truncated to the new size | |
3189 | * as well. | |
7b128766 JB |
3190 | * |
3191 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3192 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3193 | */ |
8082510e YZ |
3194 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3195 | struct btrfs_root *root, | |
3196 | struct inode *inode, | |
3197 | u64 new_size, u32 min_type) | |
39279cc3 | 3198 | { |
39279cc3 | 3199 | struct btrfs_path *path; |
5f39d397 | 3200 | struct extent_buffer *leaf; |
39279cc3 | 3201 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3202 | struct btrfs_key key; |
3203 | struct btrfs_key found_key; | |
39279cc3 | 3204 | u64 extent_start = 0; |
db94535d | 3205 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3206 | u64 extent_offset = 0; |
39279cc3 | 3207 | u64 item_end = 0; |
8082510e YZ |
3208 | u64 mask = root->sectorsize - 1; |
3209 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3210 | int found_extent; |
3211 | int del_item; | |
85e21bac CM |
3212 | int pending_del_nr = 0; |
3213 | int pending_del_slot = 0; | |
179e29e4 | 3214 | int extent_type = -1; |
8082510e YZ |
3215 | int ret; |
3216 | int err = 0; | |
33345d01 | 3217 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3218 | |
3219 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3220 | |
0eb0e19c MF |
3221 | path = btrfs_alloc_path(); |
3222 | if (!path) | |
3223 | return -ENOMEM; | |
3224 | path->reada = -1; | |
3225 | ||
0af3d00b | 3226 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3227 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3228 | |
16cdcec7 MX |
3229 | /* |
3230 | * This function is also used to drop the items in the log tree before | |
3231 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3232 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3233 | * items. | |
3234 | */ | |
3235 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3236 | btrfs_kill_delayed_inode_items(inode); | |
3237 | ||
33345d01 | 3238 | key.objectid = ino; |
39279cc3 | 3239 | key.offset = (u64)-1; |
5f39d397 CM |
3240 | key.type = (u8)-1; |
3241 | ||
85e21bac | 3242 | search_again: |
b9473439 | 3243 | path->leave_spinning = 1; |
85e21bac | 3244 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3245 | if (ret < 0) { |
3246 | err = ret; | |
3247 | goto out; | |
3248 | } | |
d397712b | 3249 | |
85e21bac | 3250 | if (ret > 0) { |
e02119d5 CM |
3251 | /* there are no items in the tree for us to truncate, we're |
3252 | * done | |
3253 | */ | |
8082510e YZ |
3254 | if (path->slots[0] == 0) |
3255 | goto out; | |
85e21bac CM |
3256 | path->slots[0]--; |
3257 | } | |
3258 | ||
d397712b | 3259 | while (1) { |
39279cc3 | 3260 | fi = NULL; |
5f39d397 CM |
3261 | leaf = path->nodes[0]; |
3262 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3263 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3264 | |
33345d01 | 3265 | if (found_key.objectid != ino) |
39279cc3 | 3266 | break; |
5f39d397 | 3267 | |
85e21bac | 3268 | if (found_type < min_type) |
39279cc3 CM |
3269 | break; |
3270 | ||
5f39d397 | 3271 | item_end = found_key.offset; |
39279cc3 | 3272 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3273 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3274 | struct btrfs_file_extent_item); |
179e29e4 CM |
3275 | extent_type = btrfs_file_extent_type(leaf, fi); |
3276 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3277 | item_end += |
db94535d | 3278 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3279 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3280 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3281 | fi); |
39279cc3 | 3282 | } |
008630c1 | 3283 | item_end--; |
39279cc3 | 3284 | } |
8082510e YZ |
3285 | if (found_type > min_type) { |
3286 | del_item = 1; | |
3287 | } else { | |
3288 | if (item_end < new_size) | |
b888db2b | 3289 | break; |
8082510e YZ |
3290 | if (found_key.offset >= new_size) |
3291 | del_item = 1; | |
3292 | else | |
3293 | del_item = 0; | |
39279cc3 | 3294 | } |
39279cc3 | 3295 | found_extent = 0; |
39279cc3 | 3296 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3297 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3298 | goto delete; | |
3299 | ||
3300 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3301 | u64 num_dec; |
db94535d | 3302 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3303 | if (!del_item) { |
db94535d CM |
3304 | u64 orig_num_bytes = |
3305 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3306 | extent_num_bytes = new_size - |
5f39d397 | 3307 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3308 | extent_num_bytes = extent_num_bytes & |
3309 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3310 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3311 | extent_num_bytes); | |
3312 | num_dec = (orig_num_bytes - | |
9069218d | 3313 | extent_num_bytes); |
e02119d5 | 3314 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3315 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3316 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3317 | } else { |
db94535d CM |
3318 | extent_num_bytes = |
3319 | btrfs_file_extent_disk_num_bytes(leaf, | |
3320 | fi); | |
5d4f98a2 YZ |
3321 | extent_offset = found_key.offset - |
3322 | btrfs_file_extent_offset(leaf, fi); | |
3323 | ||
39279cc3 | 3324 | /* FIXME blocksize != 4096 */ |
9069218d | 3325 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3326 | if (extent_start != 0) { |
3327 | found_extent = 1; | |
e02119d5 | 3328 | if (root->ref_cows) |
a76a3cd4 | 3329 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3330 | } |
39279cc3 | 3331 | } |
9069218d | 3332 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3333 | /* |
3334 | * we can't truncate inline items that have had | |
3335 | * special encodings | |
3336 | */ | |
3337 | if (!del_item && | |
3338 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3339 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3340 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3341 | u32 size = new_size - found_key.offset; |
3342 | ||
3343 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3344 | inode_sub_bytes(inode, item_end + 1 - |
3345 | new_size); | |
e02119d5 CM |
3346 | } |
3347 | size = | |
3348 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3349 | btrfs_truncate_item(trans, root, path, |
3350 | size, 1); | |
e02119d5 | 3351 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3352 | inode_sub_bytes(inode, item_end + 1 - |
3353 | found_key.offset); | |
9069218d | 3354 | } |
39279cc3 | 3355 | } |
179e29e4 | 3356 | delete: |
39279cc3 | 3357 | if (del_item) { |
85e21bac CM |
3358 | if (!pending_del_nr) { |
3359 | /* no pending yet, add ourselves */ | |
3360 | pending_del_slot = path->slots[0]; | |
3361 | pending_del_nr = 1; | |
3362 | } else if (pending_del_nr && | |
3363 | path->slots[0] + 1 == pending_del_slot) { | |
3364 | /* hop on the pending chunk */ | |
3365 | pending_del_nr++; | |
3366 | pending_del_slot = path->slots[0]; | |
3367 | } else { | |
d397712b | 3368 | BUG(); |
85e21bac | 3369 | } |
39279cc3 CM |
3370 | } else { |
3371 | break; | |
3372 | } | |
0af3d00b JB |
3373 | if (found_extent && (root->ref_cows || |
3374 | root == root->fs_info->tree_root)) { | |
b9473439 | 3375 | btrfs_set_path_blocking(path); |
39279cc3 | 3376 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3377 | extent_num_bytes, 0, |
3378 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3379 | ino, extent_offset, 0); |
39279cc3 CM |
3380 | BUG_ON(ret); |
3381 | } | |
85e21bac | 3382 | |
8082510e YZ |
3383 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3384 | break; | |
3385 | ||
3386 | if (path->slots[0] == 0 || | |
3387 | path->slots[0] != pending_del_slot) { | |
82d5902d LZ |
3388 | if (root->ref_cows && |
3389 | BTRFS_I(inode)->location.objectid != | |
3390 | BTRFS_FREE_INO_OBJECTID) { | |
8082510e YZ |
3391 | err = -EAGAIN; |
3392 | goto out; | |
3393 | } | |
3394 | if (pending_del_nr) { | |
3395 | ret = btrfs_del_items(trans, root, path, | |
3396 | pending_del_slot, | |
3397 | pending_del_nr); | |
79787eaa JM |
3398 | if (ret) { |
3399 | btrfs_abort_transaction(trans, | |
3400 | root, ret); | |
3401 | goto error; | |
3402 | } | |
8082510e YZ |
3403 | pending_del_nr = 0; |
3404 | } | |
b3b4aa74 | 3405 | btrfs_release_path(path); |
85e21bac | 3406 | goto search_again; |
8082510e YZ |
3407 | } else { |
3408 | path->slots[0]--; | |
85e21bac | 3409 | } |
39279cc3 | 3410 | } |
8082510e | 3411 | out: |
85e21bac CM |
3412 | if (pending_del_nr) { |
3413 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3414 | pending_del_nr); | |
79787eaa JM |
3415 | if (ret) |
3416 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3417 | } |
79787eaa | 3418 | error: |
39279cc3 | 3419 | btrfs_free_path(path); |
8082510e | 3420 | return err; |
39279cc3 CM |
3421 | } |
3422 | ||
3423 | /* | |
3424 | * taken from block_truncate_page, but does cow as it zeros out | |
3425 | * any bytes left in the last page in the file. | |
3426 | */ | |
3427 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3428 | { | |
3429 | struct inode *inode = mapping->host; | |
db94535d | 3430 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3431 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3432 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3433 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3434 | char *kaddr; |
db94535d | 3435 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3436 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3437 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3438 | struct page *page; | |
3b16a4e3 | 3439 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3440 | int ret = 0; |
a52d9a80 | 3441 | u64 page_start; |
e6dcd2dc | 3442 | u64 page_end; |
39279cc3 CM |
3443 | |
3444 | if ((offset & (blocksize - 1)) == 0) | |
3445 | goto out; | |
0ca1f7ce | 3446 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3447 | if (ret) |
3448 | goto out; | |
39279cc3 CM |
3449 | |
3450 | ret = -ENOMEM; | |
211c17f5 | 3451 | again: |
3b16a4e3 | 3452 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3453 | if (!page) { |
0ca1f7ce | 3454 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3455 | goto out; |
5d5e103a | 3456 | } |
e6dcd2dc CM |
3457 | |
3458 | page_start = page_offset(page); | |
3459 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3460 | ||
39279cc3 | 3461 | if (!PageUptodate(page)) { |
9ebefb18 | 3462 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3463 | lock_page(page); |
211c17f5 CM |
3464 | if (page->mapping != mapping) { |
3465 | unlock_page(page); | |
3466 | page_cache_release(page); | |
3467 | goto again; | |
3468 | } | |
39279cc3 CM |
3469 | if (!PageUptodate(page)) { |
3470 | ret = -EIO; | |
89642229 | 3471 | goto out_unlock; |
39279cc3 CM |
3472 | } |
3473 | } | |
211c17f5 | 3474 | wait_on_page_writeback(page); |
e6dcd2dc | 3475 | |
d0082371 | 3476 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3477 | set_page_extent_mapped(page); |
3478 | ||
3479 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3480 | if (ordered) { | |
2ac55d41 JB |
3481 | unlock_extent_cached(io_tree, page_start, page_end, |
3482 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3483 | unlock_page(page); |
3484 | page_cache_release(page); | |
eb84ae03 | 3485 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3486 | btrfs_put_ordered_extent(ordered); |
3487 | goto again; | |
3488 | } | |
3489 | ||
2ac55d41 | 3490 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3491 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3492 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3493 | |
2ac55d41 JB |
3494 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3495 | &cached_state); | |
9ed74f2d | 3496 | if (ret) { |
2ac55d41 JB |
3497 | unlock_extent_cached(io_tree, page_start, page_end, |
3498 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3499 | goto out_unlock; |
3500 | } | |
3501 | ||
e6dcd2dc CM |
3502 | ret = 0; |
3503 | if (offset != PAGE_CACHE_SIZE) { | |
3504 | kaddr = kmap(page); | |
3505 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3506 | flush_dcache_page(page); | |
3507 | kunmap(page); | |
3508 | } | |
247e743c | 3509 | ClearPageChecked(page); |
e6dcd2dc | 3510 | set_page_dirty(page); |
2ac55d41 JB |
3511 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3512 | GFP_NOFS); | |
39279cc3 | 3513 | |
89642229 | 3514 | out_unlock: |
5d5e103a | 3515 | if (ret) |
0ca1f7ce | 3516 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3517 | unlock_page(page); |
3518 | page_cache_release(page); | |
3519 | out: | |
3520 | return ret; | |
3521 | } | |
3522 | ||
695a0d0d JB |
3523 | /* |
3524 | * This function puts in dummy file extents for the area we're creating a hole | |
3525 | * for. So if we are truncating this file to a larger size we need to insert | |
3526 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3527 | * the range between oldsize and size | |
3528 | */ | |
a41ad394 | 3529 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3530 | { |
9036c102 YZ |
3531 | struct btrfs_trans_handle *trans; |
3532 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3533 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3534 | struct extent_map *em = NULL; |
2ac55d41 | 3535 | struct extent_state *cached_state = NULL; |
9036c102 | 3536 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3537 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3538 | u64 block_end = (size + mask) & ~mask; |
3539 | u64 last_byte; | |
3540 | u64 cur_offset; | |
3541 | u64 hole_size; | |
9ed74f2d | 3542 | int err = 0; |
39279cc3 | 3543 | |
9036c102 YZ |
3544 | if (size <= hole_start) |
3545 | return 0; | |
3546 | ||
9036c102 YZ |
3547 | while (1) { |
3548 | struct btrfs_ordered_extent *ordered; | |
3549 | btrfs_wait_ordered_range(inode, hole_start, | |
3550 | block_end - hole_start); | |
2ac55d41 | 3551 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3552 | &cached_state); |
9036c102 YZ |
3553 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3554 | if (!ordered) | |
3555 | break; | |
2ac55d41 JB |
3556 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3557 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3558 | btrfs_put_ordered_extent(ordered); |
3559 | } | |
39279cc3 | 3560 | |
9036c102 YZ |
3561 | cur_offset = hole_start; |
3562 | while (1) { | |
3563 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3564 | block_end - cur_offset, 0); | |
79787eaa JM |
3565 | if (IS_ERR(em)) { |
3566 | err = PTR_ERR(em); | |
3567 | break; | |
3568 | } | |
9036c102 YZ |
3569 | last_byte = min(extent_map_end(em), block_end); |
3570 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3571 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3572 | u64 hint_byte = 0; |
9036c102 | 3573 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3574 | |
3642320e | 3575 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3576 | if (IS_ERR(trans)) { |
3577 | err = PTR_ERR(trans); | |
9ed74f2d | 3578 | break; |
a22285a6 | 3579 | } |
8082510e YZ |
3580 | |
3581 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3582 | cur_offset + hole_size, | |
3583 | &hint_byte, 1); | |
5b397377 | 3584 | if (err) { |
79787eaa | 3585 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3586 | btrfs_end_transaction(trans, root); |
3893e33b | 3587 | break; |
5b397377 | 3588 | } |
8082510e | 3589 | |
9036c102 | 3590 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3591 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3592 | 0, hole_size, 0, hole_size, |
3593 | 0, 0, 0); | |
5b397377 | 3594 | if (err) { |
79787eaa | 3595 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3596 | btrfs_end_transaction(trans, root); |
3893e33b | 3597 | break; |
5b397377 | 3598 | } |
8082510e | 3599 | |
9036c102 YZ |
3600 | btrfs_drop_extent_cache(inode, hole_start, |
3601 | last_byte - 1, 0); | |
8082510e | 3602 | |
3642320e | 3603 | btrfs_update_inode(trans, root, inode); |
8082510e | 3604 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3605 | } |
3606 | free_extent_map(em); | |
a22285a6 | 3607 | em = NULL; |
9036c102 | 3608 | cur_offset = last_byte; |
8082510e | 3609 | if (cur_offset >= block_end) |
9036c102 YZ |
3610 | break; |
3611 | } | |
1832a6d5 | 3612 | |
a22285a6 | 3613 | free_extent_map(em); |
2ac55d41 JB |
3614 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3615 | GFP_NOFS); | |
9036c102 YZ |
3616 | return err; |
3617 | } | |
39279cc3 | 3618 | |
a41ad394 | 3619 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3620 | { |
f4a2f4c5 MX |
3621 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3622 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3623 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3624 | int ret; |
3625 | ||
a41ad394 | 3626 | if (newsize == oldsize) |
8082510e YZ |
3627 | return 0; |
3628 | ||
a41ad394 | 3629 | if (newsize > oldsize) { |
a41ad394 JB |
3630 | truncate_pagecache(inode, oldsize, newsize); |
3631 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3632 | if (ret) |
8082510e | 3633 | return ret; |
8082510e | 3634 | |
f4a2f4c5 MX |
3635 | trans = btrfs_start_transaction(root, 1); |
3636 | if (IS_ERR(trans)) | |
3637 | return PTR_ERR(trans); | |
3638 | ||
3639 | i_size_write(inode, newsize); | |
3640 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3641 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3642 | btrfs_end_transaction(trans, root); |
a41ad394 | 3643 | } else { |
8082510e | 3644 | |
a41ad394 JB |
3645 | /* |
3646 | * We're truncating a file that used to have good data down to | |
3647 | * zero. Make sure it gets into the ordered flush list so that | |
3648 | * any new writes get down to disk quickly. | |
3649 | */ | |
3650 | if (newsize == 0) | |
72ac3c0d JB |
3651 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3652 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3653 | |
a41ad394 JB |
3654 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3655 | truncate_setsize(inode, newsize); | |
3656 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3657 | } |
3658 | ||
a41ad394 | 3659 | return ret; |
8082510e YZ |
3660 | } |
3661 | ||
9036c102 YZ |
3662 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3663 | { | |
3664 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3665 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3666 | int err; |
39279cc3 | 3667 | |
b83cc969 LZ |
3668 | if (btrfs_root_readonly(root)) |
3669 | return -EROFS; | |
3670 | ||
9036c102 YZ |
3671 | err = inode_change_ok(inode, attr); |
3672 | if (err) | |
3673 | return err; | |
2bf5a725 | 3674 | |
5a3f23d5 | 3675 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3676 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3677 | if (err) |
3678 | return err; | |
39279cc3 | 3679 | } |
9036c102 | 3680 | |
1025774c CH |
3681 | if (attr->ia_valid) { |
3682 | setattr_copy(inode, attr); | |
0c4d2d95 | 3683 | inode_inc_iversion(inode); |
22c44fe6 | 3684 | err = btrfs_dirty_inode(inode); |
1025774c | 3685 | |
22c44fe6 | 3686 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3687 | err = btrfs_acl_chmod(inode); |
3688 | } | |
33268eaf | 3689 | |
39279cc3 CM |
3690 | return err; |
3691 | } | |
61295eb8 | 3692 | |
bd555975 | 3693 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3694 | { |
3695 | struct btrfs_trans_handle *trans; | |
3696 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3697 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3698 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3699 | unsigned long nr; |
39279cc3 CM |
3700 | int ret; |
3701 | ||
1abe9b8a | 3702 | trace_btrfs_inode_evict(inode); |
3703 | ||
39279cc3 | 3704 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3705 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
2cf8572d | 3706 | btrfs_is_free_space_inode(root, inode))) |
bd555975 AV |
3707 | goto no_delete; |
3708 | ||
39279cc3 | 3709 | if (is_bad_inode(inode)) { |
7b128766 | 3710 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3711 | goto no_delete; |
3712 | } | |
bd555975 | 3713 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3714 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3715 | |
c71bf099 | 3716 | if (root->fs_info->log_root_recovering) { |
8a35d95f JB |
3717 | BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3718 | &BTRFS_I(inode)->runtime_flags)); | |
c71bf099 YZ |
3719 | goto no_delete; |
3720 | } | |
3721 | ||
76dda93c YZ |
3722 | if (inode->i_nlink > 0) { |
3723 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3724 | goto no_delete; | |
3725 | } | |
3726 | ||
4289a667 JB |
3727 | rsv = btrfs_alloc_block_rsv(root); |
3728 | if (!rsv) { | |
3729 | btrfs_orphan_del(NULL, inode); | |
3730 | goto no_delete; | |
3731 | } | |
4a338542 | 3732 | rsv->size = min_size; |
726c35fa | 3733 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3734 | |
dbe674a9 | 3735 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3736 | |
4289a667 JB |
3737 | /* |
3738 | * This is a bit simpler than btrfs_truncate since | |
3739 | * | |
3740 | * 1) We've already reserved our space for our orphan item in the | |
3741 | * unlink. | |
3742 | * 2) We're going to delete the inode item, so we don't need to update | |
3743 | * it at all. | |
3744 | * | |
3745 | * So we just need to reserve some slack space in case we add bytes when | |
3746 | * doing the truncate. | |
3747 | */ | |
8082510e | 3748 | while (1) { |
aa38a711 | 3749 | ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size); |
726c35fa JB |
3750 | |
3751 | /* | |
3752 | * Try and steal from the global reserve since we will | |
3753 | * likely not use this space anyway, we want to try as | |
3754 | * hard as possible to get this to work. | |
3755 | */ | |
3756 | if (ret) | |
3757 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3758 | |
d68fc57b | 3759 | if (ret) { |
4289a667 | 3760 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3761 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3762 | btrfs_orphan_del(NULL, inode); |
3763 | btrfs_free_block_rsv(root, rsv); | |
3764 | goto no_delete; | |
d68fc57b | 3765 | } |
7b128766 | 3766 | |
4289a667 JB |
3767 | trans = btrfs_start_transaction(root, 0); |
3768 | if (IS_ERR(trans)) { | |
3769 | btrfs_orphan_del(NULL, inode); | |
3770 | btrfs_free_block_rsv(root, rsv); | |
3771 | goto no_delete; | |
d68fc57b | 3772 | } |
7b128766 | 3773 | |
4289a667 JB |
3774 | trans->block_rsv = rsv; |
3775 | ||
d68fc57b | 3776 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3777 | if (ret != -EAGAIN) |
3778 | break; | |
85e21bac | 3779 | |
8082510e YZ |
3780 | nr = trans->blocks_used; |
3781 | btrfs_end_transaction(trans, root); | |
3782 | trans = NULL; | |
3783 | btrfs_btree_balance_dirty(root, nr); | |
3784 | } | |
5f39d397 | 3785 | |
4289a667 JB |
3786 | btrfs_free_block_rsv(root, rsv); |
3787 | ||
8082510e | 3788 | if (ret == 0) { |
4289a667 | 3789 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3790 | ret = btrfs_orphan_del(trans, inode); |
3791 | BUG_ON(ret); | |
3792 | } | |
54aa1f4d | 3793 | |
4289a667 | 3794 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3795 | if (!(root == root->fs_info->tree_root || |
3796 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3797 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3798 | |
d3c2fdcf | 3799 | nr = trans->blocks_used; |
54aa1f4d | 3800 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3801 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3802 | no_delete: |
bd555975 | 3803 | end_writeback(inode); |
8082510e | 3804 | return; |
39279cc3 CM |
3805 | } |
3806 | ||
3807 | /* | |
3808 | * this returns the key found in the dir entry in the location pointer. | |
3809 | * If no dir entries were found, location->objectid is 0. | |
3810 | */ | |
3811 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3812 | struct btrfs_key *location) | |
3813 | { | |
3814 | const char *name = dentry->d_name.name; | |
3815 | int namelen = dentry->d_name.len; | |
3816 | struct btrfs_dir_item *di; | |
3817 | struct btrfs_path *path; | |
3818 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3819 | int ret = 0; |
39279cc3 CM |
3820 | |
3821 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3822 | if (!path) |
3823 | return -ENOMEM; | |
3954401f | 3824 | |
33345d01 | 3825 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3826 | namelen, 0); |
0d9f7f3e Y |
3827 | if (IS_ERR(di)) |
3828 | ret = PTR_ERR(di); | |
d397712b | 3829 | |
c704005d | 3830 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3831 | goto out_err; |
d397712b | 3832 | |
5f39d397 | 3833 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3834 | out: |
39279cc3 CM |
3835 | btrfs_free_path(path); |
3836 | return ret; | |
3954401f CM |
3837 | out_err: |
3838 | location->objectid = 0; | |
3839 | goto out; | |
39279cc3 CM |
3840 | } |
3841 | ||
3842 | /* | |
3843 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3844 | * needs to be changed to reflect the root directory of the tree root. This | |
3845 | * is kind of like crossing a mount point. | |
3846 | */ | |
3847 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3848 | struct inode *dir, |
3849 | struct dentry *dentry, | |
3850 | struct btrfs_key *location, | |
3851 | struct btrfs_root **sub_root) | |
39279cc3 | 3852 | { |
4df27c4d YZ |
3853 | struct btrfs_path *path; |
3854 | struct btrfs_root *new_root; | |
3855 | struct btrfs_root_ref *ref; | |
3856 | struct extent_buffer *leaf; | |
3857 | int ret; | |
3858 | int err = 0; | |
39279cc3 | 3859 | |
4df27c4d YZ |
3860 | path = btrfs_alloc_path(); |
3861 | if (!path) { | |
3862 | err = -ENOMEM; | |
3863 | goto out; | |
3864 | } | |
39279cc3 | 3865 | |
4df27c4d YZ |
3866 | err = -ENOENT; |
3867 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3868 | BTRFS_I(dir)->root->root_key.objectid, | |
3869 | location->objectid); | |
3870 | if (ret) { | |
3871 | if (ret < 0) | |
3872 | err = ret; | |
3873 | goto out; | |
3874 | } | |
39279cc3 | 3875 | |
4df27c4d YZ |
3876 | leaf = path->nodes[0]; |
3877 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3878 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3879 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3880 | goto out; | |
39279cc3 | 3881 | |
4df27c4d YZ |
3882 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3883 | (unsigned long)(ref + 1), | |
3884 | dentry->d_name.len); | |
3885 | if (ret) | |
3886 | goto out; | |
3887 | ||
b3b4aa74 | 3888 | btrfs_release_path(path); |
4df27c4d YZ |
3889 | |
3890 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3891 | if (IS_ERR(new_root)) { | |
3892 | err = PTR_ERR(new_root); | |
3893 | goto out; | |
3894 | } | |
3895 | ||
3896 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3897 | err = -ENOENT; | |
3898 | goto out; | |
3899 | } | |
3900 | ||
3901 | *sub_root = new_root; | |
3902 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3903 | location->type = BTRFS_INODE_ITEM_KEY; | |
3904 | location->offset = 0; | |
3905 | err = 0; | |
3906 | out: | |
3907 | btrfs_free_path(path); | |
3908 | return err; | |
39279cc3 CM |
3909 | } |
3910 | ||
5d4f98a2 YZ |
3911 | static void inode_tree_add(struct inode *inode) |
3912 | { | |
3913 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3914 | struct btrfs_inode *entry; | |
03e860bd FNP |
3915 | struct rb_node **p; |
3916 | struct rb_node *parent; | |
33345d01 | 3917 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
3918 | again: |
3919 | p = &root->inode_tree.rb_node; | |
3920 | parent = NULL; | |
5d4f98a2 | 3921 | |
1d3382cb | 3922 | if (inode_unhashed(inode)) |
76dda93c YZ |
3923 | return; |
3924 | ||
5d4f98a2 YZ |
3925 | spin_lock(&root->inode_lock); |
3926 | while (*p) { | |
3927 | parent = *p; | |
3928 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3929 | ||
33345d01 | 3930 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3931 | p = &parent->rb_left; |
33345d01 | 3932 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3933 | p = &parent->rb_right; |
5d4f98a2 YZ |
3934 | else { |
3935 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3936 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3937 | rb_erase(parent, &root->inode_tree); |
3938 | RB_CLEAR_NODE(parent); | |
3939 | spin_unlock(&root->inode_lock); | |
3940 | goto again; | |
5d4f98a2 YZ |
3941 | } |
3942 | } | |
3943 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3944 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3945 | spin_unlock(&root->inode_lock); | |
3946 | } | |
3947 | ||
3948 | static void inode_tree_del(struct inode *inode) | |
3949 | { | |
3950 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3951 | int empty = 0; |
5d4f98a2 | 3952 | |
03e860bd | 3953 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3954 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3955 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3956 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3957 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3958 | } |
03e860bd | 3959 | spin_unlock(&root->inode_lock); |
76dda93c | 3960 | |
0af3d00b JB |
3961 | /* |
3962 | * Free space cache has inodes in the tree root, but the tree root has a | |
3963 | * root_refs of 0, so this could end up dropping the tree root as a | |
3964 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3965 | * make sure we don't drop it. | |
3966 | */ | |
3967 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3968 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3969 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3970 | spin_lock(&root->inode_lock); | |
3971 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3972 | spin_unlock(&root->inode_lock); | |
3973 | if (empty) | |
3974 | btrfs_add_dead_root(root); | |
3975 | } | |
3976 | } | |
3977 | ||
143bede5 | 3978 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
3979 | { |
3980 | struct rb_node *node; | |
3981 | struct rb_node *prev; | |
3982 | struct btrfs_inode *entry; | |
3983 | struct inode *inode; | |
3984 | u64 objectid = 0; | |
3985 | ||
3986 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3987 | ||
3988 | spin_lock(&root->inode_lock); | |
3989 | again: | |
3990 | node = root->inode_tree.rb_node; | |
3991 | prev = NULL; | |
3992 | while (node) { | |
3993 | prev = node; | |
3994 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3995 | ||
33345d01 | 3996 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 3997 | node = node->rb_left; |
33345d01 | 3998 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
3999 | node = node->rb_right; |
4000 | else | |
4001 | break; | |
4002 | } | |
4003 | if (!node) { | |
4004 | while (prev) { | |
4005 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4006 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4007 | node = prev; |
4008 | break; | |
4009 | } | |
4010 | prev = rb_next(prev); | |
4011 | } | |
4012 | } | |
4013 | while (node) { | |
4014 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4015 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4016 | inode = igrab(&entry->vfs_inode); |
4017 | if (inode) { | |
4018 | spin_unlock(&root->inode_lock); | |
4019 | if (atomic_read(&inode->i_count) > 1) | |
4020 | d_prune_aliases(inode); | |
4021 | /* | |
45321ac5 | 4022 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4023 | * the inode cache when its usage count |
4024 | * hits zero. | |
4025 | */ | |
4026 | iput(inode); | |
4027 | cond_resched(); | |
4028 | spin_lock(&root->inode_lock); | |
4029 | goto again; | |
4030 | } | |
4031 | ||
4032 | if (cond_resched_lock(&root->inode_lock)) | |
4033 | goto again; | |
4034 | ||
4035 | node = rb_next(node); | |
4036 | } | |
4037 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4038 | } |
4039 | ||
e02119d5 CM |
4040 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4041 | { | |
4042 | struct btrfs_iget_args *args = p; | |
4043 | inode->i_ino = args->ino; | |
e02119d5 | 4044 | BTRFS_I(inode)->root = args->root; |
6a63209f | 4045 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
4046 | return 0; |
4047 | } | |
4048 | ||
4049 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4050 | { | |
4051 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4052 | return args->ino == btrfs_ino(inode) && |
d397712b | 4053 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4054 | } |
4055 | ||
5d4f98a2 YZ |
4056 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4057 | u64 objectid, | |
4058 | struct btrfs_root *root) | |
39279cc3 CM |
4059 | { |
4060 | struct inode *inode; | |
4061 | struct btrfs_iget_args args; | |
4062 | args.ino = objectid; | |
4063 | args.root = root; | |
4064 | ||
4065 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4066 | btrfs_init_locked_inode, | |
4067 | (void *)&args); | |
4068 | return inode; | |
4069 | } | |
4070 | ||
1a54ef8c BR |
4071 | /* Get an inode object given its location and corresponding root. |
4072 | * Returns in *is_new if the inode was read from disk | |
4073 | */ | |
4074 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4075 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4076 | { |
4077 | struct inode *inode; | |
4078 | ||
4079 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4080 | if (!inode) | |
5d4f98a2 | 4081 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4082 | |
4083 | if (inode->i_state & I_NEW) { | |
4084 | BTRFS_I(inode)->root = root; | |
4085 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4086 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4087 | if (!is_bad_inode(inode)) { |
4088 | inode_tree_add(inode); | |
4089 | unlock_new_inode(inode); | |
4090 | if (new) | |
4091 | *new = 1; | |
4092 | } else { | |
e0b6d65b ST |
4093 | unlock_new_inode(inode); |
4094 | iput(inode); | |
4095 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4096 | } |
4097 | } | |
4098 | ||
1a54ef8c BR |
4099 | return inode; |
4100 | } | |
4101 | ||
4df27c4d YZ |
4102 | static struct inode *new_simple_dir(struct super_block *s, |
4103 | struct btrfs_key *key, | |
4104 | struct btrfs_root *root) | |
4105 | { | |
4106 | struct inode *inode = new_inode(s); | |
4107 | ||
4108 | if (!inode) | |
4109 | return ERR_PTR(-ENOMEM); | |
4110 | ||
4df27c4d YZ |
4111 | BTRFS_I(inode)->root = root; |
4112 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4113 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4114 | |
4115 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4116 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4117 | inode->i_fop = &simple_dir_operations; |
4118 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4119 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4120 | ||
4121 | return inode; | |
4122 | } | |
4123 | ||
3de4586c | 4124 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4125 | { |
d397712b | 4126 | struct inode *inode; |
4df27c4d | 4127 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4128 | struct btrfs_root *sub_root = root; |
4129 | struct btrfs_key location; | |
76dda93c | 4130 | int index; |
b4aff1f8 | 4131 | int ret = 0; |
39279cc3 CM |
4132 | |
4133 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4134 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4135 | |
b4aff1f8 JB |
4136 | if (unlikely(d_need_lookup(dentry))) { |
4137 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4138 | kfree(dentry->d_fsdata); | |
4139 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4140 | /* This thing is hashed, drop it for now */ |
4141 | d_drop(dentry); | |
b4aff1f8 JB |
4142 | } else { |
4143 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4144 | } | |
5f39d397 | 4145 | |
39279cc3 CM |
4146 | if (ret < 0) |
4147 | return ERR_PTR(ret); | |
5f39d397 | 4148 | |
4df27c4d YZ |
4149 | if (location.objectid == 0) |
4150 | return NULL; | |
4151 | ||
4152 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4153 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4154 | return inode; |
4155 | } | |
4156 | ||
4157 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4158 | ||
76dda93c | 4159 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4160 | ret = fixup_tree_root_location(root, dir, dentry, |
4161 | &location, &sub_root); | |
4162 | if (ret < 0) { | |
4163 | if (ret != -ENOENT) | |
4164 | inode = ERR_PTR(ret); | |
4165 | else | |
4166 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4167 | } else { | |
73f73415 | 4168 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4169 | } |
76dda93c YZ |
4170 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4171 | ||
34d19bad | 4172 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4173 | down_read(&root->fs_info->cleanup_work_sem); |
4174 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4175 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4176 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4177 | if (ret) |
4178 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4179 | } |
4180 | ||
3de4586c CM |
4181 | return inode; |
4182 | } | |
4183 | ||
fe15ce44 | 4184 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4185 | { |
4186 | struct btrfs_root *root; | |
848cce0d | 4187 | struct inode *inode = dentry->d_inode; |
76dda93c | 4188 | |
848cce0d LZ |
4189 | if (!inode && !IS_ROOT(dentry)) |
4190 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4191 | |
848cce0d LZ |
4192 | if (inode) { |
4193 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4194 | if (btrfs_root_refs(&root->root_item) == 0) |
4195 | return 1; | |
848cce0d LZ |
4196 | |
4197 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4198 | return 1; | |
efefb143 | 4199 | } |
76dda93c YZ |
4200 | return 0; |
4201 | } | |
4202 | ||
b4aff1f8 JB |
4203 | static void btrfs_dentry_release(struct dentry *dentry) |
4204 | { | |
4205 | if (dentry->d_fsdata) | |
4206 | kfree(dentry->d_fsdata); | |
4207 | } | |
4208 | ||
3de4586c CM |
4209 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4210 | struct nameidata *nd) | |
4211 | { | |
a66e7cc6 JB |
4212 | struct dentry *ret; |
4213 | ||
4214 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4215 | if (unlikely(d_need_lookup(dentry))) { | |
4216 | spin_lock(&dentry->d_lock); | |
4217 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4218 | spin_unlock(&dentry->d_lock); | |
4219 | } | |
4220 | return ret; | |
39279cc3 CM |
4221 | } |
4222 | ||
16cdcec7 | 4223 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4224 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4225 | }; | |
4226 | ||
cbdf5a24 DW |
4227 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4228 | filldir_t filldir) | |
39279cc3 | 4229 | { |
6da6abae | 4230 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4231 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4232 | struct btrfs_item *item; | |
4233 | struct btrfs_dir_item *di; | |
4234 | struct btrfs_key key; | |
5f39d397 | 4235 | struct btrfs_key found_key; |
39279cc3 | 4236 | struct btrfs_path *path; |
16cdcec7 MX |
4237 | struct list_head ins_list; |
4238 | struct list_head del_list; | |
39279cc3 | 4239 | int ret; |
5f39d397 | 4240 | struct extent_buffer *leaf; |
39279cc3 | 4241 | int slot; |
39279cc3 CM |
4242 | unsigned char d_type; |
4243 | int over = 0; | |
4244 | u32 di_cur; | |
4245 | u32 di_total; | |
4246 | u32 di_len; | |
4247 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4248 | char tmp_name[32]; |
4249 | char *name_ptr; | |
4250 | int name_len; | |
16cdcec7 | 4251 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4252 | |
4253 | /* FIXME, use a real flag for deciding about the key type */ | |
4254 | if (root->fs_info->tree_root == root) | |
4255 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4256 | |
3954401f CM |
4257 | /* special case for "." */ |
4258 | if (filp->f_pos == 0) { | |
3765fefa HS |
4259 | over = filldir(dirent, ".", 1, |
4260 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4261 | if (over) |
4262 | return 0; | |
4263 | filp->f_pos = 1; | |
4264 | } | |
3954401f CM |
4265 | /* special case for .., just use the back ref */ |
4266 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4267 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4268 | over = filldir(dirent, "..", 2, |
3765fefa | 4269 | filp->f_pos, pino, DT_DIR); |
3954401f | 4270 | if (over) |
49593bfa | 4271 | return 0; |
3954401f CM |
4272 | filp->f_pos = 2; |
4273 | } | |
49593bfa | 4274 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4275 | if (!path) |
4276 | return -ENOMEM; | |
ff5714cc | 4277 | |
026fd317 | 4278 | path->reada = 1; |
49593bfa | 4279 | |
16cdcec7 MX |
4280 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4281 | INIT_LIST_HEAD(&ins_list); | |
4282 | INIT_LIST_HEAD(&del_list); | |
4283 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4284 | } | |
4285 | ||
39279cc3 CM |
4286 | btrfs_set_key_type(&key, key_type); |
4287 | key.offset = filp->f_pos; | |
33345d01 | 4288 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4289 | |
39279cc3 CM |
4290 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4291 | if (ret < 0) | |
4292 | goto err; | |
49593bfa DW |
4293 | |
4294 | while (1) { | |
5f39d397 | 4295 | leaf = path->nodes[0]; |
39279cc3 | 4296 | slot = path->slots[0]; |
b9e03af0 LZ |
4297 | if (slot >= btrfs_header_nritems(leaf)) { |
4298 | ret = btrfs_next_leaf(root, path); | |
4299 | if (ret < 0) | |
4300 | goto err; | |
4301 | else if (ret > 0) | |
4302 | break; | |
4303 | continue; | |
39279cc3 | 4304 | } |
3de4586c | 4305 | |
5f39d397 CM |
4306 | item = btrfs_item_nr(leaf, slot); |
4307 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4308 | ||
4309 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4310 | break; |
5f39d397 | 4311 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4312 | break; |
5f39d397 | 4313 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4314 | goto next; |
16cdcec7 MX |
4315 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4316 | btrfs_should_delete_dir_index(&del_list, | |
4317 | found_key.offset)) | |
4318 | goto next; | |
5f39d397 CM |
4319 | |
4320 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4321 | is_curr = 1; |
49593bfa | 4322 | |
39279cc3 CM |
4323 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4324 | di_cur = 0; | |
5f39d397 | 4325 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4326 | |
4327 | while (di_cur < di_total) { | |
5f39d397 CM |
4328 | struct btrfs_key location; |
4329 | ||
22a94d44 JB |
4330 | if (verify_dir_item(root, leaf, di)) |
4331 | break; | |
4332 | ||
5f39d397 | 4333 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4334 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4335 | name_ptr = tmp_name; |
4336 | } else { | |
4337 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4338 | if (!name_ptr) { |
4339 | ret = -ENOMEM; | |
4340 | goto err; | |
4341 | } | |
5f39d397 CM |
4342 | } |
4343 | read_extent_buffer(leaf, name_ptr, | |
4344 | (unsigned long)(di + 1), name_len); | |
4345 | ||
4346 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4347 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4348 | |
fede766f | 4349 | |
3de4586c | 4350 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4351 | * skip it. |
4352 | * | |
4353 | * In contrast to old kernels, we insert the snapshot's | |
4354 | * dir item and dir index after it has been created, so | |
4355 | * we won't find a reference to our own snapshot. We | |
4356 | * still keep the following code for backward | |
4357 | * compatibility. | |
3de4586c CM |
4358 | */ |
4359 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4360 | location.objectid == root->root_key.objectid) { | |
4361 | over = 0; | |
4362 | goto skip; | |
4363 | } | |
5f39d397 | 4364 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4365 | found_key.offset, location.objectid, |
39279cc3 | 4366 | d_type); |
5f39d397 | 4367 | |
3de4586c | 4368 | skip: |
5f39d397 CM |
4369 | if (name_ptr != tmp_name) |
4370 | kfree(name_ptr); | |
4371 | ||
39279cc3 CM |
4372 | if (over) |
4373 | goto nopos; | |
5103e947 | 4374 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4375 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4376 | di_cur += di_len; |
4377 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4378 | } | |
b9e03af0 LZ |
4379 | next: |
4380 | path->slots[0]++; | |
39279cc3 | 4381 | } |
49593bfa | 4382 | |
16cdcec7 MX |
4383 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4384 | if (is_curr) | |
4385 | filp->f_pos++; | |
4386 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4387 | &ins_list); | |
4388 | if (ret) | |
4389 | goto nopos; | |
4390 | } | |
4391 | ||
49593bfa | 4392 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4393 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4394 | /* |
4395 | * 32-bit glibc will use getdents64, but then strtol - | |
4396 | * so the last number we can serve is this. | |
4397 | */ | |
4398 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4399 | else |
4400 | filp->f_pos++; | |
39279cc3 CM |
4401 | nopos: |
4402 | ret = 0; | |
4403 | err: | |
16cdcec7 MX |
4404 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4405 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4406 | btrfs_free_path(path); |
39279cc3 CM |
4407 | return ret; |
4408 | } | |
4409 | ||
a9185b41 | 4410 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4411 | { |
4412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4413 | struct btrfs_trans_handle *trans; | |
4414 | int ret = 0; | |
0af3d00b | 4415 | bool nolock = false; |
39279cc3 | 4416 | |
72ac3c0d | 4417 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4418 | return 0; |
4419 | ||
2cf8572d | 4420 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode)) |
82d5902d | 4421 | nolock = true; |
0af3d00b | 4422 | |
a9185b41 | 4423 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4424 | if (nolock) |
7a7eaa40 | 4425 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4426 | else |
7a7eaa40 | 4427 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4428 | if (IS_ERR(trans)) |
4429 | return PTR_ERR(trans); | |
0af3d00b JB |
4430 | if (nolock) |
4431 | ret = btrfs_end_transaction_nolock(trans, root); | |
4432 | else | |
4433 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4434 | } |
4435 | return ret; | |
4436 | } | |
4437 | ||
4438 | /* | |
54aa1f4d | 4439 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4440 | * inode changes. But, it is most likely to find the inode in cache. |
4441 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4442 | * to keep or drop this code. | |
4443 | */ | |
22c44fe6 | 4444 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4445 | { |
4446 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4447 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4448 | int ret; |
4449 | ||
72ac3c0d | 4450 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4451 | return 0; |
39279cc3 | 4452 | |
7a7eaa40 | 4453 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4454 | if (IS_ERR(trans)) |
4455 | return PTR_ERR(trans); | |
8929ecfa YZ |
4456 | |
4457 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4458 | if (ret && ret == -ENOSPC) { |
4459 | /* whoops, lets try again with the full transaction */ | |
4460 | btrfs_end_transaction(trans, root); | |
4461 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4462 | if (IS_ERR(trans)) |
4463 | return PTR_ERR(trans); | |
8929ecfa | 4464 | |
94b60442 | 4465 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4466 | } |
39279cc3 | 4467 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4468 | if (BTRFS_I(inode)->delayed_node) |
4469 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4470 | |
4471 | return ret; | |
4472 | } | |
4473 | ||
4474 | /* | |
4475 | * This is a copy of file_update_time. We need this so we can return error on | |
4476 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4477 | */ | |
4478 | int btrfs_update_time(struct file *file) | |
4479 | { | |
4480 | struct inode *inode = file->f_path.dentry->d_inode; | |
4481 | struct timespec now; | |
4482 | int ret; | |
4483 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; | |
4484 | ||
4485 | /* First try to exhaust all avenues to not sync */ | |
4486 | if (IS_NOCMTIME(inode)) | |
4487 | return 0; | |
4488 | ||
4489 | now = current_fs_time(inode->i_sb); | |
4490 | if (!timespec_equal(&inode->i_mtime, &now)) | |
4491 | sync_it = S_MTIME; | |
4492 | ||
4493 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4494 | sync_it |= S_CTIME; | |
4495 | ||
4496 | if (IS_I_VERSION(inode)) | |
4497 | sync_it |= S_VERSION; | |
4498 | ||
4499 | if (!sync_it) | |
4500 | return 0; | |
4501 | ||
4502 | /* Finally allowed to write? Takes lock. */ | |
4503 | if (mnt_want_write_file(file)) | |
4504 | return 0; | |
4505 | ||
4506 | /* Only change inode inside the lock region */ | |
4507 | if (sync_it & S_VERSION) | |
4508 | inode_inc_iversion(inode); | |
4509 | if (sync_it & S_CTIME) | |
4510 | inode->i_ctime = now; | |
4511 | if (sync_it & S_MTIME) | |
4512 | inode->i_mtime = now; | |
4513 | ret = btrfs_dirty_inode(inode); | |
4514 | if (!ret) | |
4515 | mark_inode_dirty_sync(inode); | |
4516 | mnt_drop_write(file->f_path.mnt); | |
4517 | return ret; | |
39279cc3 CM |
4518 | } |
4519 | ||
d352ac68 CM |
4520 | /* |
4521 | * find the highest existing sequence number in a directory | |
4522 | * and then set the in-memory index_cnt variable to reflect | |
4523 | * free sequence numbers | |
4524 | */ | |
aec7477b JB |
4525 | static int btrfs_set_inode_index_count(struct inode *inode) |
4526 | { | |
4527 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4528 | struct btrfs_key key, found_key; | |
4529 | struct btrfs_path *path; | |
4530 | struct extent_buffer *leaf; | |
4531 | int ret; | |
4532 | ||
33345d01 | 4533 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4534 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4535 | key.offset = (u64)-1; | |
4536 | ||
4537 | path = btrfs_alloc_path(); | |
4538 | if (!path) | |
4539 | return -ENOMEM; | |
4540 | ||
4541 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4542 | if (ret < 0) | |
4543 | goto out; | |
4544 | /* FIXME: we should be able to handle this */ | |
4545 | if (ret == 0) | |
4546 | goto out; | |
4547 | ret = 0; | |
4548 | ||
4549 | /* | |
4550 | * MAGIC NUMBER EXPLANATION: | |
4551 | * since we search a directory based on f_pos we have to start at 2 | |
4552 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4553 | * else has to start at 2 | |
4554 | */ | |
4555 | if (path->slots[0] == 0) { | |
4556 | BTRFS_I(inode)->index_cnt = 2; | |
4557 | goto out; | |
4558 | } | |
4559 | ||
4560 | path->slots[0]--; | |
4561 | ||
4562 | leaf = path->nodes[0]; | |
4563 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4564 | ||
33345d01 | 4565 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4566 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4567 | BTRFS_I(inode)->index_cnt = 2; | |
4568 | goto out; | |
4569 | } | |
4570 | ||
4571 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4572 | out: | |
4573 | btrfs_free_path(path); | |
4574 | return ret; | |
4575 | } | |
4576 | ||
d352ac68 CM |
4577 | /* |
4578 | * helper to find a free sequence number in a given directory. This current | |
4579 | * code is very simple, later versions will do smarter things in the btree | |
4580 | */ | |
3de4586c | 4581 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4582 | { |
4583 | int ret = 0; | |
4584 | ||
4585 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4586 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4587 | if (ret) { | |
4588 | ret = btrfs_set_inode_index_count(dir); | |
4589 | if (ret) | |
4590 | return ret; | |
4591 | } | |
aec7477b JB |
4592 | } |
4593 | ||
00e4e6b3 | 4594 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4595 | BTRFS_I(dir)->index_cnt++; |
4596 | ||
4597 | return ret; | |
4598 | } | |
4599 | ||
39279cc3 CM |
4600 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4601 | struct btrfs_root *root, | |
aec7477b | 4602 | struct inode *dir, |
9c58309d | 4603 | const char *name, int name_len, |
175a4eb7 AV |
4604 | u64 ref_objectid, u64 objectid, |
4605 | umode_t mode, u64 *index) | |
39279cc3 CM |
4606 | { |
4607 | struct inode *inode; | |
5f39d397 | 4608 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4609 | struct btrfs_key *location; |
5f39d397 | 4610 | struct btrfs_path *path; |
9c58309d CM |
4611 | struct btrfs_inode_ref *ref; |
4612 | struct btrfs_key key[2]; | |
4613 | u32 sizes[2]; | |
4614 | unsigned long ptr; | |
39279cc3 CM |
4615 | int ret; |
4616 | int owner; | |
4617 | ||
5f39d397 | 4618 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4619 | if (!path) |
4620 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4621 | |
39279cc3 | 4622 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4623 | if (!inode) { |
4624 | btrfs_free_path(path); | |
39279cc3 | 4625 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4626 | } |
39279cc3 | 4627 | |
581bb050 LZ |
4628 | /* |
4629 | * we have to initialize this early, so we can reclaim the inode | |
4630 | * number if we fail afterwards in this function. | |
4631 | */ | |
4632 | inode->i_ino = objectid; | |
4633 | ||
aec7477b | 4634 | if (dir) { |
1abe9b8a | 4635 | trace_btrfs_inode_request(dir); |
4636 | ||
3de4586c | 4637 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4638 | if (ret) { |
8fb27640 | 4639 | btrfs_free_path(path); |
09771430 | 4640 | iput(inode); |
aec7477b | 4641 | return ERR_PTR(ret); |
09771430 | 4642 | } |
aec7477b JB |
4643 | } |
4644 | /* | |
4645 | * index_cnt is ignored for everything but a dir, | |
4646 | * btrfs_get_inode_index_count has an explanation for the magic | |
4647 | * number | |
4648 | */ | |
4649 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4650 | BTRFS_I(inode)->root = root; |
e02119d5 | 4651 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4652 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4653 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4654 | |
569254b0 | 4655 | if (S_ISDIR(mode)) |
39279cc3 CM |
4656 | owner = 0; |
4657 | else | |
4658 | owner = 1; | |
9c58309d CM |
4659 | |
4660 | key[0].objectid = objectid; | |
4661 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4662 | key[0].offset = 0; | |
4663 | ||
4664 | key[1].objectid = objectid; | |
4665 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4666 | key[1].offset = ref_objectid; | |
4667 | ||
4668 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4669 | sizes[1] = name_len + sizeof(*ref); | |
4670 | ||
b9473439 | 4671 | path->leave_spinning = 1; |
9c58309d CM |
4672 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4673 | if (ret != 0) | |
5f39d397 CM |
4674 | goto fail; |
4675 | ||
ecc11fab | 4676 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4677 | inode_set_bytes(inode, 0); |
39279cc3 | 4678 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4679 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4680 | struct btrfs_inode_item); | |
e02119d5 | 4681 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4682 | |
4683 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4684 | struct btrfs_inode_ref); | |
4685 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4686 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4687 | ptr = (unsigned long)(ref + 1); |
4688 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4689 | ||
5f39d397 CM |
4690 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4691 | btrfs_free_path(path); | |
4692 | ||
39279cc3 CM |
4693 | location = &BTRFS_I(inode)->location; |
4694 | location->objectid = objectid; | |
39279cc3 CM |
4695 | location->offset = 0; |
4696 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4697 | ||
6cbff00f CH |
4698 | btrfs_inherit_iflags(inode, dir); |
4699 | ||
569254b0 | 4700 | if (S_ISREG(mode)) { |
94272164 CM |
4701 | if (btrfs_test_opt(root, NODATASUM)) |
4702 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4703 | if (btrfs_test_opt(root, NODATACOW) || |
4704 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4705 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4706 | } | |
4707 | ||
39279cc3 | 4708 | insert_inode_hash(inode); |
5d4f98a2 | 4709 | inode_tree_add(inode); |
1abe9b8a | 4710 | |
4711 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4712 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4713 | |
39279cc3 | 4714 | return inode; |
5f39d397 | 4715 | fail: |
aec7477b JB |
4716 | if (dir) |
4717 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4718 | btrfs_free_path(path); |
09771430 | 4719 | iput(inode); |
5f39d397 | 4720 | return ERR_PTR(ret); |
39279cc3 CM |
4721 | } |
4722 | ||
4723 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4724 | { | |
4725 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4726 | } | |
4727 | ||
d352ac68 CM |
4728 | /* |
4729 | * utility function to add 'inode' into 'parent_inode' with | |
4730 | * a give name and a given sequence number. | |
4731 | * if 'add_backref' is true, also insert a backref from the | |
4732 | * inode to the parent directory. | |
4733 | */ | |
e02119d5 CM |
4734 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4735 | struct inode *parent_inode, struct inode *inode, | |
4736 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4737 | { |
4df27c4d | 4738 | int ret = 0; |
39279cc3 | 4739 | struct btrfs_key key; |
e02119d5 | 4740 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4741 | u64 ino = btrfs_ino(inode); |
4742 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4743 | |
33345d01 | 4744 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4745 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4746 | } else { | |
33345d01 | 4747 | key.objectid = ino; |
4df27c4d YZ |
4748 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4749 | key.offset = 0; | |
4750 | } | |
4751 | ||
33345d01 | 4752 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4753 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4754 | key.objectid, root->root_key.objectid, | |
33345d01 | 4755 | parent_ino, index, name, name_len); |
4df27c4d | 4756 | } else if (add_backref) { |
33345d01 LZ |
4757 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4758 | parent_ino, index); | |
4df27c4d | 4759 | } |
39279cc3 | 4760 | |
79787eaa JM |
4761 | /* Nothing to clean up yet */ |
4762 | if (ret) | |
4763 | return ret; | |
4df27c4d | 4764 | |
79787eaa JM |
4765 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4766 | parent_inode, &key, | |
4767 | btrfs_inode_type(inode), index); | |
4768 | if (ret == -EEXIST) | |
4769 | goto fail_dir_item; | |
4770 | else if (ret) { | |
4771 | btrfs_abort_transaction(trans, root, ret); | |
4772 | return ret; | |
39279cc3 | 4773 | } |
79787eaa JM |
4774 | |
4775 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4776 | name_len * 2); | |
0c4d2d95 | 4777 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4778 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4779 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4780 | if (ret) | |
4781 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4782 | return ret; |
fe66a05a CM |
4783 | |
4784 | fail_dir_item: | |
4785 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4786 | u64 local_index; | |
4787 | int err; | |
4788 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4789 | key.objectid, root->root_key.objectid, | |
4790 | parent_ino, &local_index, name, name_len); | |
4791 | ||
4792 | } else if (add_backref) { | |
4793 | u64 local_index; | |
4794 | int err; | |
4795 | ||
4796 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
4797 | ino, parent_ino, &local_index); | |
4798 | } | |
4799 | return ret; | |
39279cc3 CM |
4800 | } |
4801 | ||
4802 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4803 | struct inode *dir, struct dentry *dentry, |
4804 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4805 | { |
a1b075d2 JB |
4806 | int err = btrfs_add_link(trans, dir, inode, |
4807 | dentry->d_name.name, dentry->d_name.len, | |
4808 | backref, index); | |
39279cc3 CM |
4809 | if (err > 0) |
4810 | err = -EEXIST; | |
4811 | return err; | |
4812 | } | |
4813 | ||
618e21d5 | 4814 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 4815 | umode_t mode, dev_t rdev) |
618e21d5 JB |
4816 | { |
4817 | struct btrfs_trans_handle *trans; | |
4818 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4819 | struct inode *inode = NULL; |
618e21d5 JB |
4820 | int err; |
4821 | int drop_inode = 0; | |
4822 | u64 objectid; | |
1832a6d5 | 4823 | unsigned long nr = 0; |
00e4e6b3 | 4824 | u64 index = 0; |
618e21d5 JB |
4825 | |
4826 | if (!new_valid_dev(rdev)) | |
4827 | return -EINVAL; | |
4828 | ||
9ed74f2d JB |
4829 | /* |
4830 | * 2 for inode item and ref | |
4831 | * 2 for dir items | |
4832 | * 1 for xattr if selinux is on | |
4833 | */ | |
a22285a6 YZ |
4834 | trans = btrfs_start_transaction(root, 5); |
4835 | if (IS_ERR(trans)) | |
4836 | return PTR_ERR(trans); | |
1832a6d5 | 4837 | |
581bb050 LZ |
4838 | err = btrfs_find_free_ino(root, &objectid); |
4839 | if (err) | |
4840 | goto out_unlock; | |
4841 | ||
aec7477b | 4842 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4843 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4844 | mode, &index); |
7cf96da3 TI |
4845 | if (IS_ERR(inode)) { |
4846 | err = PTR_ERR(inode); | |
618e21d5 | 4847 | goto out_unlock; |
7cf96da3 | 4848 | } |
618e21d5 | 4849 | |
2a7dba39 | 4850 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4851 | if (err) { |
4852 | drop_inode = 1; | |
4853 | goto out_unlock; | |
4854 | } | |
4855 | ||
ad19db71 CS |
4856 | /* |
4857 | * If the active LSM wants to access the inode during | |
4858 | * d_instantiate it needs these. Smack checks to see | |
4859 | * if the filesystem supports xattrs by looking at the | |
4860 | * ops vector. | |
4861 | */ | |
4862 | ||
4863 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4864 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4865 | if (err) |
4866 | drop_inode = 1; | |
4867 | else { | |
618e21d5 | 4868 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4869 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4870 | d_instantiate(dentry, inode); |
618e21d5 | 4871 | } |
618e21d5 | 4872 | out_unlock: |
d3c2fdcf | 4873 | nr = trans->blocks_used; |
7ad85bb7 | 4874 | btrfs_end_transaction(trans, root); |
a22285a6 | 4875 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4876 | if (drop_inode) { |
4877 | inode_dec_link_count(inode); | |
4878 | iput(inode); | |
4879 | } | |
618e21d5 JB |
4880 | return err; |
4881 | } | |
4882 | ||
39279cc3 | 4883 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4acdaf27 | 4884 | umode_t mode, struct nameidata *nd) |
39279cc3 CM |
4885 | { |
4886 | struct btrfs_trans_handle *trans; | |
4887 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4888 | struct inode *inode = NULL; |
39279cc3 | 4889 | int drop_inode = 0; |
a22285a6 | 4890 | int err; |
1832a6d5 | 4891 | unsigned long nr = 0; |
39279cc3 | 4892 | u64 objectid; |
00e4e6b3 | 4893 | u64 index = 0; |
39279cc3 | 4894 | |
9ed74f2d JB |
4895 | /* |
4896 | * 2 for inode item and ref | |
4897 | * 2 for dir items | |
4898 | * 1 for xattr if selinux is on | |
4899 | */ | |
a22285a6 YZ |
4900 | trans = btrfs_start_transaction(root, 5); |
4901 | if (IS_ERR(trans)) | |
4902 | return PTR_ERR(trans); | |
9ed74f2d | 4903 | |
581bb050 LZ |
4904 | err = btrfs_find_free_ino(root, &objectid); |
4905 | if (err) | |
4906 | goto out_unlock; | |
4907 | ||
aec7477b | 4908 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4909 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4910 | mode, &index); |
7cf96da3 TI |
4911 | if (IS_ERR(inode)) { |
4912 | err = PTR_ERR(inode); | |
39279cc3 | 4913 | goto out_unlock; |
7cf96da3 | 4914 | } |
39279cc3 | 4915 | |
2a7dba39 | 4916 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4917 | if (err) { |
4918 | drop_inode = 1; | |
4919 | goto out_unlock; | |
4920 | } | |
4921 | ||
ad19db71 CS |
4922 | /* |
4923 | * If the active LSM wants to access the inode during | |
4924 | * d_instantiate it needs these. Smack checks to see | |
4925 | * if the filesystem supports xattrs by looking at the | |
4926 | * ops vector. | |
4927 | */ | |
4928 | inode->i_fop = &btrfs_file_operations; | |
4929 | inode->i_op = &btrfs_file_inode_operations; | |
4930 | ||
a1b075d2 | 4931 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4932 | if (err) |
4933 | drop_inode = 1; | |
4934 | else { | |
4935 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4936 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 4937 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
08c422c2 | 4938 | d_instantiate(dentry, inode); |
39279cc3 | 4939 | } |
39279cc3 | 4940 | out_unlock: |
d3c2fdcf | 4941 | nr = trans->blocks_used; |
7ad85bb7 | 4942 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4943 | if (drop_inode) { |
4944 | inode_dec_link_count(inode); | |
4945 | iput(inode); | |
4946 | } | |
d3c2fdcf | 4947 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4948 | return err; |
4949 | } | |
4950 | ||
4951 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4952 | struct dentry *dentry) | |
4953 | { | |
4954 | struct btrfs_trans_handle *trans; | |
4955 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4956 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4957 | u64 index; |
1832a6d5 | 4958 | unsigned long nr = 0; |
39279cc3 CM |
4959 | int err; |
4960 | int drop_inode = 0; | |
4961 | ||
4a8be425 TH |
4962 | /* do not allow sys_link's with other subvols of the same device */ |
4963 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4964 | return -EXDEV; |
4a8be425 | 4965 | |
c055e99e AV |
4966 | if (inode->i_nlink == ~0U) |
4967 | return -EMLINK; | |
4a8be425 | 4968 | |
3de4586c | 4969 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4970 | if (err) |
4971 | goto fail; | |
4972 | ||
a22285a6 | 4973 | /* |
7e6b6465 | 4974 | * 2 items for inode and inode ref |
a22285a6 | 4975 | * 2 items for dir items |
7e6b6465 | 4976 | * 1 item for parent inode |
a22285a6 | 4977 | */ |
7e6b6465 | 4978 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4979 | if (IS_ERR(trans)) { |
4980 | err = PTR_ERR(trans); | |
4981 | goto fail; | |
4982 | } | |
5f39d397 | 4983 | |
3153495d | 4984 | btrfs_inc_nlink(inode); |
0c4d2d95 | 4985 | inode_inc_iversion(inode); |
3153495d | 4986 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 4987 | ihold(inode); |
aec7477b | 4988 | |
a1b075d2 | 4989 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4990 | |
a5719521 | 4991 | if (err) { |
54aa1f4d | 4992 | drop_inode = 1; |
a5719521 | 4993 | } else { |
10d9f309 | 4994 | struct dentry *parent = dentry->d_parent; |
a5719521 | 4995 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
4996 | if (err) |
4997 | goto fail; | |
08c422c2 | 4998 | d_instantiate(dentry, inode); |
6a912213 | 4999 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5000 | } |
39279cc3 | 5001 | |
d3c2fdcf | 5002 | nr = trans->blocks_used; |
7ad85bb7 | 5003 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5004 | fail: |
39279cc3 CM |
5005 | if (drop_inode) { |
5006 | inode_dec_link_count(inode); | |
5007 | iput(inode); | |
5008 | } | |
d3c2fdcf | 5009 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5010 | return err; |
5011 | } | |
5012 | ||
18bb1db3 | 5013 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5014 | { |
b9d86667 | 5015 | struct inode *inode = NULL; |
39279cc3 CM |
5016 | struct btrfs_trans_handle *trans; |
5017 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5018 | int err = 0; | |
5019 | int drop_on_err = 0; | |
b9d86667 | 5020 | u64 objectid = 0; |
00e4e6b3 | 5021 | u64 index = 0; |
d3c2fdcf | 5022 | unsigned long nr = 1; |
39279cc3 | 5023 | |
9ed74f2d JB |
5024 | /* |
5025 | * 2 items for inode and ref | |
5026 | * 2 items for dir items | |
5027 | * 1 for xattr if selinux is on | |
5028 | */ | |
a22285a6 YZ |
5029 | trans = btrfs_start_transaction(root, 5); |
5030 | if (IS_ERR(trans)) | |
5031 | return PTR_ERR(trans); | |
39279cc3 | 5032 | |
581bb050 LZ |
5033 | err = btrfs_find_free_ino(root, &objectid); |
5034 | if (err) | |
5035 | goto out_fail; | |
5036 | ||
aec7477b | 5037 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5038 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5039 | S_IFDIR | mode, &index); |
39279cc3 CM |
5040 | if (IS_ERR(inode)) { |
5041 | err = PTR_ERR(inode); | |
5042 | goto out_fail; | |
5043 | } | |
5f39d397 | 5044 | |
39279cc3 | 5045 | drop_on_err = 1; |
33268eaf | 5046 | |
2a7dba39 | 5047 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5048 | if (err) |
5049 | goto out_fail; | |
5050 | ||
39279cc3 CM |
5051 | inode->i_op = &btrfs_dir_inode_operations; |
5052 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5053 | |
dbe674a9 | 5054 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5055 | err = btrfs_update_inode(trans, root, inode); |
5056 | if (err) | |
5057 | goto out_fail; | |
5f39d397 | 5058 | |
a1b075d2 JB |
5059 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5060 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5061 | if (err) |
5062 | goto out_fail; | |
5f39d397 | 5063 | |
39279cc3 CM |
5064 | d_instantiate(dentry, inode); |
5065 | drop_on_err = 0; | |
39279cc3 CM |
5066 | |
5067 | out_fail: | |
d3c2fdcf | 5068 | nr = trans->blocks_used; |
7ad85bb7 | 5069 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5070 | if (drop_on_err) |
5071 | iput(inode); | |
d3c2fdcf | 5072 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5073 | return err; |
5074 | } | |
5075 | ||
d352ac68 CM |
5076 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5077 | * and an extent that you want to insert, deal with overlap and insert | |
5078 | * the new extent into the tree. | |
5079 | */ | |
3b951516 CM |
5080 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5081 | struct extent_map *existing, | |
e6dcd2dc CM |
5082 | struct extent_map *em, |
5083 | u64 map_start, u64 map_len) | |
3b951516 CM |
5084 | { |
5085 | u64 start_diff; | |
3b951516 | 5086 | |
e6dcd2dc CM |
5087 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5088 | start_diff = map_start - em->start; | |
5089 | em->start = map_start; | |
5090 | em->len = map_len; | |
c8b97818 CM |
5091 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5092 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5093 | em->block_start += start_diff; |
c8b97818 CM |
5094 | em->block_len -= start_diff; |
5095 | } | |
e6dcd2dc | 5096 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5097 | } |
5098 | ||
c8b97818 CM |
5099 | static noinline int uncompress_inline(struct btrfs_path *path, |
5100 | struct inode *inode, struct page *page, | |
5101 | size_t pg_offset, u64 extent_offset, | |
5102 | struct btrfs_file_extent_item *item) | |
5103 | { | |
5104 | int ret; | |
5105 | struct extent_buffer *leaf = path->nodes[0]; | |
5106 | char *tmp; | |
5107 | size_t max_size; | |
5108 | unsigned long inline_size; | |
5109 | unsigned long ptr; | |
261507a0 | 5110 | int compress_type; |
c8b97818 CM |
5111 | |
5112 | WARN_ON(pg_offset != 0); | |
261507a0 | 5113 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5114 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5115 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5116 | btrfs_item_nr(leaf, path->slots[0])); | |
5117 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5118 | if (!tmp) |
5119 | return -ENOMEM; | |
c8b97818 CM |
5120 | ptr = btrfs_file_extent_inline_start(item); |
5121 | ||
5122 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5123 | ||
5b050f04 | 5124 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5125 | ret = btrfs_decompress(compress_type, tmp, page, |
5126 | extent_offset, inline_size, max_size); | |
c8b97818 | 5127 | if (ret) { |
7ac687d9 | 5128 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5129 | unsigned long copy_size = min_t(u64, |
5130 | PAGE_CACHE_SIZE - pg_offset, | |
5131 | max_size - extent_offset); | |
5132 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5133 | kunmap_atomic(kaddr); |
c8b97818 CM |
5134 | } |
5135 | kfree(tmp); | |
5136 | return 0; | |
5137 | } | |
5138 | ||
d352ac68 CM |
5139 | /* |
5140 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5141 | * the ugly parts come from merging extents from the disk with the in-ram |
5142 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5143 | * where the in-ram extents might be locked pending data=ordered completion. |
5144 | * | |
5145 | * This also copies inline extents directly into the page. | |
5146 | */ | |
d397712b | 5147 | |
a52d9a80 | 5148 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5149 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5150 | int create) |
5151 | { | |
5152 | int ret; | |
5153 | int err = 0; | |
db94535d | 5154 | u64 bytenr; |
a52d9a80 CM |
5155 | u64 extent_start = 0; |
5156 | u64 extent_end = 0; | |
33345d01 | 5157 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5158 | u32 found_type; |
f421950f | 5159 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5160 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5161 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5162 | struct extent_buffer *leaf; |
5163 | struct btrfs_key found_key; | |
a52d9a80 CM |
5164 | struct extent_map *em = NULL; |
5165 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5166 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5167 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5168 | int compress_type; |
a52d9a80 | 5169 | |
a52d9a80 | 5170 | again: |
890871be | 5171 | read_lock(&em_tree->lock); |
d1310b2e | 5172 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5173 | if (em) |
5174 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5175 | read_unlock(&em_tree->lock); |
d1310b2e | 5176 | |
a52d9a80 | 5177 | if (em) { |
e1c4b745 CM |
5178 | if (em->start > start || em->start + em->len <= start) |
5179 | free_extent_map(em); | |
5180 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5181 | free_extent_map(em); |
5182 | else | |
5183 | goto out; | |
a52d9a80 | 5184 | } |
172ddd60 | 5185 | em = alloc_extent_map(); |
a52d9a80 | 5186 | if (!em) { |
d1310b2e CM |
5187 | err = -ENOMEM; |
5188 | goto out; | |
a52d9a80 | 5189 | } |
e6dcd2dc | 5190 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5191 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5192 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5193 | em->len = (u64)-1; |
c8b97818 | 5194 | em->block_len = (u64)-1; |
f421950f CM |
5195 | |
5196 | if (!path) { | |
5197 | path = btrfs_alloc_path(); | |
026fd317 JB |
5198 | if (!path) { |
5199 | err = -ENOMEM; | |
5200 | goto out; | |
5201 | } | |
5202 | /* | |
5203 | * Chances are we'll be called again, so go ahead and do | |
5204 | * readahead | |
5205 | */ | |
5206 | path->reada = 1; | |
f421950f CM |
5207 | } |
5208 | ||
179e29e4 CM |
5209 | ret = btrfs_lookup_file_extent(trans, root, path, |
5210 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5211 | if (ret < 0) { |
5212 | err = ret; | |
5213 | goto out; | |
5214 | } | |
5215 | ||
5216 | if (ret != 0) { | |
5217 | if (path->slots[0] == 0) | |
5218 | goto not_found; | |
5219 | path->slots[0]--; | |
5220 | } | |
5221 | ||
5f39d397 CM |
5222 | leaf = path->nodes[0]; |
5223 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5224 | struct btrfs_file_extent_item); |
a52d9a80 | 5225 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5226 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5227 | found_type = btrfs_key_type(&found_key); | |
5228 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5229 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5230 | goto not_found; | |
5231 | } | |
5232 | ||
5f39d397 CM |
5233 | found_type = btrfs_file_extent_type(leaf, item); |
5234 | extent_start = found_key.offset; | |
261507a0 | 5235 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5236 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5237 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5238 | extent_end = extent_start + |
db94535d | 5239 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5240 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5241 | size_t size; | |
5242 | size = btrfs_file_extent_inline_len(leaf, item); | |
5243 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5244 | ~((u64)root->sectorsize - 1); | |
5245 | } | |
5246 | ||
5247 | if (start >= extent_end) { | |
5248 | path->slots[0]++; | |
5249 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5250 | ret = btrfs_next_leaf(root, path); | |
5251 | if (ret < 0) { | |
5252 | err = ret; | |
5253 | goto out; | |
a52d9a80 | 5254 | } |
9036c102 YZ |
5255 | if (ret > 0) |
5256 | goto not_found; | |
5257 | leaf = path->nodes[0]; | |
a52d9a80 | 5258 | } |
9036c102 YZ |
5259 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5260 | if (found_key.objectid != objectid || | |
5261 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5262 | goto not_found; | |
5263 | if (start + len <= found_key.offset) | |
5264 | goto not_found; | |
5265 | em->start = start; | |
5266 | em->len = found_key.offset - start; | |
5267 | goto not_found_em; | |
5268 | } | |
5269 | ||
d899e052 YZ |
5270 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5271 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5272 | em->start = extent_start; |
5273 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5274 | em->orig_start = extent_start - |
5275 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5276 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5277 | if (bytenr == 0) { | |
5f39d397 | 5278 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5279 | goto insert; |
5280 | } | |
261507a0 | 5281 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5282 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5283 | em->compress_type = compress_type; |
c8b97818 CM |
5284 | em->block_start = bytenr; |
5285 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5286 | item); | |
5287 | } else { | |
5288 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5289 | em->block_start = bytenr; | |
5290 | em->block_len = em->len; | |
d899e052 YZ |
5291 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5292 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5293 | } |
a52d9a80 CM |
5294 | goto insert; |
5295 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5296 | unsigned long ptr; |
a52d9a80 | 5297 | char *map; |
3326d1b0 CM |
5298 | size_t size; |
5299 | size_t extent_offset; | |
5300 | size_t copy_size; | |
a52d9a80 | 5301 | |
689f9346 | 5302 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5303 | if (!page || create) { |
689f9346 | 5304 | em->start = extent_start; |
9036c102 | 5305 | em->len = extent_end - extent_start; |
689f9346 Y |
5306 | goto out; |
5307 | } | |
5f39d397 | 5308 | |
9036c102 YZ |
5309 | size = btrfs_file_extent_inline_len(leaf, item); |
5310 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5311 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5312 | size - extent_offset); |
3326d1b0 | 5313 | em->start = extent_start + extent_offset; |
70dec807 CM |
5314 | em->len = (copy_size + root->sectorsize - 1) & |
5315 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5316 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5317 | if (compress_type) { |
c8b97818 | 5318 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5319 | em->compress_type = compress_type; |
5320 | } | |
689f9346 | 5321 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5322 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5323 | if (btrfs_file_extent_compression(leaf, item) != |
5324 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5325 | ret = uncompress_inline(path, inode, page, |
5326 | pg_offset, | |
5327 | extent_offset, item); | |
79787eaa | 5328 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5329 | } else { |
5330 | map = kmap(page); | |
5331 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5332 | copy_size); | |
93c82d57 CM |
5333 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5334 | memset(map + pg_offset + copy_size, 0, | |
5335 | PAGE_CACHE_SIZE - pg_offset - | |
5336 | copy_size); | |
5337 | } | |
c8b97818 CM |
5338 | kunmap(page); |
5339 | } | |
179e29e4 CM |
5340 | flush_dcache_page(page); |
5341 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5342 | BUG(); |
179e29e4 CM |
5343 | if (!trans) { |
5344 | kunmap(page); | |
5345 | free_extent_map(em); | |
5346 | em = NULL; | |
ff5714cc | 5347 | |
b3b4aa74 | 5348 | btrfs_release_path(path); |
7a7eaa40 | 5349 | trans = btrfs_join_transaction(root); |
ff5714cc | 5350 | |
3612b495 TI |
5351 | if (IS_ERR(trans)) |
5352 | return ERR_CAST(trans); | |
179e29e4 CM |
5353 | goto again; |
5354 | } | |
c8b97818 | 5355 | map = kmap(page); |
70dec807 | 5356 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5357 | copy_size); |
c8b97818 | 5358 | kunmap(page); |
179e29e4 | 5359 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5360 | } |
d1310b2e | 5361 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5362 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5363 | goto insert; |
5364 | } else { | |
d397712b | 5365 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5366 | WARN_ON(1); |
5367 | } | |
5368 | not_found: | |
5369 | em->start = start; | |
d1310b2e | 5370 | em->len = len; |
a52d9a80 | 5371 | not_found_em: |
5f39d397 | 5372 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5373 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5374 | insert: |
b3b4aa74 | 5375 | btrfs_release_path(path); |
d1310b2e | 5376 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5377 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5378 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5379 | (unsigned long long)em->len, | |
5380 | (unsigned long long)start, | |
5381 | (unsigned long long)len); | |
a52d9a80 CM |
5382 | err = -EIO; |
5383 | goto out; | |
5384 | } | |
d1310b2e CM |
5385 | |
5386 | err = 0; | |
890871be | 5387 | write_lock(&em_tree->lock); |
a52d9a80 | 5388 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5389 | /* it is possible that someone inserted the extent into the tree |
5390 | * while we had the lock dropped. It is also possible that | |
5391 | * an overlapping map exists in the tree | |
5392 | */ | |
a52d9a80 | 5393 | if (ret == -EEXIST) { |
3b951516 | 5394 | struct extent_map *existing; |
e6dcd2dc CM |
5395 | |
5396 | ret = 0; | |
5397 | ||
3b951516 | 5398 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5399 | if (existing && (existing->start > start || |
5400 | existing->start + existing->len <= start)) { | |
5401 | free_extent_map(existing); | |
5402 | existing = NULL; | |
5403 | } | |
3b951516 CM |
5404 | if (!existing) { |
5405 | existing = lookup_extent_mapping(em_tree, em->start, | |
5406 | em->len); | |
5407 | if (existing) { | |
5408 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5409 | em, start, |
5410 | root->sectorsize); | |
3b951516 CM |
5411 | free_extent_map(existing); |
5412 | if (err) { | |
5413 | free_extent_map(em); | |
5414 | em = NULL; | |
5415 | } | |
5416 | } else { | |
5417 | err = -EIO; | |
3b951516 CM |
5418 | free_extent_map(em); |
5419 | em = NULL; | |
5420 | } | |
5421 | } else { | |
5422 | free_extent_map(em); | |
5423 | em = existing; | |
e6dcd2dc | 5424 | err = 0; |
a52d9a80 | 5425 | } |
a52d9a80 | 5426 | } |
890871be | 5427 | write_unlock(&em_tree->lock); |
a52d9a80 | 5428 | out: |
1abe9b8a | 5429 | |
5430 | trace_btrfs_get_extent(root, em); | |
5431 | ||
f421950f CM |
5432 | if (path) |
5433 | btrfs_free_path(path); | |
a52d9a80 CM |
5434 | if (trans) { |
5435 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5436 | if (!err) |
a52d9a80 CM |
5437 | err = ret; |
5438 | } | |
a52d9a80 CM |
5439 | if (err) { |
5440 | free_extent_map(em); | |
a52d9a80 CM |
5441 | return ERR_PTR(err); |
5442 | } | |
79787eaa | 5443 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5444 | return em; |
5445 | } | |
5446 | ||
ec29ed5b CM |
5447 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5448 | size_t pg_offset, u64 start, u64 len, | |
5449 | int create) | |
5450 | { | |
5451 | struct extent_map *em; | |
5452 | struct extent_map *hole_em = NULL; | |
5453 | u64 range_start = start; | |
5454 | u64 end; | |
5455 | u64 found; | |
5456 | u64 found_end; | |
5457 | int err = 0; | |
5458 | ||
5459 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5460 | if (IS_ERR(em)) | |
5461 | return em; | |
5462 | if (em) { | |
5463 | /* | |
5464 | * if our em maps to a hole, there might | |
5465 | * actually be delalloc bytes behind it | |
5466 | */ | |
5467 | if (em->block_start != EXTENT_MAP_HOLE) | |
5468 | return em; | |
5469 | else | |
5470 | hole_em = em; | |
5471 | } | |
5472 | ||
5473 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5474 | end = start + len; | |
5475 | if (end < start) | |
5476 | end = (u64)-1; | |
5477 | else | |
5478 | end -= 1; | |
5479 | ||
5480 | em = NULL; | |
5481 | ||
5482 | /* ok, we didn't find anything, lets look for delalloc */ | |
5483 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5484 | end, len, EXTENT_DELALLOC, 1); | |
5485 | found_end = range_start + found; | |
5486 | if (found_end < range_start) | |
5487 | found_end = (u64)-1; | |
5488 | ||
5489 | /* | |
5490 | * we didn't find anything useful, return | |
5491 | * the original results from get_extent() | |
5492 | */ | |
5493 | if (range_start > end || found_end <= start) { | |
5494 | em = hole_em; | |
5495 | hole_em = NULL; | |
5496 | goto out; | |
5497 | } | |
5498 | ||
5499 | /* adjust the range_start to make sure it doesn't | |
5500 | * go backwards from the start they passed in | |
5501 | */ | |
5502 | range_start = max(start,range_start); | |
5503 | found = found_end - range_start; | |
5504 | ||
5505 | if (found > 0) { | |
5506 | u64 hole_start = start; | |
5507 | u64 hole_len = len; | |
5508 | ||
172ddd60 | 5509 | em = alloc_extent_map(); |
ec29ed5b CM |
5510 | if (!em) { |
5511 | err = -ENOMEM; | |
5512 | goto out; | |
5513 | } | |
5514 | /* | |
5515 | * when btrfs_get_extent can't find anything it | |
5516 | * returns one huge hole | |
5517 | * | |
5518 | * make sure what it found really fits our range, and | |
5519 | * adjust to make sure it is based on the start from | |
5520 | * the caller | |
5521 | */ | |
5522 | if (hole_em) { | |
5523 | u64 calc_end = extent_map_end(hole_em); | |
5524 | ||
5525 | if (calc_end <= start || (hole_em->start > end)) { | |
5526 | free_extent_map(hole_em); | |
5527 | hole_em = NULL; | |
5528 | } else { | |
5529 | hole_start = max(hole_em->start, start); | |
5530 | hole_len = calc_end - hole_start; | |
5531 | } | |
5532 | } | |
5533 | em->bdev = NULL; | |
5534 | if (hole_em && range_start > hole_start) { | |
5535 | /* our hole starts before our delalloc, so we | |
5536 | * have to return just the parts of the hole | |
5537 | * that go until the delalloc starts | |
5538 | */ | |
5539 | em->len = min(hole_len, | |
5540 | range_start - hole_start); | |
5541 | em->start = hole_start; | |
5542 | em->orig_start = hole_start; | |
5543 | /* | |
5544 | * don't adjust block start at all, | |
5545 | * it is fixed at EXTENT_MAP_HOLE | |
5546 | */ | |
5547 | em->block_start = hole_em->block_start; | |
5548 | em->block_len = hole_len; | |
5549 | } else { | |
5550 | em->start = range_start; | |
5551 | em->len = found; | |
5552 | em->orig_start = range_start; | |
5553 | em->block_start = EXTENT_MAP_DELALLOC; | |
5554 | em->block_len = found; | |
5555 | } | |
5556 | } else if (hole_em) { | |
5557 | return hole_em; | |
5558 | } | |
5559 | out: | |
5560 | ||
5561 | free_extent_map(hole_em); | |
5562 | if (err) { | |
5563 | free_extent_map(em); | |
5564 | return ERR_PTR(err); | |
5565 | } | |
5566 | return em; | |
5567 | } | |
5568 | ||
4b46fce2 | 5569 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5570 | struct extent_map *em, |
4b46fce2 JB |
5571 | u64 start, u64 len) |
5572 | { | |
5573 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5574 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5575 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5576 | struct btrfs_key ins; | |
5577 | u64 alloc_hint; | |
5578 | int ret; | |
16d299ac | 5579 | bool insert = false; |
4b46fce2 | 5580 | |
16d299ac JB |
5581 | /* |
5582 | * Ok if the extent map we looked up is a hole and is for the exact | |
5583 | * range we want, there is no reason to allocate a new one, however if | |
5584 | * it is not right then we need to free this one and drop the cache for | |
5585 | * our range. | |
5586 | */ | |
5587 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5588 | em->len != len) { | |
5589 | free_extent_map(em); | |
5590 | em = NULL; | |
5591 | insert = true; | |
5592 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5593 | } | |
4b46fce2 | 5594 | |
7a7eaa40 | 5595 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5596 | if (IS_ERR(trans)) |
5597 | return ERR_CAST(trans); | |
4b46fce2 | 5598 | |
4cb5300b CM |
5599 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5600 | btrfs_add_inode_defrag(trans, inode); | |
5601 | ||
4b46fce2 JB |
5602 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5603 | ||
5604 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5605 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5606 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5607 | if (ret) { |
5608 | em = ERR_PTR(ret); | |
5609 | goto out; | |
5610 | } | |
5611 | ||
4b46fce2 | 5612 | if (!em) { |
172ddd60 | 5613 | em = alloc_extent_map(); |
16d299ac JB |
5614 | if (!em) { |
5615 | em = ERR_PTR(-ENOMEM); | |
5616 | goto out; | |
5617 | } | |
4b46fce2 JB |
5618 | } |
5619 | ||
5620 | em->start = start; | |
5621 | em->orig_start = em->start; | |
5622 | em->len = ins.offset; | |
5623 | ||
5624 | em->block_start = ins.objectid; | |
5625 | em->block_len = ins.offset; | |
5626 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5627 | |
5628 | /* | |
5629 | * We need to do this because if we're using the original em we searched | |
5630 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5631 | */ | |
5632 | em->flags = 0; | |
4b46fce2 JB |
5633 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5634 | ||
16d299ac | 5635 | while (insert) { |
4b46fce2 JB |
5636 | write_lock(&em_tree->lock); |
5637 | ret = add_extent_mapping(em_tree, em); | |
5638 | write_unlock(&em_tree->lock); | |
5639 | if (ret != -EEXIST) | |
5640 | break; | |
5641 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5642 | } | |
5643 | ||
5644 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5645 | ins.offset, ins.offset, 0); | |
5646 | if (ret) { | |
5647 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5648 | em = ERR_PTR(ret); | |
5649 | } | |
5650 | out: | |
5651 | btrfs_end_transaction(trans, root); | |
5652 | return em; | |
5653 | } | |
5654 | ||
46bfbb5c CM |
5655 | /* |
5656 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5657 | * block must be cow'd | |
5658 | */ | |
5659 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5660 | struct inode *inode, u64 offset, u64 len) | |
5661 | { | |
5662 | struct btrfs_path *path; | |
5663 | int ret; | |
5664 | struct extent_buffer *leaf; | |
5665 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5666 | struct btrfs_file_extent_item *fi; | |
5667 | struct btrfs_key key; | |
5668 | u64 disk_bytenr; | |
5669 | u64 backref_offset; | |
5670 | u64 extent_end; | |
5671 | u64 num_bytes; | |
5672 | int slot; | |
5673 | int found_type; | |
5674 | ||
5675 | path = btrfs_alloc_path(); | |
5676 | if (!path) | |
5677 | return -ENOMEM; | |
5678 | ||
33345d01 | 5679 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5680 | offset, 0); |
5681 | if (ret < 0) | |
5682 | goto out; | |
5683 | ||
5684 | slot = path->slots[0]; | |
5685 | if (ret == 1) { | |
5686 | if (slot == 0) { | |
5687 | /* can't find the item, must cow */ | |
5688 | ret = 0; | |
5689 | goto out; | |
5690 | } | |
5691 | slot--; | |
5692 | } | |
5693 | ret = 0; | |
5694 | leaf = path->nodes[0]; | |
5695 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5696 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5697 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5698 | /* not our file or wrong item type, must cow */ | |
5699 | goto out; | |
5700 | } | |
5701 | ||
5702 | if (key.offset > offset) { | |
5703 | /* Wrong offset, must cow */ | |
5704 | goto out; | |
5705 | } | |
5706 | ||
5707 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5708 | found_type = btrfs_file_extent_type(leaf, fi); | |
5709 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5710 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5711 | /* not a regular extent, must cow */ | |
5712 | goto out; | |
5713 | } | |
5714 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5715 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5716 | ||
5717 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5718 | if (extent_end < offset + len) { | |
5719 | /* extent doesn't include our full range, must cow */ | |
5720 | goto out; | |
5721 | } | |
5722 | ||
5723 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5724 | goto out; | |
5725 | ||
5726 | /* | |
5727 | * look for other files referencing this extent, if we | |
5728 | * find any we must cow | |
5729 | */ | |
33345d01 | 5730 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5731 | key.offset - backref_offset, disk_bytenr)) |
5732 | goto out; | |
5733 | ||
5734 | /* | |
5735 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5736 | * in this extent we are about to write. If there | |
5737 | * are any csums in that range we have to cow in order | |
5738 | * to keep the csums correct | |
5739 | */ | |
5740 | disk_bytenr += backref_offset; | |
5741 | disk_bytenr += offset - key.offset; | |
5742 | num_bytes = min(offset + len, extent_end) - offset; | |
5743 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5744 | goto out; | |
5745 | /* | |
5746 | * all of the above have passed, it is safe to overwrite this extent | |
5747 | * without cow | |
5748 | */ | |
5749 | ret = 1; | |
5750 | out: | |
5751 | btrfs_free_path(path); | |
5752 | return ret; | |
5753 | } | |
5754 | ||
4b46fce2 JB |
5755 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5756 | struct buffer_head *bh_result, int create) | |
5757 | { | |
5758 | struct extent_map *em; | |
5759 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5760 | u64 start = iblock << inode->i_blkbits; | |
5761 | u64 len = bh_result->b_size; | |
46bfbb5c | 5762 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5763 | |
5764 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5765 | if (IS_ERR(em)) | |
5766 | return PTR_ERR(em); | |
5767 | ||
5768 | /* | |
5769 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5770 | * io. INLINE is special, and we could probably kludge it in here, but | |
5771 | * it's still buffered so for safety lets just fall back to the generic | |
5772 | * buffered path. | |
5773 | * | |
5774 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5775 | * decompress it, so there will be buffering required no matter what we | |
5776 | * do, so go ahead and fallback to buffered. | |
5777 | * | |
5778 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5779 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5780 | * the generic code. | |
5781 | */ | |
5782 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5783 | em->block_start == EXTENT_MAP_INLINE) { | |
5784 | free_extent_map(em); | |
5785 | return -ENOTBLK; | |
5786 | } | |
5787 | ||
5788 | /* Just a good old fashioned hole, return */ | |
5789 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5790 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5791 | free_extent_map(em); | |
5792 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5793 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
d0082371 | 5794 | start + root->sectorsize - 1); |
4b46fce2 JB |
5795 | return 0; |
5796 | } | |
5797 | ||
5798 | /* | |
5799 | * We don't allocate a new extent in the following cases | |
5800 | * | |
5801 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5802 | * existing extent. | |
5803 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5804 | * just use the extent. | |
5805 | * | |
5806 | */ | |
46bfbb5c CM |
5807 | if (!create) { |
5808 | len = em->len - (start - em->start); | |
4b46fce2 | 5809 | goto map; |
46bfbb5c | 5810 | } |
4b46fce2 JB |
5811 | |
5812 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5813 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5814 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5815 | int type; |
5816 | int ret; | |
46bfbb5c | 5817 | u64 block_start; |
4b46fce2 JB |
5818 | |
5819 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5820 | type = BTRFS_ORDERED_PREALLOC; | |
5821 | else | |
5822 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5823 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5824 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5825 | |
5826 | /* | |
5827 | * we're not going to log anything, but we do need | |
5828 | * to make sure the current transaction stays open | |
5829 | * while we look for nocow cross refs | |
5830 | */ | |
7a7eaa40 | 5831 | trans = btrfs_join_transaction(root); |
3612b495 | 5832 | if (IS_ERR(trans)) |
46bfbb5c CM |
5833 | goto must_cow; |
5834 | ||
5835 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5836 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5837 | block_start, len, len, type); | |
5838 | btrfs_end_transaction(trans, root); | |
5839 | if (ret) { | |
5840 | free_extent_map(em); | |
5841 | return ret; | |
5842 | } | |
5843 | goto unlock; | |
4b46fce2 | 5844 | } |
46bfbb5c | 5845 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5846 | } |
46bfbb5c CM |
5847 | must_cow: |
5848 | /* | |
5849 | * this will cow the extent, reset the len in case we changed | |
5850 | * it above | |
5851 | */ | |
5852 | len = bh_result->b_size; | |
16d299ac | 5853 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5854 | if (IS_ERR(em)) |
5855 | return PTR_ERR(em); | |
5856 | len = min(len, em->len - (start - em->start)); | |
5857 | unlock: | |
4845e44f CM |
5858 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5859 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5860 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5861 | map: |
5862 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5863 | inode->i_blkbits; | |
46bfbb5c | 5864 | bh_result->b_size = len; |
4b46fce2 JB |
5865 | bh_result->b_bdev = em->bdev; |
5866 | set_buffer_mapped(bh_result); | |
5867 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5868 | set_buffer_new(bh_result); | |
5869 | ||
5870 | free_extent_map(em); | |
5871 | ||
5872 | return 0; | |
5873 | } | |
5874 | ||
5875 | struct btrfs_dio_private { | |
5876 | struct inode *inode; | |
5877 | u64 logical_offset; | |
5878 | u64 disk_bytenr; | |
5879 | u64 bytes; | |
5880 | u32 *csums; | |
5881 | void *private; | |
e65e1535 MX |
5882 | |
5883 | /* number of bios pending for this dio */ | |
5884 | atomic_t pending_bios; | |
5885 | ||
5886 | /* IO errors */ | |
5887 | int errors; | |
5888 | ||
5889 | struct bio *orig_bio; | |
4b46fce2 JB |
5890 | }; |
5891 | ||
5892 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5893 | { | |
e65e1535 | 5894 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5895 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5896 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5897 | struct inode *inode = dip->inode; |
5898 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5899 | u64 start; | |
5900 | u32 *private = dip->csums; | |
5901 | ||
5902 | start = dip->logical_offset; | |
5903 | do { | |
5904 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5905 | struct page *page = bvec->bv_page; | |
5906 | char *kaddr; | |
5907 | u32 csum = ~(u32)0; | |
5908 | unsigned long flags; | |
5909 | ||
5910 | local_irq_save(flags); | |
7ac687d9 | 5911 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
5912 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
5913 | csum, bvec->bv_len); | |
5914 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 5915 | kunmap_atomic(kaddr); |
4b46fce2 JB |
5916 | local_irq_restore(flags); |
5917 | ||
5918 | flush_dcache_page(bvec->bv_page); | |
5919 | if (csum != *private) { | |
33345d01 | 5920 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 5921 | " %llu csum %u private %u\n", |
33345d01 LZ |
5922 | (unsigned long long)btrfs_ino(inode), |
5923 | (unsigned long long)start, | |
4b46fce2 JB |
5924 | csum, *private); |
5925 | err = -EIO; | |
5926 | } | |
5927 | } | |
5928 | ||
5929 | start += bvec->bv_len; | |
5930 | private++; | |
5931 | bvec++; | |
5932 | } while (bvec <= bvec_end); | |
5933 | ||
5934 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 5935 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
5936 | bio->bi_private = dip->private; |
5937 | ||
5938 | kfree(dip->csums); | |
5939 | kfree(dip); | |
c0da7aa1 JB |
5940 | |
5941 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5942 | if (err) | |
5943 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5944 | dio_end_io(bio, err); |
5945 | } | |
5946 | ||
5947 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5948 | { | |
5949 | struct btrfs_dio_private *dip = bio->bi_private; | |
5950 | struct inode *inode = dip->inode; | |
5951 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 5952 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
5953 | u64 ordered_offset = dip->logical_offset; |
5954 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5955 | int ret; |
5956 | ||
5957 | if (err) | |
5958 | goto out_done; | |
163cf09c CM |
5959 | again: |
5960 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5961 | &ordered_offset, | |
5fd02043 | 5962 | ordered_bytes, !err); |
4b46fce2 | 5963 | if (!ret) |
163cf09c | 5964 | goto out_test; |
4b46fce2 | 5965 | |
5fd02043 JB |
5966 | ordered->work.func = finish_ordered_fn; |
5967 | ordered->work.flags = 0; | |
5968 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
5969 | &ordered->work); | |
163cf09c CM |
5970 | out_test: |
5971 | /* | |
5972 | * our bio might span multiple ordered extents. If we haven't | |
5973 | * completed the accounting for the whole dio, go back and try again | |
5974 | */ | |
5975 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5976 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5977 | ordered_offset; | |
5fd02043 | 5978 | ordered = NULL; |
163cf09c CM |
5979 | goto again; |
5980 | } | |
4b46fce2 JB |
5981 | out_done: |
5982 | bio->bi_private = dip->private; | |
5983 | ||
4b46fce2 | 5984 | kfree(dip); |
c0da7aa1 JB |
5985 | |
5986 | /* If we had an error make sure to clear the uptodate flag */ | |
5987 | if (err) | |
5988 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5989 | dio_end_io(bio, err); |
5990 | } | |
5991 | ||
eaf25d93 CM |
5992 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5993 | struct bio *bio, int mirror_num, | |
5994 | unsigned long bio_flags, u64 offset) | |
5995 | { | |
5996 | int ret; | |
5997 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5998 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 5999 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6000 | return 0; |
6001 | } | |
6002 | ||
e65e1535 MX |
6003 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6004 | { | |
6005 | struct btrfs_dio_private *dip = bio->bi_private; | |
6006 | ||
6007 | if (err) { | |
33345d01 | 6008 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6009 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6010 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6011 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6012 | dip->errors = 1; |
6013 | ||
6014 | /* | |
6015 | * before atomic variable goto zero, we must make sure | |
6016 | * dip->errors is perceived to be set. | |
6017 | */ | |
6018 | smp_mb__before_atomic_dec(); | |
6019 | } | |
6020 | ||
6021 | /* if there are more bios still pending for this dio, just exit */ | |
6022 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6023 | goto out; | |
6024 | ||
6025 | if (dip->errors) | |
6026 | bio_io_error(dip->orig_bio); | |
6027 | else { | |
6028 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6029 | bio_endio(dip->orig_bio, 0); | |
6030 | } | |
6031 | out: | |
6032 | bio_put(bio); | |
6033 | } | |
6034 | ||
6035 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6036 | u64 first_sector, gfp_t gfp_flags) | |
6037 | { | |
6038 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6039 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6040 | } | |
6041 | ||
6042 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6043 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 6044 | u32 *csums, int async_submit) |
e65e1535 MX |
6045 | { |
6046 | int write = rw & REQ_WRITE; | |
6047 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6048 | int ret; | |
6049 | ||
6050 | bio_get(bio); | |
5fd02043 JB |
6051 | |
6052 | if (!write) { | |
6053 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6054 | if (ret) | |
6055 | goto err; | |
6056 | } | |
e65e1535 | 6057 | |
1ae39938 JB |
6058 | if (skip_sum) |
6059 | goto map; | |
6060 | ||
6061 | if (write && async_submit) { | |
e65e1535 MX |
6062 | ret = btrfs_wq_submit_bio(root->fs_info, |
6063 | inode, rw, bio, 0, 0, | |
6064 | file_offset, | |
6065 | __btrfs_submit_bio_start_direct_io, | |
6066 | __btrfs_submit_bio_done); | |
6067 | goto err; | |
1ae39938 JB |
6068 | } else if (write) { |
6069 | /* | |
6070 | * If we aren't doing async submit, calculate the csum of the | |
6071 | * bio now. | |
6072 | */ | |
6073 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6074 | if (ret) | |
6075 | goto err; | |
c2db1073 TI |
6076 | } else if (!skip_sum) { |
6077 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 6078 | file_offset, csums); |
c2db1073 TI |
6079 | if (ret) |
6080 | goto err; | |
6081 | } | |
e65e1535 | 6082 | |
1ae39938 JB |
6083 | map: |
6084 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6085 | err: |
6086 | bio_put(bio); | |
6087 | return ret; | |
6088 | } | |
6089 | ||
6090 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6091 | int skip_sum) | |
6092 | { | |
6093 | struct inode *inode = dip->inode; | |
6094 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6095 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6096 | struct bio *bio; | |
6097 | struct bio *orig_bio = dip->orig_bio; | |
6098 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6099 | u64 start_sector = orig_bio->bi_sector; | |
6100 | u64 file_offset = dip->logical_offset; | |
6101 | u64 submit_len = 0; | |
6102 | u64 map_length; | |
6103 | int nr_pages = 0; | |
6104 | u32 *csums = dip->csums; | |
6105 | int ret = 0; | |
1ae39938 | 6106 | int async_submit = 0; |
98bc3149 | 6107 | int write = rw & REQ_WRITE; |
e65e1535 | 6108 | |
e65e1535 MX |
6109 | map_length = orig_bio->bi_size; |
6110 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6111 | &map_length, NULL, 0); | |
6112 | if (ret) { | |
64728bbb | 6113 | bio_put(orig_bio); |
e65e1535 MX |
6114 | return -EIO; |
6115 | } | |
6116 | ||
02f57c7a JB |
6117 | if (map_length >= orig_bio->bi_size) { |
6118 | bio = orig_bio; | |
6119 | goto submit; | |
6120 | } | |
6121 | ||
1ae39938 | 6122 | async_submit = 1; |
02f57c7a JB |
6123 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6124 | if (!bio) | |
6125 | return -ENOMEM; | |
6126 | bio->bi_private = dip; | |
6127 | bio->bi_end_io = btrfs_end_dio_bio; | |
6128 | atomic_inc(&dip->pending_bios); | |
6129 | ||
e65e1535 MX |
6130 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6131 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6132 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6133 | bvec->bv_offset) < bvec->bv_len)) { | |
6134 | /* | |
6135 | * inc the count before we submit the bio so | |
6136 | * we know the end IO handler won't happen before | |
6137 | * we inc the count. Otherwise, the dip might get freed | |
6138 | * before we're done setting it up | |
6139 | */ | |
6140 | atomic_inc(&dip->pending_bios); | |
6141 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6142 | file_offset, skip_sum, | |
1ae39938 | 6143 | csums, async_submit); |
e65e1535 MX |
6144 | if (ret) { |
6145 | bio_put(bio); | |
6146 | atomic_dec(&dip->pending_bios); | |
6147 | goto out_err; | |
6148 | } | |
6149 | ||
98bc3149 JB |
6150 | /* Write's use the ordered csums */ |
6151 | if (!write && !skip_sum) | |
e65e1535 MX |
6152 | csums = csums + nr_pages; |
6153 | start_sector += submit_len >> 9; | |
6154 | file_offset += submit_len; | |
6155 | ||
6156 | submit_len = 0; | |
6157 | nr_pages = 0; | |
6158 | ||
6159 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6160 | start_sector, GFP_NOFS); | |
6161 | if (!bio) | |
6162 | goto out_err; | |
6163 | bio->bi_private = dip; | |
6164 | bio->bi_end_io = btrfs_end_dio_bio; | |
6165 | ||
6166 | map_length = orig_bio->bi_size; | |
6167 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6168 | &map_length, NULL, 0); | |
6169 | if (ret) { | |
6170 | bio_put(bio); | |
6171 | goto out_err; | |
6172 | } | |
6173 | } else { | |
6174 | submit_len += bvec->bv_len; | |
6175 | nr_pages ++; | |
6176 | bvec++; | |
6177 | } | |
6178 | } | |
6179 | ||
02f57c7a | 6180 | submit: |
e65e1535 | 6181 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6182 | csums, async_submit); |
e65e1535 MX |
6183 | if (!ret) |
6184 | return 0; | |
6185 | ||
6186 | bio_put(bio); | |
6187 | out_err: | |
6188 | dip->errors = 1; | |
6189 | /* | |
6190 | * before atomic variable goto zero, we must | |
6191 | * make sure dip->errors is perceived to be set. | |
6192 | */ | |
6193 | smp_mb__before_atomic_dec(); | |
6194 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6195 | bio_io_error(dip->orig_bio); | |
6196 | ||
6197 | /* bio_end_io() will handle error, so we needn't return it */ | |
6198 | return 0; | |
6199 | } | |
6200 | ||
4b46fce2 JB |
6201 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6202 | loff_t file_offset) | |
6203 | { | |
6204 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6205 | struct btrfs_dio_private *dip; | |
6206 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6207 | int skip_sum; |
7b6d91da | 6208 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6209 | int ret = 0; |
6210 | ||
6211 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6212 | ||
6213 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6214 | if (!dip) { | |
6215 | ret = -ENOMEM; | |
6216 | goto free_ordered; | |
6217 | } | |
6218 | dip->csums = NULL; | |
6219 | ||
98bc3149 JB |
6220 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6221 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6222 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6223 | if (!dip->csums) { | |
b4966b77 | 6224 | kfree(dip); |
4b46fce2 JB |
6225 | ret = -ENOMEM; |
6226 | goto free_ordered; | |
6227 | } | |
6228 | } | |
6229 | ||
6230 | dip->private = bio->bi_private; | |
6231 | dip->inode = inode; | |
6232 | dip->logical_offset = file_offset; | |
6233 | ||
4b46fce2 JB |
6234 | dip->bytes = 0; |
6235 | do { | |
6236 | dip->bytes += bvec->bv_len; | |
6237 | bvec++; | |
6238 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6239 | ||
46bfbb5c | 6240 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6241 | bio->bi_private = dip; |
e65e1535 MX |
6242 | dip->errors = 0; |
6243 | dip->orig_bio = bio; | |
6244 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6245 | |
6246 | if (write) | |
6247 | bio->bi_end_io = btrfs_endio_direct_write; | |
6248 | else | |
6249 | bio->bi_end_io = btrfs_endio_direct_read; | |
6250 | ||
e65e1535 MX |
6251 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6252 | if (!ret) | |
eaf25d93 | 6253 | return; |
4b46fce2 JB |
6254 | free_ordered: |
6255 | /* | |
6256 | * If this is a write, we need to clean up the reserved space and kill | |
6257 | * the ordered extent. | |
6258 | */ | |
6259 | if (write) { | |
6260 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6261 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6262 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6263 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6264 | btrfs_free_reserved_extent(root, ordered->start, | |
6265 | ordered->disk_len); | |
6266 | btrfs_put_ordered_extent(ordered); | |
6267 | btrfs_put_ordered_extent(ordered); | |
6268 | } | |
6269 | bio_endio(bio, ret); | |
6270 | } | |
6271 | ||
5a5f79b5 CM |
6272 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6273 | const struct iovec *iov, loff_t offset, | |
6274 | unsigned long nr_segs) | |
6275 | { | |
6276 | int seg; | |
a1b75f7d | 6277 | int i; |
5a5f79b5 CM |
6278 | size_t size; |
6279 | unsigned long addr; | |
6280 | unsigned blocksize_mask = root->sectorsize - 1; | |
6281 | ssize_t retval = -EINVAL; | |
6282 | loff_t end = offset; | |
6283 | ||
6284 | if (offset & blocksize_mask) | |
6285 | goto out; | |
6286 | ||
6287 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6288 | for (seg = 0; seg < nr_segs; seg++) { | |
6289 | addr = (unsigned long)iov[seg].iov_base; | |
6290 | size = iov[seg].iov_len; | |
6291 | end += size; | |
a1b75f7d | 6292 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6293 | goto out; |
a1b75f7d JB |
6294 | |
6295 | /* If this is a write we don't need to check anymore */ | |
6296 | if (rw & WRITE) | |
6297 | continue; | |
6298 | ||
6299 | /* | |
6300 | * Check to make sure we don't have duplicate iov_base's in this | |
6301 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6302 | * when reading back. | |
6303 | */ | |
6304 | for (i = seg + 1; i < nr_segs; i++) { | |
6305 | if (iov[seg].iov_base == iov[i].iov_base) | |
6306 | goto out; | |
6307 | } | |
5a5f79b5 CM |
6308 | } |
6309 | retval = 0; | |
6310 | out: | |
6311 | return retval; | |
6312 | } | |
16432985 CM |
6313 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6314 | const struct iovec *iov, loff_t offset, | |
6315 | unsigned long nr_segs) | |
6316 | { | |
4b46fce2 JB |
6317 | struct file *file = iocb->ki_filp; |
6318 | struct inode *inode = file->f_mapping->host; | |
6319 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6320 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6321 | u64 lockstart, lockend; |
6322 | ssize_t ret; | |
4845e44f CM |
6323 | int writing = rw & WRITE; |
6324 | int write_bits = 0; | |
3f7c579c | 6325 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6326 | |
5a5f79b5 CM |
6327 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6328 | offset, nr_segs)) { | |
6329 | return 0; | |
6330 | } | |
6331 | ||
4b46fce2 | 6332 | lockstart = offset; |
3f7c579c CM |
6333 | lockend = offset + count - 1; |
6334 | ||
6335 | if (writing) { | |
6336 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6337 | if (ret) | |
6338 | goto out; | |
6339 | } | |
4845e44f | 6340 | |
4b46fce2 | 6341 | while (1) { |
4845e44f | 6342 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 6343 | 0, &cached_state); |
4b46fce2 JB |
6344 | /* |
6345 | * We're concerned with the entire range that we're going to be | |
6346 | * doing DIO to, so we need to make sure theres no ordered | |
6347 | * extents in this range. | |
6348 | */ | |
6349 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6350 | lockend - lockstart + 1); | |
6351 | if (!ordered) | |
6352 | break; | |
4845e44f CM |
6353 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6354 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6355 | btrfs_start_ordered_extent(inode, ordered, 1); |
6356 | btrfs_put_ordered_extent(ordered); | |
6357 | cond_resched(); | |
6358 | } | |
6359 | ||
4845e44f CM |
6360 | /* |
6361 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6362 | * the dirty or uptodate bits | |
6363 | */ | |
6364 | if (writing) { | |
6365 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6366 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
3fbe5c02 | 6367 | EXTENT_DELALLOC, NULL, &cached_state, |
4845e44f CM |
6368 | GFP_NOFS); |
6369 | if (ret) { | |
6370 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6371 | lockend, EXTENT_LOCKED | write_bits, | |
6372 | 1, 0, &cached_state, GFP_NOFS); | |
6373 | goto out; | |
6374 | } | |
6375 | } | |
6376 | ||
6377 | free_extent_state(cached_state); | |
6378 | cached_state = NULL; | |
6379 | ||
5a5f79b5 CM |
6380 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6381 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6382 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6383 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6384 | |
6385 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6386 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6387 | offset + iov_length(iov, nr_segs) - 1, | |
6388 | EXTENT_LOCKED | write_bits, 1, 0, | |
6389 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6390 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6391 | /* | |
6392 | * We're falling back to buffered, unlock the section we didn't | |
6393 | * do IO on. | |
6394 | */ | |
4845e44f CM |
6395 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6396 | offset + iov_length(iov, nr_segs) - 1, | |
6397 | EXTENT_LOCKED | write_bits, 1, 0, | |
6398 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6399 | } |
4845e44f CM |
6400 | out: |
6401 | free_extent_state(cached_state); | |
4b46fce2 | 6402 | return ret; |
16432985 CM |
6403 | } |
6404 | ||
1506fcc8 YS |
6405 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6406 | __u64 start, __u64 len) | |
6407 | { | |
ec29ed5b | 6408 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6409 | } |
6410 | ||
a52d9a80 | 6411 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6412 | { |
d1310b2e CM |
6413 | struct extent_io_tree *tree; |
6414 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6415 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6416 | } |
1832a6d5 | 6417 | |
a52d9a80 | 6418 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6419 | { |
d1310b2e | 6420 | struct extent_io_tree *tree; |
b888db2b CM |
6421 | |
6422 | ||
6423 | if (current->flags & PF_MEMALLOC) { | |
6424 | redirty_page_for_writepage(wbc, page); | |
6425 | unlock_page(page); | |
6426 | return 0; | |
6427 | } | |
d1310b2e | 6428 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6429 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6430 | } |
6431 | ||
f421950f CM |
6432 | int btrfs_writepages(struct address_space *mapping, |
6433 | struct writeback_control *wbc) | |
b293f02e | 6434 | { |
d1310b2e | 6435 | struct extent_io_tree *tree; |
771ed689 | 6436 | |
d1310b2e | 6437 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6438 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6439 | } | |
6440 | ||
3ab2fb5a CM |
6441 | static int |
6442 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6443 | struct list_head *pages, unsigned nr_pages) | |
6444 | { | |
d1310b2e CM |
6445 | struct extent_io_tree *tree; |
6446 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6447 | return extent_readpages(tree, mapping, pages, nr_pages, |
6448 | btrfs_get_extent); | |
6449 | } | |
e6dcd2dc | 6450 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6451 | { |
d1310b2e CM |
6452 | struct extent_io_tree *tree; |
6453 | struct extent_map_tree *map; | |
a52d9a80 | 6454 | int ret; |
8c2383c3 | 6455 | |
d1310b2e CM |
6456 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6457 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6458 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6459 | if (ret == 1) { |
6460 | ClearPagePrivate(page); | |
6461 | set_page_private(page, 0); | |
6462 | page_cache_release(page); | |
39279cc3 | 6463 | } |
a52d9a80 | 6464 | return ret; |
39279cc3 CM |
6465 | } |
6466 | ||
e6dcd2dc CM |
6467 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6468 | { | |
98509cfc CM |
6469 | if (PageWriteback(page) || PageDirty(page)) |
6470 | return 0; | |
b335b003 | 6471 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6472 | } |
6473 | ||
a52d9a80 | 6474 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6475 | { |
5fd02043 | 6476 | struct inode *inode = page->mapping->host; |
d1310b2e | 6477 | struct extent_io_tree *tree; |
e6dcd2dc | 6478 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6479 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6480 | u64 page_start = page_offset(page); |
6481 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6482 | |
8b62b72b CM |
6483 | /* |
6484 | * we have the page locked, so new writeback can't start, | |
6485 | * and the dirty bit won't be cleared while we are here. | |
6486 | * | |
6487 | * Wait for IO on this page so that we can safely clear | |
6488 | * the PagePrivate2 bit and do ordered accounting | |
6489 | */ | |
e6dcd2dc | 6490 | wait_on_page_writeback(page); |
8b62b72b | 6491 | |
5fd02043 | 6492 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6493 | if (offset) { |
6494 | btrfs_releasepage(page, GFP_NOFS); | |
6495 | return; | |
6496 | } | |
d0082371 | 6497 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
5fd02043 | 6498 | ordered = btrfs_lookup_ordered_extent(inode, |
e6dcd2dc CM |
6499 | page_offset(page)); |
6500 | if (ordered) { | |
eb84ae03 CM |
6501 | /* |
6502 | * IO on this page will never be started, so we need | |
6503 | * to account for any ordered extents now | |
6504 | */ | |
e6dcd2dc CM |
6505 | clear_extent_bit(tree, page_start, page_end, |
6506 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6507 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6508 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6509 | /* |
6510 | * whoever cleared the private bit is responsible | |
6511 | * for the finish_ordered_io | |
6512 | */ | |
5fd02043 JB |
6513 | if (TestClearPagePrivate2(page) && |
6514 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6515 | PAGE_CACHE_SIZE, 1)) { | |
6516 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6517 | } |
e6dcd2dc | 6518 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6519 | cached_state = NULL; |
d0082371 | 6520 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6521 | } |
6522 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6523 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6524 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6525 | __btrfs_releasepage(page, GFP_NOFS); |
6526 | ||
4a096752 | 6527 | ClearPageChecked(page); |
9ad6b7bc | 6528 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6529 | ClearPagePrivate(page); |
6530 | set_page_private(page, 0); | |
6531 | page_cache_release(page); | |
6532 | } | |
39279cc3 CM |
6533 | } |
6534 | ||
9ebefb18 CM |
6535 | /* |
6536 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6537 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6538 | * be careful to check for EOF conditions here. We set the page up correctly | |
6539 | * for a written page which means we get ENOSPC checking when writing into | |
6540 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6541 | * support these features. | |
6542 | * | |
6543 | * We are not allowed to take the i_mutex here so we have to play games to | |
6544 | * protect against truncate races as the page could now be beyond EOF. Because | |
6545 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6546 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6547 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6548 | * unlock the page. | |
6549 | */ | |
c2ec175c | 6550 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6551 | { |
c2ec175c | 6552 | struct page *page = vmf->page; |
6da6abae | 6553 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6554 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6555 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6556 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6557 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6558 | char *kaddr; |
6559 | unsigned long zero_start; | |
9ebefb18 | 6560 | loff_t size; |
1832a6d5 | 6561 | int ret; |
9998eb70 | 6562 | int reserved = 0; |
a52d9a80 | 6563 | u64 page_start; |
e6dcd2dc | 6564 | u64 page_end; |
9ebefb18 | 6565 | |
0ca1f7ce | 6566 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6567 | if (!ret) { |
22c44fe6 | 6568 | ret = btrfs_update_time(vma->vm_file); |
9998eb70 CM |
6569 | reserved = 1; |
6570 | } | |
56a76f82 NP |
6571 | if (ret) { |
6572 | if (ret == -ENOMEM) | |
6573 | ret = VM_FAULT_OOM; | |
6574 | else /* -ENOSPC, -EIO, etc */ | |
6575 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6576 | if (reserved) |
6577 | goto out; | |
6578 | goto out_noreserve; | |
56a76f82 | 6579 | } |
1832a6d5 | 6580 | |
56a76f82 | 6581 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6582 | again: |
9ebefb18 | 6583 | lock_page(page); |
9ebefb18 | 6584 | size = i_size_read(inode); |
e6dcd2dc CM |
6585 | page_start = page_offset(page); |
6586 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6587 | |
9ebefb18 | 6588 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6589 | (page_start >= size)) { |
9ebefb18 CM |
6590 | /* page got truncated out from underneath us */ |
6591 | goto out_unlock; | |
6592 | } | |
e6dcd2dc CM |
6593 | wait_on_page_writeback(page); |
6594 | ||
d0082371 | 6595 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6596 | set_page_extent_mapped(page); |
6597 | ||
eb84ae03 CM |
6598 | /* |
6599 | * we can't set the delalloc bits if there are pending ordered | |
6600 | * extents. Drop our locks and wait for them to finish | |
6601 | */ | |
e6dcd2dc CM |
6602 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6603 | if (ordered) { | |
2ac55d41 JB |
6604 | unlock_extent_cached(io_tree, page_start, page_end, |
6605 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6606 | unlock_page(page); |
eb84ae03 | 6607 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6608 | btrfs_put_ordered_extent(ordered); |
6609 | goto again; | |
6610 | } | |
6611 | ||
fbf19087 JB |
6612 | /* |
6613 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6614 | * if it was already dirty, so for space accounting reasons we need to | |
6615 | * clear any delalloc bits for the range we are fixing to save. There | |
6616 | * is probably a better way to do this, but for now keep consistent with | |
6617 | * prepare_pages in the normal write path. | |
6618 | */ | |
2ac55d41 | 6619 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6620 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6621 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6622 | |
2ac55d41 JB |
6623 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6624 | &cached_state); | |
9ed74f2d | 6625 | if (ret) { |
2ac55d41 JB |
6626 | unlock_extent_cached(io_tree, page_start, page_end, |
6627 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6628 | ret = VM_FAULT_SIGBUS; |
6629 | goto out_unlock; | |
6630 | } | |
e6dcd2dc | 6631 | ret = 0; |
9ebefb18 CM |
6632 | |
6633 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6634 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6635 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6636 | else |
e6dcd2dc | 6637 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6638 | |
e6dcd2dc CM |
6639 | if (zero_start != PAGE_CACHE_SIZE) { |
6640 | kaddr = kmap(page); | |
6641 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6642 | flush_dcache_page(page); | |
6643 | kunmap(page); | |
6644 | } | |
247e743c | 6645 | ClearPageChecked(page); |
e6dcd2dc | 6646 | set_page_dirty(page); |
50a9b214 | 6647 | SetPageUptodate(page); |
5a3f23d5 | 6648 | |
257c62e1 CM |
6649 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6650 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6651 | ||
2ac55d41 | 6652 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6653 | |
6654 | out_unlock: | |
50a9b214 CM |
6655 | if (!ret) |
6656 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6657 | unlock_page(page); |
1832a6d5 | 6658 | out: |
ec39e180 | 6659 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6660 | out_noreserve: |
9ebefb18 CM |
6661 | return ret; |
6662 | } | |
6663 | ||
a41ad394 | 6664 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6665 | { |
6666 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6667 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6668 | int ret; |
3893e33b | 6669 | int err = 0; |
39279cc3 | 6670 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6671 | unsigned long nr; |
dbe674a9 | 6672 | u64 mask = root->sectorsize - 1; |
07127184 | 6673 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6674 | |
5d5e103a JB |
6675 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6676 | if (ret) | |
a41ad394 | 6677 | return ret; |
8082510e | 6678 | |
4a096752 | 6679 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6680 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6681 | |
fcb80c2a JB |
6682 | /* |
6683 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6684 | * 3 things going on here | |
6685 | * | |
6686 | * 1) We need to reserve space for our orphan item and the space to | |
6687 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6688 | * orphan item because we didn't reserve space to remove it. | |
6689 | * | |
6690 | * 2) We need to reserve space to update our inode. | |
6691 | * | |
6692 | * 3) We need to have something to cache all the space that is going to | |
6693 | * be free'd up by the truncate operation, but also have some slack | |
6694 | * space reserved in case it uses space during the truncate (thank you | |
6695 | * very much snapshotting). | |
6696 | * | |
6697 | * And we need these to all be seperate. The fact is we can use alot of | |
6698 | * space doing the truncate, and we have no earthly idea how much space | |
6699 | * we will use, so we need the truncate reservation to be seperate so it | |
6700 | * doesn't end up using space reserved for updating the inode or | |
6701 | * removing the orphan item. We also need to be able to stop the | |
6702 | * transaction and start a new one, which means we need to be able to | |
6703 | * update the inode several times, and we have no idea of knowing how | |
6704 | * many times that will be, so we can't just reserve 1 item for the | |
6705 | * entirety of the opration, so that has to be done seperately as well. | |
6706 | * Then there is the orphan item, which does indeed need to be held on | |
6707 | * to for the whole operation, and we need nobody to touch this reserved | |
6708 | * space except the orphan code. | |
6709 | * | |
6710 | * So that leaves us with | |
6711 | * | |
6712 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6713 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6714 | * transaction reservation. | |
6715 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6716 | * updating the inode. | |
6717 | */ | |
6718 | rsv = btrfs_alloc_block_rsv(root); | |
6719 | if (!rsv) | |
6720 | return -ENOMEM; | |
4a338542 | 6721 | rsv->size = min_size; |
f0cd846e | 6722 | |
907cbceb | 6723 | /* |
07127184 | 6724 | * 1 for the truncate slack space |
907cbceb JB |
6725 | * 1 for the orphan item we're going to add |
6726 | * 1 for the orphan item deletion | |
6727 | * 1 for updating the inode. | |
6728 | */ | |
fcb80c2a JB |
6729 | trans = btrfs_start_transaction(root, 4); |
6730 | if (IS_ERR(trans)) { | |
6731 | err = PTR_ERR(trans); | |
6732 | goto out; | |
6733 | } | |
f0cd846e | 6734 | |
907cbceb JB |
6735 | /* Migrate the slack space for the truncate to our reserve */ |
6736 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6737 | min_size); | |
fcb80c2a | 6738 | BUG_ON(ret); |
f0cd846e JB |
6739 | |
6740 | ret = btrfs_orphan_add(trans, inode); | |
6741 | if (ret) { | |
6742 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6743 | goto out; |
f0cd846e JB |
6744 | } |
6745 | ||
5a3f23d5 CM |
6746 | /* |
6747 | * setattr is responsible for setting the ordered_data_close flag, | |
6748 | * but that is only tested during the last file release. That | |
6749 | * could happen well after the next commit, leaving a great big | |
6750 | * window where new writes may get lost if someone chooses to write | |
6751 | * to this file after truncating to zero | |
6752 | * | |
6753 | * The inode doesn't have any dirty data here, and so if we commit | |
6754 | * this is a noop. If someone immediately starts writing to the inode | |
6755 | * it is very likely we'll catch some of their writes in this | |
6756 | * transaction, and the commit will find this file on the ordered | |
6757 | * data list with good things to send down. | |
6758 | * | |
6759 | * This is a best effort solution, there is still a window where | |
6760 | * using truncate to replace the contents of the file will | |
6761 | * end up with a zero length file after a crash. | |
6762 | */ | |
72ac3c0d JB |
6763 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
6764 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
6765 | btrfs_add_ordered_operation(trans, root, inode); |
6766 | ||
8082510e | 6767 | while (1) { |
36ba022a | 6768 | ret = btrfs_block_rsv_refill(root, rsv, min_size); |
907cbceb JB |
6769 | if (ret) { |
6770 | /* | |
6771 | * This can only happen with the original transaction we | |
6772 | * started above, every other time we shouldn't have a | |
6773 | * transaction started yet. | |
6774 | */ | |
6775 | if (ret == -EAGAIN) | |
6776 | goto end_trans; | |
6777 | err = ret; | |
6778 | break; | |
6779 | } | |
6780 | ||
d68fc57b | 6781 | if (!trans) { |
907cbceb JB |
6782 | /* Just need the 1 for updating the inode */ |
6783 | trans = btrfs_start_transaction(root, 1); | |
fcb80c2a | 6784 | if (IS_ERR(trans)) { |
7041ee97 JB |
6785 | ret = err = PTR_ERR(trans); |
6786 | trans = NULL; | |
6787 | break; | |
fcb80c2a | 6788 | } |
d68fc57b YZ |
6789 | } |
6790 | ||
907cbceb JB |
6791 | trans->block_rsv = rsv; |
6792 | ||
8082510e YZ |
6793 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6794 | inode->i_size, | |
6795 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6796 | if (ret != -EAGAIN) { |
6797 | err = ret; | |
8082510e | 6798 | break; |
3893e33b | 6799 | } |
39279cc3 | 6800 | |
fcb80c2a | 6801 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6802 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6803 | if (ret) { |
6804 | err = ret; | |
6805 | break; | |
6806 | } | |
907cbceb | 6807 | end_trans: |
8082510e YZ |
6808 | nr = trans->blocks_used; |
6809 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6810 | trans = NULL; |
8082510e | 6811 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6812 | } |
6813 | ||
6814 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6815 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6816 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6817 | if (ret) |
6818 | err = ret; | |
ded5db9d JB |
6819 | } else if (ret && inode->i_nlink > 0) { |
6820 | /* | |
6821 | * Failed to do the truncate, remove us from the in memory | |
6822 | * orphan list. | |
6823 | */ | |
6824 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6825 | } |
6826 | ||
917c16b2 CM |
6827 | if (trans) { |
6828 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
6829 | ret = btrfs_update_inode(trans, root, inode); | |
6830 | if (ret && !err) | |
6831 | err = ret; | |
7b128766 | 6832 | |
917c16b2 | 6833 | nr = trans->blocks_used; |
7ad85bb7 | 6834 | ret = btrfs_end_transaction(trans, root); |
917c16b2 CM |
6835 | btrfs_btree_balance_dirty(root, nr); |
6836 | } | |
fcb80c2a JB |
6837 | |
6838 | out: | |
6839 | btrfs_free_block_rsv(root, rsv); | |
6840 | ||
3893e33b JB |
6841 | if (ret && !err) |
6842 | err = ret; | |
a41ad394 | 6843 | |
3893e33b | 6844 | return err; |
39279cc3 CM |
6845 | } |
6846 | ||
d352ac68 CM |
6847 | /* |
6848 | * create a new subvolume directory/inode (helper for the ioctl). | |
6849 | */ | |
d2fb3437 | 6850 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6851 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6852 | { |
39279cc3 | 6853 | struct inode *inode; |
76dda93c | 6854 | int err; |
00e4e6b3 | 6855 | u64 index = 0; |
39279cc3 | 6856 | |
12fc9d09 FA |
6857 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
6858 | new_dirid, new_dirid, | |
6859 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
6860 | &index); | |
54aa1f4d | 6861 | if (IS_ERR(inode)) |
f46b5a66 | 6862 | return PTR_ERR(inode); |
39279cc3 CM |
6863 | inode->i_op = &btrfs_dir_inode_operations; |
6864 | inode->i_fop = &btrfs_dir_file_operations; | |
6865 | ||
bfe86848 | 6866 | set_nlink(inode, 1); |
dbe674a9 | 6867 | btrfs_i_size_write(inode, 0); |
3b96362c | 6868 | |
76dda93c | 6869 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 6870 | |
76dda93c | 6871 | iput(inode); |
ce598979 | 6872 | return err; |
39279cc3 CM |
6873 | } |
6874 | ||
39279cc3 CM |
6875 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6876 | { | |
6877 | struct btrfs_inode *ei; | |
2ead6ae7 | 6878 | struct inode *inode; |
39279cc3 CM |
6879 | |
6880 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6881 | if (!ei) | |
6882 | return NULL; | |
2ead6ae7 YZ |
6883 | |
6884 | ei->root = NULL; | |
6885 | ei->space_info = NULL; | |
6886 | ei->generation = 0; | |
15ee9bc7 | 6887 | ei->last_trans = 0; |
257c62e1 | 6888 | ei->last_sub_trans = 0; |
e02119d5 | 6889 | ei->logged_trans = 0; |
2ead6ae7 | 6890 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
6891 | ei->disk_i_size = 0; |
6892 | ei->flags = 0; | |
7709cde3 | 6893 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
6894 | ei->index_cnt = (u64)-1; |
6895 | ei->last_unlink_trans = 0; | |
6896 | ||
9e0baf60 JB |
6897 | spin_lock_init(&ei->lock); |
6898 | ei->outstanding_extents = 0; | |
6899 | ei->reserved_extents = 0; | |
2ead6ae7 | 6900 | |
72ac3c0d | 6901 | ei->runtime_flags = 0; |
261507a0 | 6902 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 6903 | |
16cdcec7 MX |
6904 | ei->delayed_node = NULL; |
6905 | ||
2ead6ae7 | 6906 | inode = &ei->vfs_inode; |
a8067e02 | 6907 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
6908 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
6909 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
6910 | ei->io_tree.track_uptodate = 1; |
6911 | ei->io_failure_tree.track_uptodate = 1; | |
2ead6ae7 | 6912 | mutex_init(&ei->log_mutex); |
f248679e | 6913 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 6914 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 6915 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6916 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6917 | RB_CLEAR_NODE(&ei->rb_node); |
6918 | ||
6919 | return inode; | |
39279cc3 CM |
6920 | } |
6921 | ||
fa0d7e3d NP |
6922 | static void btrfs_i_callback(struct rcu_head *head) |
6923 | { | |
6924 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
6925 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
6926 | } | |
6927 | ||
39279cc3 CM |
6928 | void btrfs_destroy_inode(struct inode *inode) |
6929 | { | |
e6dcd2dc | 6930 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6931 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6932 | ||
39279cc3 CM |
6933 | WARN_ON(!list_empty(&inode->i_dentry)); |
6934 | WARN_ON(inode->i_data.nrpages); | |
9e0baf60 JB |
6935 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
6936 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
6937 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
6938 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 6939 | |
a6dbd429 JB |
6940 | /* |
6941 | * This can happen where we create an inode, but somebody else also | |
6942 | * created the same inode and we need to destroy the one we already | |
6943 | * created. | |
6944 | */ | |
6945 | if (!root) | |
6946 | goto free; | |
6947 | ||
5a3f23d5 CM |
6948 | /* |
6949 | * Make sure we're properly removed from the ordered operation | |
6950 | * lists. | |
6951 | */ | |
6952 | smp_mb(); | |
6953 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6954 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6955 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6956 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6957 | } | |
6958 | ||
8a35d95f JB |
6959 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
6960 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
6961 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
6962 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 6963 | atomic_dec(&root->orphan_inodes); |
7b128766 | 6964 | } |
7b128766 | 6965 | |
d397712b | 6966 | while (1) { |
e6dcd2dc CM |
6967 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6968 | if (!ordered) | |
6969 | break; | |
6970 | else { | |
d397712b CM |
6971 | printk(KERN_ERR "btrfs found ordered " |
6972 | "extent %llu %llu on inode cleanup\n", | |
6973 | (unsigned long long)ordered->file_offset, | |
6974 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6975 | btrfs_remove_ordered_extent(inode, ordered); |
6976 | btrfs_put_ordered_extent(ordered); | |
6977 | btrfs_put_ordered_extent(ordered); | |
6978 | } | |
6979 | } | |
5d4f98a2 | 6980 | inode_tree_del(inode); |
5b21f2ed | 6981 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6982 | free: |
16cdcec7 | 6983 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 6984 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6985 | } |
6986 | ||
45321ac5 | 6987 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6988 | { |
6989 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6990 | |
0af3d00b | 6991 | if (btrfs_root_refs(&root->root_item) == 0 && |
2cf8572d | 6992 | !btrfs_is_free_space_inode(root, inode)) |
45321ac5 | 6993 | return 1; |
76dda93c | 6994 | else |
45321ac5 | 6995 | return generic_drop_inode(inode); |
76dda93c YZ |
6996 | } |
6997 | ||
0ee0fda0 | 6998 | static void init_once(void *foo) |
39279cc3 CM |
6999 | { |
7000 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7001 | ||
7002 | inode_init_once(&ei->vfs_inode); | |
7003 | } | |
7004 | ||
7005 | void btrfs_destroy_cachep(void) | |
7006 | { | |
7007 | if (btrfs_inode_cachep) | |
7008 | kmem_cache_destroy(btrfs_inode_cachep); | |
7009 | if (btrfs_trans_handle_cachep) | |
7010 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7011 | if (btrfs_transaction_cachep) | |
7012 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7013 | if (btrfs_path_cachep) |
7014 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7015 | if (btrfs_free_space_cachep) |
7016 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
7017 | } |
7018 | ||
7019 | int btrfs_init_cachep(void) | |
7020 | { | |
9601e3f6 CH |
7021 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
7022 | sizeof(struct btrfs_inode), 0, | |
7023 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7024 | if (!btrfs_inode_cachep) |
7025 | goto fail; | |
9601e3f6 CH |
7026 | |
7027 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
7028 | sizeof(struct btrfs_trans_handle), 0, | |
7029 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7030 | if (!btrfs_trans_handle_cachep) |
7031 | goto fail; | |
9601e3f6 CH |
7032 | |
7033 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
7034 | sizeof(struct btrfs_transaction), 0, | |
7035 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7036 | if (!btrfs_transaction_cachep) |
7037 | goto fail; | |
9601e3f6 CH |
7038 | |
7039 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
7040 | sizeof(struct btrfs_path), 0, | |
7041 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7042 | if (!btrfs_path_cachep) |
7043 | goto fail; | |
9601e3f6 | 7044 | |
dc89e982 JB |
7045 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
7046 | sizeof(struct btrfs_free_space), 0, | |
7047 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7048 | if (!btrfs_free_space_cachep) | |
7049 | goto fail; | |
7050 | ||
39279cc3 CM |
7051 | return 0; |
7052 | fail: | |
7053 | btrfs_destroy_cachep(); | |
7054 | return -ENOMEM; | |
7055 | } | |
7056 | ||
7057 | static int btrfs_getattr(struct vfsmount *mnt, | |
7058 | struct dentry *dentry, struct kstat *stat) | |
7059 | { | |
7060 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
7061 | u32 blocksize = inode->i_sb->s_blocksize; |
7062 | ||
39279cc3 | 7063 | generic_fillattr(inode, stat); |
0ee5dc67 | 7064 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7065 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
7066 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
7067 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
7068 | return 0; |
7069 | } | |
7070 | ||
75e7cb7f LB |
7071 | /* |
7072 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7073 | * directory. | |
7074 | */ | |
7075 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7076 | { | |
7077 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7078 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7079 | ||
7080 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7081 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7082 | else | |
7083 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7084 | ||
7085 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
7086 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
7087 | else | |
7088 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
7089 | } | |
7090 | ||
d397712b CM |
7091 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7092 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7093 | { |
7094 | struct btrfs_trans_handle *trans; | |
7095 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7096 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7097 | struct inode *new_inode = new_dentry->d_inode; |
7098 | struct inode *old_inode = old_dentry->d_inode; | |
7099 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7100 | u64 index = 0; |
4df27c4d | 7101 | u64 root_objectid; |
39279cc3 | 7102 | int ret; |
33345d01 | 7103 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7104 | |
33345d01 | 7105 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7106 | return -EPERM; |
7107 | ||
4df27c4d | 7108 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7109 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7110 | return -EXDEV; |
7111 | ||
33345d01 LZ |
7112 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7113 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7114 | return -ENOTEMPTY; |
5f39d397 | 7115 | |
4df27c4d YZ |
7116 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7117 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7118 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7119 | /* |
7120 | * we're using rename to replace one file with another. | |
7121 | * and the replacement file is large. Start IO on it now so | |
7122 | * we don't add too much work to the end of the transaction | |
7123 | */ | |
4baf8c92 | 7124 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7125 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7126 | filemap_flush(old_inode->i_mapping); | |
7127 | ||
76dda93c | 7128 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7129 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7130 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7131 | /* |
7132 | * We want to reserve the absolute worst case amount of items. So if | |
7133 | * both inodes are subvols and we need to unlink them then that would | |
7134 | * require 4 item modifications, but if they are both normal inodes it | |
7135 | * would require 5 item modifications, so we'll assume their normal | |
7136 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7137 | * should cover the worst case number of items we'll modify. | |
7138 | */ | |
7139 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7140 | if (IS_ERR(trans)) { |
7141 | ret = PTR_ERR(trans); | |
7142 | goto out_notrans; | |
7143 | } | |
76dda93c | 7144 | |
4df27c4d YZ |
7145 | if (dest != root) |
7146 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7147 | |
a5719521 YZ |
7148 | ret = btrfs_set_inode_index(new_dir, &index); |
7149 | if (ret) | |
7150 | goto out_fail; | |
5a3f23d5 | 7151 | |
33345d01 | 7152 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7153 | /* force full log commit if subvolume involved. */ |
7154 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7155 | } else { | |
a5719521 YZ |
7156 | ret = btrfs_insert_inode_ref(trans, dest, |
7157 | new_dentry->d_name.name, | |
7158 | new_dentry->d_name.len, | |
33345d01 LZ |
7159 | old_ino, |
7160 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7161 | if (ret) |
7162 | goto out_fail; | |
4df27c4d YZ |
7163 | /* |
7164 | * this is an ugly little race, but the rename is required | |
7165 | * to make sure that if we crash, the inode is either at the | |
7166 | * old name or the new one. pinning the log transaction lets | |
7167 | * us make sure we don't allow a log commit to come in after | |
7168 | * we unlink the name but before we add the new name back in. | |
7169 | */ | |
7170 | btrfs_pin_log_trans(root); | |
7171 | } | |
5a3f23d5 CM |
7172 | /* |
7173 | * make sure the inode gets flushed if it is replacing | |
7174 | * something. | |
7175 | */ | |
33345d01 | 7176 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7177 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7178 | |
0c4d2d95 JB |
7179 | inode_inc_iversion(old_dir); |
7180 | inode_inc_iversion(new_dir); | |
7181 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7182 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7183 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7184 | old_inode->i_ctime = ctime; | |
5f39d397 | 7185 | |
12fcfd22 CM |
7186 | if (old_dentry->d_parent != new_dentry->d_parent) |
7187 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7188 | ||
33345d01 | 7189 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7190 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7191 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7192 | old_dentry->d_name.name, | |
7193 | old_dentry->d_name.len); | |
7194 | } else { | |
92986796 AV |
7195 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7196 | old_dentry->d_inode, | |
7197 | old_dentry->d_name.name, | |
7198 | old_dentry->d_name.len); | |
7199 | if (!ret) | |
7200 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7201 | } |
79787eaa JM |
7202 | if (ret) { |
7203 | btrfs_abort_transaction(trans, root, ret); | |
7204 | goto out_fail; | |
7205 | } | |
39279cc3 CM |
7206 | |
7207 | if (new_inode) { | |
0c4d2d95 | 7208 | inode_inc_iversion(new_inode); |
39279cc3 | 7209 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7210 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7211 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7212 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7213 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7214 | root_objectid, | |
7215 | new_dentry->d_name.name, | |
7216 | new_dentry->d_name.len); | |
7217 | BUG_ON(new_inode->i_nlink == 0); | |
7218 | } else { | |
7219 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7220 | new_dentry->d_inode, | |
7221 | new_dentry->d_name.name, | |
7222 | new_dentry->d_name.len); | |
7223 | } | |
79787eaa | 7224 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7225 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7226 | BUG_ON(ret); |
7b128766 | 7227 | } |
79787eaa JM |
7228 | if (ret) { |
7229 | btrfs_abort_transaction(trans, root, ret); | |
7230 | goto out_fail; | |
7231 | } | |
39279cc3 | 7232 | } |
aec7477b | 7233 | |
75e7cb7f LB |
7234 | fixup_inode_flags(new_dir, old_inode); |
7235 | ||
4df27c4d YZ |
7236 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7237 | new_dentry->d_name.name, | |
a5719521 | 7238 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7239 | if (ret) { |
7240 | btrfs_abort_transaction(trans, root, ret); | |
7241 | goto out_fail; | |
7242 | } | |
39279cc3 | 7243 | |
33345d01 | 7244 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7245 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7246 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7247 | btrfs_end_log_trans(root); |
7248 | } | |
39279cc3 | 7249 | out_fail: |
7ad85bb7 | 7250 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7251 | out_notrans: |
33345d01 | 7252 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7253 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7254 | |
39279cc3 CM |
7255 | return ret; |
7256 | } | |
7257 | ||
d352ac68 CM |
7258 | /* |
7259 | * some fairly slow code that needs optimization. This walks the list | |
7260 | * of all the inodes with pending delalloc and forces them to disk. | |
7261 | */ | |
24bbcf04 | 7262 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7263 | { |
7264 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7265 | struct btrfs_inode *binode; | |
5b21f2ed | 7266 | struct inode *inode; |
ea8c2819 | 7267 | |
c146afad YZ |
7268 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7269 | return -EROFS; | |
7270 | ||
75eff68e | 7271 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7272 | while (!list_empty(head)) { |
ea8c2819 CM |
7273 | binode = list_entry(head->next, struct btrfs_inode, |
7274 | delalloc_inodes); | |
5b21f2ed ZY |
7275 | inode = igrab(&binode->vfs_inode); |
7276 | if (!inode) | |
7277 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7278 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7279 | if (inode) { |
8c8bee1d | 7280 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7281 | if (delay_iput) |
7282 | btrfs_add_delayed_iput(inode); | |
7283 | else | |
7284 | iput(inode); | |
5b21f2ed ZY |
7285 | } |
7286 | cond_resched(); | |
75eff68e | 7287 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7288 | } |
75eff68e | 7289 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7290 | |
7291 | /* the filemap_flush will queue IO into the worker threads, but | |
7292 | * we have to make sure the IO is actually started and that | |
7293 | * ordered extents get created before we return | |
7294 | */ | |
7295 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7296 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7297 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7298 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7299 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7300 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7301 | } |
7302 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7303 | return 0; |
7304 | } | |
7305 | ||
39279cc3 CM |
7306 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7307 | const char *symname) | |
7308 | { | |
7309 | struct btrfs_trans_handle *trans; | |
7310 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7311 | struct btrfs_path *path; | |
7312 | struct btrfs_key key; | |
1832a6d5 | 7313 | struct inode *inode = NULL; |
39279cc3 CM |
7314 | int err; |
7315 | int drop_inode = 0; | |
7316 | u64 objectid; | |
00e4e6b3 | 7317 | u64 index = 0 ; |
39279cc3 CM |
7318 | int name_len; |
7319 | int datasize; | |
5f39d397 | 7320 | unsigned long ptr; |
39279cc3 | 7321 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7322 | struct extent_buffer *leaf; |
1832a6d5 | 7323 | unsigned long nr = 0; |
39279cc3 CM |
7324 | |
7325 | name_len = strlen(symname) + 1; | |
7326 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7327 | return -ENAMETOOLONG; | |
1832a6d5 | 7328 | |
9ed74f2d JB |
7329 | /* |
7330 | * 2 items for inode item and ref | |
7331 | * 2 items for dir items | |
7332 | * 1 item for xattr if selinux is on | |
7333 | */ | |
a22285a6 YZ |
7334 | trans = btrfs_start_transaction(root, 5); |
7335 | if (IS_ERR(trans)) | |
7336 | return PTR_ERR(trans); | |
1832a6d5 | 7337 | |
581bb050 LZ |
7338 | err = btrfs_find_free_ino(root, &objectid); |
7339 | if (err) | |
7340 | goto out_unlock; | |
7341 | ||
aec7477b | 7342 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7343 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7344 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7345 | if (IS_ERR(inode)) { |
7346 | err = PTR_ERR(inode); | |
39279cc3 | 7347 | goto out_unlock; |
7cf96da3 | 7348 | } |
39279cc3 | 7349 | |
2a7dba39 | 7350 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7351 | if (err) { |
7352 | drop_inode = 1; | |
7353 | goto out_unlock; | |
7354 | } | |
7355 | ||
ad19db71 CS |
7356 | /* |
7357 | * If the active LSM wants to access the inode during | |
7358 | * d_instantiate it needs these. Smack checks to see | |
7359 | * if the filesystem supports xattrs by looking at the | |
7360 | * ops vector. | |
7361 | */ | |
7362 | inode->i_fop = &btrfs_file_operations; | |
7363 | inode->i_op = &btrfs_file_inode_operations; | |
7364 | ||
a1b075d2 | 7365 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7366 | if (err) |
7367 | drop_inode = 1; | |
7368 | else { | |
7369 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7370 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7371 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7372 | } |
39279cc3 CM |
7373 | if (drop_inode) |
7374 | goto out_unlock; | |
7375 | ||
7376 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7377 | if (!path) { |
7378 | err = -ENOMEM; | |
7379 | drop_inode = 1; | |
7380 | goto out_unlock; | |
7381 | } | |
33345d01 | 7382 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7383 | key.offset = 0; |
39279cc3 CM |
7384 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7385 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7386 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7387 | datasize); | |
54aa1f4d CM |
7388 | if (err) { |
7389 | drop_inode = 1; | |
b0839166 | 7390 | btrfs_free_path(path); |
54aa1f4d CM |
7391 | goto out_unlock; |
7392 | } | |
5f39d397 CM |
7393 | leaf = path->nodes[0]; |
7394 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7395 | struct btrfs_file_extent_item); | |
7396 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7397 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7398 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7399 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7400 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7401 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7402 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7403 | ||
39279cc3 | 7404 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7405 | write_extent_buffer(leaf, symname, ptr, name_len); |
7406 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7407 | btrfs_free_path(path); |
5f39d397 | 7408 | |
39279cc3 CM |
7409 | inode->i_op = &btrfs_symlink_inode_operations; |
7410 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7411 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7412 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7413 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7414 | err = btrfs_update_inode(trans, root, inode); |
7415 | if (err) | |
7416 | drop_inode = 1; | |
39279cc3 CM |
7417 | |
7418 | out_unlock: | |
08c422c2 AV |
7419 | if (!err) |
7420 | d_instantiate(dentry, inode); | |
d3c2fdcf | 7421 | nr = trans->blocks_used; |
7ad85bb7 | 7422 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7423 | if (drop_inode) { |
7424 | inode_dec_link_count(inode); | |
7425 | iput(inode); | |
7426 | } | |
d3c2fdcf | 7427 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7428 | return err; |
7429 | } | |
16432985 | 7430 | |
0af3d00b JB |
7431 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7432 | u64 start, u64 num_bytes, u64 min_size, | |
7433 | loff_t actual_len, u64 *alloc_hint, | |
7434 | struct btrfs_trans_handle *trans) | |
d899e052 | 7435 | { |
d899e052 YZ |
7436 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7437 | struct btrfs_key ins; | |
d899e052 | 7438 | u64 cur_offset = start; |
55a61d1d | 7439 | u64 i_size; |
d899e052 | 7440 | int ret = 0; |
0af3d00b | 7441 | bool own_trans = true; |
d899e052 | 7442 | |
0af3d00b JB |
7443 | if (trans) |
7444 | own_trans = false; | |
d899e052 | 7445 | while (num_bytes > 0) { |
0af3d00b JB |
7446 | if (own_trans) { |
7447 | trans = btrfs_start_transaction(root, 3); | |
7448 | if (IS_ERR(trans)) { | |
7449 | ret = PTR_ERR(trans); | |
7450 | break; | |
7451 | } | |
5a303d5d YZ |
7452 | } |
7453 | ||
efa56464 | 7454 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7455 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7456 | if (ret) { |
0af3d00b JB |
7457 | if (own_trans) |
7458 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7459 | break; |
d899e052 | 7460 | } |
5a303d5d | 7461 | |
d899e052 YZ |
7462 | ret = insert_reserved_file_extent(trans, inode, |
7463 | cur_offset, ins.objectid, | |
7464 | ins.offset, ins.offset, | |
920bbbfb | 7465 | ins.offset, 0, 0, 0, |
d899e052 | 7466 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7467 | if (ret) { |
7468 | btrfs_abort_transaction(trans, root, ret); | |
7469 | if (own_trans) | |
7470 | btrfs_end_transaction(trans, root); | |
7471 | break; | |
7472 | } | |
a1ed835e CM |
7473 | btrfs_drop_extent_cache(inode, cur_offset, |
7474 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7475 | |
d899e052 YZ |
7476 | num_bytes -= ins.offset; |
7477 | cur_offset += ins.offset; | |
efa56464 | 7478 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7479 | |
0c4d2d95 | 7480 | inode_inc_iversion(inode); |
d899e052 | 7481 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7482 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7483 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7484 | (actual_len > inode->i_size) && |
7485 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7486 | if (cur_offset > actual_len) |
55a61d1d | 7487 | i_size = actual_len; |
d1ea6a61 | 7488 | else |
55a61d1d JB |
7489 | i_size = cur_offset; |
7490 | i_size_write(inode, i_size); | |
7491 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7492 | } |
7493 | ||
d899e052 | 7494 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7495 | |
7496 | if (ret) { | |
7497 | btrfs_abort_transaction(trans, root, ret); | |
7498 | if (own_trans) | |
7499 | btrfs_end_transaction(trans, root); | |
7500 | break; | |
7501 | } | |
d899e052 | 7502 | |
0af3d00b JB |
7503 | if (own_trans) |
7504 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7505 | } |
d899e052 YZ |
7506 | return ret; |
7507 | } | |
7508 | ||
0af3d00b JB |
7509 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7510 | u64 start, u64 num_bytes, u64 min_size, | |
7511 | loff_t actual_len, u64 *alloc_hint) | |
7512 | { | |
7513 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7514 | min_size, actual_len, alloc_hint, | |
7515 | NULL); | |
7516 | } | |
7517 | ||
7518 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7519 | struct btrfs_trans_handle *trans, int mode, | |
7520 | u64 start, u64 num_bytes, u64 min_size, | |
7521 | loff_t actual_len, u64 *alloc_hint) | |
7522 | { | |
7523 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7524 | min_size, actual_len, alloc_hint, trans); | |
7525 | } | |
7526 | ||
e6dcd2dc CM |
7527 | static int btrfs_set_page_dirty(struct page *page) |
7528 | { | |
e6dcd2dc CM |
7529 | return __set_page_dirty_nobuffers(page); |
7530 | } | |
7531 | ||
10556cb2 | 7532 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7533 | { |
b83cc969 | 7534 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7535 | umode_t mode = inode->i_mode; |
b83cc969 | 7536 | |
cb6db4e5 JM |
7537 | if (mask & MAY_WRITE && |
7538 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7539 | if (btrfs_root_readonly(root)) | |
7540 | return -EROFS; | |
7541 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7542 | return -EACCES; | |
7543 | } | |
2830ba7f | 7544 | return generic_permission(inode, mask); |
fdebe2bd | 7545 | } |
39279cc3 | 7546 | |
6e1d5dcc | 7547 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7548 | .getattr = btrfs_getattr, |
39279cc3 CM |
7549 | .lookup = btrfs_lookup, |
7550 | .create = btrfs_create, | |
7551 | .unlink = btrfs_unlink, | |
7552 | .link = btrfs_link, | |
7553 | .mkdir = btrfs_mkdir, | |
7554 | .rmdir = btrfs_rmdir, | |
7555 | .rename = btrfs_rename, | |
7556 | .symlink = btrfs_symlink, | |
7557 | .setattr = btrfs_setattr, | |
618e21d5 | 7558 | .mknod = btrfs_mknod, |
95819c05 CH |
7559 | .setxattr = btrfs_setxattr, |
7560 | .getxattr = btrfs_getxattr, | |
5103e947 | 7561 | .listxattr = btrfs_listxattr, |
95819c05 | 7562 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7563 | .permission = btrfs_permission, |
4e34e719 | 7564 | .get_acl = btrfs_get_acl, |
39279cc3 | 7565 | }; |
6e1d5dcc | 7566 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7567 | .lookup = btrfs_lookup, |
fdebe2bd | 7568 | .permission = btrfs_permission, |
4e34e719 | 7569 | .get_acl = btrfs_get_acl, |
39279cc3 | 7570 | }; |
76dda93c | 7571 | |
828c0950 | 7572 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7573 | .llseek = generic_file_llseek, |
7574 | .read = generic_read_dir, | |
cbdf5a24 | 7575 | .readdir = btrfs_real_readdir, |
34287aa3 | 7576 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7577 | #ifdef CONFIG_COMPAT |
34287aa3 | 7578 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7579 | #endif |
6bf13c0c | 7580 | .release = btrfs_release_file, |
e02119d5 | 7581 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7582 | }; |
7583 | ||
d1310b2e | 7584 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7585 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7586 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7587 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7588 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7589 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7590 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7591 | .set_bit_hook = btrfs_set_bit_hook, |
7592 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7593 | .merge_extent_hook = btrfs_merge_extent_hook, |
7594 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7595 | }; |
7596 | ||
35054394 CM |
7597 | /* |
7598 | * btrfs doesn't support the bmap operation because swapfiles | |
7599 | * use bmap to make a mapping of extents in the file. They assume | |
7600 | * these extents won't change over the life of the file and they | |
7601 | * use the bmap result to do IO directly to the drive. | |
7602 | * | |
7603 | * the btrfs bmap call would return logical addresses that aren't | |
7604 | * suitable for IO and they also will change frequently as COW | |
7605 | * operations happen. So, swapfile + btrfs == corruption. | |
7606 | * | |
7607 | * For now we're avoiding this by dropping bmap. | |
7608 | */ | |
7f09410b | 7609 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7610 | .readpage = btrfs_readpage, |
7611 | .writepage = btrfs_writepage, | |
b293f02e | 7612 | .writepages = btrfs_writepages, |
3ab2fb5a | 7613 | .readpages = btrfs_readpages, |
16432985 | 7614 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7615 | .invalidatepage = btrfs_invalidatepage, |
7616 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7617 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7618 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7619 | }; |
7620 | ||
7f09410b | 7621 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7622 | .readpage = btrfs_readpage, |
7623 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7624 | .invalidatepage = btrfs_invalidatepage, |
7625 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7626 | }; |
7627 | ||
6e1d5dcc | 7628 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7629 | .getattr = btrfs_getattr, |
7630 | .setattr = btrfs_setattr, | |
95819c05 CH |
7631 | .setxattr = btrfs_setxattr, |
7632 | .getxattr = btrfs_getxattr, | |
5103e947 | 7633 | .listxattr = btrfs_listxattr, |
95819c05 | 7634 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7635 | .permission = btrfs_permission, |
1506fcc8 | 7636 | .fiemap = btrfs_fiemap, |
4e34e719 | 7637 | .get_acl = btrfs_get_acl, |
39279cc3 | 7638 | }; |
6e1d5dcc | 7639 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7640 | .getattr = btrfs_getattr, |
7641 | .setattr = btrfs_setattr, | |
fdebe2bd | 7642 | .permission = btrfs_permission, |
95819c05 CH |
7643 | .setxattr = btrfs_setxattr, |
7644 | .getxattr = btrfs_getxattr, | |
33268eaf | 7645 | .listxattr = btrfs_listxattr, |
95819c05 | 7646 | .removexattr = btrfs_removexattr, |
4e34e719 | 7647 | .get_acl = btrfs_get_acl, |
618e21d5 | 7648 | }; |
6e1d5dcc | 7649 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7650 | .readlink = generic_readlink, |
7651 | .follow_link = page_follow_link_light, | |
7652 | .put_link = page_put_link, | |
f209561a | 7653 | .getattr = btrfs_getattr, |
22c44fe6 | 7654 | .setattr = btrfs_setattr, |
fdebe2bd | 7655 | .permission = btrfs_permission, |
0279b4cd JO |
7656 | .setxattr = btrfs_setxattr, |
7657 | .getxattr = btrfs_getxattr, | |
7658 | .listxattr = btrfs_listxattr, | |
7659 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7660 | .get_acl = btrfs_get_acl, |
39279cc3 | 7661 | }; |
76dda93c | 7662 | |
82d339d9 | 7663 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7664 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7665 | .d_release = btrfs_dentry_release, |
76dda93c | 7666 | }; |