]>
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); |
79787eaa JM |
260 | if (ret) { |
261 | btrfs_abort_transaction(trans, root, ret); | |
262 | return ret; | |
263 | } | |
0ca1f7ce | 264 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 265 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
266 | return 0; |
267 | } | |
268 | ||
771ed689 CM |
269 | struct async_extent { |
270 | u64 start; | |
271 | u64 ram_size; | |
272 | u64 compressed_size; | |
273 | struct page **pages; | |
274 | unsigned long nr_pages; | |
261507a0 | 275 | int compress_type; |
771ed689 CM |
276 | struct list_head list; |
277 | }; | |
278 | ||
279 | struct async_cow { | |
280 | struct inode *inode; | |
281 | struct btrfs_root *root; | |
282 | struct page *locked_page; | |
283 | u64 start; | |
284 | u64 end; | |
285 | struct list_head extents; | |
286 | struct btrfs_work work; | |
287 | }; | |
288 | ||
289 | static noinline int add_async_extent(struct async_cow *cow, | |
290 | u64 start, u64 ram_size, | |
291 | u64 compressed_size, | |
292 | struct page **pages, | |
261507a0 LZ |
293 | unsigned long nr_pages, |
294 | int compress_type) | |
771ed689 CM |
295 | { |
296 | struct async_extent *async_extent; | |
297 | ||
298 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 299 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
300 | async_extent->start = start; |
301 | async_extent->ram_size = ram_size; | |
302 | async_extent->compressed_size = compressed_size; | |
303 | async_extent->pages = pages; | |
304 | async_extent->nr_pages = nr_pages; | |
261507a0 | 305 | async_extent->compress_type = compress_type; |
771ed689 CM |
306 | list_add_tail(&async_extent->list, &cow->extents); |
307 | return 0; | |
308 | } | |
309 | ||
d352ac68 | 310 | /* |
771ed689 CM |
311 | * we create compressed extents in two phases. The first |
312 | * phase compresses a range of pages that have already been | |
313 | * locked (both pages and state bits are locked). | |
c8b97818 | 314 | * |
771ed689 CM |
315 | * This is done inside an ordered work queue, and the compression |
316 | * is spread across many cpus. The actual IO submission is step | |
317 | * two, and the ordered work queue takes care of making sure that | |
318 | * happens in the same order things were put onto the queue by | |
319 | * writepages and friends. | |
c8b97818 | 320 | * |
771ed689 CM |
321 | * If this code finds it can't get good compression, it puts an |
322 | * entry onto the work queue to write the uncompressed bytes. This | |
323 | * makes sure that both compressed inodes and uncompressed inodes | |
324 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 325 | */ |
771ed689 CM |
326 | static noinline int compress_file_range(struct inode *inode, |
327 | struct page *locked_page, | |
328 | u64 start, u64 end, | |
329 | struct async_cow *async_cow, | |
330 | int *num_added) | |
b888db2b CM |
331 | { |
332 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
333 | struct btrfs_trans_handle *trans; | |
db94535d | 334 | u64 num_bytes; |
db94535d | 335 | u64 blocksize = root->sectorsize; |
c8b97818 | 336 | u64 actual_end; |
42dc7bab | 337 | u64 isize = i_size_read(inode); |
e6dcd2dc | 338 | int ret = 0; |
c8b97818 CM |
339 | struct page **pages = NULL; |
340 | unsigned long nr_pages; | |
341 | unsigned long nr_pages_ret = 0; | |
342 | unsigned long total_compressed = 0; | |
343 | unsigned long total_in = 0; | |
344 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 345 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
346 | int i; |
347 | int will_compress; | |
261507a0 | 348 | int compress_type = root->fs_info->compress_type; |
b888db2b | 349 | |
4cb13e5d LB |
350 | /* if this is a small write inside eof, kick off a defrag */ |
351 | if ((end - start + 1) < 16 * 1024 && | |
352 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
353 | btrfs_add_inode_defrag(NULL, inode); |
354 | ||
42dc7bab | 355 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
356 | again: |
357 | will_compress = 0; | |
358 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
359 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 360 | |
f03d9301 CM |
361 | /* |
362 | * we don't want to send crud past the end of i_size through | |
363 | * compression, that's just a waste of CPU time. So, if the | |
364 | * end of the file is before the start of our current | |
365 | * requested range of bytes, we bail out to the uncompressed | |
366 | * cleanup code that can deal with all of this. | |
367 | * | |
368 | * It isn't really the fastest way to fix things, but this is a | |
369 | * very uncommon corner. | |
370 | */ | |
371 | if (actual_end <= start) | |
372 | goto cleanup_and_bail_uncompressed; | |
373 | ||
c8b97818 CM |
374 | total_compressed = actual_end - start; |
375 | ||
376 | /* we want to make sure that amount of ram required to uncompress | |
377 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
378 | * of a compressed extent to 128k. This is a crucial number |
379 | * because it also controls how easily we can spread reads across | |
380 | * cpus for decompression. | |
381 | * | |
382 | * We also want to make sure the amount of IO required to do | |
383 | * a random read is reasonably small, so we limit the size of | |
384 | * a compressed extent to 128k. | |
c8b97818 CM |
385 | */ |
386 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 387 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 388 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
389 | total_in = 0; |
390 | ret = 0; | |
db94535d | 391 | |
771ed689 CM |
392 | /* |
393 | * we do compression for mount -o compress and when the | |
394 | * inode has not been flagged as nocompress. This flag can | |
395 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 396 | */ |
6cbff00f | 397 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 398 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
399 | (BTRFS_I(inode)->force_compress) || |
400 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 401 | WARN_ON(pages); |
cfbc246e | 402 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
403 | if (!pages) { |
404 | /* just bail out to the uncompressed code */ | |
405 | goto cont; | |
406 | } | |
c8b97818 | 407 | |
261507a0 LZ |
408 | if (BTRFS_I(inode)->force_compress) |
409 | compress_type = BTRFS_I(inode)->force_compress; | |
410 | ||
411 | ret = btrfs_compress_pages(compress_type, | |
412 | inode->i_mapping, start, | |
413 | total_compressed, pages, | |
414 | nr_pages, &nr_pages_ret, | |
415 | &total_in, | |
416 | &total_compressed, | |
417 | max_compressed); | |
c8b97818 CM |
418 | |
419 | if (!ret) { | |
420 | unsigned long offset = total_compressed & | |
421 | (PAGE_CACHE_SIZE - 1); | |
422 | struct page *page = pages[nr_pages_ret - 1]; | |
423 | char *kaddr; | |
424 | ||
425 | /* zero the tail end of the last page, we might be | |
426 | * sending it down to disk | |
427 | */ | |
428 | if (offset) { | |
7ac687d9 | 429 | kaddr = kmap_atomic(page); |
c8b97818 CM |
430 | memset(kaddr + offset, 0, |
431 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 432 | kunmap_atomic(kaddr); |
c8b97818 CM |
433 | } |
434 | will_compress = 1; | |
435 | } | |
436 | } | |
560f7d75 | 437 | cont: |
c8b97818 | 438 | if (start == 0) { |
7a7eaa40 | 439 | trans = btrfs_join_transaction(root); |
79787eaa JM |
440 | if (IS_ERR(trans)) { |
441 | ret = PTR_ERR(trans); | |
442 | trans = NULL; | |
443 | goto cleanup_and_out; | |
444 | } | |
0ca1f7ce | 445 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 446 | |
c8b97818 | 447 | /* lets try to make an inline extent */ |
771ed689 | 448 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 449 | /* we didn't compress the entire range, try |
771ed689 | 450 | * to make an uncompressed inline extent. |
c8b97818 CM |
451 | */ |
452 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 453 | start, end, 0, 0, NULL); |
c8b97818 | 454 | } else { |
771ed689 | 455 | /* try making a compressed inline extent */ |
c8b97818 CM |
456 | ret = cow_file_range_inline(trans, root, inode, |
457 | start, end, | |
fe3f566c LZ |
458 | total_compressed, |
459 | compress_type, pages); | |
c8b97818 | 460 | } |
79787eaa | 461 | if (ret <= 0) { |
771ed689 | 462 | /* |
79787eaa JM |
463 | * inline extent creation worked or returned error, |
464 | * we don't need to create any more async work items. | |
465 | * Unlock and free up our temp pages. | |
771ed689 | 466 | */ |
c8b97818 | 467 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
468 | &BTRFS_I(inode)->io_tree, |
469 | start, end, NULL, | |
470 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 471 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 472 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
473 | |
474 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
475 | goto free_pages_out; |
476 | } | |
c2167754 | 477 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
478 | } |
479 | ||
480 | if (will_compress) { | |
481 | /* | |
482 | * we aren't doing an inline extent round the compressed size | |
483 | * up to a block size boundary so the allocator does sane | |
484 | * things | |
485 | */ | |
486 | total_compressed = (total_compressed + blocksize - 1) & | |
487 | ~(blocksize - 1); | |
488 | ||
489 | /* | |
490 | * one last check to make sure the compression is really a | |
491 | * win, compare the page count read with the blocks on disk | |
492 | */ | |
493 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
494 | ~(PAGE_CACHE_SIZE - 1); | |
495 | if (total_compressed >= total_in) { | |
496 | will_compress = 0; | |
497 | } else { | |
c8b97818 CM |
498 | num_bytes = total_in; |
499 | } | |
500 | } | |
501 | if (!will_compress && pages) { | |
502 | /* | |
503 | * the compression code ran but failed to make things smaller, | |
504 | * free any pages it allocated and our page pointer array | |
505 | */ | |
506 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 507 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
508 | page_cache_release(pages[i]); |
509 | } | |
510 | kfree(pages); | |
511 | pages = NULL; | |
512 | total_compressed = 0; | |
513 | nr_pages_ret = 0; | |
514 | ||
515 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
516 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
517 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 518 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 519 | } |
c8b97818 | 520 | } |
771ed689 CM |
521 | if (will_compress) { |
522 | *num_added += 1; | |
c8b97818 | 523 | |
771ed689 CM |
524 | /* the async work queues will take care of doing actual |
525 | * allocation on disk for these compressed pages, | |
526 | * and will submit them to the elevator. | |
527 | */ | |
528 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
529 | total_compressed, pages, nr_pages_ret, |
530 | compress_type); | |
179e29e4 | 531 | |
24ae6365 | 532 | if (start + num_bytes < end) { |
771ed689 CM |
533 | start += num_bytes; |
534 | pages = NULL; | |
535 | cond_resched(); | |
536 | goto again; | |
537 | } | |
538 | } else { | |
f03d9301 | 539 | cleanup_and_bail_uncompressed: |
771ed689 CM |
540 | /* |
541 | * No compression, but we still need to write the pages in | |
542 | * the file we've been given so far. redirty the locked | |
543 | * page if it corresponds to our extent and set things up | |
544 | * for the async work queue to run cow_file_range to do | |
545 | * the normal delalloc dance | |
546 | */ | |
547 | if (page_offset(locked_page) >= start && | |
548 | page_offset(locked_page) <= end) { | |
549 | __set_page_dirty_nobuffers(locked_page); | |
550 | /* unlocked later on in the async handlers */ | |
551 | } | |
261507a0 LZ |
552 | add_async_extent(async_cow, start, end - start + 1, |
553 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
554 | *num_added += 1; |
555 | } | |
3b951516 | 556 | |
771ed689 | 557 | out: |
79787eaa | 558 | return ret; |
771ed689 CM |
559 | |
560 | free_pages_out: | |
561 | for (i = 0; i < nr_pages_ret; i++) { | |
562 | WARN_ON(pages[i]->mapping); | |
563 | page_cache_release(pages[i]); | |
564 | } | |
d397712b | 565 | kfree(pages); |
771ed689 CM |
566 | |
567 | goto out; | |
79787eaa JM |
568 | |
569 | cleanup_and_out: | |
570 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
571 | start, end, NULL, | |
572 | EXTENT_CLEAR_UNLOCK_PAGE | | |
573 | EXTENT_CLEAR_DIRTY | | |
574 | EXTENT_CLEAR_DELALLOC | | |
575 | EXTENT_SET_WRITEBACK | | |
576 | EXTENT_END_WRITEBACK); | |
577 | if (!trans || IS_ERR(trans)) | |
578 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
579 | else | |
580 | btrfs_abort_transaction(trans, root, ret); | |
581 | goto free_pages_out; | |
771ed689 CM |
582 | } |
583 | ||
584 | /* | |
585 | * phase two of compressed writeback. This is the ordered portion | |
586 | * of the code, which only gets called in the order the work was | |
587 | * queued. We walk all the async extents created by compress_file_range | |
588 | * and send them down to the disk. | |
589 | */ | |
590 | static noinline int submit_compressed_extents(struct inode *inode, | |
591 | struct async_cow *async_cow) | |
592 | { | |
593 | struct async_extent *async_extent; | |
594 | u64 alloc_hint = 0; | |
595 | struct btrfs_trans_handle *trans; | |
596 | struct btrfs_key ins; | |
597 | struct extent_map *em; | |
598 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
599 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
600 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 601 | int ret = 0; |
771ed689 CM |
602 | |
603 | if (list_empty(&async_cow->extents)) | |
604 | return 0; | |
605 | ||
771ed689 | 606 | |
d397712b | 607 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
608 | async_extent = list_entry(async_cow->extents.next, |
609 | struct async_extent, list); | |
610 | list_del(&async_extent->list); | |
c8b97818 | 611 | |
771ed689 CM |
612 | io_tree = &BTRFS_I(inode)->io_tree; |
613 | ||
f5a84ee3 | 614 | retry: |
771ed689 CM |
615 | /* did the compression code fall back to uncompressed IO? */ |
616 | if (!async_extent->pages) { | |
617 | int page_started = 0; | |
618 | unsigned long nr_written = 0; | |
619 | ||
620 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 621 | async_extent->start + |
d0082371 | 622 | async_extent->ram_size - 1); |
771ed689 CM |
623 | |
624 | /* allocate blocks */ | |
f5a84ee3 JB |
625 | ret = cow_file_range(inode, async_cow->locked_page, |
626 | async_extent->start, | |
627 | async_extent->start + | |
628 | async_extent->ram_size - 1, | |
629 | &page_started, &nr_written, 0); | |
771ed689 | 630 | |
79787eaa JM |
631 | /* JDM XXX */ |
632 | ||
771ed689 CM |
633 | /* |
634 | * if page_started, cow_file_range inserted an | |
635 | * inline extent and took care of all the unlocking | |
636 | * and IO for us. Otherwise, we need to submit | |
637 | * all those pages down to the drive. | |
638 | */ | |
f5a84ee3 | 639 | if (!page_started && !ret) |
771ed689 CM |
640 | extent_write_locked_range(io_tree, |
641 | inode, async_extent->start, | |
d397712b | 642 | async_extent->start + |
771ed689 CM |
643 | async_extent->ram_size - 1, |
644 | btrfs_get_extent, | |
645 | WB_SYNC_ALL); | |
646 | kfree(async_extent); | |
647 | cond_resched(); | |
648 | continue; | |
649 | } | |
650 | ||
651 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 652 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 653 | |
7a7eaa40 | 654 | trans = btrfs_join_transaction(root); |
79787eaa JM |
655 | if (IS_ERR(trans)) { |
656 | ret = PTR_ERR(trans); | |
657 | } else { | |
658 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
659 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
660 | async_extent->compressed_size, |
661 | async_extent->compressed_size, | |
81c9ad23 | 662 | 0, alloc_hint, &ins, 1); |
79787eaa JM |
663 | if (ret) |
664 | btrfs_abort_transaction(trans, root, ret); | |
665 | btrfs_end_transaction(trans, root); | |
666 | } | |
c2167754 | 667 | |
f5a84ee3 JB |
668 | if (ret) { |
669 | int i; | |
670 | for (i = 0; i < async_extent->nr_pages; i++) { | |
671 | WARN_ON(async_extent->pages[i]->mapping); | |
672 | page_cache_release(async_extent->pages[i]); | |
673 | } | |
674 | kfree(async_extent->pages); | |
675 | async_extent->nr_pages = 0; | |
676 | async_extent->pages = NULL; | |
677 | unlock_extent(io_tree, async_extent->start, | |
678 | async_extent->start + | |
d0082371 | 679 | async_extent->ram_size - 1); |
79787eaa JM |
680 | if (ret == -ENOSPC) |
681 | goto retry; | |
682 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
683 | } |
684 | ||
c2167754 YZ |
685 | /* |
686 | * here we're doing allocation and writeback of the | |
687 | * compressed pages | |
688 | */ | |
689 | btrfs_drop_extent_cache(inode, async_extent->start, | |
690 | async_extent->start + | |
691 | async_extent->ram_size - 1, 0); | |
692 | ||
172ddd60 | 693 | em = alloc_extent_map(); |
79787eaa | 694 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
695 | em->start = async_extent->start; |
696 | em->len = async_extent->ram_size; | |
445a6944 | 697 | em->orig_start = em->start; |
c8b97818 | 698 | |
771ed689 CM |
699 | em->block_start = ins.objectid; |
700 | em->block_len = ins.offset; | |
701 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 702 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
703 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
704 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
705 | ||
d397712b | 706 | while (1) { |
890871be | 707 | write_lock(&em_tree->lock); |
771ed689 | 708 | ret = add_extent_mapping(em_tree, em); |
890871be | 709 | write_unlock(&em_tree->lock); |
771ed689 CM |
710 | if (ret != -EEXIST) { |
711 | free_extent_map(em); | |
712 | break; | |
713 | } | |
714 | btrfs_drop_extent_cache(inode, async_extent->start, | |
715 | async_extent->start + | |
716 | async_extent->ram_size - 1, 0); | |
717 | } | |
718 | ||
261507a0 LZ |
719 | ret = btrfs_add_ordered_extent_compress(inode, |
720 | async_extent->start, | |
721 | ins.objectid, | |
722 | async_extent->ram_size, | |
723 | ins.offset, | |
724 | BTRFS_ORDERED_COMPRESSED, | |
725 | async_extent->compress_type); | |
79787eaa | 726 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 727 | |
771ed689 CM |
728 | /* |
729 | * clear dirty, set writeback and unlock the pages. | |
730 | */ | |
731 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
732 | &BTRFS_I(inode)->io_tree, |
733 | async_extent->start, | |
734 | async_extent->start + | |
735 | async_extent->ram_size - 1, | |
736 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
737 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 738 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 739 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
740 | |
741 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
742 | async_extent->start, |
743 | async_extent->ram_size, | |
744 | ins.objectid, | |
745 | ins.offset, async_extent->pages, | |
746 | async_extent->nr_pages); | |
771ed689 | 747 | |
79787eaa | 748 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
749 | alloc_hint = ins.objectid + ins.offset; |
750 | kfree(async_extent); | |
751 | cond_resched(); | |
752 | } | |
79787eaa JM |
753 | ret = 0; |
754 | out: | |
755 | return ret; | |
756 | out_free: | |
757 | kfree(async_extent); | |
758 | goto out; | |
771ed689 CM |
759 | } |
760 | ||
4b46fce2 JB |
761 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
762 | u64 num_bytes) | |
763 | { | |
764 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
765 | struct extent_map *em; | |
766 | u64 alloc_hint = 0; | |
767 | ||
768 | read_lock(&em_tree->lock); | |
769 | em = search_extent_mapping(em_tree, start, num_bytes); | |
770 | if (em) { | |
771 | /* | |
772 | * if block start isn't an actual block number then find the | |
773 | * first block in this inode and use that as a hint. If that | |
774 | * block is also bogus then just don't worry about it. | |
775 | */ | |
776 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
777 | free_extent_map(em); | |
778 | em = search_extent_mapping(em_tree, 0, 0); | |
779 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
780 | alloc_hint = em->block_start; | |
781 | if (em) | |
782 | free_extent_map(em); | |
783 | } else { | |
784 | alloc_hint = em->block_start; | |
785 | free_extent_map(em); | |
786 | } | |
787 | } | |
788 | read_unlock(&em_tree->lock); | |
789 | ||
790 | return alloc_hint; | |
791 | } | |
792 | ||
771ed689 CM |
793 | /* |
794 | * when extent_io.c finds a delayed allocation range in the file, | |
795 | * the call backs end up in this code. The basic idea is to | |
796 | * allocate extents on disk for the range, and create ordered data structs | |
797 | * in ram to track those extents. | |
798 | * | |
799 | * locked_page is the page that writepage had locked already. We use | |
800 | * it to make sure we don't do extra locks or unlocks. | |
801 | * | |
802 | * *page_started is set to one if we unlock locked_page and do everything | |
803 | * required to start IO on it. It may be clean and already done with | |
804 | * IO when we return. | |
805 | */ | |
806 | static noinline int cow_file_range(struct inode *inode, | |
807 | struct page *locked_page, | |
808 | u64 start, u64 end, int *page_started, | |
809 | unsigned long *nr_written, | |
810 | int unlock) | |
811 | { | |
812 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
813 | struct btrfs_trans_handle *trans; | |
814 | u64 alloc_hint = 0; | |
815 | u64 num_bytes; | |
816 | unsigned long ram_size; | |
817 | u64 disk_num_bytes; | |
818 | u64 cur_alloc_size; | |
819 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
820 | struct btrfs_key ins; |
821 | struct extent_map *em; | |
822 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
823 | int ret = 0; | |
824 | ||
2cf8572d | 825 | BUG_ON(btrfs_is_free_space_inode(root, inode)); |
7a7eaa40 | 826 | trans = btrfs_join_transaction(root); |
79787eaa JM |
827 | if (IS_ERR(trans)) { |
828 | extent_clear_unlock_delalloc(inode, | |
829 | &BTRFS_I(inode)->io_tree, | |
830 | start, end, NULL, | |
831 | EXTENT_CLEAR_UNLOCK_PAGE | | |
832 | EXTENT_CLEAR_UNLOCK | | |
833 | EXTENT_CLEAR_DELALLOC | | |
834 | EXTENT_CLEAR_DIRTY | | |
835 | EXTENT_SET_WRITEBACK | | |
836 | EXTENT_END_WRITEBACK); | |
837 | return PTR_ERR(trans); | |
838 | } | |
0ca1f7ce | 839 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 840 | |
771ed689 CM |
841 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
842 | num_bytes = max(blocksize, num_bytes); | |
843 | disk_num_bytes = num_bytes; | |
844 | ret = 0; | |
845 | ||
4cb5300b | 846 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
847 | if (num_bytes < 64 * 1024 && |
848 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
849 | btrfs_add_inode_defrag(trans, inode); |
850 | ||
771ed689 CM |
851 | if (start == 0) { |
852 | /* lets try to make an inline extent */ | |
853 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 854 | start, end, 0, 0, NULL); |
771ed689 CM |
855 | if (ret == 0) { |
856 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
857 | &BTRFS_I(inode)->io_tree, |
858 | start, end, NULL, | |
859 | EXTENT_CLEAR_UNLOCK_PAGE | | |
860 | EXTENT_CLEAR_UNLOCK | | |
861 | EXTENT_CLEAR_DELALLOC | | |
862 | EXTENT_CLEAR_DIRTY | | |
863 | EXTENT_SET_WRITEBACK | | |
864 | EXTENT_END_WRITEBACK); | |
c2167754 | 865 | |
771ed689 CM |
866 | *nr_written = *nr_written + |
867 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
868 | *page_started = 1; | |
771ed689 | 869 | goto out; |
79787eaa JM |
870 | } else if (ret < 0) { |
871 | btrfs_abort_transaction(trans, root, ret); | |
872 | goto out_unlock; | |
771ed689 CM |
873 | } |
874 | } | |
875 | ||
876 | BUG_ON(disk_num_bytes > | |
6c41761f | 877 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 878 | |
4b46fce2 | 879 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
880 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
881 | ||
d397712b | 882 | while (disk_num_bytes > 0) { |
a791e35e CM |
883 | unsigned long op; |
884 | ||
287a0ab9 | 885 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 886 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 887 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 888 | &ins, 1); |
79787eaa JM |
889 | if (ret < 0) { |
890 | btrfs_abort_transaction(trans, root, ret); | |
891 | goto out_unlock; | |
892 | } | |
d397712b | 893 | |
172ddd60 | 894 | em = alloc_extent_map(); |
79787eaa | 895 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 896 | em->start = start; |
445a6944 | 897 | em->orig_start = em->start; |
771ed689 CM |
898 | ram_size = ins.offset; |
899 | em->len = ins.offset; | |
c8b97818 | 900 | |
e6dcd2dc | 901 | em->block_start = ins.objectid; |
c8b97818 | 902 | em->block_len = ins.offset; |
e6dcd2dc | 903 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 904 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 905 | |
d397712b | 906 | while (1) { |
890871be | 907 | write_lock(&em_tree->lock); |
e6dcd2dc | 908 | ret = add_extent_mapping(em_tree, em); |
890871be | 909 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
910 | if (ret != -EEXIST) { |
911 | free_extent_map(em); | |
912 | break; | |
913 | } | |
914 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 915 | start + ram_size - 1, 0); |
e6dcd2dc CM |
916 | } |
917 | ||
98d20f67 | 918 | cur_alloc_size = ins.offset; |
e6dcd2dc | 919 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 920 | ram_size, cur_alloc_size, 0); |
79787eaa | 921 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 922 | |
17d217fe YZ |
923 | if (root->root_key.objectid == |
924 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
925 | ret = btrfs_reloc_clone_csums(inode, start, | |
926 | cur_alloc_size); | |
79787eaa JM |
927 | if (ret) { |
928 | btrfs_abort_transaction(trans, root, ret); | |
929 | goto out_unlock; | |
930 | } | |
17d217fe YZ |
931 | } |
932 | ||
d397712b | 933 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 934 | break; |
d397712b | 935 | |
c8b97818 CM |
936 | /* we're not doing compressed IO, don't unlock the first |
937 | * page (which the caller expects to stay locked), don't | |
938 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
939 | * |
940 | * Do set the Private2 bit so we know this page was properly | |
941 | * setup for writepage | |
c8b97818 | 942 | */ |
a791e35e CM |
943 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
944 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
945 | EXTENT_SET_PRIVATE2; | |
946 | ||
c8b97818 CM |
947 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
948 | start, start + ram_size - 1, | |
a791e35e | 949 | locked_page, op); |
c8b97818 | 950 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
951 | num_bytes -= cur_alloc_size; |
952 | alloc_hint = ins.objectid + ins.offset; | |
953 | start += cur_alloc_size; | |
b888db2b | 954 | } |
771ed689 | 955 | ret = 0; |
79787eaa | 956 | out: |
b888db2b | 957 | btrfs_end_transaction(trans, root); |
c8b97818 | 958 | |
be20aa9d | 959 | return ret; |
79787eaa JM |
960 | out_unlock: |
961 | extent_clear_unlock_delalloc(inode, | |
962 | &BTRFS_I(inode)->io_tree, | |
963 | start, end, NULL, | |
964 | EXTENT_CLEAR_UNLOCK_PAGE | | |
965 | EXTENT_CLEAR_UNLOCK | | |
966 | EXTENT_CLEAR_DELALLOC | | |
967 | EXTENT_CLEAR_DIRTY | | |
968 | EXTENT_SET_WRITEBACK | | |
969 | EXTENT_END_WRITEBACK); | |
970 | ||
971 | goto out; | |
771ed689 | 972 | } |
c8b97818 | 973 | |
771ed689 CM |
974 | /* |
975 | * work queue call back to started compression on a file and pages | |
976 | */ | |
977 | static noinline void async_cow_start(struct btrfs_work *work) | |
978 | { | |
979 | struct async_cow *async_cow; | |
980 | int num_added = 0; | |
981 | async_cow = container_of(work, struct async_cow, work); | |
982 | ||
983 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
984 | async_cow->start, async_cow->end, async_cow, | |
985 | &num_added); | |
986 | if (num_added == 0) | |
987 | async_cow->inode = NULL; | |
988 | } | |
989 | ||
990 | /* | |
991 | * work queue call back to submit previously compressed pages | |
992 | */ | |
993 | static noinline void async_cow_submit(struct btrfs_work *work) | |
994 | { | |
995 | struct async_cow *async_cow; | |
996 | struct btrfs_root *root; | |
997 | unsigned long nr_pages; | |
998 | ||
999 | async_cow = container_of(work, struct async_cow, work); | |
1000 | ||
1001 | root = async_cow->root; | |
1002 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1003 | PAGE_CACHE_SHIFT; | |
1004 | ||
1005 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
1006 | ||
1007 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1008 | 5 * 1042 * 1024 && | |
1009 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
1010 | wake_up(&root->fs_info->async_submit_wait); | |
1011 | ||
d397712b | 1012 | if (async_cow->inode) |
771ed689 | 1013 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1014 | } |
c8b97818 | 1015 | |
771ed689 CM |
1016 | static noinline void async_cow_free(struct btrfs_work *work) |
1017 | { | |
1018 | struct async_cow *async_cow; | |
1019 | async_cow = container_of(work, struct async_cow, work); | |
1020 | kfree(async_cow); | |
1021 | } | |
1022 | ||
1023 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1024 | u64 start, u64 end, int *page_started, | |
1025 | unsigned long *nr_written) | |
1026 | { | |
1027 | struct async_cow *async_cow; | |
1028 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1029 | unsigned long nr_pages; | |
1030 | u64 cur_end; | |
1031 | int limit = 10 * 1024 * 1042; | |
1032 | ||
a3429ab7 CM |
1033 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1034 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1035 | while (start < end) { |
771ed689 | 1036 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1037 | BUG_ON(!async_cow); /* -ENOMEM */ |
771ed689 CM |
1038 | async_cow->inode = inode; |
1039 | async_cow->root = root; | |
1040 | async_cow->locked_page = locked_page; | |
1041 | async_cow->start = start; | |
1042 | ||
6cbff00f | 1043 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1044 | cur_end = end; |
1045 | else | |
1046 | cur_end = min(end, start + 512 * 1024 - 1); | |
1047 | ||
1048 | async_cow->end = cur_end; | |
1049 | INIT_LIST_HEAD(&async_cow->extents); | |
1050 | ||
1051 | async_cow->work.func = async_cow_start; | |
1052 | async_cow->work.ordered_func = async_cow_submit; | |
1053 | async_cow->work.ordered_free = async_cow_free; | |
1054 | async_cow->work.flags = 0; | |
1055 | ||
771ed689 CM |
1056 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1057 | PAGE_CACHE_SHIFT; | |
1058 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1059 | ||
1060 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1061 | &async_cow->work); | |
1062 | ||
1063 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1064 | wait_event(root->fs_info->async_submit_wait, | |
1065 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1066 | limit)); | |
1067 | } | |
1068 | ||
d397712b | 1069 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1070 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1071 | wait_event(root->fs_info->async_submit_wait, | |
1072 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1073 | 0)); | |
1074 | } | |
1075 | ||
1076 | *nr_written += nr_pages; | |
1077 | start = cur_end + 1; | |
1078 | } | |
1079 | *page_started = 1; | |
1080 | return 0; | |
be20aa9d CM |
1081 | } |
1082 | ||
d397712b | 1083 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1084 | u64 bytenr, u64 num_bytes) |
1085 | { | |
1086 | int ret; | |
1087 | struct btrfs_ordered_sum *sums; | |
1088 | LIST_HEAD(list); | |
1089 | ||
07d400a6 | 1090 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1091 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1092 | if (ret == 0 && list_empty(&list)) |
1093 | return 0; | |
1094 | ||
1095 | while (!list_empty(&list)) { | |
1096 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1097 | list_del(&sums->list); | |
1098 | kfree(sums); | |
1099 | } | |
1100 | return 1; | |
1101 | } | |
1102 | ||
d352ac68 CM |
1103 | /* |
1104 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1105 | * of the extents that exist in the file, and COWs the file as required. | |
1106 | * | |
1107 | * If no cow copies or snapshots exist, we write directly to the existing | |
1108 | * blocks on disk | |
1109 | */ | |
7f366cfe CM |
1110 | static noinline int run_delalloc_nocow(struct inode *inode, |
1111 | struct page *locked_page, | |
771ed689 CM |
1112 | u64 start, u64 end, int *page_started, int force, |
1113 | unsigned long *nr_written) | |
be20aa9d | 1114 | { |
be20aa9d | 1115 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1116 | struct btrfs_trans_handle *trans; |
be20aa9d | 1117 | struct extent_buffer *leaf; |
be20aa9d | 1118 | struct btrfs_path *path; |
80ff3856 | 1119 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1120 | struct btrfs_key found_key; |
80ff3856 YZ |
1121 | u64 cow_start; |
1122 | u64 cur_offset; | |
1123 | u64 extent_end; | |
5d4f98a2 | 1124 | u64 extent_offset; |
80ff3856 YZ |
1125 | u64 disk_bytenr; |
1126 | u64 num_bytes; | |
1127 | int extent_type; | |
79787eaa | 1128 | int ret, err; |
d899e052 | 1129 | int type; |
80ff3856 YZ |
1130 | int nocow; |
1131 | int check_prev = 1; | |
82d5902d | 1132 | bool nolock; |
33345d01 | 1133 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1134 | |
1135 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1136 | if (!path) |
1137 | return -ENOMEM; | |
82d5902d | 1138 | |
2cf8572d | 1139 | nolock = btrfs_is_free_space_inode(root, inode); |
82d5902d LZ |
1140 | |
1141 | if (nolock) | |
7a7eaa40 | 1142 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1143 | else |
7a7eaa40 | 1144 | trans = btrfs_join_transaction(root); |
ff5714cc | 1145 | |
79787eaa JM |
1146 | if (IS_ERR(trans)) { |
1147 | btrfs_free_path(path); | |
1148 | return PTR_ERR(trans); | |
1149 | } | |
1150 | ||
74b21075 | 1151 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1152 | |
80ff3856 YZ |
1153 | cow_start = (u64)-1; |
1154 | cur_offset = start; | |
1155 | while (1) { | |
33345d01 | 1156 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1157 | cur_offset, 0); |
79787eaa JM |
1158 | if (ret < 0) { |
1159 | btrfs_abort_transaction(trans, root, ret); | |
1160 | goto error; | |
1161 | } | |
80ff3856 YZ |
1162 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1163 | leaf = path->nodes[0]; | |
1164 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1165 | path->slots[0] - 1); | |
33345d01 | 1166 | if (found_key.objectid == ino && |
80ff3856 YZ |
1167 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1168 | path->slots[0]--; | |
1169 | } | |
1170 | check_prev = 0; | |
1171 | next_slot: | |
1172 | leaf = path->nodes[0]; | |
1173 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1174 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1175 | if (ret < 0) { |
1176 | btrfs_abort_transaction(trans, root, ret); | |
1177 | goto error; | |
1178 | } | |
80ff3856 YZ |
1179 | if (ret > 0) |
1180 | break; | |
1181 | leaf = path->nodes[0]; | |
1182 | } | |
be20aa9d | 1183 | |
80ff3856 YZ |
1184 | nocow = 0; |
1185 | disk_bytenr = 0; | |
17d217fe | 1186 | num_bytes = 0; |
80ff3856 YZ |
1187 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1188 | ||
33345d01 | 1189 | if (found_key.objectid > ino || |
80ff3856 YZ |
1190 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1191 | found_key.offset > end) | |
1192 | break; | |
1193 | ||
1194 | if (found_key.offset > cur_offset) { | |
1195 | extent_end = found_key.offset; | |
e9061e21 | 1196 | extent_type = 0; |
80ff3856 YZ |
1197 | goto out_check; |
1198 | } | |
1199 | ||
1200 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1201 | struct btrfs_file_extent_item); | |
1202 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1203 | ||
d899e052 YZ |
1204 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1205 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1206 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1207 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1208 | extent_end = found_key.offset + |
1209 | btrfs_file_extent_num_bytes(leaf, fi); | |
1210 | if (extent_end <= start) { | |
1211 | path->slots[0]++; | |
1212 | goto next_slot; | |
1213 | } | |
17d217fe YZ |
1214 | if (disk_bytenr == 0) |
1215 | goto out_check; | |
80ff3856 YZ |
1216 | if (btrfs_file_extent_compression(leaf, fi) || |
1217 | btrfs_file_extent_encryption(leaf, fi) || | |
1218 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1219 | goto out_check; | |
d899e052 YZ |
1220 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1221 | goto out_check; | |
d2fb3437 | 1222 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1223 | goto out_check; |
33345d01 | 1224 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1225 | found_key.offset - |
1226 | extent_offset, disk_bytenr)) | |
17d217fe | 1227 | goto out_check; |
5d4f98a2 | 1228 | disk_bytenr += extent_offset; |
17d217fe YZ |
1229 | disk_bytenr += cur_offset - found_key.offset; |
1230 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1231 | /* | |
1232 | * force cow if csum exists in the range. | |
1233 | * this ensure that csum for a given extent are | |
1234 | * either valid or do not exist. | |
1235 | */ | |
1236 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1237 | goto out_check; | |
80ff3856 YZ |
1238 | nocow = 1; |
1239 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1240 | extent_end = found_key.offset + | |
1241 | btrfs_file_extent_inline_len(leaf, fi); | |
1242 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1243 | } else { | |
1244 | BUG_ON(1); | |
1245 | } | |
1246 | out_check: | |
1247 | if (extent_end <= start) { | |
1248 | path->slots[0]++; | |
1249 | goto next_slot; | |
1250 | } | |
1251 | if (!nocow) { | |
1252 | if (cow_start == (u64)-1) | |
1253 | cow_start = cur_offset; | |
1254 | cur_offset = extent_end; | |
1255 | if (cur_offset > end) | |
1256 | break; | |
1257 | path->slots[0]++; | |
1258 | goto next_slot; | |
7ea394f1 YZ |
1259 | } |
1260 | ||
b3b4aa74 | 1261 | btrfs_release_path(path); |
80ff3856 YZ |
1262 | if (cow_start != (u64)-1) { |
1263 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1264 | found_key.offset - 1, page_started, |
1265 | nr_written, 1); | |
79787eaa JM |
1266 | if (ret) { |
1267 | btrfs_abort_transaction(trans, root, ret); | |
1268 | goto error; | |
1269 | } | |
80ff3856 | 1270 | cow_start = (u64)-1; |
7ea394f1 | 1271 | } |
80ff3856 | 1272 | |
d899e052 YZ |
1273 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1274 | struct extent_map *em; | |
1275 | struct extent_map_tree *em_tree; | |
1276 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1277 | em = alloc_extent_map(); |
79787eaa | 1278 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1279 | em->start = cur_offset; |
445a6944 | 1280 | em->orig_start = em->start; |
d899e052 YZ |
1281 | em->len = num_bytes; |
1282 | em->block_len = num_bytes; | |
1283 | em->block_start = disk_bytenr; | |
1284 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1285 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1286 | while (1) { | |
890871be | 1287 | write_lock(&em_tree->lock); |
d899e052 | 1288 | ret = add_extent_mapping(em_tree, em); |
890871be | 1289 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1290 | if (ret != -EEXIST) { |
1291 | free_extent_map(em); | |
1292 | break; | |
1293 | } | |
1294 | btrfs_drop_extent_cache(inode, em->start, | |
1295 | em->start + em->len - 1, 0); | |
1296 | } | |
1297 | type = BTRFS_ORDERED_PREALLOC; | |
1298 | } else { | |
1299 | type = BTRFS_ORDERED_NOCOW; | |
1300 | } | |
80ff3856 YZ |
1301 | |
1302 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1303 | num_bytes, num_bytes, type); |
79787eaa | 1304 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1305 | |
efa56464 YZ |
1306 | if (root->root_key.objectid == |
1307 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1308 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1309 | num_bytes); | |
79787eaa JM |
1310 | if (ret) { |
1311 | btrfs_abort_transaction(trans, root, ret); | |
1312 | goto error; | |
1313 | } | |
efa56464 YZ |
1314 | } |
1315 | ||
d899e052 | 1316 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1317 | cur_offset, cur_offset + num_bytes - 1, |
1318 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1319 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1320 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1321 | cur_offset = extent_end; |
1322 | if (cur_offset > end) | |
1323 | break; | |
be20aa9d | 1324 | } |
b3b4aa74 | 1325 | btrfs_release_path(path); |
80ff3856 YZ |
1326 | |
1327 | if (cur_offset <= end && cow_start == (u64)-1) | |
1328 | cow_start = cur_offset; | |
1329 | if (cow_start != (u64)-1) { | |
1330 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1331 | page_started, nr_written, 1); |
79787eaa JM |
1332 | if (ret) { |
1333 | btrfs_abort_transaction(trans, root, ret); | |
1334 | goto error; | |
1335 | } | |
80ff3856 YZ |
1336 | } |
1337 | ||
79787eaa | 1338 | error: |
0cb59c99 | 1339 | if (nolock) { |
79787eaa | 1340 | err = btrfs_end_transaction_nolock(trans, root); |
0cb59c99 | 1341 | } else { |
79787eaa | 1342 | err = btrfs_end_transaction(trans, root); |
0cb59c99 | 1343 | } |
79787eaa JM |
1344 | if (!ret) |
1345 | ret = err; | |
1346 | ||
7ea394f1 | 1347 | btrfs_free_path(path); |
79787eaa | 1348 | return ret; |
be20aa9d CM |
1349 | } |
1350 | ||
d352ac68 CM |
1351 | /* |
1352 | * extent_io.c call back to do delayed allocation processing | |
1353 | */ | |
c8b97818 | 1354 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1355 | u64 start, u64 end, int *page_started, |
1356 | unsigned long *nr_written) | |
be20aa9d | 1357 | { |
be20aa9d | 1358 | int ret; |
7f366cfe | 1359 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1360 | |
6cbff00f | 1361 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1362 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1363 | page_started, 1, nr_written); |
6cbff00f | 1364 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1365 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1366 | page_started, 0, nr_written); |
1e701a32 | 1367 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1368 | !(BTRFS_I(inode)->force_compress) && |
1369 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1370 | ret = cow_file_range(inode, locked_page, start, end, |
1371 | page_started, nr_written, 1); | |
be20aa9d | 1372 | else |
771ed689 | 1373 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1374 | page_started, nr_written); |
b888db2b CM |
1375 | return ret; |
1376 | } | |
1377 | ||
1bf85046 JM |
1378 | static void btrfs_split_extent_hook(struct inode *inode, |
1379 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1380 | { |
0ca1f7ce | 1381 | /* not delalloc, ignore it */ |
9ed74f2d | 1382 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1383 | return; |
9ed74f2d | 1384 | |
9e0baf60 JB |
1385 | spin_lock(&BTRFS_I(inode)->lock); |
1386 | BTRFS_I(inode)->outstanding_extents++; | |
1387 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1388 | } |
1389 | ||
1390 | /* | |
1391 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1392 | * extents so we can keep track of new extents that are just merged onto old | |
1393 | * extents, such as when we are doing sequential writes, so we can properly | |
1394 | * account for the metadata space we'll need. | |
1395 | */ | |
1bf85046 JM |
1396 | static void btrfs_merge_extent_hook(struct inode *inode, |
1397 | struct extent_state *new, | |
1398 | struct extent_state *other) | |
9ed74f2d | 1399 | { |
9ed74f2d JB |
1400 | /* not delalloc, ignore it */ |
1401 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1402 | return; |
9ed74f2d | 1403 | |
9e0baf60 JB |
1404 | spin_lock(&BTRFS_I(inode)->lock); |
1405 | BTRFS_I(inode)->outstanding_extents--; | |
1406 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1407 | } |
1408 | ||
d352ac68 CM |
1409 | /* |
1410 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1411 | * bytes in this file, and to maintain the list of inodes that | |
1412 | * have pending delalloc work to be done. | |
1413 | */ | |
1bf85046 JM |
1414 | static void btrfs_set_bit_hook(struct inode *inode, |
1415 | struct extent_state *state, int *bits) | |
291d673e | 1416 | { |
9ed74f2d | 1417 | |
75eff68e CM |
1418 | /* |
1419 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1420 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1421 | * bit, which is only set or cleared with irqs on |
1422 | */ | |
0ca1f7ce | 1423 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1424 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1425 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1426 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
9ed74f2d | 1427 | |
9e0baf60 | 1428 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1429 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1430 | } else { |
1431 | spin_lock(&BTRFS_I(inode)->lock); | |
1432 | BTRFS_I(inode)->outstanding_extents++; | |
1433 | spin_unlock(&BTRFS_I(inode)->lock); | |
1434 | } | |
287a0ab9 | 1435 | |
75eff68e | 1436 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1437 | BTRFS_I(inode)->delalloc_bytes += len; |
1438 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1439 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1440 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1441 | &root->fs_info->delalloc_inodes); | |
1442 | } | |
75eff68e | 1443 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1444 | } |
291d673e CM |
1445 | } |
1446 | ||
d352ac68 CM |
1447 | /* |
1448 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1449 | */ | |
1bf85046 JM |
1450 | static void btrfs_clear_bit_hook(struct inode *inode, |
1451 | struct extent_state *state, int *bits) | |
291d673e | 1452 | { |
75eff68e CM |
1453 | /* |
1454 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1455 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1456 | * bit, which is only set or cleared with irqs on |
1457 | */ | |
0ca1f7ce | 1458 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1459 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1460 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1461 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
bcbfce8a | 1462 | |
9e0baf60 | 1463 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1464 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1465 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1466 | spin_lock(&BTRFS_I(inode)->lock); | |
1467 | BTRFS_I(inode)->outstanding_extents--; | |
1468 | spin_unlock(&BTRFS_I(inode)->lock); | |
1469 | } | |
0ca1f7ce YZ |
1470 | |
1471 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1472 | btrfs_delalloc_release_metadata(inode, len); | |
1473 | ||
0cb59c99 JB |
1474 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1475 | && do_list) | |
0ca1f7ce | 1476 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1477 | |
75eff68e | 1478 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1479 | root->fs_info->delalloc_bytes -= len; |
1480 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1481 | ||
0cb59c99 | 1482 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1483 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1484 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1485 | } | |
75eff68e | 1486 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1487 | } |
291d673e CM |
1488 | } |
1489 | ||
d352ac68 CM |
1490 | /* |
1491 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1492 | * we don't create bios that span stripes or chunks | |
1493 | */ | |
239b14b3 | 1494 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1495 | size_t size, struct bio *bio, |
1496 | unsigned long bio_flags) | |
239b14b3 CM |
1497 | { |
1498 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1499 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1500 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1501 | u64 length = 0; |
1502 | u64 map_length; | |
239b14b3 CM |
1503 | int ret; |
1504 | ||
771ed689 CM |
1505 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1506 | return 0; | |
1507 | ||
f2d8d74d | 1508 | length = bio->bi_size; |
239b14b3 CM |
1509 | map_tree = &root->fs_info->mapping_tree; |
1510 | map_length = length; | |
cea9e445 | 1511 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1512 | &map_length, NULL, 0); |
3444a972 JM |
1513 | /* Will always return 0 or 1 with map_multi == NULL */ |
1514 | BUG_ON(ret < 0); | |
d397712b | 1515 | if (map_length < length + size) |
239b14b3 | 1516 | return 1; |
3444a972 | 1517 | return 0; |
239b14b3 CM |
1518 | } |
1519 | ||
d352ac68 CM |
1520 | /* |
1521 | * in order to insert checksums into the metadata in large chunks, | |
1522 | * we wait until bio submission time. All the pages in the bio are | |
1523 | * checksummed and sums are attached onto the ordered extent record. | |
1524 | * | |
1525 | * At IO completion time the cums attached on the ordered extent record | |
1526 | * are inserted into the btree | |
1527 | */ | |
d397712b CM |
1528 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1529 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1530 | unsigned long bio_flags, |
1531 | u64 bio_offset) | |
065631f6 | 1532 | { |
065631f6 | 1533 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1534 | int ret = 0; |
e015640f | 1535 | |
d20f7043 | 1536 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1537 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1538 | return 0; |
1539 | } | |
e015640f | 1540 | |
4a69a410 CM |
1541 | /* |
1542 | * in order to insert checksums into the metadata in large chunks, | |
1543 | * we wait until bio submission time. All the pages in the bio are | |
1544 | * checksummed and sums are attached onto the ordered extent record. | |
1545 | * | |
1546 | * At IO completion time the cums attached on the ordered extent record | |
1547 | * are inserted into the btree | |
1548 | */ | |
b2950863 | 1549 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1550 | int mirror_num, unsigned long bio_flags, |
1551 | u64 bio_offset) | |
4a69a410 CM |
1552 | { |
1553 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1554 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1555 | } |
1556 | ||
d352ac68 | 1557 | /* |
cad321ad CM |
1558 | * extent_io.c submission hook. This does the right thing for csum calculation |
1559 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1560 | */ |
b2950863 | 1561 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1562 | int mirror_num, unsigned long bio_flags, |
1563 | u64 bio_offset) | |
44b8bd7e CM |
1564 | { |
1565 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1566 | int ret = 0; | |
19b9bdb0 | 1567 | int skip_sum; |
0417341e | 1568 | int metadata = 0; |
44b8bd7e | 1569 | |
6cbff00f | 1570 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1571 | |
2cf8572d | 1572 | if (btrfs_is_free_space_inode(root, inode)) |
0417341e JM |
1573 | metadata = 2; |
1574 | ||
7b6d91da | 1575 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1576 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1577 | if (ret) | |
1578 | return ret; | |
1579 | ||
d20f7043 | 1580 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1581 | return btrfs_submit_compressed_read(inode, bio, |
1582 | mirror_num, bio_flags); | |
c2db1073 TI |
1583 | } else if (!skip_sum) { |
1584 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1585 | if (ret) | |
1586 | return ret; | |
1587 | } | |
4d1b5fb4 | 1588 | goto mapit; |
19b9bdb0 | 1589 | } else if (!skip_sum) { |
17d217fe YZ |
1590 | /* csum items have already been cloned */ |
1591 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1592 | goto mapit; | |
19b9bdb0 CM |
1593 | /* we're doing a write, do the async checksumming */ |
1594 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1595 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1596 | bio_flags, bio_offset, |
1597 | __btrfs_submit_bio_start, | |
4a69a410 | 1598 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1599 | } |
1600 | ||
0b86a832 | 1601 | mapit: |
8b712842 | 1602 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1603 | } |
6885f308 | 1604 | |
d352ac68 CM |
1605 | /* |
1606 | * given a list of ordered sums record them in the inode. This happens | |
1607 | * at IO completion time based on sums calculated at bio submission time. | |
1608 | */ | |
ba1da2f4 | 1609 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1610 | struct inode *inode, u64 file_offset, |
1611 | struct list_head *list) | |
1612 | { | |
e6dcd2dc CM |
1613 | struct btrfs_ordered_sum *sum; |
1614 | ||
c6e30871 | 1615 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1616 | btrfs_csum_file_blocks(trans, |
1617 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1618 | } |
1619 | return 0; | |
1620 | } | |
1621 | ||
2ac55d41 JB |
1622 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1623 | struct extent_state **cached_state) | |
ea8c2819 | 1624 | { |
d397712b | 1625 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1626 | WARN_ON(1); |
ea8c2819 | 1627 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1628 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1629 | } |
1630 | ||
d352ac68 | 1631 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1632 | struct btrfs_writepage_fixup { |
1633 | struct page *page; | |
1634 | struct btrfs_work work; | |
1635 | }; | |
1636 | ||
b2950863 | 1637 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1638 | { |
1639 | struct btrfs_writepage_fixup *fixup; | |
1640 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1641 | struct extent_state *cached_state = NULL; |
247e743c CM |
1642 | struct page *page; |
1643 | struct inode *inode; | |
1644 | u64 page_start; | |
1645 | u64 page_end; | |
87826df0 | 1646 | int ret; |
247e743c CM |
1647 | |
1648 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1649 | page = fixup->page; | |
4a096752 | 1650 | again: |
247e743c CM |
1651 | lock_page(page); |
1652 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1653 | ClearPageChecked(page); | |
1654 | goto out_page; | |
1655 | } | |
1656 | ||
1657 | inode = page->mapping->host; | |
1658 | page_start = page_offset(page); | |
1659 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1660 | ||
2ac55d41 | 1661 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1662 | &cached_state); |
4a096752 CM |
1663 | |
1664 | /* already ordered? We're done */ | |
8b62b72b | 1665 | if (PagePrivate2(page)) |
247e743c | 1666 | goto out; |
4a096752 CM |
1667 | |
1668 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1669 | if (ordered) { | |
2ac55d41 JB |
1670 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1671 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1672 | unlock_page(page); |
1673 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1674 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1675 | goto again; |
1676 | } | |
247e743c | 1677 | |
87826df0 JM |
1678 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1679 | if (ret) { | |
1680 | mapping_set_error(page->mapping, ret); | |
1681 | end_extent_writepage(page, ret, page_start, page_end); | |
1682 | ClearPageChecked(page); | |
1683 | goto out; | |
1684 | } | |
1685 | ||
2ac55d41 | 1686 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1687 | ClearPageChecked(page); |
87826df0 | 1688 | set_page_dirty(page); |
247e743c | 1689 | out: |
2ac55d41 JB |
1690 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1691 | &cached_state, GFP_NOFS); | |
247e743c CM |
1692 | out_page: |
1693 | unlock_page(page); | |
1694 | page_cache_release(page); | |
b897abec | 1695 | kfree(fixup); |
247e743c CM |
1696 | } |
1697 | ||
1698 | /* | |
1699 | * There are a few paths in the higher layers of the kernel that directly | |
1700 | * set the page dirty bit without asking the filesystem if it is a | |
1701 | * good idea. This causes problems because we want to make sure COW | |
1702 | * properly happens and the data=ordered rules are followed. | |
1703 | * | |
c8b97818 | 1704 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1705 | * hasn't been properly setup for IO. We kick off an async process |
1706 | * to fix it up. The async helper will wait for ordered extents, set | |
1707 | * the delalloc bit and make it safe to write the page. | |
1708 | */ | |
b2950863 | 1709 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1710 | { |
1711 | struct inode *inode = page->mapping->host; | |
1712 | struct btrfs_writepage_fixup *fixup; | |
1713 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1714 | |
8b62b72b CM |
1715 | /* this page is properly in the ordered list */ |
1716 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1717 | return 0; |
1718 | ||
1719 | if (PageChecked(page)) | |
1720 | return -EAGAIN; | |
1721 | ||
1722 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1723 | if (!fixup) | |
1724 | return -EAGAIN; | |
f421950f | 1725 | |
247e743c CM |
1726 | SetPageChecked(page); |
1727 | page_cache_get(page); | |
1728 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1729 | fixup->page = page; | |
1730 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1731 | return -EBUSY; |
247e743c CM |
1732 | } |
1733 | ||
d899e052 YZ |
1734 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1735 | struct inode *inode, u64 file_pos, | |
1736 | u64 disk_bytenr, u64 disk_num_bytes, | |
1737 | u64 num_bytes, u64 ram_bytes, | |
1738 | u8 compression, u8 encryption, | |
1739 | u16 other_encoding, int extent_type) | |
1740 | { | |
1741 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1742 | struct btrfs_file_extent_item *fi; | |
1743 | struct btrfs_path *path; | |
1744 | struct extent_buffer *leaf; | |
1745 | struct btrfs_key ins; | |
1746 | u64 hint; | |
1747 | int ret; | |
1748 | ||
1749 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1750 | if (!path) |
1751 | return -ENOMEM; | |
d899e052 | 1752 | |
b9473439 | 1753 | path->leave_spinning = 1; |
a1ed835e CM |
1754 | |
1755 | /* | |
1756 | * we may be replacing one extent in the tree with another. | |
1757 | * The new extent is pinned in the extent map, and we don't want | |
1758 | * to drop it from the cache until it is completely in the btree. | |
1759 | * | |
1760 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1761 | * the caller is expected to unpin it and allow it to be merged | |
1762 | * with the others. | |
1763 | */ | |
920bbbfb YZ |
1764 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1765 | &hint, 0); | |
79787eaa JM |
1766 | if (ret) |
1767 | goto out; | |
d899e052 | 1768 | |
33345d01 | 1769 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1770 | ins.offset = file_pos; |
1771 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1772 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1773 | if (ret) |
1774 | goto out; | |
d899e052 YZ |
1775 | leaf = path->nodes[0]; |
1776 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1777 | struct btrfs_file_extent_item); | |
1778 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1779 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1780 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1781 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1782 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1783 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1784 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1785 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1786 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1787 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1788 | |
1789 | btrfs_unlock_up_safe(path, 1); | |
1790 | btrfs_set_lock_blocking(leaf); | |
1791 | ||
d899e052 YZ |
1792 | btrfs_mark_buffer_dirty(leaf); |
1793 | ||
1794 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1795 | |
1796 | ins.objectid = disk_bytenr; | |
1797 | ins.offset = disk_num_bytes; | |
1798 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1799 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1800 | root->root_key.objectid, | |
33345d01 | 1801 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1802 | out: |
d899e052 | 1803 | btrfs_free_path(path); |
b9473439 | 1804 | |
79787eaa | 1805 | return ret; |
d899e052 YZ |
1806 | } |
1807 | ||
5d13a98f CM |
1808 | /* |
1809 | * helper function for btrfs_finish_ordered_io, this | |
1810 | * just reads in some of the csum leaves to prime them into ram | |
1811 | * before we start the transaction. It limits the amount of btree | |
1812 | * reads required while inside the transaction. | |
1813 | */ | |
d352ac68 CM |
1814 | /* as ordered data IO finishes, this gets called so we can finish |
1815 | * an ordered extent if the range of bytes in the file it covers are | |
1816 | * fully written. | |
1817 | */ | |
5fd02043 | 1818 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1819 | { |
5fd02043 | 1820 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1821 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1822 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1823 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1824 | struct extent_state *cached_state = NULL; |
261507a0 | 1825 | int compress_type = 0; |
e6dcd2dc | 1826 | int ret; |
82d5902d | 1827 | bool nolock; |
e6dcd2dc | 1828 | |
2cf8572d | 1829 | nolock = btrfs_is_free_space_inode(root, inode); |
0cb59c99 | 1830 | |
5fd02043 JB |
1831 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1832 | ret = -EIO; | |
1833 | goto out; | |
1834 | } | |
1835 | ||
c2167754 | 1836 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1837 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
c2167754 YZ |
1838 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1839 | if (!ret) { | |
0cb59c99 | 1840 | if (nolock) |
7a7eaa40 | 1841 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1842 | else |
7a7eaa40 | 1843 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1844 | if (IS_ERR(trans)) |
1845 | return PTR_ERR(trans); | |
0ca1f7ce | 1846 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1847 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1848 | if (ret) /* -ENOMEM or corruption */ |
1849 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1850 | } |
1851 | goto out; | |
1852 | } | |
e6dcd2dc | 1853 | |
2ac55d41 JB |
1854 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1855 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1856 | 0, &cached_state); |
e6dcd2dc | 1857 | |
0cb59c99 | 1858 | if (nolock) |
7a7eaa40 | 1859 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1860 | else |
7a7eaa40 | 1861 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1862 | if (IS_ERR(trans)) { |
1863 | ret = PTR_ERR(trans); | |
1864 | trans = NULL; | |
1865 | goto out_unlock; | |
1866 | } | |
0ca1f7ce | 1867 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1868 | |
c8b97818 | 1869 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1870 | compress_type = ordered_extent->compress_type; |
d899e052 | 1871 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1872 | BUG_ON(compress_type); |
920bbbfb | 1873 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1874 | ordered_extent->file_offset, |
1875 | ordered_extent->file_offset + | |
1876 | ordered_extent->len); | |
d899e052 | 1877 | } else { |
0af3d00b | 1878 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1879 | ret = insert_reserved_file_extent(trans, inode, |
1880 | ordered_extent->file_offset, | |
1881 | ordered_extent->start, | |
1882 | ordered_extent->disk_len, | |
1883 | ordered_extent->len, | |
1884 | ordered_extent->len, | |
261507a0 | 1885 | compress_type, 0, 0, |
d899e052 | 1886 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1887 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1888 | ordered_extent->file_offset, | |
1889 | ordered_extent->len); | |
d899e052 | 1890 | } |
5fd02043 | 1891 | |
79787eaa JM |
1892 | if (ret < 0) { |
1893 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1894 | goto out_unlock; |
79787eaa | 1895 | } |
2ac55d41 | 1896 | |
e6dcd2dc CM |
1897 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1898 | &ordered_extent->list); | |
1899 | ||
1ef30be1 | 1900 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1901 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1902 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1903 | if (ret) { /* -ENOMEM or corruption */ |
1904 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1905 | goto out_unlock; |
79787eaa | 1906 | } |
1ef30be1 JB |
1907 | } |
1908 | ret = 0; | |
5fd02043 JB |
1909 | out_unlock: |
1910 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
1911 | ordered_extent->file_offset + | |
1912 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 1913 | out: |
5b0e95bf | 1914 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1915 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
5b0e95bf JB |
1916 | if (trans) { |
1917 | if (nolock) | |
0cb59c99 | 1918 | btrfs_end_transaction_nolock(trans, root); |
5b0e95bf | 1919 | else |
0cb59c99 JB |
1920 | btrfs_end_transaction(trans, root); |
1921 | } | |
1922 | ||
5fd02043 JB |
1923 | if (ret) |
1924 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, | |
1925 | ordered_extent->file_offset + | |
1926 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
1927 | ||
1928 | /* | |
1929 | * This needs to be dont to make sure anybody waiting knows we are done | |
1930 | * upating everything for this ordered extent. | |
1931 | */ | |
1932 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
1933 | ||
e6dcd2dc CM |
1934 | /* once for us */ |
1935 | btrfs_put_ordered_extent(ordered_extent); | |
1936 | /* once for the tree */ | |
1937 | btrfs_put_ordered_extent(ordered_extent); | |
1938 | ||
5fd02043 JB |
1939 | return ret; |
1940 | } | |
1941 | ||
1942 | static void finish_ordered_fn(struct btrfs_work *work) | |
1943 | { | |
1944 | struct btrfs_ordered_extent *ordered_extent; | |
1945 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
1946 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
1947 | } |
1948 | ||
b2950863 | 1949 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1950 | struct extent_state *state, int uptodate) |
1951 | { | |
5fd02043 JB |
1952 | struct inode *inode = page->mapping->host; |
1953 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1954 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
1955 | struct btrfs_workers *workers; | |
1956 | ||
1abe9b8a | 1957 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1958 | ||
8b62b72b | 1959 | ClearPagePrivate2(page); |
5fd02043 JB |
1960 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1961 | end - start + 1, uptodate)) | |
1962 | return 0; | |
1963 | ||
1964 | ordered_extent->work.func = finish_ordered_fn; | |
1965 | ordered_extent->work.flags = 0; | |
1966 | ||
1967 | if (btrfs_is_free_space_inode(root, inode)) | |
1968 | workers = &root->fs_info->endio_freespace_worker; | |
1969 | else | |
1970 | workers = &root->fs_info->endio_write_workers; | |
1971 | btrfs_queue_worker(workers, &ordered_extent->work); | |
1972 | ||
1973 | return 0; | |
211f90e6 CM |
1974 | } |
1975 | ||
d352ac68 CM |
1976 | /* |
1977 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
1978 | * if there's a match, we allow the bio to finish. If not, the code in |
1979 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 1980 | */ |
b2950863 | 1981 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 1982 | struct extent_state *state, int mirror) |
07157aac | 1983 | { |
35ebb934 | 1984 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1985 | struct inode *inode = page->mapping->host; |
d1310b2e | 1986 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1987 | char *kaddr; |
aadfeb6e | 1988 | u64 private = ~(u32)0; |
07157aac | 1989 | int ret; |
ff79f819 CM |
1990 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1991 | u32 csum = ~(u32)0; | |
d1310b2e | 1992 | |
d20f7043 CM |
1993 | if (PageChecked(page)) { |
1994 | ClearPageChecked(page); | |
1995 | goto good; | |
1996 | } | |
6cbff00f CH |
1997 | |
1998 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 1999 | goto good; |
17d217fe YZ |
2000 | |
2001 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2002 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2003 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2004 | GFP_NOFS); | |
b6cda9bc | 2005 | return 0; |
17d217fe | 2006 | } |
d20f7043 | 2007 | |
c2e639f0 | 2008 | if (state && state->start == start) { |
70dec807 CM |
2009 | private = state->private; |
2010 | ret = 0; | |
2011 | } else { | |
2012 | ret = get_state_private(io_tree, start, &private); | |
2013 | } | |
7ac687d9 | 2014 | kaddr = kmap_atomic(page); |
d397712b | 2015 | if (ret) |
07157aac | 2016 | goto zeroit; |
d397712b | 2017 | |
ff79f819 CM |
2018 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2019 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2020 | if (csum != private) |
07157aac | 2021 | goto zeroit; |
d397712b | 2022 | |
7ac687d9 | 2023 | kunmap_atomic(kaddr); |
d20f7043 | 2024 | good: |
07157aac CM |
2025 | return 0; |
2026 | ||
2027 | zeroit: | |
945d8962 | 2028 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2029 | "private %llu\n", |
2030 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2031 | (unsigned long long)start, csum, |
2032 | (unsigned long long)private); | |
db94535d CM |
2033 | memset(kaddr + offset, 1, end - start + 1); |
2034 | flush_dcache_page(page); | |
7ac687d9 | 2035 | kunmap_atomic(kaddr); |
3b951516 CM |
2036 | if (private == 0) |
2037 | return 0; | |
7e38326f | 2038 | return -EIO; |
07157aac | 2039 | } |
b888db2b | 2040 | |
24bbcf04 YZ |
2041 | struct delayed_iput { |
2042 | struct list_head list; | |
2043 | struct inode *inode; | |
2044 | }; | |
2045 | ||
79787eaa JM |
2046 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2047 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2048 | void btrfs_add_delayed_iput(struct inode *inode) |
2049 | { | |
2050 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2051 | struct delayed_iput *delayed; | |
2052 | ||
2053 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2054 | return; | |
2055 | ||
2056 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2057 | delayed->inode = inode; | |
2058 | ||
2059 | spin_lock(&fs_info->delayed_iput_lock); | |
2060 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2061 | spin_unlock(&fs_info->delayed_iput_lock); | |
2062 | } | |
2063 | ||
2064 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2065 | { | |
2066 | LIST_HEAD(list); | |
2067 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2068 | struct delayed_iput *delayed; | |
2069 | int empty; | |
2070 | ||
2071 | spin_lock(&fs_info->delayed_iput_lock); | |
2072 | empty = list_empty(&fs_info->delayed_iputs); | |
2073 | spin_unlock(&fs_info->delayed_iput_lock); | |
2074 | if (empty) | |
2075 | return; | |
2076 | ||
2077 | down_read(&root->fs_info->cleanup_work_sem); | |
2078 | spin_lock(&fs_info->delayed_iput_lock); | |
2079 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2080 | spin_unlock(&fs_info->delayed_iput_lock); | |
2081 | ||
2082 | while (!list_empty(&list)) { | |
2083 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2084 | list_del(&delayed->list); | |
2085 | iput(delayed->inode); | |
2086 | kfree(delayed); | |
2087 | } | |
2088 | up_read(&root->fs_info->cleanup_work_sem); | |
2089 | } | |
2090 | ||
d68fc57b YZ |
2091 | enum btrfs_orphan_cleanup_state { |
2092 | ORPHAN_CLEANUP_STARTED = 1, | |
2093 | ORPHAN_CLEANUP_DONE = 2, | |
2094 | }; | |
2095 | ||
2096 | /* | |
42b2aa86 | 2097 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2098 | * files in the subvolume, it removes orphan item and frees block_rsv |
2099 | * structure. | |
2100 | */ | |
2101 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2102 | struct btrfs_root *root) | |
2103 | { | |
90290e19 | 2104 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2105 | int ret; |
2106 | ||
2107 | if (!list_empty(&root->orphan_list) || | |
2108 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2109 | return; | |
2110 | ||
90290e19 JB |
2111 | spin_lock(&root->orphan_lock); |
2112 | if (!list_empty(&root->orphan_list)) { | |
2113 | spin_unlock(&root->orphan_lock); | |
2114 | return; | |
2115 | } | |
2116 | ||
2117 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2118 | spin_unlock(&root->orphan_lock); | |
2119 | return; | |
2120 | } | |
2121 | ||
2122 | block_rsv = root->orphan_block_rsv; | |
2123 | root->orphan_block_rsv = NULL; | |
2124 | spin_unlock(&root->orphan_lock); | |
2125 | ||
d68fc57b YZ |
2126 | if (root->orphan_item_inserted && |
2127 | btrfs_root_refs(&root->root_item) > 0) { | |
2128 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2129 | root->root_key.objectid); | |
2130 | BUG_ON(ret); | |
2131 | root->orphan_item_inserted = 0; | |
2132 | } | |
2133 | ||
90290e19 JB |
2134 | if (block_rsv) { |
2135 | WARN_ON(block_rsv->size > 0); | |
2136 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2137 | } |
2138 | } | |
2139 | ||
7b128766 JB |
2140 | /* |
2141 | * This creates an orphan entry for the given inode in case something goes | |
2142 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2143 | * |
2144 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2145 | * this function. | |
7b128766 JB |
2146 | */ |
2147 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2148 | { | |
2149 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2150 | struct btrfs_block_rsv *block_rsv = NULL; |
2151 | int reserve = 0; | |
2152 | int insert = 0; | |
2153 | int ret; | |
7b128766 | 2154 | |
d68fc57b YZ |
2155 | if (!root->orphan_block_rsv) { |
2156 | block_rsv = btrfs_alloc_block_rsv(root); | |
b532402e TI |
2157 | if (!block_rsv) |
2158 | return -ENOMEM; | |
d68fc57b | 2159 | } |
7b128766 | 2160 | |
d68fc57b YZ |
2161 | spin_lock(&root->orphan_lock); |
2162 | if (!root->orphan_block_rsv) { | |
2163 | root->orphan_block_rsv = block_rsv; | |
2164 | } else if (block_rsv) { | |
2165 | btrfs_free_block_rsv(root, block_rsv); | |
2166 | block_rsv = NULL; | |
7b128766 | 2167 | } |
7b128766 | 2168 | |
d68fc57b YZ |
2169 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2170 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2171 | #if 0 | |
2172 | /* | |
2173 | * For proper ENOSPC handling, we should do orphan | |
2174 | * cleanup when mounting. But this introduces backward | |
2175 | * compatibility issue. | |
2176 | */ | |
2177 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2178 | insert = 2; | |
2179 | else | |
2180 | insert = 1; | |
2181 | #endif | |
2182 | insert = 1; | |
7b128766 JB |
2183 | } |
2184 | ||
d68fc57b YZ |
2185 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2186 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2187 | reserve = 1; | |
2188 | } | |
2189 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2190 | |
d68fc57b YZ |
2191 | /* grab metadata reservation from transaction handle */ |
2192 | if (reserve) { | |
2193 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2194 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2195 | } |
7b128766 | 2196 | |
d68fc57b YZ |
2197 | /* insert an orphan item to track this unlinked/truncated file */ |
2198 | if (insert >= 1) { | |
33345d01 | 2199 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa JM |
2200 | if (ret && ret != -EEXIST) { |
2201 | btrfs_abort_transaction(trans, root, ret); | |
2202 | return ret; | |
2203 | } | |
2204 | ret = 0; | |
d68fc57b YZ |
2205 | } |
2206 | ||
2207 | /* insert an orphan item to track subvolume contains orphan files */ | |
2208 | if (insert >= 2) { | |
2209 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2210 | root->root_key.objectid); | |
79787eaa JM |
2211 | if (ret && ret != -EEXIST) { |
2212 | btrfs_abort_transaction(trans, root, ret); | |
2213 | return ret; | |
2214 | } | |
d68fc57b YZ |
2215 | } |
2216 | return 0; | |
7b128766 JB |
2217 | } |
2218 | ||
2219 | /* | |
2220 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2221 | * item for this particular inode. | |
2222 | */ | |
2223 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2224 | { | |
2225 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2226 | int delete_item = 0; |
2227 | int release_rsv = 0; | |
7b128766 JB |
2228 | int ret = 0; |
2229 | ||
d68fc57b YZ |
2230 | spin_lock(&root->orphan_lock); |
2231 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2232 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2233 | delete_item = 1; | |
7b128766 JB |
2234 | } |
2235 | ||
d68fc57b YZ |
2236 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2237 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2238 | release_rsv = 1; | |
7b128766 | 2239 | } |
d68fc57b | 2240 | spin_unlock(&root->orphan_lock); |
7b128766 | 2241 | |
d68fc57b | 2242 | if (trans && delete_item) { |
33345d01 | 2243 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2244 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2245 | } |
7b128766 | 2246 | |
d68fc57b YZ |
2247 | if (release_rsv) |
2248 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2249 | |
d68fc57b | 2250 | return 0; |
7b128766 JB |
2251 | } |
2252 | ||
2253 | /* | |
2254 | * this cleans up any orphans that may be left on the list from the last use | |
2255 | * of this root. | |
2256 | */ | |
66b4ffd1 | 2257 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2258 | { |
2259 | struct btrfs_path *path; | |
2260 | struct extent_buffer *leaf; | |
7b128766 JB |
2261 | struct btrfs_key key, found_key; |
2262 | struct btrfs_trans_handle *trans; | |
2263 | struct inode *inode; | |
8f6d7f4f | 2264 | u64 last_objectid = 0; |
7b128766 JB |
2265 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2266 | ||
d68fc57b | 2267 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2268 | return 0; |
c71bf099 YZ |
2269 | |
2270 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2271 | if (!path) { |
2272 | ret = -ENOMEM; | |
2273 | goto out; | |
2274 | } | |
7b128766 JB |
2275 | path->reada = -1; |
2276 | ||
2277 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2278 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2279 | key.offset = (u64)-1; | |
2280 | ||
7b128766 JB |
2281 | while (1) { |
2282 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2283 | if (ret < 0) |
2284 | goto out; | |
7b128766 JB |
2285 | |
2286 | /* | |
2287 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2288 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2289 | * find the key and see if we have stuff that matches |
2290 | */ | |
2291 | if (ret > 0) { | |
66b4ffd1 | 2292 | ret = 0; |
7b128766 JB |
2293 | if (path->slots[0] == 0) |
2294 | break; | |
2295 | path->slots[0]--; | |
2296 | } | |
2297 | ||
2298 | /* pull out the item */ | |
2299 | leaf = path->nodes[0]; | |
7b128766 JB |
2300 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2301 | ||
2302 | /* make sure the item matches what we want */ | |
2303 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2304 | break; | |
2305 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2306 | break; | |
2307 | ||
2308 | /* release the path since we're done with it */ | |
b3b4aa74 | 2309 | btrfs_release_path(path); |
7b128766 JB |
2310 | |
2311 | /* | |
2312 | * this is where we are basically btrfs_lookup, without the | |
2313 | * crossing root thing. we store the inode number in the | |
2314 | * offset of the orphan item. | |
2315 | */ | |
8f6d7f4f JB |
2316 | |
2317 | if (found_key.offset == last_objectid) { | |
2318 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2319 | "stopping orphan cleanup\n"); | |
2320 | ret = -EINVAL; | |
2321 | goto out; | |
2322 | } | |
2323 | ||
2324 | last_objectid = found_key.offset; | |
2325 | ||
5d4f98a2 YZ |
2326 | found_key.objectid = found_key.offset; |
2327 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2328 | found_key.offset = 0; | |
73f73415 | 2329 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2330 | ret = PTR_RET(inode); |
2331 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2332 | goto out; |
7b128766 | 2333 | |
f8e9e0b0 AJ |
2334 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2335 | struct btrfs_root *dead_root; | |
2336 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2337 | int is_dead_root = 0; | |
2338 | ||
2339 | /* | |
2340 | * this is an orphan in the tree root. Currently these | |
2341 | * could come from 2 sources: | |
2342 | * a) a snapshot deletion in progress | |
2343 | * b) a free space cache inode | |
2344 | * We need to distinguish those two, as the snapshot | |
2345 | * orphan must not get deleted. | |
2346 | * find_dead_roots already ran before us, so if this | |
2347 | * is a snapshot deletion, we should find the root | |
2348 | * in the dead_roots list | |
2349 | */ | |
2350 | spin_lock(&fs_info->trans_lock); | |
2351 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2352 | root_list) { | |
2353 | if (dead_root->root_key.objectid == | |
2354 | found_key.objectid) { | |
2355 | is_dead_root = 1; | |
2356 | break; | |
2357 | } | |
2358 | } | |
2359 | spin_unlock(&fs_info->trans_lock); | |
2360 | if (is_dead_root) { | |
2361 | /* prevent this orphan from being found again */ | |
2362 | key.offset = found_key.objectid - 1; | |
2363 | continue; | |
2364 | } | |
2365 | } | |
7b128766 | 2366 | /* |
a8c9e576 JB |
2367 | * Inode is already gone but the orphan item is still there, |
2368 | * kill the orphan item. | |
7b128766 | 2369 | */ |
a8c9e576 JB |
2370 | if (ret == -ESTALE) { |
2371 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2372 | if (IS_ERR(trans)) { |
2373 | ret = PTR_ERR(trans); | |
2374 | goto out; | |
2375 | } | |
a8c9e576 JB |
2376 | ret = btrfs_del_orphan_item(trans, root, |
2377 | found_key.objectid); | |
79787eaa | 2378 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2379 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2380 | continue; |
2381 | } | |
2382 | ||
a8c9e576 JB |
2383 | /* |
2384 | * add this inode to the orphan list so btrfs_orphan_del does | |
2385 | * the proper thing when we hit it | |
2386 | */ | |
2387 | spin_lock(&root->orphan_lock); | |
2388 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2389 | spin_unlock(&root->orphan_lock); | |
2390 | ||
7b128766 JB |
2391 | /* if we have links, this was a truncate, lets do that */ |
2392 | if (inode->i_nlink) { | |
a41ad394 JB |
2393 | if (!S_ISREG(inode->i_mode)) { |
2394 | WARN_ON(1); | |
2395 | iput(inode); | |
2396 | continue; | |
2397 | } | |
7b128766 | 2398 | nr_truncate++; |
66b4ffd1 | 2399 | ret = btrfs_truncate(inode); |
7b128766 JB |
2400 | } else { |
2401 | nr_unlink++; | |
2402 | } | |
2403 | ||
2404 | /* this will do delete_inode and everything for us */ | |
2405 | iput(inode); | |
66b4ffd1 JB |
2406 | if (ret) |
2407 | goto out; | |
7b128766 | 2408 | } |
3254c876 MX |
2409 | /* release the path since we're done with it */ |
2410 | btrfs_release_path(path); | |
2411 | ||
d68fc57b YZ |
2412 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2413 | ||
2414 | if (root->orphan_block_rsv) | |
2415 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2416 | (u64)-1); | |
2417 | ||
2418 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2419 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2420 | if (!IS_ERR(trans)) |
2421 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2422 | } |
7b128766 JB |
2423 | |
2424 | if (nr_unlink) | |
2425 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2426 | if (nr_truncate) | |
2427 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2428 | |
2429 | out: | |
2430 | if (ret) | |
2431 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2432 | btrfs_free_path(path); | |
2433 | return ret; | |
7b128766 JB |
2434 | } |
2435 | ||
46a53cca CM |
2436 | /* |
2437 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2438 | * don't find any xattrs, we know there can't be any acls. | |
2439 | * | |
2440 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2441 | */ | |
2442 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2443 | int slot, u64 objectid) | |
2444 | { | |
2445 | u32 nritems = btrfs_header_nritems(leaf); | |
2446 | struct btrfs_key found_key; | |
2447 | int scanned = 0; | |
2448 | ||
2449 | slot++; | |
2450 | while (slot < nritems) { | |
2451 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2452 | ||
2453 | /* we found a different objectid, there must not be acls */ | |
2454 | if (found_key.objectid != objectid) | |
2455 | return 0; | |
2456 | ||
2457 | /* we found an xattr, assume we've got an acl */ | |
2458 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2459 | return 1; | |
2460 | ||
2461 | /* | |
2462 | * we found a key greater than an xattr key, there can't | |
2463 | * be any acls later on | |
2464 | */ | |
2465 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2466 | return 0; | |
2467 | ||
2468 | slot++; | |
2469 | scanned++; | |
2470 | ||
2471 | /* | |
2472 | * it goes inode, inode backrefs, xattrs, extents, | |
2473 | * so if there are a ton of hard links to an inode there can | |
2474 | * be a lot of backrefs. Don't waste time searching too hard, | |
2475 | * this is just an optimization | |
2476 | */ | |
2477 | if (scanned >= 8) | |
2478 | break; | |
2479 | } | |
2480 | /* we hit the end of the leaf before we found an xattr or | |
2481 | * something larger than an xattr. We have to assume the inode | |
2482 | * has acls | |
2483 | */ | |
2484 | return 1; | |
2485 | } | |
2486 | ||
d352ac68 CM |
2487 | /* |
2488 | * read an inode from the btree into the in-memory inode | |
2489 | */ | |
5d4f98a2 | 2490 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2491 | { |
2492 | struct btrfs_path *path; | |
5f39d397 | 2493 | struct extent_buffer *leaf; |
39279cc3 | 2494 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2495 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2496 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2497 | struct btrfs_key location; | |
46a53cca | 2498 | int maybe_acls; |
618e21d5 | 2499 | u32 rdev; |
39279cc3 | 2500 | int ret; |
2f7e33d4 MX |
2501 | bool filled = false; |
2502 | ||
2503 | ret = btrfs_fill_inode(inode, &rdev); | |
2504 | if (!ret) | |
2505 | filled = true; | |
39279cc3 CM |
2506 | |
2507 | path = btrfs_alloc_path(); | |
1748f843 MF |
2508 | if (!path) |
2509 | goto make_bad; | |
2510 | ||
d90c7321 | 2511 | path->leave_spinning = 1; |
39279cc3 | 2512 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2513 | |
39279cc3 | 2514 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2515 | if (ret) |
39279cc3 | 2516 | goto make_bad; |
39279cc3 | 2517 | |
5f39d397 | 2518 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2519 | |
2520 | if (filled) | |
2521 | goto cache_acl; | |
2522 | ||
5f39d397 CM |
2523 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2524 | struct btrfs_inode_item); | |
5f39d397 | 2525 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2526 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
5f39d397 CM |
2527 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); |
2528 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2529 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2530 | |
2531 | tspec = btrfs_inode_atime(inode_item); | |
2532 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2533 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2534 | ||
2535 | tspec = btrfs_inode_mtime(inode_item); | |
2536 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2537 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2538 | ||
2539 | tspec = btrfs_inode_ctime(inode_item); | |
2540 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2541 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2542 | ||
a76a3cd4 | 2543 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2544 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
0c4d2d95 | 2545 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2546 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2547 | inode->i_rdev = 0; |
5f39d397 CM |
2548 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2549 | ||
aec7477b | 2550 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2551 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2552 | cache_acl: |
46a53cca CM |
2553 | /* |
2554 | * try to precache a NULL acl entry for files that don't have | |
2555 | * any xattrs or acls | |
2556 | */ | |
33345d01 LZ |
2557 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2558 | btrfs_ino(inode)); | |
72c04902 AV |
2559 | if (!maybe_acls) |
2560 | cache_no_acl(inode); | |
46a53cca | 2561 | |
39279cc3 | 2562 | btrfs_free_path(path); |
39279cc3 | 2563 | |
39279cc3 | 2564 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2565 | case S_IFREG: |
2566 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2567 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2568 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2569 | inode->i_fop = &btrfs_file_operations; |
2570 | inode->i_op = &btrfs_file_inode_operations; | |
2571 | break; | |
2572 | case S_IFDIR: | |
2573 | inode->i_fop = &btrfs_dir_file_operations; | |
2574 | if (root == root->fs_info->tree_root) | |
2575 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2576 | else | |
2577 | inode->i_op = &btrfs_dir_inode_operations; | |
2578 | break; | |
2579 | case S_IFLNK: | |
2580 | inode->i_op = &btrfs_symlink_inode_operations; | |
2581 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2582 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2583 | break; |
618e21d5 | 2584 | default: |
0279b4cd | 2585 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2586 | init_special_inode(inode, inode->i_mode, rdev); |
2587 | break; | |
39279cc3 | 2588 | } |
6cbff00f CH |
2589 | |
2590 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2591 | return; |
2592 | ||
2593 | make_bad: | |
39279cc3 | 2594 | btrfs_free_path(path); |
39279cc3 CM |
2595 | make_bad_inode(inode); |
2596 | } | |
2597 | ||
d352ac68 CM |
2598 | /* |
2599 | * given a leaf and an inode, copy the inode fields into the leaf | |
2600 | */ | |
e02119d5 CM |
2601 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2602 | struct extent_buffer *leaf, | |
5f39d397 | 2603 | struct btrfs_inode_item *item, |
39279cc3 CM |
2604 | struct inode *inode) |
2605 | { | |
5f39d397 CM |
2606 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2607 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2608 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2609 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2610 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2611 | ||
2612 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2613 | inode->i_atime.tv_sec); | |
2614 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2615 | inode->i_atime.tv_nsec); | |
2616 | ||
2617 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2618 | inode->i_mtime.tv_sec); | |
2619 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2620 | inode->i_mtime.tv_nsec); | |
2621 | ||
2622 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2623 | inode->i_ctime.tv_sec); | |
2624 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2625 | inode->i_ctime.tv_nsec); | |
2626 | ||
a76a3cd4 | 2627 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2628 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
0c4d2d95 | 2629 | btrfs_set_inode_sequence(leaf, item, inode->i_version); |
e02119d5 | 2630 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2631 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2632 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2633 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2634 | } |
2635 | ||
d352ac68 CM |
2636 | /* |
2637 | * copy everything in the in-memory inode into the btree. | |
2638 | */ | |
2115133f | 2639 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2640 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2641 | { |
2642 | struct btrfs_inode_item *inode_item; | |
2643 | struct btrfs_path *path; | |
5f39d397 | 2644 | struct extent_buffer *leaf; |
39279cc3 CM |
2645 | int ret; |
2646 | ||
2647 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2648 | if (!path) |
2649 | return -ENOMEM; | |
2650 | ||
b9473439 | 2651 | path->leave_spinning = 1; |
16cdcec7 MX |
2652 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2653 | 1); | |
39279cc3 CM |
2654 | if (ret) { |
2655 | if (ret > 0) | |
2656 | ret = -ENOENT; | |
2657 | goto failed; | |
2658 | } | |
2659 | ||
b4ce94de | 2660 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2661 | leaf = path->nodes[0]; |
2662 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2663 | struct btrfs_inode_item); |
39279cc3 | 2664 | |
e02119d5 | 2665 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2666 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2667 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2668 | ret = 0; |
2669 | failed: | |
39279cc3 CM |
2670 | btrfs_free_path(path); |
2671 | return ret; | |
2672 | } | |
2673 | ||
2115133f CM |
2674 | /* |
2675 | * copy everything in the in-memory inode into the btree. | |
2676 | */ | |
2677 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2678 | struct btrfs_root *root, struct inode *inode) | |
2679 | { | |
2680 | int ret; | |
2681 | ||
2682 | /* | |
2683 | * If the inode is a free space inode, we can deadlock during commit | |
2684 | * if we put it into the delayed code. | |
2685 | * | |
2686 | * The data relocation inode should also be directly updated | |
2687 | * without delay | |
2688 | */ | |
2689 | if (!btrfs_is_free_space_inode(root, inode) | |
2690 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
2691 | ret = btrfs_delayed_update_inode(trans, root, inode); | |
2692 | if (!ret) | |
2693 | btrfs_set_inode_last_trans(trans, inode); | |
2694 | return ret; | |
2695 | } | |
2696 | ||
2697 | return btrfs_update_inode_item(trans, root, inode); | |
2698 | } | |
2699 | ||
2700 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, | |
2701 | struct btrfs_root *root, struct inode *inode) | |
2702 | { | |
2703 | int ret; | |
2704 | ||
2705 | ret = btrfs_update_inode(trans, root, inode); | |
2706 | if (ret == -ENOSPC) | |
2707 | return btrfs_update_inode_item(trans, root, inode); | |
2708 | return ret; | |
2709 | } | |
2710 | ||
d352ac68 CM |
2711 | /* |
2712 | * unlink helper that gets used here in inode.c and in the tree logging | |
2713 | * recovery code. It remove a link in a directory with a given name, and | |
2714 | * also drops the back refs in the inode to the directory | |
2715 | */ | |
92986796 AV |
2716 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2717 | struct btrfs_root *root, | |
2718 | struct inode *dir, struct inode *inode, | |
2719 | const char *name, int name_len) | |
39279cc3 CM |
2720 | { |
2721 | struct btrfs_path *path; | |
39279cc3 | 2722 | int ret = 0; |
5f39d397 | 2723 | struct extent_buffer *leaf; |
39279cc3 | 2724 | struct btrfs_dir_item *di; |
5f39d397 | 2725 | struct btrfs_key key; |
aec7477b | 2726 | u64 index; |
33345d01 LZ |
2727 | u64 ino = btrfs_ino(inode); |
2728 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2729 | |
2730 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2731 | if (!path) { |
2732 | ret = -ENOMEM; | |
554233a6 | 2733 | goto out; |
54aa1f4d CM |
2734 | } |
2735 | ||
b9473439 | 2736 | path->leave_spinning = 1; |
33345d01 | 2737 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2738 | name, name_len, -1); |
2739 | if (IS_ERR(di)) { | |
2740 | ret = PTR_ERR(di); | |
2741 | goto err; | |
2742 | } | |
2743 | if (!di) { | |
2744 | ret = -ENOENT; | |
2745 | goto err; | |
2746 | } | |
5f39d397 CM |
2747 | leaf = path->nodes[0]; |
2748 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2749 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2750 | if (ret) |
2751 | goto err; | |
b3b4aa74 | 2752 | btrfs_release_path(path); |
39279cc3 | 2753 | |
33345d01 LZ |
2754 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2755 | dir_ino, &index); | |
aec7477b | 2756 | if (ret) { |
d397712b | 2757 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2758 | "inode %llu parent %llu\n", name_len, name, |
2759 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2760 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2761 | goto err; |
2762 | } | |
2763 | ||
16cdcec7 | 2764 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2765 | if (ret) { |
2766 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2767 | goto err; |
79787eaa | 2768 | } |
39279cc3 | 2769 | |
e02119d5 | 2770 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2771 | inode, dir_ino); |
79787eaa JM |
2772 | if (ret != 0 && ret != -ENOENT) { |
2773 | btrfs_abort_transaction(trans, root, ret); | |
2774 | goto err; | |
2775 | } | |
e02119d5 CM |
2776 | |
2777 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2778 | dir, index); | |
6418c961 CM |
2779 | if (ret == -ENOENT) |
2780 | ret = 0; | |
39279cc3 CM |
2781 | err: |
2782 | btrfs_free_path(path); | |
e02119d5 CM |
2783 | if (ret) |
2784 | goto out; | |
2785 | ||
2786 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2787 | inode_inc_iversion(inode); |
2788 | inode_inc_iversion(dir); | |
e02119d5 CM |
2789 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
2790 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2791 | out: |
39279cc3 CM |
2792 | return ret; |
2793 | } | |
2794 | ||
92986796 AV |
2795 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2796 | struct btrfs_root *root, | |
2797 | struct inode *dir, struct inode *inode, | |
2798 | const char *name, int name_len) | |
2799 | { | |
2800 | int ret; | |
2801 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2802 | if (!ret) { | |
2803 | btrfs_drop_nlink(inode); | |
2804 | ret = btrfs_update_inode(trans, root, inode); | |
2805 | } | |
2806 | return ret; | |
2807 | } | |
2808 | ||
2809 | ||
a22285a6 YZ |
2810 | /* helper to check if there is any shared block in the path */ |
2811 | static int check_path_shared(struct btrfs_root *root, | |
2812 | struct btrfs_path *path) | |
39279cc3 | 2813 | { |
a22285a6 YZ |
2814 | struct extent_buffer *eb; |
2815 | int level; | |
0e4dcbef | 2816 | u64 refs = 1; |
5df6a9f6 | 2817 | |
a22285a6 | 2818 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2819 | int ret; |
2820 | ||
a22285a6 YZ |
2821 | if (!path->nodes[level]) |
2822 | break; | |
2823 | eb = path->nodes[level]; | |
2824 | if (!btrfs_block_can_be_shared(root, eb)) | |
2825 | continue; | |
2826 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2827 | &refs, NULL); | |
2828 | if (refs > 1) | |
2829 | return 1; | |
5df6a9f6 | 2830 | } |
dedefd72 | 2831 | return 0; |
39279cc3 CM |
2832 | } |
2833 | ||
a22285a6 YZ |
2834 | /* |
2835 | * helper to start transaction for unlink and rmdir. | |
2836 | * | |
2837 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2838 | * so in enospc case, we should make sure they will free space before | |
2839 | * allowing them to use the global metadata reservation. | |
2840 | */ | |
2841 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2842 | struct dentry *dentry) | |
4df27c4d | 2843 | { |
39279cc3 | 2844 | struct btrfs_trans_handle *trans; |
a22285a6 | 2845 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2846 | struct btrfs_path *path; |
a22285a6 | 2847 | struct btrfs_inode_ref *ref; |
4df27c4d | 2848 | struct btrfs_dir_item *di; |
7b128766 | 2849 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2850 | u64 index; |
a22285a6 YZ |
2851 | int check_link = 1; |
2852 | int err = -ENOSPC; | |
4df27c4d | 2853 | int ret; |
33345d01 LZ |
2854 | u64 ino = btrfs_ino(inode); |
2855 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2856 | |
e70bea5f JB |
2857 | /* |
2858 | * 1 for the possible orphan item | |
2859 | * 1 for the dir item | |
2860 | * 1 for the dir index | |
2861 | * 1 for the inode ref | |
2862 | * 1 for the inode ref in the tree log | |
2863 | * 2 for the dir entries in the log | |
2864 | * 1 for the inode | |
2865 | */ | |
2866 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2867 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2868 | return trans; | |
4df27c4d | 2869 | |
33345d01 | 2870 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2871 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2872 | |
a22285a6 YZ |
2873 | /* check if there is someone else holds reference */ |
2874 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2875 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2876 | |
a22285a6 YZ |
2877 | if (atomic_read(&inode->i_count) > 2) |
2878 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2879 | |
a22285a6 YZ |
2880 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2881 | return ERR_PTR(-ENOSPC); | |
2882 | ||
2883 | path = btrfs_alloc_path(); | |
2884 | if (!path) { | |
2885 | root->fs_info->enospc_unlink = 0; | |
2886 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2887 | } |
2888 | ||
3880a1b4 JB |
2889 | /* 1 for the orphan item */ |
2890 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2891 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2892 | btrfs_free_path(path); |
2893 | root->fs_info->enospc_unlink = 0; | |
2894 | return trans; | |
2895 | } | |
4df27c4d | 2896 | |
a22285a6 YZ |
2897 | path->skip_locking = 1; |
2898 | path->search_commit_root = 1; | |
4df27c4d | 2899 | |
a22285a6 YZ |
2900 | ret = btrfs_lookup_inode(trans, root, path, |
2901 | &BTRFS_I(dir)->location, 0); | |
2902 | if (ret < 0) { | |
2903 | err = ret; | |
2904 | goto out; | |
2905 | } | |
2906 | if (ret == 0) { | |
2907 | if (check_path_shared(root, path)) | |
2908 | goto out; | |
2909 | } else { | |
2910 | check_link = 0; | |
5df6a9f6 | 2911 | } |
b3b4aa74 | 2912 | btrfs_release_path(path); |
a22285a6 YZ |
2913 | |
2914 | ret = btrfs_lookup_inode(trans, root, path, | |
2915 | &BTRFS_I(inode)->location, 0); | |
2916 | if (ret < 0) { | |
2917 | err = ret; | |
2918 | goto out; | |
2919 | } | |
2920 | if (ret == 0) { | |
2921 | if (check_path_shared(root, path)) | |
2922 | goto out; | |
2923 | } else { | |
2924 | check_link = 0; | |
2925 | } | |
b3b4aa74 | 2926 | btrfs_release_path(path); |
a22285a6 YZ |
2927 | |
2928 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2929 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2930 | ino, (u64)-1, 0); |
a22285a6 YZ |
2931 | if (ret < 0) { |
2932 | err = ret; | |
2933 | goto out; | |
2934 | } | |
79787eaa | 2935 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
2936 | if (check_path_shared(root, path)) |
2937 | goto out; | |
b3b4aa74 | 2938 | btrfs_release_path(path); |
a22285a6 YZ |
2939 | } |
2940 | ||
2941 | if (!check_link) { | |
2942 | err = 0; | |
2943 | goto out; | |
2944 | } | |
2945 | ||
33345d01 | 2946 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2947 | dentry->d_name.name, dentry->d_name.len, 0); |
2948 | if (IS_ERR(di)) { | |
2949 | err = PTR_ERR(di); | |
2950 | goto out; | |
2951 | } | |
2952 | if (di) { | |
2953 | if (check_path_shared(root, path)) | |
2954 | goto out; | |
2955 | } else { | |
2956 | err = 0; | |
2957 | goto out; | |
2958 | } | |
b3b4aa74 | 2959 | btrfs_release_path(path); |
a22285a6 YZ |
2960 | |
2961 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2962 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 2963 | ino, dir_ino, 0); |
a22285a6 YZ |
2964 | if (IS_ERR(ref)) { |
2965 | err = PTR_ERR(ref); | |
2966 | goto out; | |
2967 | } | |
79787eaa | 2968 | BUG_ON(!ref); /* Logic error */ |
a22285a6 YZ |
2969 | if (check_path_shared(root, path)) |
2970 | goto out; | |
2971 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 2972 | btrfs_release_path(path); |
a22285a6 | 2973 | |
16cdcec7 MX |
2974 | /* |
2975 | * This is a commit root search, if we can lookup inode item and other | |
2976 | * relative items in the commit root, it means the transaction of | |
2977 | * dir/file creation has been committed, and the dir index item that we | |
2978 | * delay to insert has also been inserted into the commit root. So | |
2979 | * we needn't worry about the delayed insertion of the dir index item | |
2980 | * here. | |
2981 | */ | |
33345d01 | 2982 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
2983 | dentry->d_name.name, dentry->d_name.len, 0); |
2984 | if (IS_ERR(di)) { | |
2985 | err = PTR_ERR(di); | |
2986 | goto out; | |
2987 | } | |
2988 | BUG_ON(ret == -ENOENT); | |
2989 | if (check_path_shared(root, path)) | |
2990 | goto out; | |
2991 | ||
2992 | err = 0; | |
2993 | out: | |
2994 | btrfs_free_path(path); | |
3880a1b4 JB |
2995 | /* Migrate the orphan reservation over */ |
2996 | if (!err) | |
2997 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
2998 | &root->fs_info->global_block_rsv, | |
5a77d76c | 2999 | trans->bytes_reserved); |
3880a1b4 | 3000 | |
a22285a6 YZ |
3001 | if (err) { |
3002 | btrfs_end_transaction(trans, root); | |
3003 | root->fs_info->enospc_unlink = 0; | |
3004 | return ERR_PTR(err); | |
3005 | } | |
3006 | ||
3007 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3008 | return trans; | |
3009 | } | |
3010 | ||
3011 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3012 | struct btrfs_root *root) | |
3013 | { | |
3014 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
5a77d76c JB |
3015 | btrfs_block_rsv_release(root, trans->block_rsv, |
3016 | trans->bytes_reserved); | |
3017 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3018 | BUG_ON(!root->fs_info->enospc_unlink); |
3019 | root->fs_info->enospc_unlink = 0; | |
3020 | } | |
7ad85bb7 | 3021 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3022 | } |
3023 | ||
3024 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3025 | { | |
3026 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3027 | struct btrfs_trans_handle *trans; | |
3028 | struct inode *inode = dentry->d_inode; | |
3029 | int ret; | |
3030 | unsigned long nr = 0; | |
3031 | ||
3032 | trans = __unlink_start_trans(dir, dentry); | |
3033 | if (IS_ERR(trans)) | |
3034 | return PTR_ERR(trans); | |
5f39d397 | 3035 | |
12fcfd22 CM |
3036 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3037 | ||
e02119d5 CM |
3038 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3039 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3040 | if (ret) |
3041 | goto out; | |
7b128766 | 3042 | |
a22285a6 | 3043 | if (inode->i_nlink == 0) { |
7b128766 | 3044 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3045 | if (ret) |
3046 | goto out; | |
a22285a6 | 3047 | } |
7b128766 | 3048 | |
b532402e | 3049 | out: |
d3c2fdcf | 3050 | nr = trans->blocks_used; |
a22285a6 | 3051 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3052 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3053 | return ret; |
3054 | } | |
3055 | ||
4df27c4d YZ |
3056 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3057 | struct btrfs_root *root, | |
3058 | struct inode *dir, u64 objectid, | |
3059 | const char *name, int name_len) | |
3060 | { | |
3061 | struct btrfs_path *path; | |
3062 | struct extent_buffer *leaf; | |
3063 | struct btrfs_dir_item *di; | |
3064 | struct btrfs_key key; | |
3065 | u64 index; | |
3066 | int ret; | |
33345d01 | 3067 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3068 | |
3069 | path = btrfs_alloc_path(); | |
3070 | if (!path) | |
3071 | return -ENOMEM; | |
3072 | ||
33345d01 | 3073 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3074 | name, name_len, -1); |
79787eaa JM |
3075 | if (IS_ERR_OR_NULL(di)) { |
3076 | if (!di) | |
3077 | ret = -ENOENT; | |
3078 | else | |
3079 | ret = PTR_ERR(di); | |
3080 | goto out; | |
3081 | } | |
4df27c4d YZ |
3082 | |
3083 | leaf = path->nodes[0]; | |
3084 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3085 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3086 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3087 | if (ret) { |
3088 | btrfs_abort_transaction(trans, root, ret); | |
3089 | goto out; | |
3090 | } | |
b3b4aa74 | 3091 | btrfs_release_path(path); |
4df27c4d YZ |
3092 | |
3093 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3094 | objectid, root->root_key.objectid, | |
33345d01 | 3095 | dir_ino, &index, name, name_len); |
4df27c4d | 3096 | if (ret < 0) { |
79787eaa JM |
3097 | if (ret != -ENOENT) { |
3098 | btrfs_abort_transaction(trans, root, ret); | |
3099 | goto out; | |
3100 | } | |
33345d01 | 3101 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3102 | name, name_len); |
79787eaa JM |
3103 | if (IS_ERR_OR_NULL(di)) { |
3104 | if (!di) | |
3105 | ret = -ENOENT; | |
3106 | else | |
3107 | ret = PTR_ERR(di); | |
3108 | btrfs_abort_transaction(trans, root, ret); | |
3109 | goto out; | |
3110 | } | |
4df27c4d YZ |
3111 | |
3112 | leaf = path->nodes[0]; | |
3113 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3114 | btrfs_release_path(path); |
4df27c4d YZ |
3115 | index = key.offset; |
3116 | } | |
945d8962 | 3117 | btrfs_release_path(path); |
4df27c4d | 3118 | |
16cdcec7 | 3119 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3120 | if (ret) { |
3121 | btrfs_abort_transaction(trans, root, ret); | |
3122 | goto out; | |
3123 | } | |
4df27c4d YZ |
3124 | |
3125 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3126 | inode_inc_iversion(dir); |
4df27c4d YZ |
3127 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
3128 | ret = btrfs_update_inode(trans, root, dir); | |
79787eaa JM |
3129 | if (ret) |
3130 | btrfs_abort_transaction(trans, root, ret); | |
3131 | out: | |
71d7aed0 | 3132 | btrfs_free_path(path); |
79787eaa | 3133 | return ret; |
4df27c4d YZ |
3134 | } |
3135 | ||
39279cc3 CM |
3136 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3137 | { | |
3138 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3139 | int err = 0; |
39279cc3 | 3140 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3141 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3142 | unsigned long nr = 0; |
39279cc3 | 3143 | |
3394e160 | 3144 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 3145 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3146 | return -ENOTEMPTY; |
3147 | ||
a22285a6 YZ |
3148 | trans = __unlink_start_trans(dir, dentry); |
3149 | if (IS_ERR(trans)) | |
5df6a9f6 | 3150 | return PTR_ERR(trans); |
5df6a9f6 | 3151 | |
33345d01 | 3152 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3153 | err = btrfs_unlink_subvol(trans, root, dir, |
3154 | BTRFS_I(inode)->location.objectid, | |
3155 | dentry->d_name.name, | |
3156 | dentry->d_name.len); | |
3157 | goto out; | |
3158 | } | |
3159 | ||
7b128766 JB |
3160 | err = btrfs_orphan_add(trans, inode); |
3161 | if (err) | |
4df27c4d | 3162 | goto out; |
7b128766 | 3163 | |
39279cc3 | 3164 | /* now the directory is empty */ |
e02119d5 CM |
3165 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3166 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3167 | if (!err) |
dbe674a9 | 3168 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3169 | out: |
d3c2fdcf | 3170 | nr = trans->blocks_used; |
a22285a6 | 3171 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3172 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3173 | |
39279cc3 CM |
3174 | return err; |
3175 | } | |
3176 | ||
39279cc3 CM |
3177 | /* |
3178 | * this can truncate away extent items, csum items and directory items. | |
3179 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3180 | * any higher than new_size |
39279cc3 CM |
3181 | * |
3182 | * csum items that cross the new i_size are truncated to the new size | |
3183 | * as well. | |
7b128766 JB |
3184 | * |
3185 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3186 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3187 | */ |
8082510e YZ |
3188 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3189 | struct btrfs_root *root, | |
3190 | struct inode *inode, | |
3191 | u64 new_size, u32 min_type) | |
39279cc3 | 3192 | { |
39279cc3 | 3193 | struct btrfs_path *path; |
5f39d397 | 3194 | struct extent_buffer *leaf; |
39279cc3 | 3195 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3196 | struct btrfs_key key; |
3197 | struct btrfs_key found_key; | |
39279cc3 | 3198 | u64 extent_start = 0; |
db94535d | 3199 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3200 | u64 extent_offset = 0; |
39279cc3 | 3201 | u64 item_end = 0; |
8082510e YZ |
3202 | u64 mask = root->sectorsize - 1; |
3203 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3204 | int found_extent; |
3205 | int del_item; | |
85e21bac CM |
3206 | int pending_del_nr = 0; |
3207 | int pending_del_slot = 0; | |
179e29e4 | 3208 | int extent_type = -1; |
8082510e YZ |
3209 | int ret; |
3210 | int err = 0; | |
33345d01 | 3211 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3212 | |
3213 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3214 | |
0eb0e19c MF |
3215 | path = btrfs_alloc_path(); |
3216 | if (!path) | |
3217 | return -ENOMEM; | |
3218 | path->reada = -1; | |
3219 | ||
0af3d00b | 3220 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3221 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3222 | |
16cdcec7 MX |
3223 | /* |
3224 | * This function is also used to drop the items in the log tree before | |
3225 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3226 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3227 | * items. | |
3228 | */ | |
3229 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3230 | btrfs_kill_delayed_inode_items(inode); | |
3231 | ||
33345d01 | 3232 | key.objectid = ino; |
39279cc3 | 3233 | key.offset = (u64)-1; |
5f39d397 CM |
3234 | key.type = (u8)-1; |
3235 | ||
85e21bac | 3236 | search_again: |
b9473439 | 3237 | path->leave_spinning = 1; |
85e21bac | 3238 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3239 | if (ret < 0) { |
3240 | err = ret; | |
3241 | goto out; | |
3242 | } | |
d397712b | 3243 | |
85e21bac | 3244 | if (ret > 0) { |
e02119d5 CM |
3245 | /* there are no items in the tree for us to truncate, we're |
3246 | * done | |
3247 | */ | |
8082510e YZ |
3248 | if (path->slots[0] == 0) |
3249 | goto out; | |
85e21bac CM |
3250 | path->slots[0]--; |
3251 | } | |
3252 | ||
d397712b | 3253 | while (1) { |
39279cc3 | 3254 | fi = NULL; |
5f39d397 CM |
3255 | leaf = path->nodes[0]; |
3256 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3257 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3258 | |
33345d01 | 3259 | if (found_key.objectid != ino) |
39279cc3 | 3260 | break; |
5f39d397 | 3261 | |
85e21bac | 3262 | if (found_type < min_type) |
39279cc3 CM |
3263 | break; |
3264 | ||
5f39d397 | 3265 | item_end = found_key.offset; |
39279cc3 | 3266 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3267 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3268 | struct btrfs_file_extent_item); |
179e29e4 CM |
3269 | extent_type = btrfs_file_extent_type(leaf, fi); |
3270 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3271 | item_end += |
db94535d | 3272 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3273 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3274 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3275 | fi); |
39279cc3 | 3276 | } |
008630c1 | 3277 | item_end--; |
39279cc3 | 3278 | } |
8082510e YZ |
3279 | if (found_type > min_type) { |
3280 | del_item = 1; | |
3281 | } else { | |
3282 | if (item_end < new_size) | |
b888db2b | 3283 | break; |
8082510e YZ |
3284 | if (found_key.offset >= new_size) |
3285 | del_item = 1; | |
3286 | else | |
3287 | del_item = 0; | |
39279cc3 | 3288 | } |
39279cc3 | 3289 | found_extent = 0; |
39279cc3 | 3290 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3291 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3292 | goto delete; | |
3293 | ||
3294 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3295 | u64 num_dec; |
db94535d | 3296 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3297 | if (!del_item) { |
db94535d CM |
3298 | u64 orig_num_bytes = |
3299 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3300 | extent_num_bytes = new_size - |
5f39d397 | 3301 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3302 | extent_num_bytes = extent_num_bytes & |
3303 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3304 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3305 | extent_num_bytes); | |
3306 | num_dec = (orig_num_bytes - | |
9069218d | 3307 | extent_num_bytes); |
e02119d5 | 3308 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3309 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3310 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3311 | } else { |
db94535d CM |
3312 | extent_num_bytes = |
3313 | btrfs_file_extent_disk_num_bytes(leaf, | |
3314 | fi); | |
5d4f98a2 YZ |
3315 | extent_offset = found_key.offset - |
3316 | btrfs_file_extent_offset(leaf, fi); | |
3317 | ||
39279cc3 | 3318 | /* FIXME blocksize != 4096 */ |
9069218d | 3319 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3320 | if (extent_start != 0) { |
3321 | found_extent = 1; | |
e02119d5 | 3322 | if (root->ref_cows) |
a76a3cd4 | 3323 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3324 | } |
39279cc3 | 3325 | } |
9069218d | 3326 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3327 | /* |
3328 | * we can't truncate inline items that have had | |
3329 | * special encodings | |
3330 | */ | |
3331 | if (!del_item && | |
3332 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3333 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3334 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3335 | u32 size = new_size - found_key.offset; |
3336 | ||
3337 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3338 | inode_sub_bytes(inode, item_end + 1 - |
3339 | new_size); | |
e02119d5 CM |
3340 | } |
3341 | size = | |
3342 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3343 | btrfs_truncate_item(trans, root, path, |
3344 | size, 1); | |
e02119d5 | 3345 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3346 | inode_sub_bytes(inode, item_end + 1 - |
3347 | found_key.offset); | |
9069218d | 3348 | } |
39279cc3 | 3349 | } |
179e29e4 | 3350 | delete: |
39279cc3 | 3351 | if (del_item) { |
85e21bac CM |
3352 | if (!pending_del_nr) { |
3353 | /* no pending yet, add ourselves */ | |
3354 | pending_del_slot = path->slots[0]; | |
3355 | pending_del_nr = 1; | |
3356 | } else if (pending_del_nr && | |
3357 | path->slots[0] + 1 == pending_del_slot) { | |
3358 | /* hop on the pending chunk */ | |
3359 | pending_del_nr++; | |
3360 | pending_del_slot = path->slots[0]; | |
3361 | } else { | |
d397712b | 3362 | BUG(); |
85e21bac | 3363 | } |
39279cc3 CM |
3364 | } else { |
3365 | break; | |
3366 | } | |
0af3d00b JB |
3367 | if (found_extent && (root->ref_cows || |
3368 | root == root->fs_info->tree_root)) { | |
b9473439 | 3369 | btrfs_set_path_blocking(path); |
39279cc3 | 3370 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3371 | extent_num_bytes, 0, |
3372 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3373 | ino, extent_offset, 0); |
39279cc3 CM |
3374 | BUG_ON(ret); |
3375 | } | |
85e21bac | 3376 | |
8082510e YZ |
3377 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3378 | break; | |
3379 | ||
3380 | if (path->slots[0] == 0 || | |
3381 | path->slots[0] != pending_del_slot) { | |
82d5902d LZ |
3382 | if (root->ref_cows && |
3383 | BTRFS_I(inode)->location.objectid != | |
3384 | BTRFS_FREE_INO_OBJECTID) { | |
8082510e YZ |
3385 | err = -EAGAIN; |
3386 | goto out; | |
3387 | } | |
3388 | if (pending_del_nr) { | |
3389 | ret = btrfs_del_items(trans, root, path, | |
3390 | pending_del_slot, | |
3391 | pending_del_nr); | |
79787eaa JM |
3392 | if (ret) { |
3393 | btrfs_abort_transaction(trans, | |
3394 | root, ret); | |
3395 | goto error; | |
3396 | } | |
8082510e YZ |
3397 | pending_del_nr = 0; |
3398 | } | |
b3b4aa74 | 3399 | btrfs_release_path(path); |
85e21bac | 3400 | goto search_again; |
8082510e YZ |
3401 | } else { |
3402 | path->slots[0]--; | |
85e21bac | 3403 | } |
39279cc3 | 3404 | } |
8082510e | 3405 | out: |
85e21bac CM |
3406 | if (pending_del_nr) { |
3407 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3408 | pending_del_nr); | |
79787eaa JM |
3409 | if (ret) |
3410 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3411 | } |
79787eaa | 3412 | error: |
39279cc3 | 3413 | btrfs_free_path(path); |
8082510e | 3414 | return err; |
39279cc3 CM |
3415 | } |
3416 | ||
3417 | /* | |
3418 | * taken from block_truncate_page, but does cow as it zeros out | |
3419 | * any bytes left in the last page in the file. | |
3420 | */ | |
3421 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3422 | { | |
3423 | struct inode *inode = mapping->host; | |
db94535d | 3424 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3425 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3426 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3427 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3428 | char *kaddr; |
db94535d | 3429 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3430 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3431 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3432 | struct page *page; | |
3b16a4e3 | 3433 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3434 | int ret = 0; |
a52d9a80 | 3435 | u64 page_start; |
e6dcd2dc | 3436 | u64 page_end; |
39279cc3 CM |
3437 | |
3438 | if ((offset & (blocksize - 1)) == 0) | |
3439 | goto out; | |
0ca1f7ce | 3440 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3441 | if (ret) |
3442 | goto out; | |
39279cc3 CM |
3443 | |
3444 | ret = -ENOMEM; | |
211c17f5 | 3445 | again: |
3b16a4e3 | 3446 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3447 | if (!page) { |
0ca1f7ce | 3448 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3449 | goto out; |
5d5e103a | 3450 | } |
e6dcd2dc CM |
3451 | |
3452 | page_start = page_offset(page); | |
3453 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3454 | ||
39279cc3 | 3455 | if (!PageUptodate(page)) { |
9ebefb18 | 3456 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3457 | lock_page(page); |
211c17f5 CM |
3458 | if (page->mapping != mapping) { |
3459 | unlock_page(page); | |
3460 | page_cache_release(page); | |
3461 | goto again; | |
3462 | } | |
39279cc3 CM |
3463 | if (!PageUptodate(page)) { |
3464 | ret = -EIO; | |
89642229 | 3465 | goto out_unlock; |
39279cc3 CM |
3466 | } |
3467 | } | |
211c17f5 | 3468 | wait_on_page_writeback(page); |
e6dcd2dc | 3469 | |
d0082371 | 3470 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3471 | set_page_extent_mapped(page); |
3472 | ||
3473 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3474 | if (ordered) { | |
2ac55d41 JB |
3475 | unlock_extent_cached(io_tree, page_start, page_end, |
3476 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3477 | unlock_page(page); |
3478 | page_cache_release(page); | |
eb84ae03 | 3479 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3480 | btrfs_put_ordered_extent(ordered); |
3481 | goto again; | |
3482 | } | |
3483 | ||
2ac55d41 | 3484 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3485 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3486 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3487 | |
2ac55d41 JB |
3488 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3489 | &cached_state); | |
9ed74f2d | 3490 | if (ret) { |
2ac55d41 JB |
3491 | unlock_extent_cached(io_tree, page_start, page_end, |
3492 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3493 | goto out_unlock; |
3494 | } | |
3495 | ||
e6dcd2dc CM |
3496 | ret = 0; |
3497 | if (offset != PAGE_CACHE_SIZE) { | |
3498 | kaddr = kmap(page); | |
3499 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3500 | flush_dcache_page(page); | |
3501 | kunmap(page); | |
3502 | } | |
247e743c | 3503 | ClearPageChecked(page); |
e6dcd2dc | 3504 | set_page_dirty(page); |
2ac55d41 JB |
3505 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3506 | GFP_NOFS); | |
39279cc3 | 3507 | |
89642229 | 3508 | out_unlock: |
5d5e103a | 3509 | if (ret) |
0ca1f7ce | 3510 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3511 | unlock_page(page); |
3512 | page_cache_release(page); | |
3513 | out: | |
3514 | return ret; | |
3515 | } | |
3516 | ||
695a0d0d JB |
3517 | /* |
3518 | * This function puts in dummy file extents for the area we're creating a hole | |
3519 | * for. So if we are truncating this file to a larger size we need to insert | |
3520 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3521 | * the range between oldsize and size | |
3522 | */ | |
a41ad394 | 3523 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3524 | { |
9036c102 YZ |
3525 | struct btrfs_trans_handle *trans; |
3526 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3527 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3528 | struct extent_map *em = NULL; |
2ac55d41 | 3529 | struct extent_state *cached_state = NULL; |
9036c102 | 3530 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3531 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3532 | u64 block_end = (size + mask) & ~mask; |
3533 | u64 last_byte; | |
3534 | u64 cur_offset; | |
3535 | u64 hole_size; | |
9ed74f2d | 3536 | int err = 0; |
39279cc3 | 3537 | |
9036c102 YZ |
3538 | if (size <= hole_start) |
3539 | return 0; | |
3540 | ||
9036c102 YZ |
3541 | while (1) { |
3542 | struct btrfs_ordered_extent *ordered; | |
3543 | btrfs_wait_ordered_range(inode, hole_start, | |
3544 | block_end - hole_start); | |
2ac55d41 | 3545 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3546 | &cached_state); |
9036c102 YZ |
3547 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3548 | if (!ordered) | |
3549 | break; | |
2ac55d41 JB |
3550 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3551 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3552 | btrfs_put_ordered_extent(ordered); |
3553 | } | |
39279cc3 | 3554 | |
9036c102 YZ |
3555 | cur_offset = hole_start; |
3556 | while (1) { | |
3557 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3558 | block_end - cur_offset, 0); | |
79787eaa JM |
3559 | if (IS_ERR(em)) { |
3560 | err = PTR_ERR(em); | |
3561 | break; | |
3562 | } | |
9036c102 YZ |
3563 | last_byte = min(extent_map_end(em), block_end); |
3564 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3565 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3566 | u64 hint_byte = 0; |
9036c102 | 3567 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3568 | |
3642320e | 3569 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3570 | if (IS_ERR(trans)) { |
3571 | err = PTR_ERR(trans); | |
9ed74f2d | 3572 | break; |
a22285a6 | 3573 | } |
8082510e YZ |
3574 | |
3575 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3576 | cur_offset + hole_size, | |
3577 | &hint_byte, 1); | |
5b397377 | 3578 | if (err) { |
79787eaa | 3579 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3580 | btrfs_end_transaction(trans, root); |
3893e33b | 3581 | break; |
5b397377 | 3582 | } |
8082510e | 3583 | |
9036c102 | 3584 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3585 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3586 | 0, hole_size, 0, hole_size, |
3587 | 0, 0, 0); | |
5b397377 | 3588 | if (err) { |
79787eaa | 3589 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3590 | btrfs_end_transaction(trans, root); |
3893e33b | 3591 | break; |
5b397377 | 3592 | } |
8082510e | 3593 | |
9036c102 YZ |
3594 | btrfs_drop_extent_cache(inode, hole_start, |
3595 | last_byte - 1, 0); | |
8082510e | 3596 | |
3642320e | 3597 | btrfs_update_inode(trans, root, inode); |
8082510e | 3598 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3599 | } |
3600 | free_extent_map(em); | |
a22285a6 | 3601 | em = NULL; |
9036c102 | 3602 | cur_offset = last_byte; |
8082510e | 3603 | if (cur_offset >= block_end) |
9036c102 YZ |
3604 | break; |
3605 | } | |
1832a6d5 | 3606 | |
a22285a6 | 3607 | free_extent_map(em); |
2ac55d41 JB |
3608 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3609 | GFP_NOFS); | |
9036c102 YZ |
3610 | return err; |
3611 | } | |
39279cc3 | 3612 | |
a41ad394 | 3613 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3614 | { |
f4a2f4c5 MX |
3615 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3616 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3617 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3618 | int ret; |
3619 | ||
a41ad394 | 3620 | if (newsize == oldsize) |
8082510e YZ |
3621 | return 0; |
3622 | ||
a41ad394 | 3623 | if (newsize > oldsize) { |
a41ad394 JB |
3624 | truncate_pagecache(inode, oldsize, newsize); |
3625 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3626 | if (ret) |
8082510e | 3627 | return ret; |
8082510e | 3628 | |
f4a2f4c5 MX |
3629 | trans = btrfs_start_transaction(root, 1); |
3630 | if (IS_ERR(trans)) | |
3631 | return PTR_ERR(trans); | |
3632 | ||
3633 | i_size_write(inode, newsize); | |
3634 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3635 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3636 | btrfs_end_transaction(trans, root); |
a41ad394 | 3637 | } else { |
8082510e | 3638 | |
a41ad394 JB |
3639 | /* |
3640 | * We're truncating a file that used to have good data down to | |
3641 | * zero. Make sure it gets into the ordered flush list so that | |
3642 | * any new writes get down to disk quickly. | |
3643 | */ | |
3644 | if (newsize == 0) | |
3645 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3646 | |
a41ad394 JB |
3647 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3648 | truncate_setsize(inode, newsize); | |
3649 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3650 | } |
3651 | ||
a41ad394 | 3652 | return ret; |
8082510e YZ |
3653 | } |
3654 | ||
9036c102 YZ |
3655 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3656 | { | |
3657 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3658 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3659 | int err; |
39279cc3 | 3660 | |
b83cc969 LZ |
3661 | if (btrfs_root_readonly(root)) |
3662 | return -EROFS; | |
3663 | ||
9036c102 YZ |
3664 | err = inode_change_ok(inode, attr); |
3665 | if (err) | |
3666 | return err; | |
2bf5a725 | 3667 | |
5a3f23d5 | 3668 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3669 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3670 | if (err) |
3671 | return err; | |
39279cc3 | 3672 | } |
9036c102 | 3673 | |
1025774c CH |
3674 | if (attr->ia_valid) { |
3675 | setattr_copy(inode, attr); | |
0c4d2d95 | 3676 | inode_inc_iversion(inode); |
22c44fe6 | 3677 | err = btrfs_dirty_inode(inode); |
1025774c | 3678 | |
22c44fe6 | 3679 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3680 | err = btrfs_acl_chmod(inode); |
3681 | } | |
33268eaf | 3682 | |
39279cc3 CM |
3683 | return err; |
3684 | } | |
61295eb8 | 3685 | |
bd555975 | 3686 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3687 | { |
3688 | struct btrfs_trans_handle *trans; | |
3689 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3690 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3691 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3692 | unsigned long nr; |
39279cc3 CM |
3693 | int ret; |
3694 | ||
1abe9b8a | 3695 | trace_btrfs_inode_evict(inode); |
3696 | ||
39279cc3 | 3697 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3698 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
2cf8572d | 3699 | btrfs_is_free_space_inode(root, inode))) |
bd555975 AV |
3700 | goto no_delete; |
3701 | ||
39279cc3 | 3702 | if (is_bad_inode(inode)) { |
7b128766 | 3703 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3704 | goto no_delete; |
3705 | } | |
bd555975 | 3706 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3707 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3708 | |
c71bf099 YZ |
3709 | if (root->fs_info->log_root_recovering) { |
3710 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3711 | goto no_delete; | |
3712 | } | |
3713 | ||
76dda93c YZ |
3714 | if (inode->i_nlink > 0) { |
3715 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3716 | goto no_delete; | |
3717 | } | |
3718 | ||
4289a667 JB |
3719 | rsv = btrfs_alloc_block_rsv(root); |
3720 | if (!rsv) { | |
3721 | btrfs_orphan_del(NULL, inode); | |
3722 | goto no_delete; | |
3723 | } | |
4a338542 | 3724 | rsv->size = min_size; |
726c35fa | 3725 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3726 | |
dbe674a9 | 3727 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3728 | |
4289a667 JB |
3729 | /* |
3730 | * This is a bit simpler than btrfs_truncate since | |
3731 | * | |
3732 | * 1) We've already reserved our space for our orphan item in the | |
3733 | * unlink. | |
3734 | * 2) We're going to delete the inode item, so we don't need to update | |
3735 | * it at all. | |
3736 | * | |
3737 | * So we just need to reserve some slack space in case we add bytes when | |
3738 | * doing the truncate. | |
3739 | */ | |
8082510e | 3740 | while (1) { |
aa38a711 | 3741 | ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size); |
726c35fa JB |
3742 | |
3743 | /* | |
3744 | * Try and steal from the global reserve since we will | |
3745 | * likely not use this space anyway, we want to try as | |
3746 | * hard as possible to get this to work. | |
3747 | */ | |
3748 | if (ret) | |
3749 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3750 | |
d68fc57b | 3751 | if (ret) { |
4289a667 | 3752 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3753 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3754 | btrfs_orphan_del(NULL, inode); |
3755 | btrfs_free_block_rsv(root, rsv); | |
3756 | goto no_delete; | |
d68fc57b | 3757 | } |
7b128766 | 3758 | |
4289a667 JB |
3759 | trans = btrfs_start_transaction(root, 0); |
3760 | if (IS_ERR(trans)) { | |
3761 | btrfs_orphan_del(NULL, inode); | |
3762 | btrfs_free_block_rsv(root, rsv); | |
3763 | goto no_delete; | |
d68fc57b | 3764 | } |
7b128766 | 3765 | |
4289a667 JB |
3766 | trans->block_rsv = rsv; |
3767 | ||
d68fc57b | 3768 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3769 | if (ret != -EAGAIN) |
3770 | break; | |
85e21bac | 3771 | |
8082510e YZ |
3772 | nr = trans->blocks_used; |
3773 | btrfs_end_transaction(trans, root); | |
3774 | trans = NULL; | |
3775 | btrfs_btree_balance_dirty(root, nr); | |
3776 | } | |
5f39d397 | 3777 | |
4289a667 JB |
3778 | btrfs_free_block_rsv(root, rsv); |
3779 | ||
8082510e | 3780 | if (ret == 0) { |
4289a667 | 3781 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3782 | ret = btrfs_orphan_del(trans, inode); |
3783 | BUG_ON(ret); | |
3784 | } | |
54aa1f4d | 3785 | |
4289a667 | 3786 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3787 | if (!(root == root->fs_info->tree_root || |
3788 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3789 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3790 | |
d3c2fdcf | 3791 | nr = trans->blocks_used; |
54aa1f4d | 3792 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3793 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3794 | no_delete: |
bd555975 | 3795 | end_writeback(inode); |
8082510e | 3796 | return; |
39279cc3 CM |
3797 | } |
3798 | ||
3799 | /* | |
3800 | * this returns the key found in the dir entry in the location pointer. | |
3801 | * If no dir entries were found, location->objectid is 0. | |
3802 | */ | |
3803 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3804 | struct btrfs_key *location) | |
3805 | { | |
3806 | const char *name = dentry->d_name.name; | |
3807 | int namelen = dentry->d_name.len; | |
3808 | struct btrfs_dir_item *di; | |
3809 | struct btrfs_path *path; | |
3810 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3811 | int ret = 0; |
39279cc3 CM |
3812 | |
3813 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3814 | if (!path) |
3815 | return -ENOMEM; | |
3954401f | 3816 | |
33345d01 | 3817 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3818 | namelen, 0); |
0d9f7f3e Y |
3819 | if (IS_ERR(di)) |
3820 | ret = PTR_ERR(di); | |
d397712b | 3821 | |
c704005d | 3822 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3823 | goto out_err; |
d397712b | 3824 | |
5f39d397 | 3825 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3826 | out: |
39279cc3 CM |
3827 | btrfs_free_path(path); |
3828 | return ret; | |
3954401f CM |
3829 | out_err: |
3830 | location->objectid = 0; | |
3831 | goto out; | |
39279cc3 CM |
3832 | } |
3833 | ||
3834 | /* | |
3835 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3836 | * needs to be changed to reflect the root directory of the tree root. This | |
3837 | * is kind of like crossing a mount point. | |
3838 | */ | |
3839 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3840 | struct inode *dir, |
3841 | struct dentry *dentry, | |
3842 | struct btrfs_key *location, | |
3843 | struct btrfs_root **sub_root) | |
39279cc3 | 3844 | { |
4df27c4d YZ |
3845 | struct btrfs_path *path; |
3846 | struct btrfs_root *new_root; | |
3847 | struct btrfs_root_ref *ref; | |
3848 | struct extent_buffer *leaf; | |
3849 | int ret; | |
3850 | int err = 0; | |
39279cc3 | 3851 | |
4df27c4d YZ |
3852 | path = btrfs_alloc_path(); |
3853 | if (!path) { | |
3854 | err = -ENOMEM; | |
3855 | goto out; | |
3856 | } | |
39279cc3 | 3857 | |
4df27c4d YZ |
3858 | err = -ENOENT; |
3859 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3860 | BTRFS_I(dir)->root->root_key.objectid, | |
3861 | location->objectid); | |
3862 | if (ret) { | |
3863 | if (ret < 0) | |
3864 | err = ret; | |
3865 | goto out; | |
3866 | } | |
39279cc3 | 3867 | |
4df27c4d YZ |
3868 | leaf = path->nodes[0]; |
3869 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3870 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3871 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3872 | goto out; | |
39279cc3 | 3873 | |
4df27c4d YZ |
3874 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3875 | (unsigned long)(ref + 1), | |
3876 | dentry->d_name.len); | |
3877 | if (ret) | |
3878 | goto out; | |
3879 | ||
b3b4aa74 | 3880 | btrfs_release_path(path); |
4df27c4d YZ |
3881 | |
3882 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3883 | if (IS_ERR(new_root)) { | |
3884 | err = PTR_ERR(new_root); | |
3885 | goto out; | |
3886 | } | |
3887 | ||
3888 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3889 | err = -ENOENT; | |
3890 | goto out; | |
3891 | } | |
3892 | ||
3893 | *sub_root = new_root; | |
3894 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3895 | location->type = BTRFS_INODE_ITEM_KEY; | |
3896 | location->offset = 0; | |
3897 | err = 0; | |
3898 | out: | |
3899 | btrfs_free_path(path); | |
3900 | return err; | |
39279cc3 CM |
3901 | } |
3902 | ||
5d4f98a2 YZ |
3903 | static void inode_tree_add(struct inode *inode) |
3904 | { | |
3905 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3906 | struct btrfs_inode *entry; | |
03e860bd FNP |
3907 | struct rb_node **p; |
3908 | struct rb_node *parent; | |
33345d01 | 3909 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
3910 | again: |
3911 | p = &root->inode_tree.rb_node; | |
3912 | parent = NULL; | |
5d4f98a2 | 3913 | |
1d3382cb | 3914 | if (inode_unhashed(inode)) |
76dda93c YZ |
3915 | return; |
3916 | ||
5d4f98a2 YZ |
3917 | spin_lock(&root->inode_lock); |
3918 | while (*p) { | |
3919 | parent = *p; | |
3920 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3921 | ||
33345d01 | 3922 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3923 | p = &parent->rb_left; |
33345d01 | 3924 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3925 | p = &parent->rb_right; |
5d4f98a2 YZ |
3926 | else { |
3927 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3928 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3929 | rb_erase(parent, &root->inode_tree); |
3930 | RB_CLEAR_NODE(parent); | |
3931 | spin_unlock(&root->inode_lock); | |
3932 | goto again; | |
5d4f98a2 YZ |
3933 | } |
3934 | } | |
3935 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3936 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3937 | spin_unlock(&root->inode_lock); | |
3938 | } | |
3939 | ||
3940 | static void inode_tree_del(struct inode *inode) | |
3941 | { | |
3942 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3943 | int empty = 0; |
5d4f98a2 | 3944 | |
03e860bd | 3945 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3946 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3947 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3948 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3949 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3950 | } |
03e860bd | 3951 | spin_unlock(&root->inode_lock); |
76dda93c | 3952 | |
0af3d00b JB |
3953 | /* |
3954 | * Free space cache has inodes in the tree root, but the tree root has a | |
3955 | * root_refs of 0, so this could end up dropping the tree root as a | |
3956 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3957 | * make sure we don't drop it. | |
3958 | */ | |
3959 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3960 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3961 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3962 | spin_lock(&root->inode_lock); | |
3963 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3964 | spin_unlock(&root->inode_lock); | |
3965 | if (empty) | |
3966 | btrfs_add_dead_root(root); | |
3967 | } | |
3968 | } | |
3969 | ||
143bede5 | 3970 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
3971 | { |
3972 | struct rb_node *node; | |
3973 | struct rb_node *prev; | |
3974 | struct btrfs_inode *entry; | |
3975 | struct inode *inode; | |
3976 | u64 objectid = 0; | |
3977 | ||
3978 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3979 | ||
3980 | spin_lock(&root->inode_lock); | |
3981 | again: | |
3982 | node = root->inode_tree.rb_node; | |
3983 | prev = NULL; | |
3984 | while (node) { | |
3985 | prev = node; | |
3986 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3987 | ||
33345d01 | 3988 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 3989 | node = node->rb_left; |
33345d01 | 3990 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
3991 | node = node->rb_right; |
3992 | else | |
3993 | break; | |
3994 | } | |
3995 | if (!node) { | |
3996 | while (prev) { | |
3997 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 3998 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
3999 | node = prev; |
4000 | break; | |
4001 | } | |
4002 | prev = rb_next(prev); | |
4003 | } | |
4004 | } | |
4005 | while (node) { | |
4006 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4007 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4008 | inode = igrab(&entry->vfs_inode); |
4009 | if (inode) { | |
4010 | spin_unlock(&root->inode_lock); | |
4011 | if (atomic_read(&inode->i_count) > 1) | |
4012 | d_prune_aliases(inode); | |
4013 | /* | |
45321ac5 | 4014 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4015 | * the inode cache when its usage count |
4016 | * hits zero. | |
4017 | */ | |
4018 | iput(inode); | |
4019 | cond_resched(); | |
4020 | spin_lock(&root->inode_lock); | |
4021 | goto again; | |
4022 | } | |
4023 | ||
4024 | if (cond_resched_lock(&root->inode_lock)) | |
4025 | goto again; | |
4026 | ||
4027 | node = rb_next(node); | |
4028 | } | |
4029 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4030 | } |
4031 | ||
e02119d5 CM |
4032 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4033 | { | |
4034 | struct btrfs_iget_args *args = p; | |
4035 | inode->i_ino = args->ino; | |
e02119d5 | 4036 | BTRFS_I(inode)->root = args->root; |
6a63209f | 4037 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
4038 | return 0; |
4039 | } | |
4040 | ||
4041 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4042 | { | |
4043 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4044 | return args->ino == btrfs_ino(inode) && |
d397712b | 4045 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4046 | } |
4047 | ||
5d4f98a2 YZ |
4048 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4049 | u64 objectid, | |
4050 | struct btrfs_root *root) | |
39279cc3 CM |
4051 | { |
4052 | struct inode *inode; | |
4053 | struct btrfs_iget_args args; | |
4054 | args.ino = objectid; | |
4055 | args.root = root; | |
4056 | ||
4057 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4058 | btrfs_init_locked_inode, | |
4059 | (void *)&args); | |
4060 | return inode; | |
4061 | } | |
4062 | ||
1a54ef8c BR |
4063 | /* Get an inode object given its location and corresponding root. |
4064 | * Returns in *is_new if the inode was read from disk | |
4065 | */ | |
4066 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4067 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4068 | { |
4069 | struct inode *inode; | |
4070 | ||
4071 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4072 | if (!inode) | |
5d4f98a2 | 4073 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4074 | |
4075 | if (inode->i_state & I_NEW) { | |
4076 | BTRFS_I(inode)->root = root; | |
4077 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4078 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4079 | if (!is_bad_inode(inode)) { |
4080 | inode_tree_add(inode); | |
4081 | unlock_new_inode(inode); | |
4082 | if (new) | |
4083 | *new = 1; | |
4084 | } else { | |
e0b6d65b ST |
4085 | unlock_new_inode(inode); |
4086 | iput(inode); | |
4087 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4088 | } |
4089 | } | |
4090 | ||
1a54ef8c BR |
4091 | return inode; |
4092 | } | |
4093 | ||
4df27c4d YZ |
4094 | static struct inode *new_simple_dir(struct super_block *s, |
4095 | struct btrfs_key *key, | |
4096 | struct btrfs_root *root) | |
4097 | { | |
4098 | struct inode *inode = new_inode(s); | |
4099 | ||
4100 | if (!inode) | |
4101 | return ERR_PTR(-ENOMEM); | |
4102 | ||
4df27c4d YZ |
4103 | BTRFS_I(inode)->root = root; |
4104 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
4105 | BTRFS_I(inode)->dummy_inode = 1; | |
4106 | ||
4107 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4108 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4109 | inode->i_fop = &simple_dir_operations; |
4110 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4111 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4112 | ||
4113 | return inode; | |
4114 | } | |
4115 | ||
3de4586c | 4116 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4117 | { |
d397712b | 4118 | struct inode *inode; |
4df27c4d | 4119 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4120 | struct btrfs_root *sub_root = root; |
4121 | struct btrfs_key location; | |
76dda93c | 4122 | int index; |
b4aff1f8 | 4123 | int ret = 0; |
39279cc3 CM |
4124 | |
4125 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4126 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4127 | |
b4aff1f8 JB |
4128 | if (unlikely(d_need_lookup(dentry))) { |
4129 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4130 | kfree(dentry->d_fsdata); | |
4131 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4132 | /* This thing is hashed, drop it for now */ |
4133 | d_drop(dentry); | |
b4aff1f8 JB |
4134 | } else { |
4135 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4136 | } | |
5f39d397 | 4137 | |
39279cc3 CM |
4138 | if (ret < 0) |
4139 | return ERR_PTR(ret); | |
5f39d397 | 4140 | |
4df27c4d YZ |
4141 | if (location.objectid == 0) |
4142 | return NULL; | |
4143 | ||
4144 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4145 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4146 | return inode; |
4147 | } | |
4148 | ||
4149 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4150 | ||
76dda93c | 4151 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4152 | ret = fixup_tree_root_location(root, dir, dentry, |
4153 | &location, &sub_root); | |
4154 | if (ret < 0) { | |
4155 | if (ret != -ENOENT) | |
4156 | inode = ERR_PTR(ret); | |
4157 | else | |
4158 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4159 | } else { | |
73f73415 | 4160 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4161 | } |
76dda93c YZ |
4162 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4163 | ||
34d19bad | 4164 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4165 | down_read(&root->fs_info->cleanup_work_sem); |
4166 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4167 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4168 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4169 | if (ret) |
4170 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4171 | } |
4172 | ||
3de4586c CM |
4173 | return inode; |
4174 | } | |
4175 | ||
fe15ce44 | 4176 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4177 | { |
4178 | struct btrfs_root *root; | |
848cce0d | 4179 | struct inode *inode = dentry->d_inode; |
76dda93c | 4180 | |
848cce0d LZ |
4181 | if (!inode && !IS_ROOT(dentry)) |
4182 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4183 | |
848cce0d LZ |
4184 | if (inode) { |
4185 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4186 | if (btrfs_root_refs(&root->root_item) == 0) |
4187 | return 1; | |
848cce0d LZ |
4188 | |
4189 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4190 | return 1; | |
efefb143 | 4191 | } |
76dda93c YZ |
4192 | return 0; |
4193 | } | |
4194 | ||
b4aff1f8 JB |
4195 | static void btrfs_dentry_release(struct dentry *dentry) |
4196 | { | |
4197 | if (dentry->d_fsdata) | |
4198 | kfree(dentry->d_fsdata); | |
4199 | } | |
4200 | ||
3de4586c CM |
4201 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4202 | struct nameidata *nd) | |
4203 | { | |
a66e7cc6 JB |
4204 | struct dentry *ret; |
4205 | ||
4206 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4207 | if (unlikely(d_need_lookup(dentry))) { | |
4208 | spin_lock(&dentry->d_lock); | |
4209 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4210 | spin_unlock(&dentry->d_lock); | |
4211 | } | |
4212 | return ret; | |
39279cc3 CM |
4213 | } |
4214 | ||
16cdcec7 | 4215 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4216 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4217 | }; | |
4218 | ||
cbdf5a24 DW |
4219 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4220 | filldir_t filldir) | |
39279cc3 | 4221 | { |
6da6abae | 4222 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4223 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4224 | struct btrfs_item *item; | |
4225 | struct btrfs_dir_item *di; | |
4226 | struct btrfs_key key; | |
5f39d397 | 4227 | struct btrfs_key found_key; |
39279cc3 | 4228 | struct btrfs_path *path; |
16cdcec7 MX |
4229 | struct list_head ins_list; |
4230 | struct list_head del_list; | |
39279cc3 | 4231 | int ret; |
5f39d397 | 4232 | struct extent_buffer *leaf; |
39279cc3 | 4233 | int slot; |
39279cc3 CM |
4234 | unsigned char d_type; |
4235 | int over = 0; | |
4236 | u32 di_cur; | |
4237 | u32 di_total; | |
4238 | u32 di_len; | |
4239 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4240 | char tmp_name[32]; |
4241 | char *name_ptr; | |
4242 | int name_len; | |
16cdcec7 | 4243 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4244 | |
4245 | /* FIXME, use a real flag for deciding about the key type */ | |
4246 | if (root->fs_info->tree_root == root) | |
4247 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4248 | |
3954401f CM |
4249 | /* special case for "." */ |
4250 | if (filp->f_pos == 0) { | |
3765fefa HS |
4251 | over = filldir(dirent, ".", 1, |
4252 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4253 | if (over) |
4254 | return 0; | |
4255 | filp->f_pos = 1; | |
4256 | } | |
3954401f CM |
4257 | /* special case for .., just use the back ref */ |
4258 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4259 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4260 | over = filldir(dirent, "..", 2, |
3765fefa | 4261 | filp->f_pos, pino, DT_DIR); |
3954401f | 4262 | if (over) |
49593bfa | 4263 | return 0; |
3954401f CM |
4264 | filp->f_pos = 2; |
4265 | } | |
49593bfa | 4266 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4267 | if (!path) |
4268 | return -ENOMEM; | |
ff5714cc | 4269 | |
026fd317 | 4270 | path->reada = 1; |
49593bfa | 4271 | |
16cdcec7 MX |
4272 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4273 | INIT_LIST_HEAD(&ins_list); | |
4274 | INIT_LIST_HEAD(&del_list); | |
4275 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4276 | } | |
4277 | ||
39279cc3 CM |
4278 | btrfs_set_key_type(&key, key_type); |
4279 | key.offset = filp->f_pos; | |
33345d01 | 4280 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4281 | |
39279cc3 CM |
4282 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4283 | if (ret < 0) | |
4284 | goto err; | |
49593bfa DW |
4285 | |
4286 | while (1) { | |
5f39d397 | 4287 | leaf = path->nodes[0]; |
39279cc3 | 4288 | slot = path->slots[0]; |
b9e03af0 LZ |
4289 | if (slot >= btrfs_header_nritems(leaf)) { |
4290 | ret = btrfs_next_leaf(root, path); | |
4291 | if (ret < 0) | |
4292 | goto err; | |
4293 | else if (ret > 0) | |
4294 | break; | |
4295 | continue; | |
39279cc3 | 4296 | } |
3de4586c | 4297 | |
5f39d397 CM |
4298 | item = btrfs_item_nr(leaf, slot); |
4299 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4300 | ||
4301 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4302 | break; |
5f39d397 | 4303 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4304 | break; |
5f39d397 | 4305 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4306 | goto next; |
16cdcec7 MX |
4307 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4308 | btrfs_should_delete_dir_index(&del_list, | |
4309 | found_key.offset)) | |
4310 | goto next; | |
5f39d397 CM |
4311 | |
4312 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4313 | is_curr = 1; |
49593bfa | 4314 | |
39279cc3 CM |
4315 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4316 | di_cur = 0; | |
5f39d397 | 4317 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4318 | |
4319 | while (di_cur < di_total) { | |
5f39d397 CM |
4320 | struct btrfs_key location; |
4321 | ||
22a94d44 JB |
4322 | if (verify_dir_item(root, leaf, di)) |
4323 | break; | |
4324 | ||
5f39d397 | 4325 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4326 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4327 | name_ptr = tmp_name; |
4328 | } else { | |
4329 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4330 | if (!name_ptr) { |
4331 | ret = -ENOMEM; | |
4332 | goto err; | |
4333 | } | |
5f39d397 CM |
4334 | } |
4335 | read_extent_buffer(leaf, name_ptr, | |
4336 | (unsigned long)(di + 1), name_len); | |
4337 | ||
4338 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4339 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4340 | |
fede766f | 4341 | |
3de4586c | 4342 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4343 | * skip it. |
4344 | * | |
4345 | * In contrast to old kernels, we insert the snapshot's | |
4346 | * dir item and dir index after it has been created, so | |
4347 | * we won't find a reference to our own snapshot. We | |
4348 | * still keep the following code for backward | |
4349 | * compatibility. | |
3de4586c CM |
4350 | */ |
4351 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4352 | location.objectid == root->root_key.objectid) { | |
4353 | over = 0; | |
4354 | goto skip; | |
4355 | } | |
5f39d397 | 4356 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4357 | found_key.offset, location.objectid, |
39279cc3 | 4358 | d_type); |
5f39d397 | 4359 | |
3de4586c | 4360 | skip: |
5f39d397 CM |
4361 | if (name_ptr != tmp_name) |
4362 | kfree(name_ptr); | |
4363 | ||
39279cc3 CM |
4364 | if (over) |
4365 | goto nopos; | |
5103e947 | 4366 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4367 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4368 | di_cur += di_len; |
4369 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4370 | } | |
b9e03af0 LZ |
4371 | next: |
4372 | path->slots[0]++; | |
39279cc3 | 4373 | } |
49593bfa | 4374 | |
16cdcec7 MX |
4375 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4376 | if (is_curr) | |
4377 | filp->f_pos++; | |
4378 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4379 | &ins_list); | |
4380 | if (ret) | |
4381 | goto nopos; | |
4382 | } | |
4383 | ||
49593bfa | 4384 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4385 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4386 | /* |
4387 | * 32-bit glibc will use getdents64, but then strtol - | |
4388 | * so the last number we can serve is this. | |
4389 | */ | |
4390 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4391 | else |
4392 | filp->f_pos++; | |
39279cc3 CM |
4393 | nopos: |
4394 | ret = 0; | |
4395 | err: | |
16cdcec7 MX |
4396 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4397 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4398 | btrfs_free_path(path); |
39279cc3 CM |
4399 | return ret; |
4400 | } | |
4401 | ||
a9185b41 | 4402 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4403 | { |
4404 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4405 | struct btrfs_trans_handle *trans; | |
4406 | int ret = 0; | |
0af3d00b | 4407 | bool nolock = false; |
39279cc3 | 4408 | |
8929ecfa | 4409 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4410 | return 0; |
4411 | ||
2cf8572d | 4412 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode)) |
82d5902d | 4413 | nolock = true; |
0af3d00b | 4414 | |
a9185b41 | 4415 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4416 | if (nolock) |
7a7eaa40 | 4417 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4418 | else |
7a7eaa40 | 4419 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4420 | if (IS_ERR(trans)) |
4421 | return PTR_ERR(trans); | |
0af3d00b JB |
4422 | if (nolock) |
4423 | ret = btrfs_end_transaction_nolock(trans, root); | |
4424 | else | |
4425 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4426 | } |
4427 | return ret; | |
4428 | } | |
4429 | ||
4430 | /* | |
54aa1f4d | 4431 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4432 | * inode changes. But, it is most likely to find the inode in cache. |
4433 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4434 | * to keep or drop this code. | |
4435 | */ | |
22c44fe6 | 4436 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4437 | { |
4438 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4439 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4440 | int ret; |
4441 | ||
4442 | if (BTRFS_I(inode)->dummy_inode) | |
22c44fe6 | 4443 | return 0; |
39279cc3 | 4444 | |
7a7eaa40 | 4445 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4446 | if (IS_ERR(trans)) |
4447 | return PTR_ERR(trans); | |
8929ecfa YZ |
4448 | |
4449 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4450 | if (ret && ret == -ENOSPC) { |
4451 | /* whoops, lets try again with the full transaction */ | |
4452 | btrfs_end_transaction(trans, root); | |
4453 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4454 | if (IS_ERR(trans)) |
4455 | return PTR_ERR(trans); | |
8929ecfa | 4456 | |
94b60442 | 4457 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4458 | } |
39279cc3 | 4459 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4460 | if (BTRFS_I(inode)->delayed_node) |
4461 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4462 | |
4463 | return ret; | |
4464 | } | |
4465 | ||
4466 | /* | |
4467 | * This is a copy of file_update_time. We need this so we can return error on | |
4468 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4469 | */ | |
4470 | int btrfs_update_time(struct file *file) | |
4471 | { | |
4472 | struct inode *inode = file->f_path.dentry->d_inode; | |
4473 | struct timespec now; | |
4474 | int ret; | |
4475 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; | |
4476 | ||
4477 | /* First try to exhaust all avenues to not sync */ | |
4478 | if (IS_NOCMTIME(inode)) | |
4479 | return 0; | |
4480 | ||
4481 | now = current_fs_time(inode->i_sb); | |
4482 | if (!timespec_equal(&inode->i_mtime, &now)) | |
4483 | sync_it = S_MTIME; | |
4484 | ||
4485 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4486 | sync_it |= S_CTIME; | |
4487 | ||
4488 | if (IS_I_VERSION(inode)) | |
4489 | sync_it |= S_VERSION; | |
4490 | ||
4491 | if (!sync_it) | |
4492 | return 0; | |
4493 | ||
4494 | /* Finally allowed to write? Takes lock. */ | |
4495 | if (mnt_want_write_file(file)) | |
4496 | return 0; | |
4497 | ||
4498 | /* Only change inode inside the lock region */ | |
4499 | if (sync_it & S_VERSION) | |
4500 | inode_inc_iversion(inode); | |
4501 | if (sync_it & S_CTIME) | |
4502 | inode->i_ctime = now; | |
4503 | if (sync_it & S_MTIME) | |
4504 | inode->i_mtime = now; | |
4505 | ret = btrfs_dirty_inode(inode); | |
4506 | if (!ret) | |
4507 | mark_inode_dirty_sync(inode); | |
4508 | mnt_drop_write(file->f_path.mnt); | |
4509 | return ret; | |
39279cc3 CM |
4510 | } |
4511 | ||
d352ac68 CM |
4512 | /* |
4513 | * find the highest existing sequence number in a directory | |
4514 | * and then set the in-memory index_cnt variable to reflect | |
4515 | * free sequence numbers | |
4516 | */ | |
aec7477b JB |
4517 | static int btrfs_set_inode_index_count(struct inode *inode) |
4518 | { | |
4519 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4520 | struct btrfs_key key, found_key; | |
4521 | struct btrfs_path *path; | |
4522 | struct extent_buffer *leaf; | |
4523 | int ret; | |
4524 | ||
33345d01 | 4525 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4526 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4527 | key.offset = (u64)-1; | |
4528 | ||
4529 | path = btrfs_alloc_path(); | |
4530 | if (!path) | |
4531 | return -ENOMEM; | |
4532 | ||
4533 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4534 | if (ret < 0) | |
4535 | goto out; | |
4536 | /* FIXME: we should be able to handle this */ | |
4537 | if (ret == 0) | |
4538 | goto out; | |
4539 | ret = 0; | |
4540 | ||
4541 | /* | |
4542 | * MAGIC NUMBER EXPLANATION: | |
4543 | * since we search a directory based on f_pos we have to start at 2 | |
4544 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4545 | * else has to start at 2 | |
4546 | */ | |
4547 | if (path->slots[0] == 0) { | |
4548 | BTRFS_I(inode)->index_cnt = 2; | |
4549 | goto out; | |
4550 | } | |
4551 | ||
4552 | path->slots[0]--; | |
4553 | ||
4554 | leaf = path->nodes[0]; | |
4555 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4556 | ||
33345d01 | 4557 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4558 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4559 | BTRFS_I(inode)->index_cnt = 2; | |
4560 | goto out; | |
4561 | } | |
4562 | ||
4563 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4564 | out: | |
4565 | btrfs_free_path(path); | |
4566 | return ret; | |
4567 | } | |
4568 | ||
d352ac68 CM |
4569 | /* |
4570 | * helper to find a free sequence number in a given directory. This current | |
4571 | * code is very simple, later versions will do smarter things in the btree | |
4572 | */ | |
3de4586c | 4573 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4574 | { |
4575 | int ret = 0; | |
4576 | ||
4577 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4578 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4579 | if (ret) { | |
4580 | ret = btrfs_set_inode_index_count(dir); | |
4581 | if (ret) | |
4582 | return ret; | |
4583 | } | |
aec7477b JB |
4584 | } |
4585 | ||
00e4e6b3 | 4586 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4587 | BTRFS_I(dir)->index_cnt++; |
4588 | ||
4589 | return ret; | |
4590 | } | |
4591 | ||
39279cc3 CM |
4592 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4593 | struct btrfs_root *root, | |
aec7477b | 4594 | struct inode *dir, |
9c58309d | 4595 | const char *name, int name_len, |
175a4eb7 AV |
4596 | u64 ref_objectid, u64 objectid, |
4597 | umode_t mode, u64 *index) | |
39279cc3 CM |
4598 | { |
4599 | struct inode *inode; | |
5f39d397 | 4600 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4601 | struct btrfs_key *location; |
5f39d397 | 4602 | struct btrfs_path *path; |
9c58309d CM |
4603 | struct btrfs_inode_ref *ref; |
4604 | struct btrfs_key key[2]; | |
4605 | u32 sizes[2]; | |
4606 | unsigned long ptr; | |
39279cc3 CM |
4607 | int ret; |
4608 | int owner; | |
4609 | ||
5f39d397 | 4610 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4611 | if (!path) |
4612 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4613 | |
39279cc3 | 4614 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4615 | if (!inode) { |
4616 | btrfs_free_path(path); | |
39279cc3 | 4617 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4618 | } |
39279cc3 | 4619 | |
581bb050 LZ |
4620 | /* |
4621 | * we have to initialize this early, so we can reclaim the inode | |
4622 | * number if we fail afterwards in this function. | |
4623 | */ | |
4624 | inode->i_ino = objectid; | |
4625 | ||
aec7477b | 4626 | if (dir) { |
1abe9b8a | 4627 | trace_btrfs_inode_request(dir); |
4628 | ||
3de4586c | 4629 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4630 | if (ret) { |
8fb27640 | 4631 | btrfs_free_path(path); |
09771430 | 4632 | iput(inode); |
aec7477b | 4633 | return ERR_PTR(ret); |
09771430 | 4634 | } |
aec7477b JB |
4635 | } |
4636 | /* | |
4637 | * index_cnt is ignored for everything but a dir, | |
4638 | * btrfs_get_inode_index_count has an explanation for the magic | |
4639 | * number | |
4640 | */ | |
4641 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4642 | BTRFS_I(inode)->root = root; |
e02119d5 | 4643 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4644 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4645 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4646 | |
569254b0 | 4647 | if (S_ISDIR(mode)) |
39279cc3 CM |
4648 | owner = 0; |
4649 | else | |
4650 | owner = 1; | |
9c58309d CM |
4651 | |
4652 | key[0].objectid = objectid; | |
4653 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4654 | key[0].offset = 0; | |
4655 | ||
4656 | key[1].objectid = objectid; | |
4657 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4658 | key[1].offset = ref_objectid; | |
4659 | ||
4660 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4661 | sizes[1] = name_len + sizeof(*ref); | |
4662 | ||
b9473439 | 4663 | path->leave_spinning = 1; |
9c58309d CM |
4664 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4665 | if (ret != 0) | |
5f39d397 CM |
4666 | goto fail; |
4667 | ||
ecc11fab | 4668 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4669 | inode_set_bytes(inode, 0); |
39279cc3 | 4670 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4671 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4672 | struct btrfs_inode_item); | |
e02119d5 | 4673 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4674 | |
4675 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4676 | struct btrfs_inode_ref); | |
4677 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4678 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4679 | ptr = (unsigned long)(ref + 1); |
4680 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4681 | ||
5f39d397 CM |
4682 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4683 | btrfs_free_path(path); | |
4684 | ||
39279cc3 CM |
4685 | location = &BTRFS_I(inode)->location; |
4686 | location->objectid = objectid; | |
39279cc3 CM |
4687 | location->offset = 0; |
4688 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4689 | ||
6cbff00f CH |
4690 | btrfs_inherit_iflags(inode, dir); |
4691 | ||
569254b0 | 4692 | if (S_ISREG(mode)) { |
94272164 CM |
4693 | if (btrfs_test_opt(root, NODATASUM)) |
4694 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4695 | if (btrfs_test_opt(root, NODATACOW) || |
4696 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4697 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4698 | } | |
4699 | ||
39279cc3 | 4700 | insert_inode_hash(inode); |
5d4f98a2 | 4701 | inode_tree_add(inode); |
1abe9b8a | 4702 | |
4703 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4704 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4705 | |
39279cc3 | 4706 | return inode; |
5f39d397 | 4707 | fail: |
aec7477b JB |
4708 | if (dir) |
4709 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4710 | btrfs_free_path(path); |
09771430 | 4711 | iput(inode); |
5f39d397 | 4712 | return ERR_PTR(ret); |
39279cc3 CM |
4713 | } |
4714 | ||
4715 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4716 | { | |
4717 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4718 | } | |
4719 | ||
d352ac68 CM |
4720 | /* |
4721 | * utility function to add 'inode' into 'parent_inode' with | |
4722 | * a give name and a given sequence number. | |
4723 | * if 'add_backref' is true, also insert a backref from the | |
4724 | * inode to the parent directory. | |
4725 | */ | |
e02119d5 CM |
4726 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4727 | struct inode *parent_inode, struct inode *inode, | |
4728 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4729 | { |
4df27c4d | 4730 | int ret = 0; |
39279cc3 | 4731 | struct btrfs_key key; |
e02119d5 | 4732 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4733 | u64 ino = btrfs_ino(inode); |
4734 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4735 | |
33345d01 | 4736 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4737 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4738 | } else { | |
33345d01 | 4739 | key.objectid = ino; |
4df27c4d YZ |
4740 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4741 | key.offset = 0; | |
4742 | } | |
4743 | ||
33345d01 | 4744 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4745 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4746 | key.objectid, root->root_key.objectid, | |
33345d01 | 4747 | parent_ino, index, name, name_len); |
4df27c4d | 4748 | } else if (add_backref) { |
33345d01 LZ |
4749 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4750 | parent_ino, index); | |
4df27c4d | 4751 | } |
39279cc3 | 4752 | |
79787eaa JM |
4753 | /* Nothing to clean up yet */ |
4754 | if (ret) | |
4755 | return ret; | |
4df27c4d | 4756 | |
79787eaa JM |
4757 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4758 | parent_inode, &key, | |
4759 | btrfs_inode_type(inode), index); | |
4760 | if (ret == -EEXIST) | |
4761 | goto fail_dir_item; | |
4762 | else if (ret) { | |
4763 | btrfs_abort_transaction(trans, root, ret); | |
4764 | return ret; | |
39279cc3 | 4765 | } |
79787eaa JM |
4766 | |
4767 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4768 | name_len * 2); | |
0c4d2d95 | 4769 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4770 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4771 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4772 | if (ret) | |
4773 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4774 | return ret; |
fe66a05a CM |
4775 | |
4776 | fail_dir_item: | |
4777 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4778 | u64 local_index; | |
4779 | int err; | |
4780 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4781 | key.objectid, root->root_key.objectid, | |
4782 | parent_ino, &local_index, name, name_len); | |
4783 | ||
4784 | } else if (add_backref) { | |
4785 | u64 local_index; | |
4786 | int err; | |
4787 | ||
4788 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
4789 | ino, parent_ino, &local_index); | |
4790 | } | |
4791 | return ret; | |
39279cc3 CM |
4792 | } |
4793 | ||
4794 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4795 | struct inode *dir, struct dentry *dentry, |
4796 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4797 | { |
a1b075d2 JB |
4798 | int err = btrfs_add_link(trans, dir, inode, |
4799 | dentry->d_name.name, dentry->d_name.len, | |
4800 | backref, index); | |
39279cc3 CM |
4801 | if (err > 0) |
4802 | err = -EEXIST; | |
4803 | return err; | |
4804 | } | |
4805 | ||
618e21d5 | 4806 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 4807 | umode_t mode, dev_t rdev) |
618e21d5 JB |
4808 | { |
4809 | struct btrfs_trans_handle *trans; | |
4810 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4811 | struct inode *inode = NULL; |
618e21d5 JB |
4812 | int err; |
4813 | int drop_inode = 0; | |
4814 | u64 objectid; | |
1832a6d5 | 4815 | unsigned long nr = 0; |
00e4e6b3 | 4816 | u64 index = 0; |
618e21d5 JB |
4817 | |
4818 | if (!new_valid_dev(rdev)) | |
4819 | return -EINVAL; | |
4820 | ||
9ed74f2d JB |
4821 | /* |
4822 | * 2 for inode item and ref | |
4823 | * 2 for dir items | |
4824 | * 1 for xattr if selinux is on | |
4825 | */ | |
a22285a6 YZ |
4826 | trans = btrfs_start_transaction(root, 5); |
4827 | if (IS_ERR(trans)) | |
4828 | return PTR_ERR(trans); | |
1832a6d5 | 4829 | |
581bb050 LZ |
4830 | err = btrfs_find_free_ino(root, &objectid); |
4831 | if (err) | |
4832 | goto out_unlock; | |
4833 | ||
aec7477b | 4834 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4835 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4836 | mode, &index); |
7cf96da3 TI |
4837 | if (IS_ERR(inode)) { |
4838 | err = PTR_ERR(inode); | |
618e21d5 | 4839 | goto out_unlock; |
7cf96da3 | 4840 | } |
618e21d5 | 4841 | |
2a7dba39 | 4842 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4843 | if (err) { |
4844 | drop_inode = 1; | |
4845 | goto out_unlock; | |
4846 | } | |
4847 | ||
ad19db71 CS |
4848 | /* |
4849 | * If the active LSM wants to access the inode during | |
4850 | * d_instantiate it needs these. Smack checks to see | |
4851 | * if the filesystem supports xattrs by looking at the | |
4852 | * ops vector. | |
4853 | */ | |
4854 | ||
4855 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4856 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4857 | if (err) |
4858 | drop_inode = 1; | |
4859 | else { | |
618e21d5 | 4860 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4861 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4862 | d_instantiate(dentry, inode); |
618e21d5 | 4863 | } |
618e21d5 | 4864 | out_unlock: |
d3c2fdcf | 4865 | nr = trans->blocks_used; |
7ad85bb7 | 4866 | btrfs_end_transaction(trans, root); |
a22285a6 | 4867 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4868 | if (drop_inode) { |
4869 | inode_dec_link_count(inode); | |
4870 | iput(inode); | |
4871 | } | |
618e21d5 JB |
4872 | return err; |
4873 | } | |
4874 | ||
39279cc3 | 4875 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4acdaf27 | 4876 | umode_t mode, struct nameidata *nd) |
39279cc3 CM |
4877 | { |
4878 | struct btrfs_trans_handle *trans; | |
4879 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4880 | struct inode *inode = NULL; |
39279cc3 | 4881 | int drop_inode = 0; |
a22285a6 | 4882 | int err; |
1832a6d5 | 4883 | unsigned long nr = 0; |
39279cc3 | 4884 | u64 objectid; |
00e4e6b3 | 4885 | u64 index = 0; |
39279cc3 | 4886 | |
9ed74f2d JB |
4887 | /* |
4888 | * 2 for inode item and ref | |
4889 | * 2 for dir items | |
4890 | * 1 for xattr if selinux is on | |
4891 | */ | |
a22285a6 YZ |
4892 | trans = btrfs_start_transaction(root, 5); |
4893 | if (IS_ERR(trans)) | |
4894 | return PTR_ERR(trans); | |
9ed74f2d | 4895 | |
581bb050 LZ |
4896 | err = btrfs_find_free_ino(root, &objectid); |
4897 | if (err) | |
4898 | goto out_unlock; | |
4899 | ||
aec7477b | 4900 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4901 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4902 | mode, &index); |
7cf96da3 TI |
4903 | if (IS_ERR(inode)) { |
4904 | err = PTR_ERR(inode); | |
39279cc3 | 4905 | goto out_unlock; |
7cf96da3 | 4906 | } |
39279cc3 | 4907 | |
2a7dba39 | 4908 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4909 | if (err) { |
4910 | drop_inode = 1; | |
4911 | goto out_unlock; | |
4912 | } | |
4913 | ||
ad19db71 CS |
4914 | /* |
4915 | * If the active LSM wants to access the inode during | |
4916 | * d_instantiate it needs these. Smack checks to see | |
4917 | * if the filesystem supports xattrs by looking at the | |
4918 | * ops vector. | |
4919 | */ | |
4920 | inode->i_fop = &btrfs_file_operations; | |
4921 | inode->i_op = &btrfs_file_inode_operations; | |
4922 | ||
a1b075d2 | 4923 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4924 | if (err) |
4925 | drop_inode = 1; | |
4926 | else { | |
4927 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4928 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 4929 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
08c422c2 | 4930 | d_instantiate(dentry, inode); |
39279cc3 | 4931 | } |
39279cc3 | 4932 | out_unlock: |
d3c2fdcf | 4933 | nr = trans->blocks_used; |
7ad85bb7 | 4934 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4935 | if (drop_inode) { |
4936 | inode_dec_link_count(inode); | |
4937 | iput(inode); | |
4938 | } | |
d3c2fdcf | 4939 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4940 | return err; |
4941 | } | |
4942 | ||
4943 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4944 | struct dentry *dentry) | |
4945 | { | |
4946 | struct btrfs_trans_handle *trans; | |
4947 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4948 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4949 | u64 index; |
1832a6d5 | 4950 | unsigned long nr = 0; |
39279cc3 CM |
4951 | int err; |
4952 | int drop_inode = 0; | |
4953 | ||
4a8be425 TH |
4954 | /* do not allow sys_link's with other subvols of the same device */ |
4955 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4956 | return -EXDEV; |
4a8be425 | 4957 | |
c055e99e AV |
4958 | if (inode->i_nlink == ~0U) |
4959 | return -EMLINK; | |
4a8be425 | 4960 | |
3de4586c | 4961 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4962 | if (err) |
4963 | goto fail; | |
4964 | ||
a22285a6 | 4965 | /* |
7e6b6465 | 4966 | * 2 items for inode and inode ref |
a22285a6 | 4967 | * 2 items for dir items |
7e6b6465 | 4968 | * 1 item for parent inode |
a22285a6 | 4969 | */ |
7e6b6465 | 4970 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4971 | if (IS_ERR(trans)) { |
4972 | err = PTR_ERR(trans); | |
4973 | goto fail; | |
4974 | } | |
5f39d397 | 4975 | |
3153495d | 4976 | btrfs_inc_nlink(inode); |
0c4d2d95 | 4977 | inode_inc_iversion(inode); |
3153495d | 4978 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 4979 | ihold(inode); |
aec7477b | 4980 | |
a1b075d2 | 4981 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4982 | |
a5719521 | 4983 | if (err) { |
54aa1f4d | 4984 | drop_inode = 1; |
a5719521 | 4985 | } else { |
10d9f309 | 4986 | struct dentry *parent = dentry->d_parent; |
a5719521 | 4987 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
4988 | if (err) |
4989 | goto fail; | |
08c422c2 | 4990 | d_instantiate(dentry, inode); |
6a912213 | 4991 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 4992 | } |
39279cc3 | 4993 | |
d3c2fdcf | 4994 | nr = trans->blocks_used; |
7ad85bb7 | 4995 | btrfs_end_transaction(trans, root); |
1832a6d5 | 4996 | fail: |
39279cc3 CM |
4997 | if (drop_inode) { |
4998 | inode_dec_link_count(inode); | |
4999 | iput(inode); | |
5000 | } | |
d3c2fdcf | 5001 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5002 | return err; |
5003 | } | |
5004 | ||
18bb1db3 | 5005 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5006 | { |
b9d86667 | 5007 | struct inode *inode = NULL; |
39279cc3 CM |
5008 | struct btrfs_trans_handle *trans; |
5009 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5010 | int err = 0; | |
5011 | int drop_on_err = 0; | |
b9d86667 | 5012 | u64 objectid = 0; |
00e4e6b3 | 5013 | u64 index = 0; |
d3c2fdcf | 5014 | unsigned long nr = 1; |
39279cc3 | 5015 | |
9ed74f2d JB |
5016 | /* |
5017 | * 2 items for inode and ref | |
5018 | * 2 items for dir items | |
5019 | * 1 for xattr if selinux is on | |
5020 | */ | |
a22285a6 YZ |
5021 | trans = btrfs_start_transaction(root, 5); |
5022 | if (IS_ERR(trans)) | |
5023 | return PTR_ERR(trans); | |
39279cc3 | 5024 | |
581bb050 LZ |
5025 | err = btrfs_find_free_ino(root, &objectid); |
5026 | if (err) | |
5027 | goto out_fail; | |
5028 | ||
aec7477b | 5029 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5030 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5031 | S_IFDIR | mode, &index); |
39279cc3 CM |
5032 | if (IS_ERR(inode)) { |
5033 | err = PTR_ERR(inode); | |
5034 | goto out_fail; | |
5035 | } | |
5f39d397 | 5036 | |
39279cc3 | 5037 | drop_on_err = 1; |
33268eaf | 5038 | |
2a7dba39 | 5039 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5040 | if (err) |
5041 | goto out_fail; | |
5042 | ||
39279cc3 CM |
5043 | inode->i_op = &btrfs_dir_inode_operations; |
5044 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5045 | |
dbe674a9 | 5046 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5047 | err = btrfs_update_inode(trans, root, inode); |
5048 | if (err) | |
5049 | goto out_fail; | |
5f39d397 | 5050 | |
a1b075d2 JB |
5051 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5052 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5053 | if (err) |
5054 | goto out_fail; | |
5f39d397 | 5055 | |
39279cc3 CM |
5056 | d_instantiate(dentry, inode); |
5057 | drop_on_err = 0; | |
39279cc3 CM |
5058 | |
5059 | out_fail: | |
d3c2fdcf | 5060 | nr = trans->blocks_used; |
7ad85bb7 | 5061 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5062 | if (drop_on_err) |
5063 | iput(inode); | |
d3c2fdcf | 5064 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5065 | return err; |
5066 | } | |
5067 | ||
d352ac68 CM |
5068 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5069 | * and an extent that you want to insert, deal with overlap and insert | |
5070 | * the new extent into the tree. | |
5071 | */ | |
3b951516 CM |
5072 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5073 | struct extent_map *existing, | |
e6dcd2dc CM |
5074 | struct extent_map *em, |
5075 | u64 map_start, u64 map_len) | |
3b951516 CM |
5076 | { |
5077 | u64 start_diff; | |
3b951516 | 5078 | |
e6dcd2dc CM |
5079 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5080 | start_diff = map_start - em->start; | |
5081 | em->start = map_start; | |
5082 | em->len = map_len; | |
c8b97818 CM |
5083 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5084 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5085 | em->block_start += start_diff; |
c8b97818 CM |
5086 | em->block_len -= start_diff; |
5087 | } | |
e6dcd2dc | 5088 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5089 | } |
5090 | ||
c8b97818 CM |
5091 | static noinline int uncompress_inline(struct btrfs_path *path, |
5092 | struct inode *inode, struct page *page, | |
5093 | size_t pg_offset, u64 extent_offset, | |
5094 | struct btrfs_file_extent_item *item) | |
5095 | { | |
5096 | int ret; | |
5097 | struct extent_buffer *leaf = path->nodes[0]; | |
5098 | char *tmp; | |
5099 | size_t max_size; | |
5100 | unsigned long inline_size; | |
5101 | unsigned long ptr; | |
261507a0 | 5102 | int compress_type; |
c8b97818 CM |
5103 | |
5104 | WARN_ON(pg_offset != 0); | |
261507a0 | 5105 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5106 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5107 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5108 | btrfs_item_nr(leaf, path->slots[0])); | |
5109 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5110 | if (!tmp) |
5111 | return -ENOMEM; | |
c8b97818 CM |
5112 | ptr = btrfs_file_extent_inline_start(item); |
5113 | ||
5114 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5115 | ||
5b050f04 | 5116 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5117 | ret = btrfs_decompress(compress_type, tmp, page, |
5118 | extent_offset, inline_size, max_size); | |
c8b97818 | 5119 | if (ret) { |
7ac687d9 | 5120 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5121 | unsigned long copy_size = min_t(u64, |
5122 | PAGE_CACHE_SIZE - pg_offset, | |
5123 | max_size - extent_offset); | |
5124 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5125 | kunmap_atomic(kaddr); |
c8b97818 CM |
5126 | } |
5127 | kfree(tmp); | |
5128 | return 0; | |
5129 | } | |
5130 | ||
d352ac68 CM |
5131 | /* |
5132 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5133 | * the ugly parts come from merging extents from the disk with the in-ram |
5134 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5135 | * where the in-ram extents might be locked pending data=ordered completion. |
5136 | * | |
5137 | * This also copies inline extents directly into the page. | |
5138 | */ | |
d397712b | 5139 | |
a52d9a80 | 5140 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5141 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5142 | int create) |
5143 | { | |
5144 | int ret; | |
5145 | int err = 0; | |
db94535d | 5146 | u64 bytenr; |
a52d9a80 CM |
5147 | u64 extent_start = 0; |
5148 | u64 extent_end = 0; | |
33345d01 | 5149 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5150 | u32 found_type; |
f421950f | 5151 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5152 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5153 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5154 | struct extent_buffer *leaf; |
5155 | struct btrfs_key found_key; | |
a52d9a80 CM |
5156 | struct extent_map *em = NULL; |
5157 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5158 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5159 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5160 | int compress_type; |
a52d9a80 | 5161 | |
a52d9a80 | 5162 | again: |
890871be | 5163 | read_lock(&em_tree->lock); |
d1310b2e | 5164 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5165 | if (em) |
5166 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5167 | read_unlock(&em_tree->lock); |
d1310b2e | 5168 | |
a52d9a80 | 5169 | if (em) { |
e1c4b745 CM |
5170 | if (em->start > start || em->start + em->len <= start) |
5171 | free_extent_map(em); | |
5172 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5173 | free_extent_map(em); |
5174 | else | |
5175 | goto out; | |
a52d9a80 | 5176 | } |
172ddd60 | 5177 | em = alloc_extent_map(); |
a52d9a80 | 5178 | if (!em) { |
d1310b2e CM |
5179 | err = -ENOMEM; |
5180 | goto out; | |
a52d9a80 | 5181 | } |
e6dcd2dc | 5182 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5183 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5184 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5185 | em->len = (u64)-1; |
c8b97818 | 5186 | em->block_len = (u64)-1; |
f421950f CM |
5187 | |
5188 | if (!path) { | |
5189 | path = btrfs_alloc_path(); | |
026fd317 JB |
5190 | if (!path) { |
5191 | err = -ENOMEM; | |
5192 | goto out; | |
5193 | } | |
5194 | /* | |
5195 | * Chances are we'll be called again, so go ahead and do | |
5196 | * readahead | |
5197 | */ | |
5198 | path->reada = 1; | |
f421950f CM |
5199 | } |
5200 | ||
179e29e4 CM |
5201 | ret = btrfs_lookup_file_extent(trans, root, path, |
5202 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5203 | if (ret < 0) { |
5204 | err = ret; | |
5205 | goto out; | |
5206 | } | |
5207 | ||
5208 | if (ret != 0) { | |
5209 | if (path->slots[0] == 0) | |
5210 | goto not_found; | |
5211 | path->slots[0]--; | |
5212 | } | |
5213 | ||
5f39d397 CM |
5214 | leaf = path->nodes[0]; |
5215 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5216 | struct btrfs_file_extent_item); |
a52d9a80 | 5217 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5218 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5219 | found_type = btrfs_key_type(&found_key); | |
5220 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5221 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5222 | goto not_found; | |
5223 | } | |
5224 | ||
5f39d397 CM |
5225 | found_type = btrfs_file_extent_type(leaf, item); |
5226 | extent_start = found_key.offset; | |
261507a0 | 5227 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5228 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5229 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5230 | extent_end = extent_start + |
db94535d | 5231 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5232 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5233 | size_t size; | |
5234 | size = btrfs_file_extent_inline_len(leaf, item); | |
5235 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5236 | ~((u64)root->sectorsize - 1); | |
5237 | } | |
5238 | ||
5239 | if (start >= extent_end) { | |
5240 | path->slots[0]++; | |
5241 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5242 | ret = btrfs_next_leaf(root, path); | |
5243 | if (ret < 0) { | |
5244 | err = ret; | |
5245 | goto out; | |
a52d9a80 | 5246 | } |
9036c102 YZ |
5247 | if (ret > 0) |
5248 | goto not_found; | |
5249 | leaf = path->nodes[0]; | |
a52d9a80 | 5250 | } |
9036c102 YZ |
5251 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5252 | if (found_key.objectid != objectid || | |
5253 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5254 | goto not_found; | |
5255 | if (start + len <= found_key.offset) | |
5256 | goto not_found; | |
5257 | em->start = start; | |
5258 | em->len = found_key.offset - start; | |
5259 | goto not_found_em; | |
5260 | } | |
5261 | ||
d899e052 YZ |
5262 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5263 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5264 | em->start = extent_start; |
5265 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5266 | em->orig_start = extent_start - |
5267 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5268 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5269 | if (bytenr == 0) { | |
5f39d397 | 5270 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5271 | goto insert; |
5272 | } | |
261507a0 | 5273 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5274 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5275 | em->compress_type = compress_type; |
c8b97818 CM |
5276 | em->block_start = bytenr; |
5277 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5278 | item); | |
5279 | } else { | |
5280 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5281 | em->block_start = bytenr; | |
5282 | em->block_len = em->len; | |
d899e052 YZ |
5283 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5284 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5285 | } |
a52d9a80 CM |
5286 | goto insert; |
5287 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5288 | unsigned long ptr; |
a52d9a80 | 5289 | char *map; |
3326d1b0 CM |
5290 | size_t size; |
5291 | size_t extent_offset; | |
5292 | size_t copy_size; | |
a52d9a80 | 5293 | |
689f9346 | 5294 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5295 | if (!page || create) { |
689f9346 | 5296 | em->start = extent_start; |
9036c102 | 5297 | em->len = extent_end - extent_start; |
689f9346 Y |
5298 | goto out; |
5299 | } | |
5f39d397 | 5300 | |
9036c102 YZ |
5301 | size = btrfs_file_extent_inline_len(leaf, item); |
5302 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5303 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5304 | size - extent_offset); |
3326d1b0 | 5305 | em->start = extent_start + extent_offset; |
70dec807 CM |
5306 | em->len = (copy_size + root->sectorsize - 1) & |
5307 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5308 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5309 | if (compress_type) { |
c8b97818 | 5310 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5311 | em->compress_type = compress_type; |
5312 | } | |
689f9346 | 5313 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5314 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5315 | if (btrfs_file_extent_compression(leaf, item) != |
5316 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5317 | ret = uncompress_inline(path, inode, page, |
5318 | pg_offset, | |
5319 | extent_offset, item); | |
79787eaa | 5320 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5321 | } else { |
5322 | map = kmap(page); | |
5323 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5324 | copy_size); | |
93c82d57 CM |
5325 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5326 | memset(map + pg_offset + copy_size, 0, | |
5327 | PAGE_CACHE_SIZE - pg_offset - | |
5328 | copy_size); | |
5329 | } | |
c8b97818 CM |
5330 | kunmap(page); |
5331 | } | |
179e29e4 CM |
5332 | flush_dcache_page(page); |
5333 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5334 | BUG(); |
179e29e4 CM |
5335 | if (!trans) { |
5336 | kunmap(page); | |
5337 | free_extent_map(em); | |
5338 | em = NULL; | |
ff5714cc | 5339 | |
b3b4aa74 | 5340 | btrfs_release_path(path); |
7a7eaa40 | 5341 | trans = btrfs_join_transaction(root); |
ff5714cc | 5342 | |
3612b495 TI |
5343 | if (IS_ERR(trans)) |
5344 | return ERR_CAST(trans); | |
179e29e4 CM |
5345 | goto again; |
5346 | } | |
c8b97818 | 5347 | map = kmap(page); |
70dec807 | 5348 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5349 | copy_size); |
c8b97818 | 5350 | kunmap(page); |
179e29e4 | 5351 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5352 | } |
d1310b2e | 5353 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5354 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5355 | goto insert; |
5356 | } else { | |
d397712b | 5357 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5358 | WARN_ON(1); |
5359 | } | |
5360 | not_found: | |
5361 | em->start = start; | |
d1310b2e | 5362 | em->len = len; |
a52d9a80 | 5363 | not_found_em: |
5f39d397 | 5364 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5365 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5366 | insert: |
b3b4aa74 | 5367 | btrfs_release_path(path); |
d1310b2e | 5368 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5369 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5370 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5371 | (unsigned long long)em->len, | |
5372 | (unsigned long long)start, | |
5373 | (unsigned long long)len); | |
a52d9a80 CM |
5374 | err = -EIO; |
5375 | goto out; | |
5376 | } | |
d1310b2e CM |
5377 | |
5378 | err = 0; | |
890871be | 5379 | write_lock(&em_tree->lock); |
a52d9a80 | 5380 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5381 | /* it is possible that someone inserted the extent into the tree |
5382 | * while we had the lock dropped. It is also possible that | |
5383 | * an overlapping map exists in the tree | |
5384 | */ | |
a52d9a80 | 5385 | if (ret == -EEXIST) { |
3b951516 | 5386 | struct extent_map *existing; |
e6dcd2dc CM |
5387 | |
5388 | ret = 0; | |
5389 | ||
3b951516 | 5390 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5391 | if (existing && (existing->start > start || |
5392 | existing->start + existing->len <= start)) { | |
5393 | free_extent_map(existing); | |
5394 | existing = NULL; | |
5395 | } | |
3b951516 CM |
5396 | if (!existing) { |
5397 | existing = lookup_extent_mapping(em_tree, em->start, | |
5398 | em->len); | |
5399 | if (existing) { | |
5400 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5401 | em, start, |
5402 | root->sectorsize); | |
3b951516 CM |
5403 | free_extent_map(existing); |
5404 | if (err) { | |
5405 | free_extent_map(em); | |
5406 | em = NULL; | |
5407 | } | |
5408 | } else { | |
5409 | err = -EIO; | |
3b951516 CM |
5410 | free_extent_map(em); |
5411 | em = NULL; | |
5412 | } | |
5413 | } else { | |
5414 | free_extent_map(em); | |
5415 | em = existing; | |
e6dcd2dc | 5416 | err = 0; |
a52d9a80 | 5417 | } |
a52d9a80 | 5418 | } |
890871be | 5419 | write_unlock(&em_tree->lock); |
a52d9a80 | 5420 | out: |
1abe9b8a | 5421 | |
5422 | trace_btrfs_get_extent(root, em); | |
5423 | ||
f421950f CM |
5424 | if (path) |
5425 | btrfs_free_path(path); | |
a52d9a80 CM |
5426 | if (trans) { |
5427 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5428 | if (!err) |
a52d9a80 CM |
5429 | err = ret; |
5430 | } | |
a52d9a80 CM |
5431 | if (err) { |
5432 | free_extent_map(em); | |
a52d9a80 CM |
5433 | return ERR_PTR(err); |
5434 | } | |
79787eaa | 5435 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5436 | return em; |
5437 | } | |
5438 | ||
ec29ed5b CM |
5439 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5440 | size_t pg_offset, u64 start, u64 len, | |
5441 | int create) | |
5442 | { | |
5443 | struct extent_map *em; | |
5444 | struct extent_map *hole_em = NULL; | |
5445 | u64 range_start = start; | |
5446 | u64 end; | |
5447 | u64 found; | |
5448 | u64 found_end; | |
5449 | int err = 0; | |
5450 | ||
5451 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5452 | if (IS_ERR(em)) | |
5453 | return em; | |
5454 | if (em) { | |
5455 | /* | |
5456 | * if our em maps to a hole, there might | |
5457 | * actually be delalloc bytes behind it | |
5458 | */ | |
5459 | if (em->block_start != EXTENT_MAP_HOLE) | |
5460 | return em; | |
5461 | else | |
5462 | hole_em = em; | |
5463 | } | |
5464 | ||
5465 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5466 | end = start + len; | |
5467 | if (end < start) | |
5468 | end = (u64)-1; | |
5469 | else | |
5470 | end -= 1; | |
5471 | ||
5472 | em = NULL; | |
5473 | ||
5474 | /* ok, we didn't find anything, lets look for delalloc */ | |
5475 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5476 | end, len, EXTENT_DELALLOC, 1); | |
5477 | found_end = range_start + found; | |
5478 | if (found_end < range_start) | |
5479 | found_end = (u64)-1; | |
5480 | ||
5481 | /* | |
5482 | * we didn't find anything useful, return | |
5483 | * the original results from get_extent() | |
5484 | */ | |
5485 | if (range_start > end || found_end <= start) { | |
5486 | em = hole_em; | |
5487 | hole_em = NULL; | |
5488 | goto out; | |
5489 | } | |
5490 | ||
5491 | /* adjust the range_start to make sure it doesn't | |
5492 | * go backwards from the start they passed in | |
5493 | */ | |
5494 | range_start = max(start,range_start); | |
5495 | found = found_end - range_start; | |
5496 | ||
5497 | if (found > 0) { | |
5498 | u64 hole_start = start; | |
5499 | u64 hole_len = len; | |
5500 | ||
172ddd60 | 5501 | em = alloc_extent_map(); |
ec29ed5b CM |
5502 | if (!em) { |
5503 | err = -ENOMEM; | |
5504 | goto out; | |
5505 | } | |
5506 | /* | |
5507 | * when btrfs_get_extent can't find anything it | |
5508 | * returns one huge hole | |
5509 | * | |
5510 | * make sure what it found really fits our range, and | |
5511 | * adjust to make sure it is based on the start from | |
5512 | * the caller | |
5513 | */ | |
5514 | if (hole_em) { | |
5515 | u64 calc_end = extent_map_end(hole_em); | |
5516 | ||
5517 | if (calc_end <= start || (hole_em->start > end)) { | |
5518 | free_extent_map(hole_em); | |
5519 | hole_em = NULL; | |
5520 | } else { | |
5521 | hole_start = max(hole_em->start, start); | |
5522 | hole_len = calc_end - hole_start; | |
5523 | } | |
5524 | } | |
5525 | em->bdev = NULL; | |
5526 | if (hole_em && range_start > hole_start) { | |
5527 | /* our hole starts before our delalloc, so we | |
5528 | * have to return just the parts of the hole | |
5529 | * that go until the delalloc starts | |
5530 | */ | |
5531 | em->len = min(hole_len, | |
5532 | range_start - hole_start); | |
5533 | em->start = hole_start; | |
5534 | em->orig_start = hole_start; | |
5535 | /* | |
5536 | * don't adjust block start at all, | |
5537 | * it is fixed at EXTENT_MAP_HOLE | |
5538 | */ | |
5539 | em->block_start = hole_em->block_start; | |
5540 | em->block_len = hole_len; | |
5541 | } else { | |
5542 | em->start = range_start; | |
5543 | em->len = found; | |
5544 | em->orig_start = range_start; | |
5545 | em->block_start = EXTENT_MAP_DELALLOC; | |
5546 | em->block_len = found; | |
5547 | } | |
5548 | } else if (hole_em) { | |
5549 | return hole_em; | |
5550 | } | |
5551 | out: | |
5552 | ||
5553 | free_extent_map(hole_em); | |
5554 | if (err) { | |
5555 | free_extent_map(em); | |
5556 | return ERR_PTR(err); | |
5557 | } | |
5558 | return em; | |
5559 | } | |
5560 | ||
4b46fce2 | 5561 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5562 | struct extent_map *em, |
4b46fce2 JB |
5563 | u64 start, u64 len) |
5564 | { | |
5565 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5566 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5567 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5568 | struct btrfs_key ins; | |
5569 | u64 alloc_hint; | |
5570 | int ret; | |
16d299ac | 5571 | bool insert = false; |
4b46fce2 | 5572 | |
16d299ac JB |
5573 | /* |
5574 | * Ok if the extent map we looked up is a hole and is for the exact | |
5575 | * range we want, there is no reason to allocate a new one, however if | |
5576 | * it is not right then we need to free this one and drop the cache for | |
5577 | * our range. | |
5578 | */ | |
5579 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5580 | em->len != len) { | |
5581 | free_extent_map(em); | |
5582 | em = NULL; | |
5583 | insert = true; | |
5584 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5585 | } | |
4b46fce2 | 5586 | |
7a7eaa40 | 5587 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5588 | if (IS_ERR(trans)) |
5589 | return ERR_CAST(trans); | |
4b46fce2 | 5590 | |
4cb5300b CM |
5591 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5592 | btrfs_add_inode_defrag(trans, inode); | |
5593 | ||
4b46fce2 JB |
5594 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5595 | ||
5596 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5597 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5598 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5599 | if (ret) { |
5600 | em = ERR_PTR(ret); | |
5601 | goto out; | |
5602 | } | |
5603 | ||
4b46fce2 | 5604 | if (!em) { |
172ddd60 | 5605 | em = alloc_extent_map(); |
16d299ac JB |
5606 | if (!em) { |
5607 | em = ERR_PTR(-ENOMEM); | |
5608 | goto out; | |
5609 | } | |
4b46fce2 JB |
5610 | } |
5611 | ||
5612 | em->start = start; | |
5613 | em->orig_start = em->start; | |
5614 | em->len = ins.offset; | |
5615 | ||
5616 | em->block_start = ins.objectid; | |
5617 | em->block_len = ins.offset; | |
5618 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5619 | |
5620 | /* | |
5621 | * We need to do this because if we're using the original em we searched | |
5622 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5623 | */ | |
5624 | em->flags = 0; | |
4b46fce2 JB |
5625 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5626 | ||
16d299ac | 5627 | while (insert) { |
4b46fce2 JB |
5628 | write_lock(&em_tree->lock); |
5629 | ret = add_extent_mapping(em_tree, em); | |
5630 | write_unlock(&em_tree->lock); | |
5631 | if (ret != -EEXIST) | |
5632 | break; | |
5633 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5634 | } | |
5635 | ||
5636 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5637 | ins.offset, ins.offset, 0); | |
5638 | if (ret) { | |
5639 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5640 | em = ERR_PTR(ret); | |
5641 | } | |
5642 | out: | |
5643 | btrfs_end_transaction(trans, root); | |
5644 | return em; | |
5645 | } | |
5646 | ||
46bfbb5c CM |
5647 | /* |
5648 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5649 | * block must be cow'd | |
5650 | */ | |
5651 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5652 | struct inode *inode, u64 offset, u64 len) | |
5653 | { | |
5654 | struct btrfs_path *path; | |
5655 | int ret; | |
5656 | struct extent_buffer *leaf; | |
5657 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5658 | struct btrfs_file_extent_item *fi; | |
5659 | struct btrfs_key key; | |
5660 | u64 disk_bytenr; | |
5661 | u64 backref_offset; | |
5662 | u64 extent_end; | |
5663 | u64 num_bytes; | |
5664 | int slot; | |
5665 | int found_type; | |
5666 | ||
5667 | path = btrfs_alloc_path(); | |
5668 | if (!path) | |
5669 | return -ENOMEM; | |
5670 | ||
33345d01 | 5671 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5672 | offset, 0); |
5673 | if (ret < 0) | |
5674 | goto out; | |
5675 | ||
5676 | slot = path->slots[0]; | |
5677 | if (ret == 1) { | |
5678 | if (slot == 0) { | |
5679 | /* can't find the item, must cow */ | |
5680 | ret = 0; | |
5681 | goto out; | |
5682 | } | |
5683 | slot--; | |
5684 | } | |
5685 | ret = 0; | |
5686 | leaf = path->nodes[0]; | |
5687 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5688 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5689 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5690 | /* not our file or wrong item type, must cow */ | |
5691 | goto out; | |
5692 | } | |
5693 | ||
5694 | if (key.offset > offset) { | |
5695 | /* Wrong offset, must cow */ | |
5696 | goto out; | |
5697 | } | |
5698 | ||
5699 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5700 | found_type = btrfs_file_extent_type(leaf, fi); | |
5701 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5702 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5703 | /* not a regular extent, must cow */ | |
5704 | goto out; | |
5705 | } | |
5706 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5707 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5708 | ||
5709 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5710 | if (extent_end < offset + len) { | |
5711 | /* extent doesn't include our full range, must cow */ | |
5712 | goto out; | |
5713 | } | |
5714 | ||
5715 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5716 | goto out; | |
5717 | ||
5718 | /* | |
5719 | * look for other files referencing this extent, if we | |
5720 | * find any we must cow | |
5721 | */ | |
33345d01 | 5722 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5723 | key.offset - backref_offset, disk_bytenr)) |
5724 | goto out; | |
5725 | ||
5726 | /* | |
5727 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5728 | * in this extent we are about to write. If there | |
5729 | * are any csums in that range we have to cow in order | |
5730 | * to keep the csums correct | |
5731 | */ | |
5732 | disk_bytenr += backref_offset; | |
5733 | disk_bytenr += offset - key.offset; | |
5734 | num_bytes = min(offset + len, extent_end) - offset; | |
5735 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5736 | goto out; | |
5737 | /* | |
5738 | * all of the above have passed, it is safe to overwrite this extent | |
5739 | * without cow | |
5740 | */ | |
5741 | ret = 1; | |
5742 | out: | |
5743 | btrfs_free_path(path); | |
5744 | return ret; | |
5745 | } | |
5746 | ||
4b46fce2 JB |
5747 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5748 | struct buffer_head *bh_result, int create) | |
5749 | { | |
5750 | struct extent_map *em; | |
5751 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5752 | u64 start = iblock << inode->i_blkbits; | |
5753 | u64 len = bh_result->b_size; | |
46bfbb5c | 5754 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5755 | |
5756 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5757 | if (IS_ERR(em)) | |
5758 | return PTR_ERR(em); | |
5759 | ||
5760 | /* | |
5761 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5762 | * io. INLINE is special, and we could probably kludge it in here, but | |
5763 | * it's still buffered so for safety lets just fall back to the generic | |
5764 | * buffered path. | |
5765 | * | |
5766 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5767 | * decompress it, so there will be buffering required no matter what we | |
5768 | * do, so go ahead and fallback to buffered. | |
5769 | * | |
5770 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5771 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5772 | * the generic code. | |
5773 | */ | |
5774 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5775 | em->block_start == EXTENT_MAP_INLINE) { | |
5776 | free_extent_map(em); | |
5777 | return -ENOTBLK; | |
5778 | } | |
5779 | ||
5780 | /* Just a good old fashioned hole, return */ | |
5781 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5782 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5783 | free_extent_map(em); | |
5784 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5785 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
d0082371 | 5786 | start + root->sectorsize - 1); |
4b46fce2 JB |
5787 | return 0; |
5788 | } | |
5789 | ||
5790 | /* | |
5791 | * We don't allocate a new extent in the following cases | |
5792 | * | |
5793 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5794 | * existing extent. | |
5795 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5796 | * just use the extent. | |
5797 | * | |
5798 | */ | |
46bfbb5c CM |
5799 | if (!create) { |
5800 | len = em->len - (start - em->start); | |
4b46fce2 | 5801 | goto map; |
46bfbb5c | 5802 | } |
4b46fce2 JB |
5803 | |
5804 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5805 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5806 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5807 | int type; |
5808 | int ret; | |
46bfbb5c | 5809 | u64 block_start; |
4b46fce2 JB |
5810 | |
5811 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5812 | type = BTRFS_ORDERED_PREALLOC; | |
5813 | else | |
5814 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5815 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5816 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5817 | |
5818 | /* | |
5819 | * we're not going to log anything, but we do need | |
5820 | * to make sure the current transaction stays open | |
5821 | * while we look for nocow cross refs | |
5822 | */ | |
7a7eaa40 | 5823 | trans = btrfs_join_transaction(root); |
3612b495 | 5824 | if (IS_ERR(trans)) |
46bfbb5c CM |
5825 | goto must_cow; |
5826 | ||
5827 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5828 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5829 | block_start, len, len, type); | |
5830 | btrfs_end_transaction(trans, root); | |
5831 | if (ret) { | |
5832 | free_extent_map(em); | |
5833 | return ret; | |
5834 | } | |
5835 | goto unlock; | |
4b46fce2 | 5836 | } |
46bfbb5c | 5837 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5838 | } |
46bfbb5c CM |
5839 | must_cow: |
5840 | /* | |
5841 | * this will cow the extent, reset the len in case we changed | |
5842 | * it above | |
5843 | */ | |
5844 | len = bh_result->b_size; | |
16d299ac | 5845 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5846 | if (IS_ERR(em)) |
5847 | return PTR_ERR(em); | |
5848 | len = min(len, em->len - (start - em->start)); | |
5849 | unlock: | |
4845e44f CM |
5850 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5851 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5852 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5853 | map: |
5854 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5855 | inode->i_blkbits; | |
46bfbb5c | 5856 | bh_result->b_size = len; |
4b46fce2 JB |
5857 | bh_result->b_bdev = em->bdev; |
5858 | set_buffer_mapped(bh_result); | |
5859 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5860 | set_buffer_new(bh_result); | |
5861 | ||
5862 | free_extent_map(em); | |
5863 | ||
5864 | return 0; | |
5865 | } | |
5866 | ||
5867 | struct btrfs_dio_private { | |
5868 | struct inode *inode; | |
5869 | u64 logical_offset; | |
5870 | u64 disk_bytenr; | |
5871 | u64 bytes; | |
5872 | u32 *csums; | |
5873 | void *private; | |
e65e1535 MX |
5874 | |
5875 | /* number of bios pending for this dio */ | |
5876 | atomic_t pending_bios; | |
5877 | ||
5878 | /* IO errors */ | |
5879 | int errors; | |
5880 | ||
5881 | struct bio *orig_bio; | |
4b46fce2 JB |
5882 | }; |
5883 | ||
5884 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5885 | { | |
e65e1535 | 5886 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5887 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5888 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5889 | struct inode *inode = dip->inode; |
5890 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5891 | u64 start; | |
5892 | u32 *private = dip->csums; | |
5893 | ||
5894 | start = dip->logical_offset; | |
5895 | do { | |
5896 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5897 | struct page *page = bvec->bv_page; | |
5898 | char *kaddr; | |
5899 | u32 csum = ~(u32)0; | |
5900 | unsigned long flags; | |
5901 | ||
5902 | local_irq_save(flags); | |
7ac687d9 | 5903 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
5904 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
5905 | csum, bvec->bv_len); | |
5906 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 5907 | kunmap_atomic(kaddr); |
4b46fce2 JB |
5908 | local_irq_restore(flags); |
5909 | ||
5910 | flush_dcache_page(bvec->bv_page); | |
5911 | if (csum != *private) { | |
33345d01 | 5912 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 5913 | " %llu csum %u private %u\n", |
33345d01 LZ |
5914 | (unsigned long long)btrfs_ino(inode), |
5915 | (unsigned long long)start, | |
4b46fce2 JB |
5916 | csum, *private); |
5917 | err = -EIO; | |
5918 | } | |
5919 | } | |
5920 | ||
5921 | start += bvec->bv_len; | |
5922 | private++; | |
5923 | bvec++; | |
5924 | } while (bvec <= bvec_end); | |
5925 | ||
5926 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 5927 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
5928 | bio->bi_private = dip->private; |
5929 | ||
5930 | kfree(dip->csums); | |
5931 | kfree(dip); | |
c0da7aa1 JB |
5932 | |
5933 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5934 | if (err) | |
5935 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5936 | dio_end_io(bio, err); |
5937 | } | |
5938 | ||
5939 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5940 | { | |
5941 | struct btrfs_dio_private *dip = bio->bi_private; | |
5942 | struct inode *inode = dip->inode; | |
5943 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 5944 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
5945 | u64 ordered_offset = dip->logical_offset; |
5946 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5947 | int ret; |
5948 | ||
5949 | if (err) | |
5950 | goto out_done; | |
163cf09c CM |
5951 | again: |
5952 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5953 | &ordered_offset, | |
5fd02043 | 5954 | ordered_bytes, !err); |
4b46fce2 | 5955 | if (!ret) |
163cf09c | 5956 | goto out_test; |
4b46fce2 | 5957 | |
5fd02043 JB |
5958 | ordered->work.func = finish_ordered_fn; |
5959 | ordered->work.flags = 0; | |
5960 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
5961 | &ordered->work); | |
163cf09c CM |
5962 | out_test: |
5963 | /* | |
5964 | * our bio might span multiple ordered extents. If we haven't | |
5965 | * completed the accounting for the whole dio, go back and try again | |
5966 | */ | |
5967 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5968 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5969 | ordered_offset; | |
5fd02043 | 5970 | ordered = NULL; |
163cf09c CM |
5971 | goto again; |
5972 | } | |
4b46fce2 JB |
5973 | out_done: |
5974 | bio->bi_private = dip->private; | |
5975 | ||
4b46fce2 | 5976 | kfree(dip); |
c0da7aa1 JB |
5977 | |
5978 | /* If we had an error make sure to clear the uptodate flag */ | |
5979 | if (err) | |
5980 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5981 | dio_end_io(bio, err); |
5982 | } | |
5983 | ||
eaf25d93 CM |
5984 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5985 | struct bio *bio, int mirror_num, | |
5986 | unsigned long bio_flags, u64 offset) | |
5987 | { | |
5988 | int ret; | |
5989 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5990 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 5991 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
5992 | return 0; |
5993 | } | |
5994 | ||
e65e1535 MX |
5995 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5996 | { | |
5997 | struct btrfs_dio_private *dip = bio->bi_private; | |
5998 | ||
5999 | if (err) { | |
33345d01 | 6000 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6001 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6002 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6003 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6004 | dip->errors = 1; |
6005 | ||
6006 | /* | |
6007 | * before atomic variable goto zero, we must make sure | |
6008 | * dip->errors is perceived to be set. | |
6009 | */ | |
6010 | smp_mb__before_atomic_dec(); | |
6011 | } | |
6012 | ||
6013 | /* if there are more bios still pending for this dio, just exit */ | |
6014 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6015 | goto out; | |
6016 | ||
6017 | if (dip->errors) | |
6018 | bio_io_error(dip->orig_bio); | |
6019 | else { | |
6020 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6021 | bio_endio(dip->orig_bio, 0); | |
6022 | } | |
6023 | out: | |
6024 | bio_put(bio); | |
6025 | } | |
6026 | ||
6027 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6028 | u64 first_sector, gfp_t gfp_flags) | |
6029 | { | |
6030 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6031 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6032 | } | |
6033 | ||
6034 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6035 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 6036 | u32 *csums, int async_submit) |
e65e1535 MX |
6037 | { |
6038 | int write = rw & REQ_WRITE; | |
6039 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6040 | int ret; | |
6041 | ||
6042 | bio_get(bio); | |
5fd02043 JB |
6043 | |
6044 | if (!write) { | |
6045 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6046 | if (ret) | |
6047 | goto err; | |
6048 | } | |
e65e1535 | 6049 | |
1ae39938 JB |
6050 | if (skip_sum) |
6051 | goto map; | |
6052 | ||
6053 | if (write && async_submit) { | |
e65e1535 MX |
6054 | ret = btrfs_wq_submit_bio(root->fs_info, |
6055 | inode, rw, bio, 0, 0, | |
6056 | file_offset, | |
6057 | __btrfs_submit_bio_start_direct_io, | |
6058 | __btrfs_submit_bio_done); | |
6059 | goto err; | |
1ae39938 JB |
6060 | } else if (write) { |
6061 | /* | |
6062 | * If we aren't doing async submit, calculate the csum of the | |
6063 | * bio now. | |
6064 | */ | |
6065 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6066 | if (ret) | |
6067 | goto err; | |
c2db1073 TI |
6068 | } else if (!skip_sum) { |
6069 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 6070 | file_offset, csums); |
c2db1073 TI |
6071 | if (ret) |
6072 | goto err; | |
6073 | } | |
e65e1535 | 6074 | |
1ae39938 JB |
6075 | map: |
6076 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6077 | err: |
6078 | bio_put(bio); | |
6079 | return ret; | |
6080 | } | |
6081 | ||
6082 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6083 | int skip_sum) | |
6084 | { | |
6085 | struct inode *inode = dip->inode; | |
6086 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6087 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6088 | struct bio *bio; | |
6089 | struct bio *orig_bio = dip->orig_bio; | |
6090 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6091 | u64 start_sector = orig_bio->bi_sector; | |
6092 | u64 file_offset = dip->logical_offset; | |
6093 | u64 submit_len = 0; | |
6094 | u64 map_length; | |
6095 | int nr_pages = 0; | |
6096 | u32 *csums = dip->csums; | |
6097 | int ret = 0; | |
1ae39938 | 6098 | int async_submit = 0; |
98bc3149 | 6099 | int write = rw & REQ_WRITE; |
e65e1535 | 6100 | |
e65e1535 MX |
6101 | map_length = orig_bio->bi_size; |
6102 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6103 | &map_length, NULL, 0); | |
6104 | if (ret) { | |
64728bbb | 6105 | bio_put(orig_bio); |
e65e1535 MX |
6106 | return -EIO; |
6107 | } | |
6108 | ||
02f57c7a JB |
6109 | if (map_length >= orig_bio->bi_size) { |
6110 | bio = orig_bio; | |
6111 | goto submit; | |
6112 | } | |
6113 | ||
1ae39938 | 6114 | async_submit = 1; |
02f57c7a JB |
6115 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6116 | if (!bio) | |
6117 | return -ENOMEM; | |
6118 | bio->bi_private = dip; | |
6119 | bio->bi_end_io = btrfs_end_dio_bio; | |
6120 | atomic_inc(&dip->pending_bios); | |
6121 | ||
e65e1535 MX |
6122 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6123 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6124 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6125 | bvec->bv_offset) < bvec->bv_len)) { | |
6126 | /* | |
6127 | * inc the count before we submit the bio so | |
6128 | * we know the end IO handler won't happen before | |
6129 | * we inc the count. Otherwise, the dip might get freed | |
6130 | * before we're done setting it up | |
6131 | */ | |
6132 | atomic_inc(&dip->pending_bios); | |
6133 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6134 | file_offset, skip_sum, | |
1ae39938 | 6135 | csums, async_submit); |
e65e1535 MX |
6136 | if (ret) { |
6137 | bio_put(bio); | |
6138 | atomic_dec(&dip->pending_bios); | |
6139 | goto out_err; | |
6140 | } | |
6141 | ||
98bc3149 JB |
6142 | /* Write's use the ordered csums */ |
6143 | if (!write && !skip_sum) | |
e65e1535 MX |
6144 | csums = csums + nr_pages; |
6145 | start_sector += submit_len >> 9; | |
6146 | file_offset += submit_len; | |
6147 | ||
6148 | submit_len = 0; | |
6149 | nr_pages = 0; | |
6150 | ||
6151 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6152 | start_sector, GFP_NOFS); | |
6153 | if (!bio) | |
6154 | goto out_err; | |
6155 | bio->bi_private = dip; | |
6156 | bio->bi_end_io = btrfs_end_dio_bio; | |
6157 | ||
6158 | map_length = orig_bio->bi_size; | |
6159 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6160 | &map_length, NULL, 0); | |
6161 | if (ret) { | |
6162 | bio_put(bio); | |
6163 | goto out_err; | |
6164 | } | |
6165 | } else { | |
6166 | submit_len += bvec->bv_len; | |
6167 | nr_pages ++; | |
6168 | bvec++; | |
6169 | } | |
6170 | } | |
6171 | ||
02f57c7a | 6172 | submit: |
e65e1535 | 6173 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6174 | csums, async_submit); |
e65e1535 MX |
6175 | if (!ret) |
6176 | return 0; | |
6177 | ||
6178 | bio_put(bio); | |
6179 | out_err: | |
6180 | dip->errors = 1; | |
6181 | /* | |
6182 | * before atomic variable goto zero, we must | |
6183 | * make sure dip->errors is perceived to be set. | |
6184 | */ | |
6185 | smp_mb__before_atomic_dec(); | |
6186 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6187 | bio_io_error(dip->orig_bio); | |
6188 | ||
6189 | /* bio_end_io() will handle error, so we needn't return it */ | |
6190 | return 0; | |
6191 | } | |
6192 | ||
4b46fce2 JB |
6193 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6194 | loff_t file_offset) | |
6195 | { | |
6196 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6197 | struct btrfs_dio_private *dip; | |
6198 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6199 | int skip_sum; |
7b6d91da | 6200 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6201 | int ret = 0; |
6202 | ||
6203 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6204 | ||
6205 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6206 | if (!dip) { | |
6207 | ret = -ENOMEM; | |
6208 | goto free_ordered; | |
6209 | } | |
6210 | dip->csums = NULL; | |
6211 | ||
98bc3149 JB |
6212 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6213 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6214 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6215 | if (!dip->csums) { | |
b4966b77 | 6216 | kfree(dip); |
4b46fce2 JB |
6217 | ret = -ENOMEM; |
6218 | goto free_ordered; | |
6219 | } | |
6220 | } | |
6221 | ||
6222 | dip->private = bio->bi_private; | |
6223 | dip->inode = inode; | |
6224 | dip->logical_offset = file_offset; | |
6225 | ||
4b46fce2 JB |
6226 | dip->bytes = 0; |
6227 | do { | |
6228 | dip->bytes += bvec->bv_len; | |
6229 | bvec++; | |
6230 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6231 | ||
46bfbb5c | 6232 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6233 | bio->bi_private = dip; |
e65e1535 MX |
6234 | dip->errors = 0; |
6235 | dip->orig_bio = bio; | |
6236 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6237 | |
6238 | if (write) | |
6239 | bio->bi_end_io = btrfs_endio_direct_write; | |
6240 | else | |
6241 | bio->bi_end_io = btrfs_endio_direct_read; | |
6242 | ||
e65e1535 MX |
6243 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6244 | if (!ret) | |
eaf25d93 | 6245 | return; |
4b46fce2 JB |
6246 | free_ordered: |
6247 | /* | |
6248 | * If this is a write, we need to clean up the reserved space and kill | |
6249 | * the ordered extent. | |
6250 | */ | |
6251 | if (write) { | |
6252 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6253 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6254 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6255 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6256 | btrfs_free_reserved_extent(root, ordered->start, | |
6257 | ordered->disk_len); | |
6258 | btrfs_put_ordered_extent(ordered); | |
6259 | btrfs_put_ordered_extent(ordered); | |
6260 | } | |
6261 | bio_endio(bio, ret); | |
6262 | } | |
6263 | ||
5a5f79b5 CM |
6264 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6265 | const struct iovec *iov, loff_t offset, | |
6266 | unsigned long nr_segs) | |
6267 | { | |
6268 | int seg; | |
a1b75f7d | 6269 | int i; |
5a5f79b5 CM |
6270 | size_t size; |
6271 | unsigned long addr; | |
6272 | unsigned blocksize_mask = root->sectorsize - 1; | |
6273 | ssize_t retval = -EINVAL; | |
6274 | loff_t end = offset; | |
6275 | ||
6276 | if (offset & blocksize_mask) | |
6277 | goto out; | |
6278 | ||
6279 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6280 | for (seg = 0; seg < nr_segs; seg++) { | |
6281 | addr = (unsigned long)iov[seg].iov_base; | |
6282 | size = iov[seg].iov_len; | |
6283 | end += size; | |
a1b75f7d | 6284 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6285 | goto out; |
a1b75f7d JB |
6286 | |
6287 | /* If this is a write we don't need to check anymore */ | |
6288 | if (rw & WRITE) | |
6289 | continue; | |
6290 | ||
6291 | /* | |
6292 | * Check to make sure we don't have duplicate iov_base's in this | |
6293 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6294 | * when reading back. | |
6295 | */ | |
6296 | for (i = seg + 1; i < nr_segs; i++) { | |
6297 | if (iov[seg].iov_base == iov[i].iov_base) | |
6298 | goto out; | |
6299 | } | |
5a5f79b5 CM |
6300 | } |
6301 | retval = 0; | |
6302 | out: | |
6303 | return retval; | |
6304 | } | |
16432985 CM |
6305 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6306 | const struct iovec *iov, loff_t offset, | |
6307 | unsigned long nr_segs) | |
6308 | { | |
4b46fce2 JB |
6309 | struct file *file = iocb->ki_filp; |
6310 | struct inode *inode = file->f_mapping->host; | |
6311 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6312 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6313 | u64 lockstart, lockend; |
6314 | ssize_t ret; | |
4845e44f CM |
6315 | int writing = rw & WRITE; |
6316 | int write_bits = 0; | |
3f7c579c | 6317 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6318 | |
5a5f79b5 CM |
6319 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6320 | offset, nr_segs)) { | |
6321 | return 0; | |
6322 | } | |
6323 | ||
4b46fce2 | 6324 | lockstart = offset; |
3f7c579c CM |
6325 | lockend = offset + count - 1; |
6326 | ||
6327 | if (writing) { | |
6328 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6329 | if (ret) | |
6330 | goto out; | |
6331 | } | |
4845e44f | 6332 | |
4b46fce2 | 6333 | while (1) { |
4845e44f | 6334 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 6335 | 0, &cached_state); |
4b46fce2 JB |
6336 | /* |
6337 | * We're concerned with the entire range that we're going to be | |
6338 | * doing DIO to, so we need to make sure theres no ordered | |
6339 | * extents in this range. | |
6340 | */ | |
6341 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6342 | lockend - lockstart + 1); | |
6343 | if (!ordered) | |
6344 | break; | |
4845e44f CM |
6345 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6346 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6347 | btrfs_start_ordered_extent(inode, ordered, 1); |
6348 | btrfs_put_ordered_extent(ordered); | |
6349 | cond_resched(); | |
6350 | } | |
6351 | ||
4845e44f CM |
6352 | /* |
6353 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6354 | * the dirty or uptodate bits | |
6355 | */ | |
6356 | if (writing) { | |
6357 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6358 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
3fbe5c02 | 6359 | EXTENT_DELALLOC, NULL, &cached_state, |
4845e44f CM |
6360 | GFP_NOFS); |
6361 | if (ret) { | |
6362 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6363 | lockend, EXTENT_LOCKED | write_bits, | |
6364 | 1, 0, &cached_state, GFP_NOFS); | |
6365 | goto out; | |
6366 | } | |
6367 | } | |
6368 | ||
6369 | free_extent_state(cached_state); | |
6370 | cached_state = NULL; | |
6371 | ||
5a5f79b5 CM |
6372 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6373 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6374 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6375 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6376 | |
6377 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6378 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6379 | offset + iov_length(iov, nr_segs) - 1, | |
6380 | EXTENT_LOCKED | write_bits, 1, 0, | |
6381 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6382 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6383 | /* | |
6384 | * We're falling back to buffered, unlock the section we didn't | |
6385 | * do IO on. | |
6386 | */ | |
4845e44f CM |
6387 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6388 | offset + iov_length(iov, nr_segs) - 1, | |
6389 | EXTENT_LOCKED | write_bits, 1, 0, | |
6390 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6391 | } |
4845e44f CM |
6392 | out: |
6393 | free_extent_state(cached_state); | |
4b46fce2 | 6394 | return ret; |
16432985 CM |
6395 | } |
6396 | ||
1506fcc8 YS |
6397 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6398 | __u64 start, __u64 len) | |
6399 | { | |
ec29ed5b | 6400 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6401 | } |
6402 | ||
a52d9a80 | 6403 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6404 | { |
d1310b2e CM |
6405 | struct extent_io_tree *tree; |
6406 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6407 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6408 | } |
1832a6d5 | 6409 | |
a52d9a80 | 6410 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6411 | { |
d1310b2e | 6412 | struct extent_io_tree *tree; |
b888db2b CM |
6413 | |
6414 | ||
6415 | if (current->flags & PF_MEMALLOC) { | |
6416 | redirty_page_for_writepage(wbc, page); | |
6417 | unlock_page(page); | |
6418 | return 0; | |
6419 | } | |
d1310b2e | 6420 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6421 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6422 | } |
6423 | ||
f421950f CM |
6424 | int btrfs_writepages(struct address_space *mapping, |
6425 | struct writeback_control *wbc) | |
b293f02e | 6426 | { |
d1310b2e | 6427 | struct extent_io_tree *tree; |
771ed689 | 6428 | |
d1310b2e | 6429 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6430 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6431 | } | |
6432 | ||
3ab2fb5a CM |
6433 | static int |
6434 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6435 | struct list_head *pages, unsigned nr_pages) | |
6436 | { | |
d1310b2e CM |
6437 | struct extent_io_tree *tree; |
6438 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6439 | return extent_readpages(tree, mapping, pages, nr_pages, |
6440 | btrfs_get_extent); | |
6441 | } | |
e6dcd2dc | 6442 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6443 | { |
d1310b2e CM |
6444 | struct extent_io_tree *tree; |
6445 | struct extent_map_tree *map; | |
a52d9a80 | 6446 | int ret; |
8c2383c3 | 6447 | |
d1310b2e CM |
6448 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6449 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6450 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6451 | if (ret == 1) { |
6452 | ClearPagePrivate(page); | |
6453 | set_page_private(page, 0); | |
6454 | page_cache_release(page); | |
39279cc3 | 6455 | } |
a52d9a80 | 6456 | return ret; |
39279cc3 CM |
6457 | } |
6458 | ||
e6dcd2dc CM |
6459 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6460 | { | |
98509cfc CM |
6461 | if (PageWriteback(page) || PageDirty(page)) |
6462 | return 0; | |
b335b003 | 6463 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6464 | } |
6465 | ||
a52d9a80 | 6466 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6467 | { |
5fd02043 | 6468 | struct inode *inode = page->mapping->host; |
d1310b2e | 6469 | struct extent_io_tree *tree; |
e6dcd2dc | 6470 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6471 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6472 | u64 page_start = page_offset(page); |
6473 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6474 | |
8b62b72b CM |
6475 | /* |
6476 | * we have the page locked, so new writeback can't start, | |
6477 | * and the dirty bit won't be cleared while we are here. | |
6478 | * | |
6479 | * Wait for IO on this page so that we can safely clear | |
6480 | * the PagePrivate2 bit and do ordered accounting | |
6481 | */ | |
e6dcd2dc | 6482 | wait_on_page_writeback(page); |
8b62b72b | 6483 | |
5fd02043 | 6484 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6485 | if (offset) { |
6486 | btrfs_releasepage(page, GFP_NOFS); | |
6487 | return; | |
6488 | } | |
d0082371 | 6489 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
5fd02043 | 6490 | ordered = btrfs_lookup_ordered_extent(inode, |
e6dcd2dc CM |
6491 | page_offset(page)); |
6492 | if (ordered) { | |
eb84ae03 CM |
6493 | /* |
6494 | * IO on this page will never be started, so we need | |
6495 | * to account for any ordered extents now | |
6496 | */ | |
e6dcd2dc CM |
6497 | clear_extent_bit(tree, page_start, page_end, |
6498 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6499 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6500 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6501 | /* |
6502 | * whoever cleared the private bit is responsible | |
6503 | * for the finish_ordered_io | |
6504 | */ | |
5fd02043 JB |
6505 | if (TestClearPagePrivate2(page) && |
6506 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6507 | PAGE_CACHE_SIZE, 1)) { | |
6508 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6509 | } |
e6dcd2dc | 6510 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6511 | cached_state = NULL; |
d0082371 | 6512 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6513 | } |
6514 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6515 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6516 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6517 | __btrfs_releasepage(page, GFP_NOFS); |
6518 | ||
4a096752 | 6519 | ClearPageChecked(page); |
9ad6b7bc | 6520 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6521 | ClearPagePrivate(page); |
6522 | set_page_private(page, 0); | |
6523 | page_cache_release(page); | |
6524 | } | |
39279cc3 CM |
6525 | } |
6526 | ||
9ebefb18 CM |
6527 | /* |
6528 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6529 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6530 | * be careful to check for EOF conditions here. We set the page up correctly | |
6531 | * for a written page which means we get ENOSPC checking when writing into | |
6532 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6533 | * support these features. | |
6534 | * | |
6535 | * We are not allowed to take the i_mutex here so we have to play games to | |
6536 | * protect against truncate races as the page could now be beyond EOF. Because | |
6537 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6538 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6539 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6540 | * unlock the page. | |
6541 | */ | |
c2ec175c | 6542 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6543 | { |
c2ec175c | 6544 | struct page *page = vmf->page; |
6da6abae | 6545 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6546 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6547 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6548 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6549 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6550 | char *kaddr; |
6551 | unsigned long zero_start; | |
9ebefb18 | 6552 | loff_t size; |
1832a6d5 | 6553 | int ret; |
9998eb70 | 6554 | int reserved = 0; |
a52d9a80 | 6555 | u64 page_start; |
e6dcd2dc | 6556 | u64 page_end; |
9ebefb18 | 6557 | |
0ca1f7ce | 6558 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6559 | if (!ret) { |
22c44fe6 | 6560 | ret = btrfs_update_time(vma->vm_file); |
9998eb70 CM |
6561 | reserved = 1; |
6562 | } | |
56a76f82 NP |
6563 | if (ret) { |
6564 | if (ret == -ENOMEM) | |
6565 | ret = VM_FAULT_OOM; | |
6566 | else /* -ENOSPC, -EIO, etc */ | |
6567 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6568 | if (reserved) |
6569 | goto out; | |
6570 | goto out_noreserve; | |
56a76f82 | 6571 | } |
1832a6d5 | 6572 | |
56a76f82 | 6573 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6574 | again: |
9ebefb18 | 6575 | lock_page(page); |
9ebefb18 | 6576 | size = i_size_read(inode); |
e6dcd2dc CM |
6577 | page_start = page_offset(page); |
6578 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6579 | |
9ebefb18 | 6580 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6581 | (page_start >= size)) { |
9ebefb18 CM |
6582 | /* page got truncated out from underneath us */ |
6583 | goto out_unlock; | |
6584 | } | |
e6dcd2dc CM |
6585 | wait_on_page_writeback(page); |
6586 | ||
d0082371 | 6587 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6588 | set_page_extent_mapped(page); |
6589 | ||
eb84ae03 CM |
6590 | /* |
6591 | * we can't set the delalloc bits if there are pending ordered | |
6592 | * extents. Drop our locks and wait for them to finish | |
6593 | */ | |
e6dcd2dc CM |
6594 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6595 | if (ordered) { | |
2ac55d41 JB |
6596 | unlock_extent_cached(io_tree, page_start, page_end, |
6597 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6598 | unlock_page(page); |
eb84ae03 | 6599 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6600 | btrfs_put_ordered_extent(ordered); |
6601 | goto again; | |
6602 | } | |
6603 | ||
fbf19087 JB |
6604 | /* |
6605 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6606 | * if it was already dirty, so for space accounting reasons we need to | |
6607 | * clear any delalloc bits for the range we are fixing to save. There | |
6608 | * is probably a better way to do this, but for now keep consistent with | |
6609 | * prepare_pages in the normal write path. | |
6610 | */ | |
2ac55d41 | 6611 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6612 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6613 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6614 | |
2ac55d41 JB |
6615 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6616 | &cached_state); | |
9ed74f2d | 6617 | if (ret) { |
2ac55d41 JB |
6618 | unlock_extent_cached(io_tree, page_start, page_end, |
6619 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6620 | ret = VM_FAULT_SIGBUS; |
6621 | goto out_unlock; | |
6622 | } | |
e6dcd2dc | 6623 | ret = 0; |
9ebefb18 CM |
6624 | |
6625 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6626 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6627 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6628 | else |
e6dcd2dc | 6629 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6630 | |
e6dcd2dc CM |
6631 | if (zero_start != PAGE_CACHE_SIZE) { |
6632 | kaddr = kmap(page); | |
6633 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6634 | flush_dcache_page(page); | |
6635 | kunmap(page); | |
6636 | } | |
247e743c | 6637 | ClearPageChecked(page); |
e6dcd2dc | 6638 | set_page_dirty(page); |
50a9b214 | 6639 | SetPageUptodate(page); |
5a3f23d5 | 6640 | |
257c62e1 CM |
6641 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6642 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6643 | ||
2ac55d41 | 6644 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6645 | |
6646 | out_unlock: | |
50a9b214 CM |
6647 | if (!ret) |
6648 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6649 | unlock_page(page); |
1832a6d5 | 6650 | out: |
ec39e180 | 6651 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6652 | out_noreserve: |
9ebefb18 CM |
6653 | return ret; |
6654 | } | |
6655 | ||
a41ad394 | 6656 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6657 | { |
6658 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6659 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6660 | int ret; |
3893e33b | 6661 | int err = 0; |
39279cc3 | 6662 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6663 | unsigned long nr; |
dbe674a9 | 6664 | u64 mask = root->sectorsize - 1; |
07127184 | 6665 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6666 | |
5d5e103a JB |
6667 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6668 | if (ret) | |
a41ad394 | 6669 | return ret; |
8082510e | 6670 | |
4a096752 | 6671 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6672 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6673 | |
fcb80c2a JB |
6674 | /* |
6675 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6676 | * 3 things going on here | |
6677 | * | |
6678 | * 1) We need to reserve space for our orphan item and the space to | |
6679 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6680 | * orphan item because we didn't reserve space to remove it. | |
6681 | * | |
6682 | * 2) We need to reserve space to update our inode. | |
6683 | * | |
6684 | * 3) We need to have something to cache all the space that is going to | |
6685 | * be free'd up by the truncate operation, but also have some slack | |
6686 | * space reserved in case it uses space during the truncate (thank you | |
6687 | * very much snapshotting). | |
6688 | * | |
6689 | * And we need these to all be seperate. The fact is we can use alot of | |
6690 | * space doing the truncate, and we have no earthly idea how much space | |
6691 | * we will use, so we need the truncate reservation to be seperate so it | |
6692 | * doesn't end up using space reserved for updating the inode or | |
6693 | * removing the orphan item. We also need to be able to stop the | |
6694 | * transaction and start a new one, which means we need to be able to | |
6695 | * update the inode several times, and we have no idea of knowing how | |
6696 | * many times that will be, so we can't just reserve 1 item for the | |
6697 | * entirety of the opration, so that has to be done seperately as well. | |
6698 | * Then there is the orphan item, which does indeed need to be held on | |
6699 | * to for the whole operation, and we need nobody to touch this reserved | |
6700 | * space except the orphan code. | |
6701 | * | |
6702 | * So that leaves us with | |
6703 | * | |
6704 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6705 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6706 | * transaction reservation. | |
6707 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6708 | * updating the inode. | |
6709 | */ | |
6710 | rsv = btrfs_alloc_block_rsv(root); | |
6711 | if (!rsv) | |
6712 | return -ENOMEM; | |
4a338542 | 6713 | rsv->size = min_size; |
f0cd846e | 6714 | |
907cbceb | 6715 | /* |
07127184 | 6716 | * 1 for the truncate slack space |
907cbceb JB |
6717 | * 1 for the orphan item we're going to add |
6718 | * 1 for the orphan item deletion | |
6719 | * 1 for updating the inode. | |
6720 | */ | |
fcb80c2a JB |
6721 | trans = btrfs_start_transaction(root, 4); |
6722 | if (IS_ERR(trans)) { | |
6723 | err = PTR_ERR(trans); | |
6724 | goto out; | |
6725 | } | |
f0cd846e | 6726 | |
907cbceb JB |
6727 | /* Migrate the slack space for the truncate to our reserve */ |
6728 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6729 | min_size); | |
fcb80c2a | 6730 | BUG_ON(ret); |
f0cd846e JB |
6731 | |
6732 | ret = btrfs_orphan_add(trans, inode); | |
6733 | if (ret) { | |
6734 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6735 | goto out; |
f0cd846e JB |
6736 | } |
6737 | ||
5a3f23d5 CM |
6738 | /* |
6739 | * setattr is responsible for setting the ordered_data_close flag, | |
6740 | * but that is only tested during the last file release. That | |
6741 | * could happen well after the next commit, leaving a great big | |
6742 | * window where new writes may get lost if someone chooses to write | |
6743 | * to this file after truncating to zero | |
6744 | * | |
6745 | * The inode doesn't have any dirty data here, and so if we commit | |
6746 | * this is a noop. If someone immediately starts writing to the inode | |
6747 | * it is very likely we'll catch some of their writes in this | |
6748 | * transaction, and the commit will find this file on the ordered | |
6749 | * data list with good things to send down. | |
6750 | * | |
6751 | * This is a best effort solution, there is still a window where | |
6752 | * using truncate to replace the contents of the file will | |
6753 | * end up with a zero length file after a crash. | |
6754 | */ | |
6755 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6756 | btrfs_add_ordered_operation(trans, root, inode); | |
6757 | ||
8082510e | 6758 | while (1) { |
36ba022a | 6759 | ret = btrfs_block_rsv_refill(root, rsv, min_size); |
907cbceb JB |
6760 | if (ret) { |
6761 | /* | |
6762 | * This can only happen with the original transaction we | |
6763 | * started above, every other time we shouldn't have a | |
6764 | * transaction started yet. | |
6765 | */ | |
6766 | if (ret == -EAGAIN) | |
6767 | goto end_trans; | |
6768 | err = ret; | |
6769 | break; | |
6770 | } | |
6771 | ||
d68fc57b | 6772 | if (!trans) { |
907cbceb JB |
6773 | /* Just need the 1 for updating the inode */ |
6774 | trans = btrfs_start_transaction(root, 1); | |
fcb80c2a | 6775 | if (IS_ERR(trans)) { |
7041ee97 JB |
6776 | ret = err = PTR_ERR(trans); |
6777 | trans = NULL; | |
6778 | break; | |
fcb80c2a | 6779 | } |
d68fc57b YZ |
6780 | } |
6781 | ||
907cbceb JB |
6782 | trans->block_rsv = rsv; |
6783 | ||
8082510e YZ |
6784 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6785 | inode->i_size, | |
6786 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6787 | if (ret != -EAGAIN) { |
6788 | err = ret; | |
8082510e | 6789 | break; |
3893e33b | 6790 | } |
39279cc3 | 6791 | |
fcb80c2a | 6792 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6793 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6794 | if (ret) { |
6795 | err = ret; | |
6796 | break; | |
6797 | } | |
907cbceb | 6798 | end_trans: |
8082510e YZ |
6799 | nr = trans->blocks_used; |
6800 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6801 | trans = NULL; |
8082510e | 6802 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6803 | } |
6804 | ||
6805 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6806 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6807 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6808 | if (ret) |
6809 | err = ret; | |
ded5db9d JB |
6810 | } else if (ret && inode->i_nlink > 0) { |
6811 | /* | |
6812 | * Failed to do the truncate, remove us from the in memory | |
6813 | * orphan list. | |
6814 | */ | |
6815 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6816 | } |
6817 | ||
917c16b2 CM |
6818 | if (trans) { |
6819 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
6820 | ret = btrfs_update_inode(trans, root, inode); | |
6821 | if (ret && !err) | |
6822 | err = ret; | |
7b128766 | 6823 | |
917c16b2 | 6824 | nr = trans->blocks_used; |
7ad85bb7 | 6825 | ret = btrfs_end_transaction(trans, root); |
917c16b2 CM |
6826 | btrfs_btree_balance_dirty(root, nr); |
6827 | } | |
fcb80c2a JB |
6828 | |
6829 | out: | |
6830 | btrfs_free_block_rsv(root, rsv); | |
6831 | ||
3893e33b JB |
6832 | if (ret && !err) |
6833 | err = ret; | |
a41ad394 | 6834 | |
3893e33b | 6835 | return err; |
39279cc3 CM |
6836 | } |
6837 | ||
d352ac68 CM |
6838 | /* |
6839 | * create a new subvolume directory/inode (helper for the ioctl). | |
6840 | */ | |
d2fb3437 | 6841 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6842 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6843 | { |
39279cc3 | 6844 | struct inode *inode; |
76dda93c | 6845 | int err; |
00e4e6b3 | 6846 | u64 index = 0; |
39279cc3 | 6847 | |
12fc9d09 FA |
6848 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
6849 | new_dirid, new_dirid, | |
6850 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
6851 | &index); | |
54aa1f4d | 6852 | if (IS_ERR(inode)) |
f46b5a66 | 6853 | return PTR_ERR(inode); |
39279cc3 CM |
6854 | inode->i_op = &btrfs_dir_inode_operations; |
6855 | inode->i_fop = &btrfs_dir_file_operations; | |
6856 | ||
bfe86848 | 6857 | set_nlink(inode, 1); |
dbe674a9 | 6858 | btrfs_i_size_write(inode, 0); |
3b96362c | 6859 | |
76dda93c | 6860 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 6861 | |
76dda93c | 6862 | iput(inode); |
ce598979 | 6863 | return err; |
39279cc3 CM |
6864 | } |
6865 | ||
39279cc3 CM |
6866 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6867 | { | |
6868 | struct btrfs_inode *ei; | |
2ead6ae7 | 6869 | struct inode *inode; |
39279cc3 CM |
6870 | |
6871 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6872 | if (!ei) | |
6873 | return NULL; | |
2ead6ae7 YZ |
6874 | |
6875 | ei->root = NULL; | |
6876 | ei->space_info = NULL; | |
6877 | ei->generation = 0; | |
15ee9bc7 | 6878 | ei->last_trans = 0; |
257c62e1 | 6879 | ei->last_sub_trans = 0; |
e02119d5 | 6880 | ei->logged_trans = 0; |
2ead6ae7 | 6881 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
6882 | ei->disk_i_size = 0; |
6883 | ei->flags = 0; | |
7709cde3 | 6884 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
6885 | ei->index_cnt = (u64)-1; |
6886 | ei->last_unlink_trans = 0; | |
6887 | ||
9e0baf60 JB |
6888 | spin_lock_init(&ei->lock); |
6889 | ei->outstanding_extents = 0; | |
6890 | ei->reserved_extents = 0; | |
2ead6ae7 YZ |
6891 | |
6892 | ei->ordered_data_close = 0; | |
d68fc57b | 6893 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6894 | ei->dummy_inode = 0; |
4cb5300b | 6895 | ei->in_defrag = 0; |
7fd2ae21 | 6896 | ei->delalloc_meta_reserved = 0; |
261507a0 | 6897 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 6898 | |
16cdcec7 MX |
6899 | ei->delayed_node = NULL; |
6900 | ||
2ead6ae7 | 6901 | inode = &ei->vfs_inode; |
a8067e02 | 6902 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
6903 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
6904 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
6905 | ei->io_tree.track_uptodate = 1; |
6906 | ei->io_failure_tree.track_uptodate = 1; | |
2ead6ae7 | 6907 | mutex_init(&ei->log_mutex); |
f248679e | 6908 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 6909 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6910 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6911 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6912 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6913 | RB_CLEAR_NODE(&ei->rb_node); |
6914 | ||
6915 | return inode; | |
39279cc3 CM |
6916 | } |
6917 | ||
fa0d7e3d NP |
6918 | static void btrfs_i_callback(struct rcu_head *head) |
6919 | { | |
6920 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
6921 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
6922 | } | |
6923 | ||
39279cc3 CM |
6924 | void btrfs_destroy_inode(struct inode *inode) |
6925 | { | |
e6dcd2dc | 6926 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6927 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6928 | ||
39279cc3 CM |
6929 | WARN_ON(!list_empty(&inode->i_dentry)); |
6930 | WARN_ON(inode->i_data.nrpages); | |
9e0baf60 JB |
6931 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
6932 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
6933 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
6934 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 6935 | |
a6dbd429 JB |
6936 | /* |
6937 | * This can happen where we create an inode, but somebody else also | |
6938 | * created the same inode and we need to destroy the one we already | |
6939 | * created. | |
6940 | */ | |
6941 | if (!root) | |
6942 | goto free; | |
6943 | ||
5a3f23d5 CM |
6944 | /* |
6945 | * Make sure we're properly removed from the ordered operation | |
6946 | * lists. | |
6947 | */ | |
6948 | smp_mb(); | |
6949 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6950 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6951 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6952 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6953 | } | |
6954 | ||
d68fc57b | 6955 | spin_lock(&root->orphan_lock); |
7b128766 | 6956 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
33345d01 LZ |
6957 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
6958 | (unsigned long long)btrfs_ino(inode)); | |
8082510e | 6959 | list_del_init(&BTRFS_I(inode)->i_orphan); |
7b128766 | 6960 | } |
d68fc57b | 6961 | spin_unlock(&root->orphan_lock); |
7b128766 | 6962 | |
d397712b | 6963 | while (1) { |
e6dcd2dc CM |
6964 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6965 | if (!ordered) | |
6966 | break; | |
6967 | else { | |
d397712b CM |
6968 | printk(KERN_ERR "btrfs found ordered " |
6969 | "extent %llu %llu on inode cleanup\n", | |
6970 | (unsigned long long)ordered->file_offset, | |
6971 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6972 | btrfs_remove_ordered_extent(inode, ordered); |
6973 | btrfs_put_ordered_extent(ordered); | |
6974 | btrfs_put_ordered_extent(ordered); | |
6975 | } | |
6976 | } | |
5d4f98a2 | 6977 | inode_tree_del(inode); |
5b21f2ed | 6978 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6979 | free: |
16cdcec7 | 6980 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 6981 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6982 | } |
6983 | ||
45321ac5 | 6984 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6985 | { |
6986 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6987 | |
0af3d00b | 6988 | if (btrfs_root_refs(&root->root_item) == 0 && |
2cf8572d | 6989 | !btrfs_is_free_space_inode(root, inode)) |
45321ac5 | 6990 | return 1; |
76dda93c | 6991 | else |
45321ac5 | 6992 | return generic_drop_inode(inode); |
76dda93c YZ |
6993 | } |
6994 | ||
0ee0fda0 | 6995 | static void init_once(void *foo) |
39279cc3 CM |
6996 | { |
6997 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6998 | ||
6999 | inode_init_once(&ei->vfs_inode); | |
7000 | } | |
7001 | ||
7002 | void btrfs_destroy_cachep(void) | |
7003 | { | |
7004 | if (btrfs_inode_cachep) | |
7005 | kmem_cache_destroy(btrfs_inode_cachep); | |
7006 | if (btrfs_trans_handle_cachep) | |
7007 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7008 | if (btrfs_transaction_cachep) | |
7009 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7010 | if (btrfs_path_cachep) |
7011 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7012 | if (btrfs_free_space_cachep) |
7013 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
7014 | } |
7015 | ||
7016 | int btrfs_init_cachep(void) | |
7017 | { | |
9601e3f6 CH |
7018 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
7019 | sizeof(struct btrfs_inode), 0, | |
7020 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7021 | if (!btrfs_inode_cachep) |
7022 | goto fail; | |
9601e3f6 CH |
7023 | |
7024 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
7025 | sizeof(struct btrfs_trans_handle), 0, | |
7026 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7027 | if (!btrfs_trans_handle_cachep) |
7028 | goto fail; | |
9601e3f6 CH |
7029 | |
7030 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
7031 | sizeof(struct btrfs_transaction), 0, | |
7032 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7033 | if (!btrfs_transaction_cachep) |
7034 | goto fail; | |
9601e3f6 CH |
7035 | |
7036 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
7037 | sizeof(struct btrfs_path), 0, | |
7038 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7039 | if (!btrfs_path_cachep) |
7040 | goto fail; | |
9601e3f6 | 7041 | |
dc89e982 JB |
7042 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
7043 | sizeof(struct btrfs_free_space), 0, | |
7044 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7045 | if (!btrfs_free_space_cachep) | |
7046 | goto fail; | |
7047 | ||
39279cc3 CM |
7048 | return 0; |
7049 | fail: | |
7050 | btrfs_destroy_cachep(); | |
7051 | return -ENOMEM; | |
7052 | } | |
7053 | ||
7054 | static int btrfs_getattr(struct vfsmount *mnt, | |
7055 | struct dentry *dentry, struct kstat *stat) | |
7056 | { | |
7057 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
7058 | u32 blocksize = inode->i_sb->s_blocksize; |
7059 | ||
39279cc3 | 7060 | generic_fillattr(inode, stat); |
0ee5dc67 | 7061 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7062 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
7063 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
7064 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
7065 | return 0; |
7066 | } | |
7067 | ||
75e7cb7f LB |
7068 | /* |
7069 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7070 | * directory. | |
7071 | */ | |
7072 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7073 | { | |
7074 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7075 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7076 | ||
7077 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7078 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7079 | else | |
7080 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7081 | ||
7082 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
7083 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
7084 | else | |
7085 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
7086 | } | |
7087 | ||
d397712b CM |
7088 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7089 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7090 | { |
7091 | struct btrfs_trans_handle *trans; | |
7092 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7093 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7094 | struct inode *new_inode = new_dentry->d_inode; |
7095 | struct inode *old_inode = old_dentry->d_inode; | |
7096 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7097 | u64 index = 0; |
4df27c4d | 7098 | u64 root_objectid; |
39279cc3 | 7099 | int ret; |
33345d01 | 7100 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7101 | |
33345d01 | 7102 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7103 | return -EPERM; |
7104 | ||
4df27c4d | 7105 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7106 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7107 | return -EXDEV; |
7108 | ||
33345d01 LZ |
7109 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7110 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7111 | return -ENOTEMPTY; |
5f39d397 | 7112 | |
4df27c4d YZ |
7113 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7114 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7115 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7116 | /* |
7117 | * we're using rename to replace one file with another. | |
7118 | * and the replacement file is large. Start IO on it now so | |
7119 | * we don't add too much work to the end of the transaction | |
7120 | */ | |
4baf8c92 | 7121 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7122 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7123 | filemap_flush(old_inode->i_mapping); | |
7124 | ||
76dda93c | 7125 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7126 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7127 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7128 | /* |
7129 | * We want to reserve the absolute worst case amount of items. So if | |
7130 | * both inodes are subvols and we need to unlink them then that would | |
7131 | * require 4 item modifications, but if they are both normal inodes it | |
7132 | * would require 5 item modifications, so we'll assume their normal | |
7133 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7134 | * should cover the worst case number of items we'll modify. | |
7135 | */ | |
7136 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7137 | if (IS_ERR(trans)) { |
7138 | ret = PTR_ERR(trans); | |
7139 | goto out_notrans; | |
7140 | } | |
76dda93c | 7141 | |
4df27c4d YZ |
7142 | if (dest != root) |
7143 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7144 | |
a5719521 YZ |
7145 | ret = btrfs_set_inode_index(new_dir, &index); |
7146 | if (ret) | |
7147 | goto out_fail; | |
5a3f23d5 | 7148 | |
33345d01 | 7149 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7150 | /* force full log commit if subvolume involved. */ |
7151 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7152 | } else { | |
a5719521 YZ |
7153 | ret = btrfs_insert_inode_ref(trans, dest, |
7154 | new_dentry->d_name.name, | |
7155 | new_dentry->d_name.len, | |
33345d01 LZ |
7156 | old_ino, |
7157 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7158 | if (ret) |
7159 | goto out_fail; | |
4df27c4d YZ |
7160 | /* |
7161 | * this is an ugly little race, but the rename is required | |
7162 | * to make sure that if we crash, the inode is either at the | |
7163 | * old name or the new one. pinning the log transaction lets | |
7164 | * us make sure we don't allow a log commit to come in after | |
7165 | * we unlink the name but before we add the new name back in. | |
7166 | */ | |
7167 | btrfs_pin_log_trans(root); | |
7168 | } | |
5a3f23d5 CM |
7169 | /* |
7170 | * make sure the inode gets flushed if it is replacing | |
7171 | * something. | |
7172 | */ | |
33345d01 | 7173 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7174 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7175 | |
0c4d2d95 JB |
7176 | inode_inc_iversion(old_dir); |
7177 | inode_inc_iversion(new_dir); | |
7178 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7179 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7180 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7181 | old_inode->i_ctime = ctime; | |
5f39d397 | 7182 | |
12fcfd22 CM |
7183 | if (old_dentry->d_parent != new_dentry->d_parent) |
7184 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7185 | ||
33345d01 | 7186 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7187 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7188 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7189 | old_dentry->d_name.name, | |
7190 | old_dentry->d_name.len); | |
7191 | } else { | |
92986796 AV |
7192 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7193 | old_dentry->d_inode, | |
7194 | old_dentry->d_name.name, | |
7195 | old_dentry->d_name.len); | |
7196 | if (!ret) | |
7197 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7198 | } |
79787eaa JM |
7199 | if (ret) { |
7200 | btrfs_abort_transaction(trans, root, ret); | |
7201 | goto out_fail; | |
7202 | } | |
39279cc3 CM |
7203 | |
7204 | if (new_inode) { | |
0c4d2d95 | 7205 | inode_inc_iversion(new_inode); |
39279cc3 | 7206 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7207 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7208 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7209 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7210 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7211 | root_objectid, | |
7212 | new_dentry->d_name.name, | |
7213 | new_dentry->d_name.len); | |
7214 | BUG_ON(new_inode->i_nlink == 0); | |
7215 | } else { | |
7216 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7217 | new_dentry->d_inode, | |
7218 | new_dentry->d_name.name, | |
7219 | new_dentry->d_name.len); | |
7220 | } | |
79787eaa | 7221 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7222 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7223 | BUG_ON(ret); |
7b128766 | 7224 | } |
79787eaa JM |
7225 | if (ret) { |
7226 | btrfs_abort_transaction(trans, root, ret); | |
7227 | goto out_fail; | |
7228 | } | |
39279cc3 | 7229 | } |
aec7477b | 7230 | |
75e7cb7f LB |
7231 | fixup_inode_flags(new_dir, old_inode); |
7232 | ||
4df27c4d YZ |
7233 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7234 | new_dentry->d_name.name, | |
a5719521 | 7235 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7236 | if (ret) { |
7237 | btrfs_abort_transaction(trans, root, ret); | |
7238 | goto out_fail; | |
7239 | } | |
39279cc3 | 7240 | |
33345d01 | 7241 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7242 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7243 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7244 | btrfs_end_log_trans(root); |
7245 | } | |
39279cc3 | 7246 | out_fail: |
7ad85bb7 | 7247 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7248 | out_notrans: |
33345d01 | 7249 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7250 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7251 | |
39279cc3 CM |
7252 | return ret; |
7253 | } | |
7254 | ||
d352ac68 CM |
7255 | /* |
7256 | * some fairly slow code that needs optimization. This walks the list | |
7257 | * of all the inodes with pending delalloc and forces them to disk. | |
7258 | */ | |
24bbcf04 | 7259 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7260 | { |
7261 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7262 | struct btrfs_inode *binode; | |
5b21f2ed | 7263 | struct inode *inode; |
ea8c2819 | 7264 | |
c146afad YZ |
7265 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7266 | return -EROFS; | |
7267 | ||
75eff68e | 7268 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7269 | while (!list_empty(head)) { |
ea8c2819 CM |
7270 | binode = list_entry(head->next, struct btrfs_inode, |
7271 | delalloc_inodes); | |
5b21f2ed ZY |
7272 | inode = igrab(&binode->vfs_inode); |
7273 | if (!inode) | |
7274 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7275 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7276 | if (inode) { |
8c8bee1d | 7277 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7278 | if (delay_iput) |
7279 | btrfs_add_delayed_iput(inode); | |
7280 | else | |
7281 | iput(inode); | |
5b21f2ed ZY |
7282 | } |
7283 | cond_resched(); | |
75eff68e | 7284 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7285 | } |
75eff68e | 7286 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7287 | |
7288 | /* the filemap_flush will queue IO into the worker threads, but | |
7289 | * we have to make sure the IO is actually started and that | |
7290 | * ordered extents get created before we return | |
7291 | */ | |
7292 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7293 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7294 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7295 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7296 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7297 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7298 | } |
7299 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7300 | return 0; |
7301 | } | |
7302 | ||
39279cc3 CM |
7303 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7304 | const char *symname) | |
7305 | { | |
7306 | struct btrfs_trans_handle *trans; | |
7307 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7308 | struct btrfs_path *path; | |
7309 | struct btrfs_key key; | |
1832a6d5 | 7310 | struct inode *inode = NULL; |
39279cc3 CM |
7311 | int err; |
7312 | int drop_inode = 0; | |
7313 | u64 objectid; | |
00e4e6b3 | 7314 | u64 index = 0 ; |
39279cc3 CM |
7315 | int name_len; |
7316 | int datasize; | |
5f39d397 | 7317 | unsigned long ptr; |
39279cc3 | 7318 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7319 | struct extent_buffer *leaf; |
1832a6d5 | 7320 | unsigned long nr = 0; |
39279cc3 CM |
7321 | |
7322 | name_len = strlen(symname) + 1; | |
7323 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7324 | return -ENAMETOOLONG; | |
1832a6d5 | 7325 | |
9ed74f2d JB |
7326 | /* |
7327 | * 2 items for inode item and ref | |
7328 | * 2 items for dir items | |
7329 | * 1 item for xattr if selinux is on | |
7330 | */ | |
a22285a6 YZ |
7331 | trans = btrfs_start_transaction(root, 5); |
7332 | if (IS_ERR(trans)) | |
7333 | return PTR_ERR(trans); | |
1832a6d5 | 7334 | |
581bb050 LZ |
7335 | err = btrfs_find_free_ino(root, &objectid); |
7336 | if (err) | |
7337 | goto out_unlock; | |
7338 | ||
aec7477b | 7339 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7340 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7341 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7342 | if (IS_ERR(inode)) { |
7343 | err = PTR_ERR(inode); | |
39279cc3 | 7344 | goto out_unlock; |
7cf96da3 | 7345 | } |
39279cc3 | 7346 | |
2a7dba39 | 7347 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7348 | if (err) { |
7349 | drop_inode = 1; | |
7350 | goto out_unlock; | |
7351 | } | |
7352 | ||
ad19db71 CS |
7353 | /* |
7354 | * If the active LSM wants to access the inode during | |
7355 | * d_instantiate it needs these. Smack checks to see | |
7356 | * if the filesystem supports xattrs by looking at the | |
7357 | * ops vector. | |
7358 | */ | |
7359 | inode->i_fop = &btrfs_file_operations; | |
7360 | inode->i_op = &btrfs_file_inode_operations; | |
7361 | ||
a1b075d2 | 7362 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7363 | if (err) |
7364 | drop_inode = 1; | |
7365 | else { | |
7366 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7367 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7368 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7369 | } |
39279cc3 CM |
7370 | if (drop_inode) |
7371 | goto out_unlock; | |
7372 | ||
7373 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7374 | if (!path) { |
7375 | err = -ENOMEM; | |
7376 | drop_inode = 1; | |
7377 | goto out_unlock; | |
7378 | } | |
33345d01 | 7379 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7380 | key.offset = 0; |
39279cc3 CM |
7381 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7382 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7383 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7384 | datasize); | |
54aa1f4d CM |
7385 | if (err) { |
7386 | drop_inode = 1; | |
b0839166 | 7387 | btrfs_free_path(path); |
54aa1f4d CM |
7388 | goto out_unlock; |
7389 | } | |
5f39d397 CM |
7390 | leaf = path->nodes[0]; |
7391 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7392 | struct btrfs_file_extent_item); | |
7393 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7394 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7395 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7396 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7397 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7398 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7399 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7400 | ||
39279cc3 | 7401 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7402 | write_extent_buffer(leaf, symname, ptr, name_len); |
7403 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7404 | btrfs_free_path(path); |
5f39d397 | 7405 | |
39279cc3 CM |
7406 | inode->i_op = &btrfs_symlink_inode_operations; |
7407 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7408 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7409 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7410 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7411 | err = btrfs_update_inode(trans, root, inode); |
7412 | if (err) | |
7413 | drop_inode = 1; | |
39279cc3 CM |
7414 | |
7415 | out_unlock: | |
08c422c2 AV |
7416 | if (!err) |
7417 | d_instantiate(dentry, inode); | |
d3c2fdcf | 7418 | nr = trans->blocks_used; |
7ad85bb7 | 7419 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7420 | if (drop_inode) { |
7421 | inode_dec_link_count(inode); | |
7422 | iput(inode); | |
7423 | } | |
d3c2fdcf | 7424 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7425 | return err; |
7426 | } | |
16432985 | 7427 | |
0af3d00b JB |
7428 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7429 | u64 start, u64 num_bytes, u64 min_size, | |
7430 | loff_t actual_len, u64 *alloc_hint, | |
7431 | struct btrfs_trans_handle *trans) | |
d899e052 | 7432 | { |
d899e052 YZ |
7433 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7434 | struct btrfs_key ins; | |
d899e052 | 7435 | u64 cur_offset = start; |
55a61d1d | 7436 | u64 i_size; |
d899e052 | 7437 | int ret = 0; |
0af3d00b | 7438 | bool own_trans = true; |
d899e052 | 7439 | |
0af3d00b JB |
7440 | if (trans) |
7441 | own_trans = false; | |
d899e052 | 7442 | while (num_bytes > 0) { |
0af3d00b JB |
7443 | if (own_trans) { |
7444 | trans = btrfs_start_transaction(root, 3); | |
7445 | if (IS_ERR(trans)) { | |
7446 | ret = PTR_ERR(trans); | |
7447 | break; | |
7448 | } | |
5a303d5d YZ |
7449 | } |
7450 | ||
efa56464 | 7451 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7452 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7453 | if (ret) { |
0af3d00b JB |
7454 | if (own_trans) |
7455 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7456 | break; |
d899e052 | 7457 | } |
5a303d5d | 7458 | |
d899e052 YZ |
7459 | ret = insert_reserved_file_extent(trans, inode, |
7460 | cur_offset, ins.objectid, | |
7461 | ins.offset, ins.offset, | |
920bbbfb | 7462 | ins.offset, 0, 0, 0, |
d899e052 | 7463 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7464 | if (ret) { |
7465 | btrfs_abort_transaction(trans, root, ret); | |
7466 | if (own_trans) | |
7467 | btrfs_end_transaction(trans, root); | |
7468 | break; | |
7469 | } | |
a1ed835e CM |
7470 | btrfs_drop_extent_cache(inode, cur_offset, |
7471 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7472 | |
d899e052 YZ |
7473 | num_bytes -= ins.offset; |
7474 | cur_offset += ins.offset; | |
efa56464 | 7475 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7476 | |
0c4d2d95 | 7477 | inode_inc_iversion(inode); |
d899e052 | 7478 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7479 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7480 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7481 | (actual_len > inode->i_size) && |
7482 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7483 | if (cur_offset > actual_len) |
55a61d1d | 7484 | i_size = actual_len; |
d1ea6a61 | 7485 | else |
55a61d1d JB |
7486 | i_size = cur_offset; |
7487 | i_size_write(inode, i_size); | |
7488 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7489 | } |
7490 | ||
d899e052 | 7491 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7492 | |
7493 | if (ret) { | |
7494 | btrfs_abort_transaction(trans, root, ret); | |
7495 | if (own_trans) | |
7496 | btrfs_end_transaction(trans, root); | |
7497 | break; | |
7498 | } | |
d899e052 | 7499 | |
0af3d00b JB |
7500 | if (own_trans) |
7501 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7502 | } |
d899e052 YZ |
7503 | return ret; |
7504 | } | |
7505 | ||
0af3d00b JB |
7506 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7507 | u64 start, u64 num_bytes, u64 min_size, | |
7508 | loff_t actual_len, u64 *alloc_hint) | |
7509 | { | |
7510 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7511 | min_size, actual_len, alloc_hint, | |
7512 | NULL); | |
7513 | } | |
7514 | ||
7515 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7516 | struct btrfs_trans_handle *trans, int mode, | |
7517 | u64 start, u64 num_bytes, u64 min_size, | |
7518 | loff_t actual_len, u64 *alloc_hint) | |
7519 | { | |
7520 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7521 | min_size, actual_len, alloc_hint, trans); | |
7522 | } | |
7523 | ||
e6dcd2dc CM |
7524 | static int btrfs_set_page_dirty(struct page *page) |
7525 | { | |
e6dcd2dc CM |
7526 | return __set_page_dirty_nobuffers(page); |
7527 | } | |
7528 | ||
10556cb2 | 7529 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7530 | { |
b83cc969 | 7531 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7532 | umode_t mode = inode->i_mode; |
b83cc969 | 7533 | |
cb6db4e5 JM |
7534 | if (mask & MAY_WRITE && |
7535 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7536 | if (btrfs_root_readonly(root)) | |
7537 | return -EROFS; | |
7538 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7539 | return -EACCES; | |
7540 | } | |
2830ba7f | 7541 | return generic_permission(inode, mask); |
fdebe2bd | 7542 | } |
39279cc3 | 7543 | |
6e1d5dcc | 7544 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7545 | .getattr = btrfs_getattr, |
39279cc3 CM |
7546 | .lookup = btrfs_lookup, |
7547 | .create = btrfs_create, | |
7548 | .unlink = btrfs_unlink, | |
7549 | .link = btrfs_link, | |
7550 | .mkdir = btrfs_mkdir, | |
7551 | .rmdir = btrfs_rmdir, | |
7552 | .rename = btrfs_rename, | |
7553 | .symlink = btrfs_symlink, | |
7554 | .setattr = btrfs_setattr, | |
618e21d5 | 7555 | .mknod = btrfs_mknod, |
95819c05 CH |
7556 | .setxattr = btrfs_setxattr, |
7557 | .getxattr = btrfs_getxattr, | |
5103e947 | 7558 | .listxattr = btrfs_listxattr, |
95819c05 | 7559 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7560 | .permission = btrfs_permission, |
4e34e719 | 7561 | .get_acl = btrfs_get_acl, |
39279cc3 | 7562 | }; |
6e1d5dcc | 7563 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7564 | .lookup = btrfs_lookup, |
fdebe2bd | 7565 | .permission = btrfs_permission, |
4e34e719 | 7566 | .get_acl = btrfs_get_acl, |
39279cc3 | 7567 | }; |
76dda93c | 7568 | |
828c0950 | 7569 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7570 | .llseek = generic_file_llseek, |
7571 | .read = generic_read_dir, | |
cbdf5a24 | 7572 | .readdir = btrfs_real_readdir, |
34287aa3 | 7573 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7574 | #ifdef CONFIG_COMPAT |
34287aa3 | 7575 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7576 | #endif |
6bf13c0c | 7577 | .release = btrfs_release_file, |
e02119d5 | 7578 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7579 | }; |
7580 | ||
d1310b2e | 7581 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7582 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7583 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7584 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7585 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7586 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7587 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7588 | .set_bit_hook = btrfs_set_bit_hook, |
7589 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7590 | .merge_extent_hook = btrfs_merge_extent_hook, |
7591 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7592 | }; |
7593 | ||
35054394 CM |
7594 | /* |
7595 | * btrfs doesn't support the bmap operation because swapfiles | |
7596 | * use bmap to make a mapping of extents in the file. They assume | |
7597 | * these extents won't change over the life of the file and they | |
7598 | * use the bmap result to do IO directly to the drive. | |
7599 | * | |
7600 | * the btrfs bmap call would return logical addresses that aren't | |
7601 | * suitable for IO and they also will change frequently as COW | |
7602 | * operations happen. So, swapfile + btrfs == corruption. | |
7603 | * | |
7604 | * For now we're avoiding this by dropping bmap. | |
7605 | */ | |
7f09410b | 7606 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7607 | .readpage = btrfs_readpage, |
7608 | .writepage = btrfs_writepage, | |
b293f02e | 7609 | .writepages = btrfs_writepages, |
3ab2fb5a | 7610 | .readpages = btrfs_readpages, |
16432985 | 7611 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7612 | .invalidatepage = btrfs_invalidatepage, |
7613 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7614 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7615 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7616 | }; |
7617 | ||
7f09410b | 7618 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7619 | .readpage = btrfs_readpage, |
7620 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7621 | .invalidatepage = btrfs_invalidatepage, |
7622 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7623 | }; |
7624 | ||
6e1d5dcc | 7625 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7626 | .getattr = btrfs_getattr, |
7627 | .setattr = btrfs_setattr, | |
95819c05 CH |
7628 | .setxattr = btrfs_setxattr, |
7629 | .getxattr = btrfs_getxattr, | |
5103e947 | 7630 | .listxattr = btrfs_listxattr, |
95819c05 | 7631 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7632 | .permission = btrfs_permission, |
1506fcc8 | 7633 | .fiemap = btrfs_fiemap, |
4e34e719 | 7634 | .get_acl = btrfs_get_acl, |
39279cc3 | 7635 | }; |
6e1d5dcc | 7636 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7637 | .getattr = btrfs_getattr, |
7638 | .setattr = btrfs_setattr, | |
fdebe2bd | 7639 | .permission = btrfs_permission, |
95819c05 CH |
7640 | .setxattr = btrfs_setxattr, |
7641 | .getxattr = btrfs_getxattr, | |
33268eaf | 7642 | .listxattr = btrfs_listxattr, |
95819c05 | 7643 | .removexattr = btrfs_removexattr, |
4e34e719 | 7644 | .get_acl = btrfs_get_acl, |
618e21d5 | 7645 | }; |
6e1d5dcc | 7646 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7647 | .readlink = generic_readlink, |
7648 | .follow_link = page_follow_link_light, | |
7649 | .put_link = page_put_link, | |
f209561a | 7650 | .getattr = btrfs_getattr, |
22c44fe6 | 7651 | .setattr = btrfs_setattr, |
fdebe2bd | 7652 | .permission = btrfs_permission, |
0279b4cd JO |
7653 | .setxattr = btrfs_setxattr, |
7654 | .getxattr = btrfs_getxattr, | |
7655 | .listxattr = btrfs_listxattr, | |
7656 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7657 | .get_acl = btrfs_get_acl, |
39279cc3 | 7658 | }; |
76dda93c | 7659 | |
82d339d9 | 7660 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7661 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7662 | .d_release = btrfs_dentry_release, |
76dda93c | 7663 | }; |