]>
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); | |
c8b97818 CM |
233 | u64 data_len = inline_len; |
234 | int ret; | |
235 | ||
236 | if (compressed_size) | |
237 | data_len = compressed_size; | |
238 | ||
239 | if (start > 0 || | |
70b99e69 | 240 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
241 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
242 | (!compressed_size && | |
243 | (actual_end & (root->sectorsize - 1)) == 0) || | |
244 | end + 1 < isize || | |
245 | data_len > root->fs_info->max_inline) { | |
246 | return 1; | |
247 | } | |
248 | ||
2671485d | 249 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
250 | if (ret) |
251 | return ret; | |
c8b97818 CM |
252 | |
253 | if (isize > actual_end) | |
254 | inline_len = min_t(u64, isize, actual_end); | |
255 | ret = insert_inline_extent(trans, root, inode, start, | |
256 | inline_len, compressed_size, | |
fe3f566c | 257 | compress_type, compressed_pages); |
2adcac1a | 258 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
259 | btrfs_abort_transaction(trans, root, ret); |
260 | return ret; | |
2adcac1a JB |
261 | } else if (ret == -ENOSPC) { |
262 | return 1; | |
79787eaa | 263 | } |
2adcac1a | 264 | |
0ca1f7ce | 265 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 266 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
267 | return 0; |
268 | } | |
269 | ||
771ed689 CM |
270 | struct async_extent { |
271 | u64 start; | |
272 | u64 ram_size; | |
273 | u64 compressed_size; | |
274 | struct page **pages; | |
275 | unsigned long nr_pages; | |
261507a0 | 276 | int compress_type; |
771ed689 CM |
277 | struct list_head list; |
278 | }; | |
279 | ||
280 | struct async_cow { | |
281 | struct inode *inode; | |
282 | struct btrfs_root *root; | |
283 | struct page *locked_page; | |
284 | u64 start; | |
285 | u64 end; | |
286 | struct list_head extents; | |
287 | struct btrfs_work work; | |
288 | }; | |
289 | ||
290 | static noinline int add_async_extent(struct async_cow *cow, | |
291 | u64 start, u64 ram_size, | |
292 | u64 compressed_size, | |
293 | struct page **pages, | |
261507a0 LZ |
294 | unsigned long nr_pages, |
295 | int compress_type) | |
771ed689 CM |
296 | { |
297 | struct async_extent *async_extent; | |
298 | ||
299 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 300 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
301 | async_extent->start = start; |
302 | async_extent->ram_size = ram_size; | |
303 | async_extent->compressed_size = compressed_size; | |
304 | async_extent->pages = pages; | |
305 | async_extent->nr_pages = nr_pages; | |
261507a0 | 306 | async_extent->compress_type = compress_type; |
771ed689 CM |
307 | list_add_tail(&async_extent->list, &cow->extents); |
308 | return 0; | |
309 | } | |
310 | ||
d352ac68 | 311 | /* |
771ed689 CM |
312 | * we create compressed extents in two phases. The first |
313 | * phase compresses a range of pages that have already been | |
314 | * locked (both pages and state bits are locked). | |
c8b97818 | 315 | * |
771ed689 CM |
316 | * This is done inside an ordered work queue, and the compression |
317 | * is spread across many cpus. The actual IO submission is step | |
318 | * two, and the ordered work queue takes care of making sure that | |
319 | * happens in the same order things were put onto the queue by | |
320 | * writepages and friends. | |
c8b97818 | 321 | * |
771ed689 CM |
322 | * If this code finds it can't get good compression, it puts an |
323 | * entry onto the work queue to write the uncompressed bytes. This | |
324 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
325 | * are written in the same order that the flusher thread sent them |
326 | * down. | |
d352ac68 | 327 | */ |
771ed689 CM |
328 | static noinline int compress_file_range(struct inode *inode, |
329 | struct page *locked_page, | |
330 | u64 start, u64 end, | |
331 | struct async_cow *async_cow, | |
332 | int *num_added) | |
b888db2b CM |
333 | { |
334 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
335 | struct btrfs_trans_handle *trans; | |
db94535d | 336 | u64 num_bytes; |
db94535d | 337 | u64 blocksize = root->sectorsize; |
c8b97818 | 338 | u64 actual_end; |
42dc7bab | 339 | u64 isize = i_size_read(inode); |
e6dcd2dc | 340 | int ret = 0; |
c8b97818 CM |
341 | struct page **pages = NULL; |
342 | unsigned long nr_pages; | |
343 | unsigned long nr_pages_ret = 0; | |
344 | unsigned long total_compressed = 0; | |
345 | unsigned long total_in = 0; | |
346 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 347 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
348 | int i; |
349 | int will_compress; | |
261507a0 | 350 | int compress_type = root->fs_info->compress_type; |
b888db2b | 351 | |
4cb13e5d LB |
352 | /* if this is a small write inside eof, kick off a defrag */ |
353 | if ((end - start + 1) < 16 * 1024 && | |
354 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
355 | btrfs_add_inode_defrag(NULL, inode); |
356 | ||
42dc7bab | 357 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
358 | again: |
359 | will_compress = 0; | |
360 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
361 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 362 | |
f03d9301 CM |
363 | /* |
364 | * we don't want to send crud past the end of i_size through | |
365 | * compression, that's just a waste of CPU time. So, if the | |
366 | * end of the file is before the start of our current | |
367 | * requested range of bytes, we bail out to the uncompressed | |
368 | * cleanup code that can deal with all of this. | |
369 | * | |
370 | * It isn't really the fastest way to fix things, but this is a | |
371 | * very uncommon corner. | |
372 | */ | |
373 | if (actual_end <= start) | |
374 | goto cleanup_and_bail_uncompressed; | |
375 | ||
c8b97818 CM |
376 | total_compressed = actual_end - start; |
377 | ||
378 | /* we want to make sure that amount of ram required to uncompress | |
379 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
380 | * of a compressed extent to 128k. This is a crucial number |
381 | * because it also controls how easily we can spread reads across | |
382 | * cpus for decompression. | |
383 | * | |
384 | * We also want to make sure the amount of IO required to do | |
385 | * a random read is reasonably small, so we limit the size of | |
386 | * a compressed extent to 128k. | |
c8b97818 CM |
387 | */ |
388 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 389 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 390 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
391 | total_in = 0; |
392 | ret = 0; | |
db94535d | 393 | |
771ed689 CM |
394 | /* |
395 | * we do compression for mount -o compress and when the | |
396 | * inode has not been flagged as nocompress. This flag can | |
397 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 398 | */ |
6cbff00f | 399 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 400 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
401 | (BTRFS_I(inode)->force_compress) || |
402 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 403 | WARN_ON(pages); |
cfbc246e | 404 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
405 | if (!pages) { |
406 | /* just bail out to the uncompressed code */ | |
407 | goto cont; | |
408 | } | |
c8b97818 | 409 | |
261507a0 LZ |
410 | if (BTRFS_I(inode)->force_compress) |
411 | compress_type = BTRFS_I(inode)->force_compress; | |
412 | ||
413 | ret = btrfs_compress_pages(compress_type, | |
414 | inode->i_mapping, start, | |
415 | total_compressed, pages, | |
416 | nr_pages, &nr_pages_ret, | |
417 | &total_in, | |
418 | &total_compressed, | |
419 | max_compressed); | |
c8b97818 CM |
420 | |
421 | if (!ret) { | |
422 | unsigned long offset = total_compressed & | |
423 | (PAGE_CACHE_SIZE - 1); | |
424 | struct page *page = pages[nr_pages_ret - 1]; | |
425 | char *kaddr; | |
426 | ||
427 | /* zero the tail end of the last page, we might be | |
428 | * sending it down to disk | |
429 | */ | |
430 | if (offset) { | |
7ac687d9 | 431 | kaddr = kmap_atomic(page); |
c8b97818 CM |
432 | memset(kaddr + offset, 0, |
433 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 434 | kunmap_atomic(kaddr); |
c8b97818 CM |
435 | } |
436 | will_compress = 1; | |
437 | } | |
438 | } | |
560f7d75 | 439 | cont: |
c8b97818 | 440 | if (start == 0) { |
7a7eaa40 | 441 | trans = btrfs_join_transaction(root); |
79787eaa JM |
442 | if (IS_ERR(trans)) { |
443 | ret = PTR_ERR(trans); | |
444 | trans = NULL; | |
445 | goto cleanup_and_out; | |
446 | } | |
0ca1f7ce | 447 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 448 | |
c8b97818 | 449 | /* lets try to make an inline extent */ |
771ed689 | 450 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 451 | /* we didn't compress the entire range, try |
771ed689 | 452 | * to make an uncompressed inline extent. |
c8b97818 CM |
453 | */ |
454 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 455 | start, end, 0, 0, NULL); |
c8b97818 | 456 | } else { |
771ed689 | 457 | /* try making a compressed inline extent */ |
c8b97818 CM |
458 | ret = cow_file_range_inline(trans, root, inode, |
459 | start, end, | |
fe3f566c LZ |
460 | total_compressed, |
461 | compress_type, pages); | |
c8b97818 | 462 | } |
79787eaa | 463 | if (ret <= 0) { |
771ed689 | 464 | /* |
79787eaa JM |
465 | * inline extent creation worked or returned error, |
466 | * we don't need to create any more async work items. | |
467 | * Unlock and free up our temp pages. | |
771ed689 | 468 | */ |
c8b97818 | 469 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
470 | &BTRFS_I(inode)->io_tree, |
471 | start, end, NULL, | |
472 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 473 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 474 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
475 | |
476 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
477 | goto free_pages_out; |
478 | } | |
c2167754 | 479 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
480 | } |
481 | ||
482 | if (will_compress) { | |
483 | /* | |
484 | * we aren't doing an inline extent round the compressed size | |
485 | * up to a block size boundary so the allocator does sane | |
486 | * things | |
487 | */ | |
488 | total_compressed = (total_compressed + blocksize - 1) & | |
489 | ~(blocksize - 1); | |
490 | ||
491 | /* | |
492 | * one last check to make sure the compression is really a | |
493 | * win, compare the page count read with the blocks on disk | |
494 | */ | |
495 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
496 | ~(PAGE_CACHE_SIZE - 1); | |
497 | if (total_compressed >= total_in) { | |
498 | will_compress = 0; | |
499 | } else { | |
c8b97818 CM |
500 | num_bytes = total_in; |
501 | } | |
502 | } | |
503 | if (!will_compress && pages) { | |
504 | /* | |
505 | * the compression code ran but failed to make things smaller, | |
506 | * free any pages it allocated and our page pointer array | |
507 | */ | |
508 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 509 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
510 | page_cache_release(pages[i]); |
511 | } | |
512 | kfree(pages); | |
513 | pages = NULL; | |
514 | total_compressed = 0; | |
515 | nr_pages_ret = 0; | |
516 | ||
517 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
518 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
519 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 520 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 521 | } |
c8b97818 | 522 | } |
771ed689 CM |
523 | if (will_compress) { |
524 | *num_added += 1; | |
c8b97818 | 525 | |
771ed689 CM |
526 | /* the async work queues will take care of doing actual |
527 | * allocation on disk for these compressed pages, | |
528 | * and will submit them to the elevator. | |
529 | */ | |
530 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
531 | total_compressed, pages, nr_pages_ret, |
532 | compress_type); | |
179e29e4 | 533 | |
24ae6365 | 534 | if (start + num_bytes < end) { |
771ed689 CM |
535 | start += num_bytes; |
536 | pages = NULL; | |
537 | cond_resched(); | |
538 | goto again; | |
539 | } | |
540 | } else { | |
f03d9301 | 541 | cleanup_and_bail_uncompressed: |
771ed689 CM |
542 | /* |
543 | * No compression, but we still need to write the pages in | |
544 | * the file we've been given so far. redirty the locked | |
545 | * page if it corresponds to our extent and set things up | |
546 | * for the async work queue to run cow_file_range to do | |
547 | * the normal delalloc dance | |
548 | */ | |
549 | if (page_offset(locked_page) >= start && | |
550 | page_offset(locked_page) <= end) { | |
551 | __set_page_dirty_nobuffers(locked_page); | |
552 | /* unlocked later on in the async handlers */ | |
553 | } | |
261507a0 LZ |
554 | add_async_extent(async_cow, start, end - start + 1, |
555 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
556 | *num_added += 1; |
557 | } | |
3b951516 | 558 | |
771ed689 | 559 | out: |
79787eaa | 560 | return ret; |
771ed689 CM |
561 | |
562 | free_pages_out: | |
563 | for (i = 0; i < nr_pages_ret; i++) { | |
564 | WARN_ON(pages[i]->mapping); | |
565 | page_cache_release(pages[i]); | |
566 | } | |
d397712b | 567 | kfree(pages); |
771ed689 CM |
568 | |
569 | goto out; | |
79787eaa JM |
570 | |
571 | cleanup_and_out: | |
572 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
573 | start, end, NULL, | |
574 | EXTENT_CLEAR_UNLOCK_PAGE | | |
575 | EXTENT_CLEAR_DIRTY | | |
576 | EXTENT_CLEAR_DELALLOC | | |
577 | EXTENT_SET_WRITEBACK | | |
578 | EXTENT_END_WRITEBACK); | |
579 | if (!trans || IS_ERR(trans)) | |
580 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
581 | else | |
582 | btrfs_abort_transaction(trans, root, ret); | |
583 | goto free_pages_out; | |
771ed689 CM |
584 | } |
585 | ||
586 | /* | |
587 | * phase two of compressed writeback. This is the ordered portion | |
588 | * of the code, which only gets called in the order the work was | |
589 | * queued. We walk all the async extents created by compress_file_range | |
590 | * and send them down to the disk. | |
591 | */ | |
592 | static noinline int submit_compressed_extents(struct inode *inode, | |
593 | struct async_cow *async_cow) | |
594 | { | |
595 | struct async_extent *async_extent; | |
596 | u64 alloc_hint = 0; | |
597 | struct btrfs_trans_handle *trans; | |
598 | struct btrfs_key ins; | |
599 | struct extent_map *em; | |
600 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
601 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
602 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 603 | int ret = 0; |
771ed689 CM |
604 | |
605 | if (list_empty(&async_cow->extents)) | |
606 | return 0; | |
607 | ||
771ed689 | 608 | |
d397712b | 609 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
610 | async_extent = list_entry(async_cow->extents.next, |
611 | struct async_extent, list); | |
612 | list_del(&async_extent->list); | |
c8b97818 | 613 | |
771ed689 CM |
614 | io_tree = &BTRFS_I(inode)->io_tree; |
615 | ||
f5a84ee3 | 616 | retry: |
771ed689 CM |
617 | /* did the compression code fall back to uncompressed IO? */ |
618 | if (!async_extent->pages) { | |
619 | int page_started = 0; | |
620 | unsigned long nr_written = 0; | |
621 | ||
622 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 623 | async_extent->start + |
d0082371 | 624 | async_extent->ram_size - 1); |
771ed689 CM |
625 | |
626 | /* allocate blocks */ | |
f5a84ee3 JB |
627 | ret = cow_file_range(inode, async_cow->locked_page, |
628 | async_extent->start, | |
629 | async_extent->start + | |
630 | async_extent->ram_size - 1, | |
631 | &page_started, &nr_written, 0); | |
771ed689 | 632 | |
79787eaa JM |
633 | /* JDM XXX */ |
634 | ||
771ed689 CM |
635 | /* |
636 | * if page_started, cow_file_range inserted an | |
637 | * inline extent and took care of all the unlocking | |
638 | * and IO for us. Otherwise, we need to submit | |
639 | * all those pages down to the drive. | |
640 | */ | |
f5a84ee3 | 641 | if (!page_started && !ret) |
771ed689 CM |
642 | extent_write_locked_range(io_tree, |
643 | inode, async_extent->start, | |
d397712b | 644 | async_extent->start + |
771ed689 CM |
645 | async_extent->ram_size - 1, |
646 | btrfs_get_extent, | |
647 | WB_SYNC_ALL); | |
648 | kfree(async_extent); | |
649 | cond_resched(); | |
650 | continue; | |
651 | } | |
652 | ||
653 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 654 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 655 | |
7a7eaa40 | 656 | trans = btrfs_join_transaction(root); |
79787eaa JM |
657 | if (IS_ERR(trans)) { |
658 | ret = PTR_ERR(trans); | |
659 | } else { | |
660 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
661 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
662 | async_extent->compressed_size, |
663 | async_extent->compressed_size, | |
81c9ad23 | 664 | 0, alloc_hint, &ins, 1); |
79787eaa JM |
665 | if (ret) |
666 | btrfs_abort_transaction(trans, root, ret); | |
667 | btrfs_end_transaction(trans, root); | |
668 | } | |
c2167754 | 669 | |
f5a84ee3 JB |
670 | if (ret) { |
671 | int i; | |
672 | for (i = 0; i < async_extent->nr_pages; i++) { | |
673 | WARN_ON(async_extent->pages[i]->mapping); | |
674 | page_cache_release(async_extent->pages[i]); | |
675 | } | |
676 | kfree(async_extent->pages); | |
677 | async_extent->nr_pages = 0; | |
678 | async_extent->pages = NULL; | |
679 | unlock_extent(io_tree, async_extent->start, | |
680 | async_extent->start + | |
d0082371 | 681 | async_extent->ram_size - 1); |
79787eaa JM |
682 | if (ret == -ENOSPC) |
683 | goto retry; | |
684 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
685 | } |
686 | ||
c2167754 YZ |
687 | /* |
688 | * here we're doing allocation and writeback of the | |
689 | * compressed pages | |
690 | */ | |
691 | btrfs_drop_extent_cache(inode, async_extent->start, | |
692 | async_extent->start + | |
693 | async_extent->ram_size - 1, 0); | |
694 | ||
172ddd60 | 695 | em = alloc_extent_map(); |
79787eaa | 696 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
697 | em->start = async_extent->start; |
698 | em->len = async_extent->ram_size; | |
445a6944 | 699 | em->orig_start = em->start; |
c8b97818 | 700 | |
771ed689 CM |
701 | em->block_start = ins.objectid; |
702 | em->block_len = ins.offset; | |
703 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 704 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
705 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
706 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
707 | ||
d397712b | 708 | while (1) { |
890871be | 709 | write_lock(&em_tree->lock); |
771ed689 | 710 | ret = add_extent_mapping(em_tree, em); |
890871be | 711 | write_unlock(&em_tree->lock); |
771ed689 CM |
712 | if (ret != -EEXIST) { |
713 | free_extent_map(em); | |
714 | break; | |
715 | } | |
716 | btrfs_drop_extent_cache(inode, async_extent->start, | |
717 | async_extent->start + | |
718 | async_extent->ram_size - 1, 0); | |
719 | } | |
720 | ||
261507a0 LZ |
721 | ret = btrfs_add_ordered_extent_compress(inode, |
722 | async_extent->start, | |
723 | ins.objectid, | |
724 | async_extent->ram_size, | |
725 | ins.offset, | |
726 | BTRFS_ORDERED_COMPRESSED, | |
727 | async_extent->compress_type); | |
79787eaa | 728 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 729 | |
771ed689 CM |
730 | /* |
731 | * clear dirty, set writeback and unlock the pages. | |
732 | */ | |
733 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
734 | &BTRFS_I(inode)->io_tree, |
735 | async_extent->start, | |
736 | async_extent->start + | |
737 | async_extent->ram_size - 1, | |
738 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
739 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 740 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 741 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
742 | |
743 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
744 | async_extent->start, |
745 | async_extent->ram_size, | |
746 | ins.objectid, | |
747 | ins.offset, async_extent->pages, | |
748 | async_extent->nr_pages); | |
771ed689 | 749 | |
79787eaa | 750 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
751 | alloc_hint = ins.objectid + ins.offset; |
752 | kfree(async_extent); | |
753 | cond_resched(); | |
754 | } | |
79787eaa JM |
755 | ret = 0; |
756 | out: | |
757 | return ret; | |
758 | out_free: | |
759 | kfree(async_extent); | |
760 | goto out; | |
771ed689 CM |
761 | } |
762 | ||
4b46fce2 JB |
763 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
764 | u64 num_bytes) | |
765 | { | |
766 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
767 | struct extent_map *em; | |
768 | u64 alloc_hint = 0; | |
769 | ||
770 | read_lock(&em_tree->lock); | |
771 | em = search_extent_mapping(em_tree, start, num_bytes); | |
772 | if (em) { | |
773 | /* | |
774 | * if block start isn't an actual block number then find the | |
775 | * first block in this inode and use that as a hint. If that | |
776 | * block is also bogus then just don't worry about it. | |
777 | */ | |
778 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
779 | free_extent_map(em); | |
780 | em = search_extent_mapping(em_tree, 0, 0); | |
781 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
782 | alloc_hint = em->block_start; | |
783 | if (em) | |
784 | free_extent_map(em); | |
785 | } else { | |
786 | alloc_hint = em->block_start; | |
787 | free_extent_map(em); | |
788 | } | |
789 | } | |
790 | read_unlock(&em_tree->lock); | |
791 | ||
792 | return alloc_hint; | |
793 | } | |
794 | ||
771ed689 CM |
795 | /* |
796 | * when extent_io.c finds a delayed allocation range in the file, | |
797 | * the call backs end up in this code. The basic idea is to | |
798 | * allocate extents on disk for the range, and create ordered data structs | |
799 | * in ram to track those extents. | |
800 | * | |
801 | * locked_page is the page that writepage had locked already. We use | |
802 | * it to make sure we don't do extra locks or unlocks. | |
803 | * | |
804 | * *page_started is set to one if we unlock locked_page and do everything | |
805 | * required to start IO on it. It may be clean and already done with | |
806 | * IO when we return. | |
807 | */ | |
808 | static noinline int cow_file_range(struct inode *inode, | |
809 | struct page *locked_page, | |
810 | u64 start, u64 end, int *page_started, | |
811 | unsigned long *nr_written, | |
812 | int unlock) | |
813 | { | |
814 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
815 | struct btrfs_trans_handle *trans; | |
816 | u64 alloc_hint = 0; | |
817 | u64 num_bytes; | |
818 | unsigned long ram_size; | |
819 | u64 disk_num_bytes; | |
820 | u64 cur_alloc_size; | |
821 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
822 | struct btrfs_key ins; |
823 | struct extent_map *em; | |
824 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
825 | int ret = 0; | |
826 | ||
83eea1f1 | 827 | BUG_ON(btrfs_is_free_space_inode(inode)); |
7a7eaa40 | 828 | trans = btrfs_join_transaction(root); |
79787eaa JM |
829 | if (IS_ERR(trans)) { |
830 | extent_clear_unlock_delalloc(inode, | |
831 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 832 | start, end, locked_page, |
79787eaa JM |
833 | EXTENT_CLEAR_UNLOCK_PAGE | |
834 | EXTENT_CLEAR_UNLOCK | | |
835 | EXTENT_CLEAR_DELALLOC | | |
836 | EXTENT_CLEAR_DIRTY | | |
837 | EXTENT_SET_WRITEBACK | | |
838 | EXTENT_END_WRITEBACK); | |
839 | return PTR_ERR(trans); | |
840 | } | |
0ca1f7ce | 841 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 842 | |
771ed689 CM |
843 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
844 | num_bytes = max(blocksize, num_bytes); | |
845 | disk_num_bytes = num_bytes; | |
846 | ret = 0; | |
847 | ||
4cb5300b | 848 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
849 | if (num_bytes < 64 * 1024 && |
850 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
851 | btrfs_add_inode_defrag(trans, inode); |
852 | ||
771ed689 CM |
853 | if (start == 0) { |
854 | /* lets try to make an inline extent */ | |
855 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 856 | start, end, 0, 0, NULL); |
771ed689 CM |
857 | if (ret == 0) { |
858 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
859 | &BTRFS_I(inode)->io_tree, |
860 | start, end, NULL, | |
861 | EXTENT_CLEAR_UNLOCK_PAGE | | |
862 | EXTENT_CLEAR_UNLOCK | | |
863 | EXTENT_CLEAR_DELALLOC | | |
864 | EXTENT_CLEAR_DIRTY | | |
865 | EXTENT_SET_WRITEBACK | | |
866 | EXTENT_END_WRITEBACK); | |
c2167754 | 867 | |
771ed689 CM |
868 | *nr_written = *nr_written + |
869 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
870 | *page_started = 1; | |
771ed689 | 871 | goto out; |
79787eaa JM |
872 | } else if (ret < 0) { |
873 | btrfs_abort_transaction(trans, root, ret); | |
874 | goto out_unlock; | |
771ed689 CM |
875 | } |
876 | } | |
877 | ||
878 | BUG_ON(disk_num_bytes > | |
6c41761f | 879 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 880 | |
4b46fce2 | 881 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
882 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
883 | ||
d397712b | 884 | while (disk_num_bytes > 0) { |
a791e35e CM |
885 | unsigned long op; |
886 | ||
287a0ab9 | 887 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 888 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 889 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 890 | &ins, 1); |
79787eaa JM |
891 | if (ret < 0) { |
892 | btrfs_abort_transaction(trans, root, ret); | |
893 | goto out_unlock; | |
894 | } | |
d397712b | 895 | |
172ddd60 | 896 | em = alloc_extent_map(); |
79787eaa | 897 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 898 | em->start = start; |
445a6944 | 899 | em->orig_start = em->start; |
771ed689 CM |
900 | ram_size = ins.offset; |
901 | em->len = ins.offset; | |
c8b97818 | 902 | |
e6dcd2dc | 903 | em->block_start = ins.objectid; |
c8b97818 | 904 | em->block_len = ins.offset; |
e6dcd2dc | 905 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 906 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 907 | |
d397712b | 908 | while (1) { |
890871be | 909 | write_lock(&em_tree->lock); |
e6dcd2dc | 910 | ret = add_extent_mapping(em_tree, em); |
890871be | 911 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
912 | if (ret != -EEXIST) { |
913 | free_extent_map(em); | |
914 | break; | |
915 | } | |
916 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 917 | start + ram_size - 1, 0); |
e6dcd2dc CM |
918 | } |
919 | ||
98d20f67 | 920 | cur_alloc_size = ins.offset; |
e6dcd2dc | 921 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 922 | ram_size, cur_alloc_size, 0); |
79787eaa | 923 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 924 | |
17d217fe YZ |
925 | if (root->root_key.objectid == |
926 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
927 | ret = btrfs_reloc_clone_csums(inode, start, | |
928 | cur_alloc_size); | |
79787eaa JM |
929 | if (ret) { |
930 | btrfs_abort_transaction(trans, root, ret); | |
931 | goto out_unlock; | |
932 | } | |
17d217fe YZ |
933 | } |
934 | ||
d397712b | 935 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 936 | break; |
d397712b | 937 | |
c8b97818 CM |
938 | /* we're not doing compressed IO, don't unlock the first |
939 | * page (which the caller expects to stay locked), don't | |
940 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
941 | * |
942 | * Do set the Private2 bit so we know this page was properly | |
943 | * setup for writepage | |
c8b97818 | 944 | */ |
a791e35e CM |
945 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
946 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
947 | EXTENT_SET_PRIVATE2; | |
948 | ||
c8b97818 CM |
949 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
950 | start, start + ram_size - 1, | |
a791e35e | 951 | locked_page, op); |
c8b97818 | 952 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
953 | num_bytes -= cur_alloc_size; |
954 | alloc_hint = ins.objectid + ins.offset; | |
955 | start += cur_alloc_size; | |
b888db2b | 956 | } |
771ed689 | 957 | ret = 0; |
79787eaa | 958 | out: |
b888db2b | 959 | btrfs_end_transaction(trans, root); |
c8b97818 | 960 | |
be20aa9d | 961 | return ret; |
79787eaa JM |
962 | out_unlock: |
963 | extent_clear_unlock_delalloc(inode, | |
964 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 965 | start, end, locked_page, |
79787eaa JM |
966 | EXTENT_CLEAR_UNLOCK_PAGE | |
967 | EXTENT_CLEAR_UNLOCK | | |
968 | EXTENT_CLEAR_DELALLOC | | |
969 | EXTENT_CLEAR_DIRTY | | |
970 | EXTENT_SET_WRITEBACK | | |
971 | EXTENT_END_WRITEBACK); | |
972 | ||
973 | goto out; | |
771ed689 | 974 | } |
c8b97818 | 975 | |
771ed689 CM |
976 | /* |
977 | * work queue call back to started compression on a file and pages | |
978 | */ | |
979 | static noinline void async_cow_start(struct btrfs_work *work) | |
980 | { | |
981 | struct async_cow *async_cow; | |
982 | int num_added = 0; | |
983 | async_cow = container_of(work, struct async_cow, work); | |
984 | ||
985 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
986 | async_cow->start, async_cow->end, async_cow, | |
987 | &num_added); | |
8180ef88 | 988 | if (num_added == 0) { |
cb77fcd8 | 989 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 990 | async_cow->inode = NULL; |
8180ef88 | 991 | } |
771ed689 CM |
992 | } |
993 | ||
994 | /* | |
995 | * work queue call back to submit previously compressed pages | |
996 | */ | |
997 | static noinline void async_cow_submit(struct btrfs_work *work) | |
998 | { | |
999 | struct async_cow *async_cow; | |
1000 | struct btrfs_root *root; | |
1001 | unsigned long nr_pages; | |
1002 | ||
1003 | async_cow = container_of(work, struct async_cow, work); | |
1004 | ||
1005 | root = async_cow->root; | |
1006 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1007 | PAGE_CACHE_SHIFT; | |
1008 | ||
66657b31 | 1009 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1010 | 5 * 1024 * 1024 && |
771ed689 CM |
1011 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1012 | wake_up(&root->fs_info->async_submit_wait); | |
1013 | ||
d397712b | 1014 | if (async_cow->inode) |
771ed689 | 1015 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1016 | } |
c8b97818 | 1017 | |
771ed689 CM |
1018 | static noinline void async_cow_free(struct btrfs_work *work) |
1019 | { | |
1020 | struct async_cow *async_cow; | |
1021 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1022 | if (async_cow->inode) |
cb77fcd8 | 1023 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1024 | kfree(async_cow); |
1025 | } | |
1026 | ||
1027 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1028 | u64 start, u64 end, int *page_started, | |
1029 | unsigned long *nr_written) | |
1030 | { | |
1031 | struct async_cow *async_cow; | |
1032 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1033 | unsigned long nr_pages; | |
1034 | u64 cur_end; | |
287082b0 | 1035 | int limit = 10 * 1024 * 1024; |
771ed689 | 1036 | |
a3429ab7 CM |
1037 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1038 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1039 | while (start < end) { |
771ed689 | 1040 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1041 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1042 | async_cow->inode = igrab(inode); |
771ed689 CM |
1043 | async_cow->root = root; |
1044 | async_cow->locked_page = locked_page; | |
1045 | async_cow->start = start; | |
1046 | ||
6cbff00f | 1047 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1048 | cur_end = end; |
1049 | else | |
1050 | cur_end = min(end, start + 512 * 1024 - 1); | |
1051 | ||
1052 | async_cow->end = cur_end; | |
1053 | INIT_LIST_HEAD(&async_cow->extents); | |
1054 | ||
1055 | async_cow->work.func = async_cow_start; | |
1056 | async_cow->work.ordered_func = async_cow_submit; | |
1057 | async_cow->work.ordered_free = async_cow_free; | |
1058 | async_cow->work.flags = 0; | |
1059 | ||
771ed689 CM |
1060 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1061 | PAGE_CACHE_SHIFT; | |
1062 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1063 | ||
1064 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1065 | &async_cow->work); | |
1066 | ||
1067 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1068 | wait_event(root->fs_info->async_submit_wait, | |
1069 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1070 | limit)); | |
1071 | } | |
1072 | ||
d397712b | 1073 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1074 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1075 | wait_event(root->fs_info->async_submit_wait, | |
1076 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1077 | 0)); | |
1078 | } | |
1079 | ||
1080 | *nr_written += nr_pages; | |
1081 | start = cur_end + 1; | |
1082 | } | |
1083 | *page_started = 1; | |
1084 | return 0; | |
be20aa9d CM |
1085 | } |
1086 | ||
d397712b | 1087 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1088 | u64 bytenr, u64 num_bytes) |
1089 | { | |
1090 | int ret; | |
1091 | struct btrfs_ordered_sum *sums; | |
1092 | LIST_HEAD(list); | |
1093 | ||
07d400a6 | 1094 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1095 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1096 | if (ret == 0 && list_empty(&list)) |
1097 | return 0; | |
1098 | ||
1099 | while (!list_empty(&list)) { | |
1100 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1101 | list_del(&sums->list); | |
1102 | kfree(sums); | |
1103 | } | |
1104 | return 1; | |
1105 | } | |
1106 | ||
d352ac68 CM |
1107 | /* |
1108 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1109 | * of the extents that exist in the file, and COWs the file as required. | |
1110 | * | |
1111 | * If no cow copies or snapshots exist, we write directly to the existing | |
1112 | * blocks on disk | |
1113 | */ | |
7f366cfe CM |
1114 | static noinline int run_delalloc_nocow(struct inode *inode, |
1115 | struct page *locked_page, | |
771ed689 CM |
1116 | u64 start, u64 end, int *page_started, int force, |
1117 | unsigned long *nr_written) | |
be20aa9d | 1118 | { |
be20aa9d | 1119 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1120 | struct btrfs_trans_handle *trans; |
be20aa9d | 1121 | struct extent_buffer *leaf; |
be20aa9d | 1122 | struct btrfs_path *path; |
80ff3856 | 1123 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1124 | struct btrfs_key found_key; |
80ff3856 YZ |
1125 | u64 cow_start; |
1126 | u64 cur_offset; | |
1127 | u64 extent_end; | |
5d4f98a2 | 1128 | u64 extent_offset; |
80ff3856 YZ |
1129 | u64 disk_bytenr; |
1130 | u64 num_bytes; | |
1131 | int extent_type; | |
79787eaa | 1132 | int ret, err; |
d899e052 | 1133 | int type; |
80ff3856 YZ |
1134 | int nocow; |
1135 | int check_prev = 1; | |
82d5902d | 1136 | bool nolock; |
33345d01 | 1137 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1138 | |
1139 | path = btrfs_alloc_path(); | |
17ca04af JB |
1140 | if (!path) { |
1141 | extent_clear_unlock_delalloc(inode, | |
1142 | &BTRFS_I(inode)->io_tree, | |
1143 | start, end, locked_page, | |
1144 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1145 | EXTENT_CLEAR_UNLOCK | | |
1146 | EXTENT_CLEAR_DELALLOC | | |
1147 | EXTENT_CLEAR_DIRTY | | |
1148 | EXTENT_SET_WRITEBACK | | |
1149 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1150 | return -ENOMEM; |
17ca04af | 1151 | } |
82d5902d | 1152 | |
83eea1f1 | 1153 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1154 | |
1155 | if (nolock) | |
7a7eaa40 | 1156 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1157 | else |
7a7eaa40 | 1158 | trans = btrfs_join_transaction(root); |
ff5714cc | 1159 | |
79787eaa | 1160 | if (IS_ERR(trans)) { |
17ca04af JB |
1161 | extent_clear_unlock_delalloc(inode, |
1162 | &BTRFS_I(inode)->io_tree, | |
1163 | start, end, locked_page, | |
1164 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1165 | EXTENT_CLEAR_UNLOCK | | |
1166 | EXTENT_CLEAR_DELALLOC | | |
1167 | EXTENT_CLEAR_DIRTY | | |
1168 | EXTENT_SET_WRITEBACK | | |
1169 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1170 | btrfs_free_path(path); |
1171 | return PTR_ERR(trans); | |
1172 | } | |
1173 | ||
74b21075 | 1174 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1175 | |
80ff3856 YZ |
1176 | cow_start = (u64)-1; |
1177 | cur_offset = start; | |
1178 | while (1) { | |
33345d01 | 1179 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1180 | cur_offset, 0); |
79787eaa JM |
1181 | if (ret < 0) { |
1182 | btrfs_abort_transaction(trans, root, ret); | |
1183 | goto error; | |
1184 | } | |
80ff3856 YZ |
1185 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1186 | leaf = path->nodes[0]; | |
1187 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1188 | path->slots[0] - 1); | |
33345d01 | 1189 | if (found_key.objectid == ino && |
80ff3856 YZ |
1190 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1191 | path->slots[0]--; | |
1192 | } | |
1193 | check_prev = 0; | |
1194 | next_slot: | |
1195 | leaf = path->nodes[0]; | |
1196 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1197 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1198 | if (ret < 0) { |
1199 | btrfs_abort_transaction(trans, root, ret); | |
1200 | goto error; | |
1201 | } | |
80ff3856 YZ |
1202 | if (ret > 0) |
1203 | break; | |
1204 | leaf = path->nodes[0]; | |
1205 | } | |
be20aa9d | 1206 | |
80ff3856 YZ |
1207 | nocow = 0; |
1208 | disk_bytenr = 0; | |
17d217fe | 1209 | num_bytes = 0; |
80ff3856 YZ |
1210 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1211 | ||
33345d01 | 1212 | if (found_key.objectid > ino || |
80ff3856 YZ |
1213 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1214 | found_key.offset > end) | |
1215 | break; | |
1216 | ||
1217 | if (found_key.offset > cur_offset) { | |
1218 | extent_end = found_key.offset; | |
e9061e21 | 1219 | extent_type = 0; |
80ff3856 YZ |
1220 | goto out_check; |
1221 | } | |
1222 | ||
1223 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1224 | struct btrfs_file_extent_item); | |
1225 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1226 | ||
d899e052 YZ |
1227 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1228 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1229 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1230 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1231 | extent_end = found_key.offset + |
1232 | btrfs_file_extent_num_bytes(leaf, fi); | |
1233 | if (extent_end <= start) { | |
1234 | path->slots[0]++; | |
1235 | goto next_slot; | |
1236 | } | |
17d217fe YZ |
1237 | if (disk_bytenr == 0) |
1238 | goto out_check; | |
80ff3856 YZ |
1239 | if (btrfs_file_extent_compression(leaf, fi) || |
1240 | btrfs_file_extent_encryption(leaf, fi) || | |
1241 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1242 | goto out_check; | |
d899e052 YZ |
1243 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1244 | goto out_check; | |
d2fb3437 | 1245 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1246 | goto out_check; |
33345d01 | 1247 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1248 | found_key.offset - |
1249 | extent_offset, disk_bytenr)) | |
17d217fe | 1250 | goto out_check; |
5d4f98a2 | 1251 | disk_bytenr += extent_offset; |
17d217fe YZ |
1252 | disk_bytenr += cur_offset - found_key.offset; |
1253 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1254 | /* | |
1255 | * force cow if csum exists in the range. | |
1256 | * this ensure that csum for a given extent are | |
1257 | * either valid or do not exist. | |
1258 | */ | |
1259 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1260 | goto out_check; | |
80ff3856 YZ |
1261 | nocow = 1; |
1262 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1263 | extent_end = found_key.offset + | |
1264 | btrfs_file_extent_inline_len(leaf, fi); | |
1265 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1266 | } else { | |
1267 | BUG_ON(1); | |
1268 | } | |
1269 | out_check: | |
1270 | if (extent_end <= start) { | |
1271 | path->slots[0]++; | |
1272 | goto next_slot; | |
1273 | } | |
1274 | if (!nocow) { | |
1275 | if (cow_start == (u64)-1) | |
1276 | cow_start = cur_offset; | |
1277 | cur_offset = extent_end; | |
1278 | if (cur_offset > end) | |
1279 | break; | |
1280 | path->slots[0]++; | |
1281 | goto next_slot; | |
7ea394f1 YZ |
1282 | } |
1283 | ||
b3b4aa74 | 1284 | btrfs_release_path(path); |
80ff3856 YZ |
1285 | if (cow_start != (u64)-1) { |
1286 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1287 | found_key.offset - 1, page_started, |
1288 | nr_written, 1); | |
79787eaa JM |
1289 | if (ret) { |
1290 | btrfs_abort_transaction(trans, root, ret); | |
1291 | goto error; | |
1292 | } | |
80ff3856 | 1293 | cow_start = (u64)-1; |
7ea394f1 | 1294 | } |
80ff3856 | 1295 | |
d899e052 YZ |
1296 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1297 | struct extent_map *em; | |
1298 | struct extent_map_tree *em_tree; | |
1299 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1300 | em = alloc_extent_map(); |
79787eaa | 1301 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1302 | em->start = cur_offset; |
445a6944 | 1303 | em->orig_start = em->start; |
d899e052 YZ |
1304 | em->len = num_bytes; |
1305 | em->block_len = num_bytes; | |
1306 | em->block_start = disk_bytenr; | |
1307 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1308 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
4e2f84e6 | 1309 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
d899e052 | 1310 | while (1) { |
890871be | 1311 | write_lock(&em_tree->lock); |
d899e052 | 1312 | ret = add_extent_mapping(em_tree, em); |
890871be | 1313 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1314 | if (ret != -EEXIST) { |
1315 | free_extent_map(em); | |
1316 | break; | |
1317 | } | |
1318 | btrfs_drop_extent_cache(inode, em->start, | |
1319 | em->start + em->len - 1, 0); | |
1320 | } | |
1321 | type = BTRFS_ORDERED_PREALLOC; | |
1322 | } else { | |
1323 | type = BTRFS_ORDERED_NOCOW; | |
1324 | } | |
80ff3856 YZ |
1325 | |
1326 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1327 | num_bytes, num_bytes, type); |
79787eaa | 1328 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1329 | |
efa56464 YZ |
1330 | if (root->root_key.objectid == |
1331 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1332 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1333 | num_bytes); | |
79787eaa JM |
1334 | if (ret) { |
1335 | btrfs_abort_transaction(trans, root, ret); | |
1336 | goto error; | |
1337 | } | |
efa56464 YZ |
1338 | } |
1339 | ||
d899e052 | 1340 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1341 | cur_offset, cur_offset + num_bytes - 1, |
1342 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1343 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1344 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1345 | cur_offset = extent_end; |
1346 | if (cur_offset > end) | |
1347 | break; | |
be20aa9d | 1348 | } |
b3b4aa74 | 1349 | btrfs_release_path(path); |
80ff3856 | 1350 | |
17ca04af | 1351 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1352 | cow_start = cur_offset; |
17ca04af JB |
1353 | cur_offset = end; |
1354 | } | |
1355 | ||
80ff3856 YZ |
1356 | if (cow_start != (u64)-1) { |
1357 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1358 | page_started, nr_written, 1); |
79787eaa JM |
1359 | if (ret) { |
1360 | btrfs_abort_transaction(trans, root, ret); | |
1361 | goto error; | |
1362 | } | |
80ff3856 YZ |
1363 | } |
1364 | ||
79787eaa | 1365 | error: |
0cb59c99 | 1366 | if (nolock) { |
79787eaa | 1367 | err = btrfs_end_transaction_nolock(trans, root); |
0cb59c99 | 1368 | } else { |
79787eaa | 1369 | err = btrfs_end_transaction(trans, root); |
0cb59c99 | 1370 | } |
79787eaa JM |
1371 | if (!ret) |
1372 | ret = err; | |
1373 | ||
17ca04af JB |
1374 | if (ret && cur_offset < end) |
1375 | extent_clear_unlock_delalloc(inode, | |
1376 | &BTRFS_I(inode)->io_tree, | |
1377 | cur_offset, end, locked_page, | |
1378 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1379 | EXTENT_CLEAR_UNLOCK | | |
1380 | EXTENT_CLEAR_DELALLOC | | |
1381 | EXTENT_CLEAR_DIRTY | | |
1382 | EXTENT_SET_WRITEBACK | | |
1383 | EXTENT_END_WRITEBACK); | |
1384 | ||
7ea394f1 | 1385 | btrfs_free_path(path); |
79787eaa | 1386 | return ret; |
be20aa9d CM |
1387 | } |
1388 | ||
d352ac68 CM |
1389 | /* |
1390 | * extent_io.c call back to do delayed allocation processing | |
1391 | */ | |
c8b97818 | 1392 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1393 | u64 start, u64 end, int *page_started, |
1394 | unsigned long *nr_written) | |
be20aa9d | 1395 | { |
be20aa9d | 1396 | int ret; |
7f366cfe | 1397 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1398 | |
7ddf5a42 | 1399 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1400 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1401 | page_started, 1, nr_written); |
7ddf5a42 | 1402 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1403 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1404 | page_started, 0, nr_written); |
7ddf5a42 JB |
1405 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1406 | !(BTRFS_I(inode)->force_compress) && | |
1407 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1408 | ret = cow_file_range(inode, locked_page, start, end, |
1409 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1410 | } else { |
1411 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1412 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1413 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1414 | page_started, nr_written); |
7ddf5a42 | 1415 | } |
b888db2b CM |
1416 | return ret; |
1417 | } | |
1418 | ||
1bf85046 JM |
1419 | static void btrfs_split_extent_hook(struct inode *inode, |
1420 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1421 | { |
0ca1f7ce | 1422 | /* not delalloc, ignore it */ |
9ed74f2d | 1423 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1424 | return; |
9ed74f2d | 1425 | |
9e0baf60 JB |
1426 | spin_lock(&BTRFS_I(inode)->lock); |
1427 | BTRFS_I(inode)->outstanding_extents++; | |
1428 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1429 | } |
1430 | ||
1431 | /* | |
1432 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1433 | * extents so we can keep track of new extents that are just merged onto old | |
1434 | * extents, such as when we are doing sequential writes, so we can properly | |
1435 | * account for the metadata space we'll need. | |
1436 | */ | |
1bf85046 JM |
1437 | static void btrfs_merge_extent_hook(struct inode *inode, |
1438 | struct extent_state *new, | |
1439 | struct extent_state *other) | |
9ed74f2d | 1440 | { |
9ed74f2d JB |
1441 | /* not delalloc, ignore it */ |
1442 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1443 | return; |
9ed74f2d | 1444 | |
9e0baf60 JB |
1445 | spin_lock(&BTRFS_I(inode)->lock); |
1446 | BTRFS_I(inode)->outstanding_extents--; | |
1447 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1448 | } |
1449 | ||
d352ac68 CM |
1450 | /* |
1451 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1452 | * bytes in this file, and to maintain the list of inodes that | |
1453 | * have pending delalloc work to be done. | |
1454 | */ | |
1bf85046 JM |
1455 | static void btrfs_set_bit_hook(struct inode *inode, |
1456 | struct extent_state *state, int *bits) | |
291d673e | 1457 | { |
9ed74f2d | 1458 | |
75eff68e CM |
1459 | /* |
1460 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1461 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1462 | * bit, which is only set or cleared with irqs on |
1463 | */ | |
0ca1f7ce | 1464 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1465 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1466 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1467 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1468 | |
9e0baf60 | 1469 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1470 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1471 | } else { |
1472 | spin_lock(&BTRFS_I(inode)->lock); | |
1473 | BTRFS_I(inode)->outstanding_extents++; | |
1474 | spin_unlock(&BTRFS_I(inode)->lock); | |
1475 | } | |
287a0ab9 | 1476 | |
75eff68e | 1477 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1478 | BTRFS_I(inode)->delalloc_bytes += len; |
1479 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1480 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1481 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1482 | &root->fs_info->delalloc_inodes); | |
1483 | } | |
75eff68e | 1484 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1485 | } |
291d673e CM |
1486 | } |
1487 | ||
d352ac68 CM |
1488 | /* |
1489 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1490 | */ | |
1bf85046 JM |
1491 | static void btrfs_clear_bit_hook(struct inode *inode, |
1492 | struct extent_state *state, int *bits) | |
291d673e | 1493 | { |
75eff68e CM |
1494 | /* |
1495 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1496 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1497 | * bit, which is only set or cleared with irqs on |
1498 | */ | |
0ca1f7ce | 1499 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1500 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1501 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1502 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1503 | |
9e0baf60 | 1504 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1505 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1506 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1507 | spin_lock(&BTRFS_I(inode)->lock); | |
1508 | BTRFS_I(inode)->outstanding_extents--; | |
1509 | spin_unlock(&BTRFS_I(inode)->lock); | |
1510 | } | |
0ca1f7ce YZ |
1511 | |
1512 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1513 | btrfs_delalloc_release_metadata(inode, len); | |
1514 | ||
0cb59c99 JB |
1515 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1516 | && do_list) | |
0ca1f7ce | 1517 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1518 | |
75eff68e | 1519 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1520 | root->fs_info->delalloc_bytes -= len; |
1521 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1522 | ||
0cb59c99 | 1523 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1524 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1525 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1526 | } | |
75eff68e | 1527 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1528 | } |
291d673e CM |
1529 | } |
1530 | ||
d352ac68 CM |
1531 | /* |
1532 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1533 | * we don't create bios that span stripes or chunks | |
1534 | */ | |
239b14b3 | 1535 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1536 | size_t size, struct bio *bio, |
1537 | unsigned long bio_flags) | |
239b14b3 CM |
1538 | { |
1539 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1540 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1541 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1542 | u64 length = 0; |
1543 | u64 map_length; | |
239b14b3 CM |
1544 | int ret; |
1545 | ||
771ed689 CM |
1546 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1547 | return 0; | |
1548 | ||
f2d8d74d | 1549 | length = bio->bi_size; |
239b14b3 CM |
1550 | map_tree = &root->fs_info->mapping_tree; |
1551 | map_length = length; | |
cea9e445 | 1552 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1553 | &map_length, NULL, 0); |
3444a972 JM |
1554 | /* Will always return 0 or 1 with map_multi == NULL */ |
1555 | BUG_ON(ret < 0); | |
d397712b | 1556 | if (map_length < length + size) |
239b14b3 | 1557 | return 1; |
3444a972 | 1558 | return 0; |
239b14b3 CM |
1559 | } |
1560 | ||
d352ac68 CM |
1561 | /* |
1562 | * in order to insert checksums into the metadata in large chunks, | |
1563 | * we wait until bio submission time. All the pages in the bio are | |
1564 | * checksummed and sums are attached onto the ordered extent record. | |
1565 | * | |
1566 | * At IO completion time the cums attached on the ordered extent record | |
1567 | * are inserted into the btree | |
1568 | */ | |
d397712b CM |
1569 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1570 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1571 | unsigned long bio_flags, |
1572 | u64 bio_offset) | |
065631f6 | 1573 | { |
065631f6 | 1574 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1575 | int ret = 0; |
e015640f | 1576 | |
d20f7043 | 1577 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1578 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1579 | return 0; |
1580 | } | |
e015640f | 1581 | |
4a69a410 CM |
1582 | /* |
1583 | * in order to insert checksums into the metadata in large chunks, | |
1584 | * we wait until bio submission time. All the pages in the bio are | |
1585 | * checksummed and sums are attached onto the ordered extent record. | |
1586 | * | |
1587 | * At IO completion time the cums attached on the ordered extent record | |
1588 | * are inserted into the btree | |
1589 | */ | |
b2950863 | 1590 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1591 | int mirror_num, unsigned long bio_flags, |
1592 | u64 bio_offset) | |
4a69a410 CM |
1593 | { |
1594 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1595 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1596 | } |
1597 | ||
d352ac68 | 1598 | /* |
cad321ad CM |
1599 | * extent_io.c submission hook. This does the right thing for csum calculation |
1600 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1601 | */ |
b2950863 | 1602 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1603 | int mirror_num, unsigned long bio_flags, |
1604 | u64 bio_offset) | |
44b8bd7e CM |
1605 | { |
1606 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1607 | int ret = 0; | |
19b9bdb0 | 1608 | int skip_sum; |
0417341e | 1609 | int metadata = 0; |
44b8bd7e | 1610 | |
6cbff00f | 1611 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1612 | |
83eea1f1 | 1613 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1614 | metadata = 2; |
1615 | ||
7b6d91da | 1616 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1617 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1618 | if (ret) | |
1619 | return ret; | |
1620 | ||
d20f7043 | 1621 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1622 | return btrfs_submit_compressed_read(inode, bio, |
1623 | mirror_num, bio_flags); | |
c2db1073 TI |
1624 | } else if (!skip_sum) { |
1625 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1626 | if (ret) | |
1627 | return ret; | |
1628 | } | |
4d1b5fb4 | 1629 | goto mapit; |
19b9bdb0 | 1630 | } else if (!skip_sum) { |
17d217fe YZ |
1631 | /* csum items have already been cloned */ |
1632 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1633 | goto mapit; | |
19b9bdb0 CM |
1634 | /* we're doing a write, do the async checksumming */ |
1635 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1636 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1637 | bio_flags, bio_offset, |
1638 | __btrfs_submit_bio_start, | |
4a69a410 | 1639 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1640 | } |
1641 | ||
0b86a832 | 1642 | mapit: |
8b712842 | 1643 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1644 | } |
6885f308 | 1645 | |
d352ac68 CM |
1646 | /* |
1647 | * given a list of ordered sums record them in the inode. This happens | |
1648 | * at IO completion time based on sums calculated at bio submission time. | |
1649 | */ | |
ba1da2f4 | 1650 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1651 | struct inode *inode, u64 file_offset, |
1652 | struct list_head *list) | |
1653 | { | |
e6dcd2dc CM |
1654 | struct btrfs_ordered_sum *sum; |
1655 | ||
c6e30871 | 1656 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1657 | btrfs_csum_file_blocks(trans, |
1658 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1659 | } |
1660 | return 0; | |
1661 | } | |
1662 | ||
2ac55d41 JB |
1663 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1664 | struct extent_state **cached_state) | |
ea8c2819 | 1665 | { |
d397712b | 1666 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1667 | WARN_ON(1); |
ea8c2819 | 1668 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1669 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1670 | } |
1671 | ||
d352ac68 | 1672 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1673 | struct btrfs_writepage_fixup { |
1674 | struct page *page; | |
1675 | struct btrfs_work work; | |
1676 | }; | |
1677 | ||
b2950863 | 1678 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1679 | { |
1680 | struct btrfs_writepage_fixup *fixup; | |
1681 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1682 | struct extent_state *cached_state = NULL; |
247e743c CM |
1683 | struct page *page; |
1684 | struct inode *inode; | |
1685 | u64 page_start; | |
1686 | u64 page_end; | |
87826df0 | 1687 | int ret; |
247e743c CM |
1688 | |
1689 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1690 | page = fixup->page; | |
4a096752 | 1691 | again: |
247e743c CM |
1692 | lock_page(page); |
1693 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1694 | ClearPageChecked(page); | |
1695 | goto out_page; | |
1696 | } | |
1697 | ||
1698 | inode = page->mapping->host; | |
1699 | page_start = page_offset(page); | |
1700 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1701 | ||
2ac55d41 | 1702 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1703 | &cached_state); |
4a096752 CM |
1704 | |
1705 | /* already ordered? We're done */ | |
8b62b72b | 1706 | if (PagePrivate2(page)) |
247e743c | 1707 | goto out; |
4a096752 CM |
1708 | |
1709 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1710 | if (ordered) { | |
2ac55d41 JB |
1711 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1712 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1713 | unlock_page(page); |
1714 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1715 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1716 | goto again; |
1717 | } | |
247e743c | 1718 | |
87826df0 JM |
1719 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1720 | if (ret) { | |
1721 | mapping_set_error(page->mapping, ret); | |
1722 | end_extent_writepage(page, ret, page_start, page_end); | |
1723 | ClearPageChecked(page); | |
1724 | goto out; | |
1725 | } | |
1726 | ||
2ac55d41 | 1727 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1728 | ClearPageChecked(page); |
87826df0 | 1729 | set_page_dirty(page); |
247e743c | 1730 | out: |
2ac55d41 JB |
1731 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1732 | &cached_state, GFP_NOFS); | |
247e743c CM |
1733 | out_page: |
1734 | unlock_page(page); | |
1735 | page_cache_release(page); | |
b897abec | 1736 | kfree(fixup); |
247e743c CM |
1737 | } |
1738 | ||
1739 | /* | |
1740 | * There are a few paths in the higher layers of the kernel that directly | |
1741 | * set the page dirty bit without asking the filesystem if it is a | |
1742 | * good idea. This causes problems because we want to make sure COW | |
1743 | * properly happens and the data=ordered rules are followed. | |
1744 | * | |
c8b97818 | 1745 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1746 | * hasn't been properly setup for IO. We kick off an async process |
1747 | * to fix it up. The async helper will wait for ordered extents, set | |
1748 | * the delalloc bit and make it safe to write the page. | |
1749 | */ | |
b2950863 | 1750 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1751 | { |
1752 | struct inode *inode = page->mapping->host; | |
1753 | struct btrfs_writepage_fixup *fixup; | |
1754 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1755 | |
8b62b72b CM |
1756 | /* this page is properly in the ordered list */ |
1757 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1758 | return 0; |
1759 | ||
1760 | if (PageChecked(page)) | |
1761 | return -EAGAIN; | |
1762 | ||
1763 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1764 | if (!fixup) | |
1765 | return -EAGAIN; | |
f421950f | 1766 | |
247e743c CM |
1767 | SetPageChecked(page); |
1768 | page_cache_get(page); | |
1769 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1770 | fixup->page = page; | |
1771 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1772 | return -EBUSY; |
247e743c CM |
1773 | } |
1774 | ||
d899e052 YZ |
1775 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1776 | struct inode *inode, u64 file_pos, | |
1777 | u64 disk_bytenr, u64 disk_num_bytes, | |
1778 | u64 num_bytes, u64 ram_bytes, | |
1779 | u8 compression, u8 encryption, | |
1780 | u16 other_encoding, int extent_type) | |
1781 | { | |
1782 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1783 | struct btrfs_file_extent_item *fi; | |
1784 | struct btrfs_path *path; | |
1785 | struct extent_buffer *leaf; | |
1786 | struct btrfs_key ins; | |
d899e052 YZ |
1787 | int ret; |
1788 | ||
1789 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1790 | if (!path) |
1791 | return -ENOMEM; | |
d899e052 | 1792 | |
b9473439 | 1793 | path->leave_spinning = 1; |
a1ed835e CM |
1794 | |
1795 | /* | |
1796 | * we may be replacing one extent in the tree with another. | |
1797 | * The new extent is pinned in the extent map, and we don't want | |
1798 | * to drop it from the cache until it is completely in the btree. | |
1799 | * | |
1800 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1801 | * the caller is expected to unpin it and allow it to be merged | |
1802 | * with the others. | |
1803 | */ | |
5dc562c5 | 1804 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1805 | file_pos + num_bytes, 0); |
79787eaa JM |
1806 | if (ret) |
1807 | goto out; | |
d899e052 | 1808 | |
33345d01 | 1809 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1810 | ins.offset = file_pos; |
1811 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1812 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1813 | if (ret) |
1814 | goto out; | |
d899e052 YZ |
1815 | leaf = path->nodes[0]; |
1816 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1817 | struct btrfs_file_extent_item); | |
1818 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1819 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1820 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1821 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1822 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1823 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1824 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1825 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1826 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1827 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1828 | |
1829 | btrfs_unlock_up_safe(path, 1); | |
1830 | btrfs_set_lock_blocking(leaf); | |
1831 | ||
d899e052 YZ |
1832 | btrfs_mark_buffer_dirty(leaf); |
1833 | ||
1834 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1835 | |
1836 | ins.objectid = disk_bytenr; | |
1837 | ins.offset = disk_num_bytes; | |
1838 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1839 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1840 | root->root_key.objectid, | |
33345d01 | 1841 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1842 | out: |
d899e052 | 1843 | btrfs_free_path(path); |
b9473439 | 1844 | |
79787eaa | 1845 | return ret; |
d899e052 YZ |
1846 | } |
1847 | ||
5d13a98f CM |
1848 | /* |
1849 | * helper function for btrfs_finish_ordered_io, this | |
1850 | * just reads in some of the csum leaves to prime them into ram | |
1851 | * before we start the transaction. It limits the amount of btree | |
1852 | * reads required while inside the transaction. | |
1853 | */ | |
d352ac68 CM |
1854 | /* as ordered data IO finishes, this gets called so we can finish |
1855 | * an ordered extent if the range of bytes in the file it covers are | |
1856 | * fully written. | |
1857 | */ | |
5fd02043 | 1858 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1859 | { |
5fd02043 | 1860 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1861 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1862 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1863 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1864 | struct extent_state *cached_state = NULL; |
261507a0 | 1865 | int compress_type = 0; |
e6dcd2dc | 1866 | int ret; |
82d5902d | 1867 | bool nolock; |
e6dcd2dc | 1868 | |
83eea1f1 | 1869 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 1870 | |
5fd02043 JB |
1871 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1872 | ret = -EIO; | |
1873 | goto out; | |
1874 | } | |
1875 | ||
c2167754 | 1876 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1877 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
c2167754 YZ |
1878 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1879 | if (!ret) { | |
0cb59c99 | 1880 | if (nolock) |
7a7eaa40 | 1881 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1882 | else |
7a7eaa40 | 1883 | trans = btrfs_join_transaction(root); |
d280e5be LB |
1884 | if (IS_ERR(trans)) { |
1885 | ret = PTR_ERR(trans); | |
1886 | trans = NULL; | |
1887 | goto out; | |
1888 | } | |
0ca1f7ce | 1889 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1890 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1891 | if (ret) /* -ENOMEM or corruption */ |
1892 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1893 | } |
1894 | goto out; | |
1895 | } | |
e6dcd2dc | 1896 | |
2ac55d41 JB |
1897 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1898 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1899 | 0, &cached_state); |
e6dcd2dc | 1900 | |
0cb59c99 | 1901 | if (nolock) |
7a7eaa40 | 1902 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1903 | else |
7a7eaa40 | 1904 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1905 | if (IS_ERR(trans)) { |
1906 | ret = PTR_ERR(trans); | |
1907 | trans = NULL; | |
1908 | goto out_unlock; | |
1909 | } | |
0ca1f7ce | 1910 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1911 | |
c8b97818 | 1912 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1913 | compress_type = ordered_extent->compress_type; |
d899e052 | 1914 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1915 | BUG_ON(compress_type); |
920bbbfb | 1916 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1917 | ordered_extent->file_offset, |
1918 | ordered_extent->file_offset + | |
1919 | ordered_extent->len); | |
d899e052 | 1920 | } else { |
0af3d00b | 1921 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1922 | ret = insert_reserved_file_extent(trans, inode, |
1923 | ordered_extent->file_offset, | |
1924 | ordered_extent->start, | |
1925 | ordered_extent->disk_len, | |
1926 | ordered_extent->len, | |
1927 | ordered_extent->len, | |
261507a0 | 1928 | compress_type, 0, 0, |
d899e052 | 1929 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 1930 | } |
5dc562c5 JB |
1931 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1932 | ordered_extent->file_offset, ordered_extent->len, | |
1933 | trans->transid); | |
79787eaa JM |
1934 | if (ret < 0) { |
1935 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1936 | goto out_unlock; |
79787eaa | 1937 | } |
2ac55d41 | 1938 | |
e6dcd2dc CM |
1939 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1940 | &ordered_extent->list); | |
1941 | ||
1ef30be1 | 1942 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1943 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1944 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1945 | if (ret) { /* -ENOMEM or corruption */ |
1946 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1947 | goto out_unlock; |
79787eaa | 1948 | } |
7c735313 JB |
1949 | } else { |
1950 | btrfs_set_inode_last_trans(trans, inode); | |
1ef30be1 JB |
1951 | } |
1952 | ret = 0; | |
5fd02043 JB |
1953 | out_unlock: |
1954 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
1955 | ordered_extent->file_offset + | |
1956 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 1957 | out: |
5b0e95bf | 1958 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1959 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
5b0e95bf JB |
1960 | if (trans) { |
1961 | if (nolock) | |
0cb59c99 | 1962 | btrfs_end_transaction_nolock(trans, root); |
5b0e95bf | 1963 | else |
0cb59c99 JB |
1964 | btrfs_end_transaction(trans, root); |
1965 | } | |
1966 | ||
5fd02043 JB |
1967 | if (ret) |
1968 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, | |
1969 | ordered_extent->file_offset + | |
1970 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
1971 | ||
1972 | /* | |
1973 | * This needs to be dont to make sure anybody waiting knows we are done | |
1974 | * upating everything for this ordered extent. | |
1975 | */ | |
1976 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
1977 | ||
e6dcd2dc CM |
1978 | /* once for us */ |
1979 | btrfs_put_ordered_extent(ordered_extent); | |
1980 | /* once for the tree */ | |
1981 | btrfs_put_ordered_extent(ordered_extent); | |
1982 | ||
5fd02043 JB |
1983 | return ret; |
1984 | } | |
1985 | ||
1986 | static void finish_ordered_fn(struct btrfs_work *work) | |
1987 | { | |
1988 | struct btrfs_ordered_extent *ordered_extent; | |
1989 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
1990 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
1991 | } |
1992 | ||
b2950863 | 1993 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1994 | struct extent_state *state, int uptodate) |
1995 | { | |
5fd02043 JB |
1996 | struct inode *inode = page->mapping->host; |
1997 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1998 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
1999 | struct btrfs_workers *workers; | |
2000 | ||
1abe9b8a | 2001 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2002 | ||
8b62b72b | 2003 | ClearPagePrivate2(page); |
5fd02043 JB |
2004 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2005 | end - start + 1, uptodate)) | |
2006 | return 0; | |
2007 | ||
2008 | ordered_extent->work.func = finish_ordered_fn; | |
2009 | ordered_extent->work.flags = 0; | |
2010 | ||
83eea1f1 | 2011 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2012 | workers = &root->fs_info->endio_freespace_worker; |
2013 | else | |
2014 | workers = &root->fs_info->endio_write_workers; | |
2015 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2016 | ||
2017 | return 0; | |
211f90e6 CM |
2018 | } |
2019 | ||
d352ac68 CM |
2020 | /* |
2021 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2022 | * if there's a match, we allow the bio to finish. If not, the code in |
2023 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2024 | */ |
b2950863 | 2025 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2026 | struct extent_state *state, int mirror) |
07157aac | 2027 | { |
35ebb934 | 2028 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 2029 | struct inode *inode = page->mapping->host; |
d1310b2e | 2030 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2031 | char *kaddr; |
aadfeb6e | 2032 | u64 private = ~(u32)0; |
07157aac | 2033 | int ret; |
ff79f819 CM |
2034 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2035 | u32 csum = ~(u32)0; | |
d1310b2e | 2036 | |
d20f7043 CM |
2037 | if (PageChecked(page)) { |
2038 | ClearPageChecked(page); | |
2039 | goto good; | |
2040 | } | |
6cbff00f CH |
2041 | |
2042 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2043 | goto good; |
17d217fe YZ |
2044 | |
2045 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2046 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2047 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2048 | GFP_NOFS); | |
b6cda9bc | 2049 | return 0; |
17d217fe | 2050 | } |
d20f7043 | 2051 | |
c2e639f0 | 2052 | if (state && state->start == start) { |
70dec807 CM |
2053 | private = state->private; |
2054 | ret = 0; | |
2055 | } else { | |
2056 | ret = get_state_private(io_tree, start, &private); | |
2057 | } | |
7ac687d9 | 2058 | kaddr = kmap_atomic(page); |
d397712b | 2059 | if (ret) |
07157aac | 2060 | goto zeroit; |
d397712b | 2061 | |
ff79f819 CM |
2062 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2063 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2064 | if (csum != private) |
07157aac | 2065 | goto zeroit; |
d397712b | 2066 | |
7ac687d9 | 2067 | kunmap_atomic(kaddr); |
d20f7043 | 2068 | good: |
07157aac CM |
2069 | return 0; |
2070 | ||
2071 | zeroit: | |
945d8962 | 2072 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2073 | "private %llu\n", |
2074 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2075 | (unsigned long long)start, csum, |
2076 | (unsigned long long)private); | |
db94535d CM |
2077 | memset(kaddr + offset, 1, end - start + 1); |
2078 | flush_dcache_page(page); | |
7ac687d9 | 2079 | kunmap_atomic(kaddr); |
3b951516 CM |
2080 | if (private == 0) |
2081 | return 0; | |
7e38326f | 2082 | return -EIO; |
07157aac | 2083 | } |
b888db2b | 2084 | |
24bbcf04 YZ |
2085 | struct delayed_iput { |
2086 | struct list_head list; | |
2087 | struct inode *inode; | |
2088 | }; | |
2089 | ||
79787eaa JM |
2090 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2091 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2092 | void btrfs_add_delayed_iput(struct inode *inode) |
2093 | { | |
2094 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2095 | struct delayed_iput *delayed; | |
2096 | ||
2097 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2098 | return; | |
2099 | ||
2100 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2101 | delayed->inode = inode; | |
2102 | ||
2103 | spin_lock(&fs_info->delayed_iput_lock); | |
2104 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2105 | spin_unlock(&fs_info->delayed_iput_lock); | |
2106 | } | |
2107 | ||
2108 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2109 | { | |
2110 | LIST_HEAD(list); | |
2111 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2112 | struct delayed_iput *delayed; | |
2113 | int empty; | |
2114 | ||
2115 | spin_lock(&fs_info->delayed_iput_lock); | |
2116 | empty = list_empty(&fs_info->delayed_iputs); | |
2117 | spin_unlock(&fs_info->delayed_iput_lock); | |
2118 | if (empty) | |
2119 | return; | |
2120 | ||
2121 | down_read(&root->fs_info->cleanup_work_sem); | |
2122 | spin_lock(&fs_info->delayed_iput_lock); | |
2123 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2124 | spin_unlock(&fs_info->delayed_iput_lock); | |
2125 | ||
2126 | while (!list_empty(&list)) { | |
2127 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2128 | list_del(&delayed->list); | |
2129 | iput(delayed->inode); | |
2130 | kfree(delayed); | |
2131 | } | |
2132 | up_read(&root->fs_info->cleanup_work_sem); | |
2133 | } | |
2134 | ||
d68fc57b YZ |
2135 | enum btrfs_orphan_cleanup_state { |
2136 | ORPHAN_CLEANUP_STARTED = 1, | |
2137 | ORPHAN_CLEANUP_DONE = 2, | |
2138 | }; | |
2139 | ||
2140 | /* | |
42b2aa86 | 2141 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2142 | * files in the subvolume, it removes orphan item and frees block_rsv |
2143 | * structure. | |
2144 | */ | |
2145 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2146 | struct btrfs_root *root) | |
2147 | { | |
90290e19 | 2148 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2149 | int ret; |
2150 | ||
8a35d95f | 2151 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2152 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2153 | return; | |
2154 | ||
90290e19 | 2155 | spin_lock(&root->orphan_lock); |
8a35d95f | 2156 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2157 | spin_unlock(&root->orphan_lock); |
2158 | return; | |
2159 | } | |
2160 | ||
2161 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2162 | spin_unlock(&root->orphan_lock); | |
2163 | return; | |
2164 | } | |
2165 | ||
2166 | block_rsv = root->orphan_block_rsv; | |
2167 | root->orphan_block_rsv = NULL; | |
2168 | spin_unlock(&root->orphan_lock); | |
2169 | ||
d68fc57b YZ |
2170 | if (root->orphan_item_inserted && |
2171 | btrfs_root_refs(&root->root_item) > 0) { | |
2172 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2173 | root->root_key.objectid); | |
2174 | BUG_ON(ret); | |
2175 | root->orphan_item_inserted = 0; | |
2176 | } | |
2177 | ||
90290e19 JB |
2178 | if (block_rsv) { |
2179 | WARN_ON(block_rsv->size > 0); | |
2180 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2181 | } |
2182 | } | |
2183 | ||
7b128766 JB |
2184 | /* |
2185 | * This creates an orphan entry for the given inode in case something goes | |
2186 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2187 | * |
2188 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2189 | * this function. | |
7b128766 JB |
2190 | */ |
2191 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2192 | { | |
2193 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2194 | struct btrfs_block_rsv *block_rsv = NULL; |
2195 | int reserve = 0; | |
2196 | int insert = 0; | |
2197 | int ret; | |
7b128766 | 2198 | |
d68fc57b YZ |
2199 | if (!root->orphan_block_rsv) { |
2200 | block_rsv = btrfs_alloc_block_rsv(root); | |
b532402e TI |
2201 | if (!block_rsv) |
2202 | return -ENOMEM; | |
d68fc57b | 2203 | } |
7b128766 | 2204 | |
d68fc57b YZ |
2205 | spin_lock(&root->orphan_lock); |
2206 | if (!root->orphan_block_rsv) { | |
2207 | root->orphan_block_rsv = block_rsv; | |
2208 | } else if (block_rsv) { | |
2209 | btrfs_free_block_rsv(root, block_rsv); | |
2210 | block_rsv = NULL; | |
7b128766 | 2211 | } |
7b128766 | 2212 | |
8a35d95f JB |
2213 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2214 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2215 | #if 0 |
2216 | /* | |
2217 | * For proper ENOSPC handling, we should do orphan | |
2218 | * cleanup when mounting. But this introduces backward | |
2219 | * compatibility issue. | |
2220 | */ | |
2221 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2222 | insert = 2; | |
2223 | else | |
2224 | insert = 1; | |
2225 | #endif | |
2226 | insert = 1; | |
321f0e70 | 2227 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2228 | } |
2229 | ||
72ac3c0d JB |
2230 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2231 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2232 | reserve = 1; |
d68fc57b | 2233 | spin_unlock(&root->orphan_lock); |
7b128766 | 2234 | |
d68fc57b YZ |
2235 | /* grab metadata reservation from transaction handle */ |
2236 | if (reserve) { | |
2237 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2238 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2239 | } |
7b128766 | 2240 | |
d68fc57b YZ |
2241 | /* insert an orphan item to track this unlinked/truncated file */ |
2242 | if (insert >= 1) { | |
33345d01 | 2243 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2244 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2245 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2246 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2247 | btrfs_abort_transaction(trans, root, ret); |
2248 | return ret; | |
2249 | } | |
2250 | ret = 0; | |
d68fc57b YZ |
2251 | } |
2252 | ||
2253 | /* insert an orphan item to track subvolume contains orphan files */ | |
2254 | if (insert >= 2) { | |
2255 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2256 | root->root_key.objectid); | |
79787eaa JM |
2257 | if (ret && ret != -EEXIST) { |
2258 | btrfs_abort_transaction(trans, root, ret); | |
2259 | return ret; | |
2260 | } | |
d68fc57b YZ |
2261 | } |
2262 | return 0; | |
7b128766 JB |
2263 | } |
2264 | ||
2265 | /* | |
2266 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2267 | * item for this particular inode. | |
2268 | */ | |
2269 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2270 | { | |
2271 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2272 | int delete_item = 0; |
2273 | int release_rsv = 0; | |
7b128766 JB |
2274 | int ret = 0; |
2275 | ||
d68fc57b | 2276 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2277 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2278 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2279 | delete_item = 1; |
7b128766 | 2280 | |
72ac3c0d JB |
2281 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2282 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2283 | release_rsv = 1; |
d68fc57b | 2284 | spin_unlock(&root->orphan_lock); |
7b128766 | 2285 | |
d68fc57b | 2286 | if (trans && delete_item) { |
33345d01 | 2287 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2288 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2289 | } |
7b128766 | 2290 | |
8a35d95f | 2291 | if (release_rsv) { |
d68fc57b | 2292 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2293 | atomic_dec(&root->orphan_inodes); |
2294 | } | |
7b128766 | 2295 | |
d68fc57b | 2296 | return 0; |
7b128766 JB |
2297 | } |
2298 | ||
2299 | /* | |
2300 | * this cleans up any orphans that may be left on the list from the last use | |
2301 | * of this root. | |
2302 | */ | |
66b4ffd1 | 2303 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2304 | { |
2305 | struct btrfs_path *path; | |
2306 | struct extent_buffer *leaf; | |
7b128766 JB |
2307 | struct btrfs_key key, found_key; |
2308 | struct btrfs_trans_handle *trans; | |
2309 | struct inode *inode; | |
8f6d7f4f | 2310 | u64 last_objectid = 0; |
7b128766 JB |
2311 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2312 | ||
d68fc57b | 2313 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2314 | return 0; |
c71bf099 YZ |
2315 | |
2316 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2317 | if (!path) { |
2318 | ret = -ENOMEM; | |
2319 | goto out; | |
2320 | } | |
7b128766 JB |
2321 | path->reada = -1; |
2322 | ||
2323 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2324 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2325 | key.offset = (u64)-1; | |
2326 | ||
7b128766 JB |
2327 | while (1) { |
2328 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2329 | if (ret < 0) |
2330 | goto out; | |
7b128766 JB |
2331 | |
2332 | /* | |
2333 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2334 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2335 | * find the key and see if we have stuff that matches |
2336 | */ | |
2337 | if (ret > 0) { | |
66b4ffd1 | 2338 | ret = 0; |
7b128766 JB |
2339 | if (path->slots[0] == 0) |
2340 | break; | |
2341 | path->slots[0]--; | |
2342 | } | |
2343 | ||
2344 | /* pull out the item */ | |
2345 | leaf = path->nodes[0]; | |
7b128766 JB |
2346 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2347 | ||
2348 | /* make sure the item matches what we want */ | |
2349 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2350 | break; | |
2351 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2352 | break; | |
2353 | ||
2354 | /* release the path since we're done with it */ | |
b3b4aa74 | 2355 | btrfs_release_path(path); |
7b128766 JB |
2356 | |
2357 | /* | |
2358 | * this is where we are basically btrfs_lookup, without the | |
2359 | * crossing root thing. we store the inode number in the | |
2360 | * offset of the orphan item. | |
2361 | */ | |
8f6d7f4f JB |
2362 | |
2363 | if (found_key.offset == last_objectid) { | |
2364 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2365 | "stopping orphan cleanup\n"); | |
2366 | ret = -EINVAL; | |
2367 | goto out; | |
2368 | } | |
2369 | ||
2370 | last_objectid = found_key.offset; | |
2371 | ||
5d4f98a2 YZ |
2372 | found_key.objectid = found_key.offset; |
2373 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2374 | found_key.offset = 0; | |
73f73415 | 2375 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2376 | ret = PTR_RET(inode); |
2377 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2378 | goto out; |
7b128766 | 2379 | |
f8e9e0b0 AJ |
2380 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2381 | struct btrfs_root *dead_root; | |
2382 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2383 | int is_dead_root = 0; | |
2384 | ||
2385 | /* | |
2386 | * this is an orphan in the tree root. Currently these | |
2387 | * could come from 2 sources: | |
2388 | * a) a snapshot deletion in progress | |
2389 | * b) a free space cache inode | |
2390 | * We need to distinguish those two, as the snapshot | |
2391 | * orphan must not get deleted. | |
2392 | * find_dead_roots already ran before us, so if this | |
2393 | * is a snapshot deletion, we should find the root | |
2394 | * in the dead_roots list | |
2395 | */ | |
2396 | spin_lock(&fs_info->trans_lock); | |
2397 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2398 | root_list) { | |
2399 | if (dead_root->root_key.objectid == | |
2400 | found_key.objectid) { | |
2401 | is_dead_root = 1; | |
2402 | break; | |
2403 | } | |
2404 | } | |
2405 | spin_unlock(&fs_info->trans_lock); | |
2406 | if (is_dead_root) { | |
2407 | /* prevent this orphan from being found again */ | |
2408 | key.offset = found_key.objectid - 1; | |
2409 | continue; | |
2410 | } | |
2411 | } | |
7b128766 | 2412 | /* |
a8c9e576 JB |
2413 | * Inode is already gone but the orphan item is still there, |
2414 | * kill the orphan item. | |
7b128766 | 2415 | */ |
a8c9e576 JB |
2416 | if (ret == -ESTALE) { |
2417 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2418 | if (IS_ERR(trans)) { |
2419 | ret = PTR_ERR(trans); | |
2420 | goto out; | |
2421 | } | |
8a35d95f JB |
2422 | printk(KERN_ERR "auto deleting %Lu\n", |
2423 | found_key.objectid); | |
a8c9e576 JB |
2424 | ret = btrfs_del_orphan_item(trans, root, |
2425 | found_key.objectid); | |
79787eaa | 2426 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2427 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2428 | continue; |
2429 | } | |
2430 | ||
a8c9e576 JB |
2431 | /* |
2432 | * add this inode to the orphan list so btrfs_orphan_del does | |
2433 | * the proper thing when we hit it | |
2434 | */ | |
8a35d95f JB |
2435 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2436 | &BTRFS_I(inode)->runtime_flags); | |
a8c9e576 | 2437 | |
7b128766 JB |
2438 | /* if we have links, this was a truncate, lets do that */ |
2439 | if (inode->i_nlink) { | |
a41ad394 JB |
2440 | if (!S_ISREG(inode->i_mode)) { |
2441 | WARN_ON(1); | |
2442 | iput(inode); | |
2443 | continue; | |
2444 | } | |
7b128766 | 2445 | nr_truncate++; |
66b4ffd1 | 2446 | ret = btrfs_truncate(inode); |
7b128766 JB |
2447 | } else { |
2448 | nr_unlink++; | |
2449 | } | |
2450 | ||
2451 | /* this will do delete_inode and everything for us */ | |
2452 | iput(inode); | |
66b4ffd1 JB |
2453 | if (ret) |
2454 | goto out; | |
7b128766 | 2455 | } |
3254c876 MX |
2456 | /* release the path since we're done with it */ |
2457 | btrfs_release_path(path); | |
2458 | ||
d68fc57b YZ |
2459 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2460 | ||
2461 | if (root->orphan_block_rsv) | |
2462 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2463 | (u64)-1); | |
2464 | ||
2465 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2466 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2467 | if (!IS_ERR(trans)) |
2468 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2469 | } |
7b128766 JB |
2470 | |
2471 | if (nr_unlink) | |
2472 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2473 | if (nr_truncate) | |
2474 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2475 | |
2476 | out: | |
2477 | if (ret) | |
2478 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2479 | btrfs_free_path(path); | |
2480 | return ret; | |
7b128766 JB |
2481 | } |
2482 | ||
46a53cca CM |
2483 | /* |
2484 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2485 | * don't find any xattrs, we know there can't be any acls. | |
2486 | * | |
2487 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2488 | */ | |
2489 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2490 | int slot, u64 objectid) | |
2491 | { | |
2492 | u32 nritems = btrfs_header_nritems(leaf); | |
2493 | struct btrfs_key found_key; | |
2494 | int scanned = 0; | |
2495 | ||
2496 | slot++; | |
2497 | while (slot < nritems) { | |
2498 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2499 | ||
2500 | /* we found a different objectid, there must not be acls */ | |
2501 | if (found_key.objectid != objectid) | |
2502 | return 0; | |
2503 | ||
2504 | /* we found an xattr, assume we've got an acl */ | |
2505 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2506 | return 1; | |
2507 | ||
2508 | /* | |
2509 | * we found a key greater than an xattr key, there can't | |
2510 | * be any acls later on | |
2511 | */ | |
2512 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2513 | return 0; | |
2514 | ||
2515 | slot++; | |
2516 | scanned++; | |
2517 | ||
2518 | /* | |
2519 | * it goes inode, inode backrefs, xattrs, extents, | |
2520 | * so if there are a ton of hard links to an inode there can | |
2521 | * be a lot of backrefs. Don't waste time searching too hard, | |
2522 | * this is just an optimization | |
2523 | */ | |
2524 | if (scanned >= 8) | |
2525 | break; | |
2526 | } | |
2527 | /* we hit the end of the leaf before we found an xattr or | |
2528 | * something larger than an xattr. We have to assume the inode | |
2529 | * has acls | |
2530 | */ | |
2531 | return 1; | |
2532 | } | |
2533 | ||
d352ac68 CM |
2534 | /* |
2535 | * read an inode from the btree into the in-memory inode | |
2536 | */ | |
5d4f98a2 | 2537 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2538 | { |
2539 | struct btrfs_path *path; | |
5f39d397 | 2540 | struct extent_buffer *leaf; |
39279cc3 | 2541 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2542 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2543 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2544 | struct btrfs_key location; | |
46a53cca | 2545 | int maybe_acls; |
618e21d5 | 2546 | u32 rdev; |
39279cc3 | 2547 | int ret; |
2f7e33d4 MX |
2548 | bool filled = false; |
2549 | ||
2550 | ret = btrfs_fill_inode(inode, &rdev); | |
2551 | if (!ret) | |
2552 | filled = true; | |
39279cc3 CM |
2553 | |
2554 | path = btrfs_alloc_path(); | |
1748f843 MF |
2555 | if (!path) |
2556 | goto make_bad; | |
2557 | ||
d90c7321 | 2558 | path->leave_spinning = 1; |
39279cc3 | 2559 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2560 | |
39279cc3 | 2561 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2562 | if (ret) |
39279cc3 | 2563 | goto make_bad; |
39279cc3 | 2564 | |
5f39d397 | 2565 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2566 | |
2567 | if (filled) | |
2568 | goto cache_acl; | |
2569 | ||
5f39d397 CM |
2570 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2571 | struct btrfs_inode_item); | |
5f39d397 | 2572 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2573 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
5f39d397 CM |
2574 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); |
2575 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2576 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2577 | |
2578 | tspec = btrfs_inode_atime(inode_item); | |
2579 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2580 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2581 | ||
2582 | tspec = btrfs_inode_mtime(inode_item); | |
2583 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2584 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2585 | ||
2586 | tspec = btrfs_inode_ctime(inode_item); | |
2587 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2588 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2589 | ||
a76a3cd4 | 2590 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2591 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
2592 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
2593 | ||
2594 | /* | |
2595 | * If we were modified in the current generation and evicted from memory | |
2596 | * and then re-read we need to do a full sync since we don't have any | |
2597 | * idea about which extents were modified before we were evicted from | |
2598 | * cache. | |
2599 | */ | |
2600 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
2601 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2602 | &BTRFS_I(inode)->runtime_flags); | |
2603 | ||
0c4d2d95 | 2604 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2605 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2606 | inode->i_rdev = 0; |
5f39d397 CM |
2607 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2608 | ||
aec7477b | 2609 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2610 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2611 | cache_acl: |
46a53cca CM |
2612 | /* |
2613 | * try to precache a NULL acl entry for files that don't have | |
2614 | * any xattrs or acls | |
2615 | */ | |
33345d01 LZ |
2616 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2617 | btrfs_ino(inode)); | |
72c04902 AV |
2618 | if (!maybe_acls) |
2619 | cache_no_acl(inode); | |
46a53cca | 2620 | |
39279cc3 | 2621 | btrfs_free_path(path); |
39279cc3 | 2622 | |
39279cc3 | 2623 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2624 | case S_IFREG: |
2625 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2626 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2627 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2628 | inode->i_fop = &btrfs_file_operations; |
2629 | inode->i_op = &btrfs_file_inode_operations; | |
2630 | break; | |
2631 | case S_IFDIR: | |
2632 | inode->i_fop = &btrfs_dir_file_operations; | |
2633 | if (root == root->fs_info->tree_root) | |
2634 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2635 | else | |
2636 | inode->i_op = &btrfs_dir_inode_operations; | |
2637 | break; | |
2638 | case S_IFLNK: | |
2639 | inode->i_op = &btrfs_symlink_inode_operations; | |
2640 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2641 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2642 | break; |
618e21d5 | 2643 | default: |
0279b4cd | 2644 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2645 | init_special_inode(inode, inode->i_mode, rdev); |
2646 | break; | |
39279cc3 | 2647 | } |
6cbff00f CH |
2648 | |
2649 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2650 | return; |
2651 | ||
2652 | make_bad: | |
39279cc3 | 2653 | btrfs_free_path(path); |
39279cc3 CM |
2654 | make_bad_inode(inode); |
2655 | } | |
2656 | ||
d352ac68 CM |
2657 | /* |
2658 | * given a leaf and an inode, copy the inode fields into the leaf | |
2659 | */ | |
e02119d5 CM |
2660 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2661 | struct extent_buffer *leaf, | |
5f39d397 | 2662 | struct btrfs_inode_item *item, |
39279cc3 CM |
2663 | struct inode *inode) |
2664 | { | |
5f39d397 CM |
2665 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2666 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2667 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2668 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2669 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2670 | ||
2671 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2672 | inode->i_atime.tv_sec); | |
2673 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2674 | inode->i_atime.tv_nsec); | |
2675 | ||
2676 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2677 | inode->i_mtime.tv_sec); | |
2678 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2679 | inode->i_mtime.tv_nsec); | |
2680 | ||
2681 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2682 | inode->i_ctime.tv_sec); | |
2683 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2684 | inode->i_ctime.tv_nsec); | |
2685 | ||
a76a3cd4 | 2686 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2687 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
0c4d2d95 | 2688 | btrfs_set_inode_sequence(leaf, item, inode->i_version); |
e02119d5 | 2689 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2690 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2691 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2692 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2693 | } |
2694 | ||
d352ac68 CM |
2695 | /* |
2696 | * copy everything in the in-memory inode into the btree. | |
2697 | */ | |
2115133f | 2698 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2699 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2700 | { |
2701 | struct btrfs_inode_item *inode_item; | |
2702 | struct btrfs_path *path; | |
5f39d397 | 2703 | struct extent_buffer *leaf; |
39279cc3 CM |
2704 | int ret; |
2705 | ||
2706 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2707 | if (!path) |
2708 | return -ENOMEM; | |
2709 | ||
b9473439 | 2710 | path->leave_spinning = 1; |
16cdcec7 MX |
2711 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2712 | 1); | |
39279cc3 CM |
2713 | if (ret) { |
2714 | if (ret > 0) | |
2715 | ret = -ENOENT; | |
2716 | goto failed; | |
2717 | } | |
2718 | ||
b4ce94de | 2719 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2720 | leaf = path->nodes[0]; |
2721 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2722 | struct btrfs_inode_item); |
39279cc3 | 2723 | |
e02119d5 | 2724 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2725 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2726 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2727 | ret = 0; |
2728 | failed: | |
39279cc3 CM |
2729 | btrfs_free_path(path); |
2730 | return ret; | |
2731 | } | |
2732 | ||
2115133f CM |
2733 | /* |
2734 | * copy everything in the in-memory inode into the btree. | |
2735 | */ | |
2736 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2737 | struct btrfs_root *root, struct inode *inode) | |
2738 | { | |
2739 | int ret; | |
2740 | ||
2741 | /* | |
2742 | * If the inode is a free space inode, we can deadlock during commit | |
2743 | * if we put it into the delayed code. | |
2744 | * | |
2745 | * The data relocation inode should also be directly updated | |
2746 | * without delay | |
2747 | */ | |
83eea1f1 | 2748 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 2749 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
2750 | btrfs_update_root_times(trans, root); |
2751 | ||
2115133f CM |
2752 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2753 | if (!ret) | |
2754 | btrfs_set_inode_last_trans(trans, inode); | |
2755 | return ret; | |
2756 | } | |
2757 | ||
2758 | return btrfs_update_inode_item(trans, root, inode); | |
2759 | } | |
2760 | ||
2761 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, | |
2762 | struct btrfs_root *root, struct inode *inode) | |
2763 | { | |
2764 | int ret; | |
2765 | ||
2766 | ret = btrfs_update_inode(trans, root, inode); | |
2767 | if (ret == -ENOSPC) | |
2768 | return btrfs_update_inode_item(trans, root, inode); | |
2769 | return ret; | |
2770 | } | |
2771 | ||
d352ac68 CM |
2772 | /* |
2773 | * unlink helper that gets used here in inode.c and in the tree logging | |
2774 | * recovery code. It remove a link in a directory with a given name, and | |
2775 | * also drops the back refs in the inode to the directory | |
2776 | */ | |
92986796 AV |
2777 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2778 | struct btrfs_root *root, | |
2779 | struct inode *dir, struct inode *inode, | |
2780 | const char *name, int name_len) | |
39279cc3 CM |
2781 | { |
2782 | struct btrfs_path *path; | |
39279cc3 | 2783 | int ret = 0; |
5f39d397 | 2784 | struct extent_buffer *leaf; |
39279cc3 | 2785 | struct btrfs_dir_item *di; |
5f39d397 | 2786 | struct btrfs_key key; |
aec7477b | 2787 | u64 index; |
33345d01 LZ |
2788 | u64 ino = btrfs_ino(inode); |
2789 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2790 | |
2791 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2792 | if (!path) { |
2793 | ret = -ENOMEM; | |
554233a6 | 2794 | goto out; |
54aa1f4d CM |
2795 | } |
2796 | ||
b9473439 | 2797 | path->leave_spinning = 1; |
33345d01 | 2798 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2799 | name, name_len, -1); |
2800 | if (IS_ERR(di)) { | |
2801 | ret = PTR_ERR(di); | |
2802 | goto err; | |
2803 | } | |
2804 | if (!di) { | |
2805 | ret = -ENOENT; | |
2806 | goto err; | |
2807 | } | |
5f39d397 CM |
2808 | leaf = path->nodes[0]; |
2809 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2810 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2811 | if (ret) |
2812 | goto err; | |
b3b4aa74 | 2813 | btrfs_release_path(path); |
39279cc3 | 2814 | |
33345d01 LZ |
2815 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2816 | dir_ino, &index); | |
aec7477b | 2817 | if (ret) { |
d397712b | 2818 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2819 | "inode %llu parent %llu\n", name_len, name, |
2820 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2821 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2822 | goto err; |
2823 | } | |
2824 | ||
16cdcec7 | 2825 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2826 | if (ret) { |
2827 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2828 | goto err; |
79787eaa | 2829 | } |
39279cc3 | 2830 | |
e02119d5 | 2831 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2832 | inode, dir_ino); |
79787eaa JM |
2833 | if (ret != 0 && ret != -ENOENT) { |
2834 | btrfs_abort_transaction(trans, root, ret); | |
2835 | goto err; | |
2836 | } | |
e02119d5 CM |
2837 | |
2838 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2839 | dir, index); | |
6418c961 CM |
2840 | if (ret == -ENOENT) |
2841 | ret = 0; | |
39279cc3 CM |
2842 | err: |
2843 | btrfs_free_path(path); | |
e02119d5 CM |
2844 | if (ret) |
2845 | goto out; | |
2846 | ||
2847 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2848 | inode_inc_iversion(inode); |
2849 | inode_inc_iversion(dir); | |
e02119d5 | 2850 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 2851 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 2852 | out: |
39279cc3 CM |
2853 | return ret; |
2854 | } | |
2855 | ||
92986796 AV |
2856 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2857 | struct btrfs_root *root, | |
2858 | struct inode *dir, struct inode *inode, | |
2859 | const char *name, int name_len) | |
2860 | { | |
2861 | int ret; | |
2862 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2863 | if (!ret) { | |
2864 | btrfs_drop_nlink(inode); | |
2865 | ret = btrfs_update_inode(trans, root, inode); | |
2866 | } | |
2867 | return ret; | |
2868 | } | |
2869 | ||
2870 | ||
a22285a6 YZ |
2871 | /* helper to check if there is any shared block in the path */ |
2872 | static int check_path_shared(struct btrfs_root *root, | |
2873 | struct btrfs_path *path) | |
39279cc3 | 2874 | { |
a22285a6 YZ |
2875 | struct extent_buffer *eb; |
2876 | int level; | |
0e4dcbef | 2877 | u64 refs = 1; |
5df6a9f6 | 2878 | |
a22285a6 | 2879 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2880 | int ret; |
2881 | ||
a22285a6 YZ |
2882 | if (!path->nodes[level]) |
2883 | break; | |
2884 | eb = path->nodes[level]; | |
2885 | if (!btrfs_block_can_be_shared(root, eb)) | |
2886 | continue; | |
2887 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2888 | &refs, NULL); | |
2889 | if (refs > 1) | |
2890 | return 1; | |
5df6a9f6 | 2891 | } |
dedefd72 | 2892 | return 0; |
39279cc3 CM |
2893 | } |
2894 | ||
a22285a6 YZ |
2895 | /* |
2896 | * helper to start transaction for unlink and rmdir. | |
2897 | * | |
2898 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2899 | * so in enospc case, we should make sure they will free space before | |
2900 | * allowing them to use the global metadata reservation. | |
2901 | */ | |
2902 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2903 | struct dentry *dentry) | |
4df27c4d | 2904 | { |
39279cc3 | 2905 | struct btrfs_trans_handle *trans; |
a22285a6 | 2906 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2907 | struct btrfs_path *path; |
a22285a6 | 2908 | struct btrfs_inode_ref *ref; |
4df27c4d | 2909 | struct btrfs_dir_item *di; |
7b128766 | 2910 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2911 | u64 index; |
a22285a6 YZ |
2912 | int check_link = 1; |
2913 | int err = -ENOSPC; | |
4df27c4d | 2914 | int ret; |
33345d01 LZ |
2915 | u64 ino = btrfs_ino(inode); |
2916 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2917 | |
e70bea5f JB |
2918 | /* |
2919 | * 1 for the possible orphan item | |
2920 | * 1 for the dir item | |
2921 | * 1 for the dir index | |
2922 | * 1 for the inode ref | |
2923 | * 1 for the inode ref in the tree log | |
2924 | * 2 for the dir entries in the log | |
2925 | * 1 for the inode | |
2926 | */ | |
2927 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2928 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2929 | return trans; | |
4df27c4d | 2930 | |
33345d01 | 2931 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2932 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2933 | |
a22285a6 YZ |
2934 | /* check if there is someone else holds reference */ |
2935 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2936 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2937 | |
a22285a6 YZ |
2938 | if (atomic_read(&inode->i_count) > 2) |
2939 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2940 | |
a22285a6 YZ |
2941 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2942 | return ERR_PTR(-ENOSPC); | |
2943 | ||
2944 | path = btrfs_alloc_path(); | |
2945 | if (!path) { | |
2946 | root->fs_info->enospc_unlink = 0; | |
2947 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2948 | } |
2949 | ||
3880a1b4 JB |
2950 | /* 1 for the orphan item */ |
2951 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2952 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2953 | btrfs_free_path(path); |
2954 | root->fs_info->enospc_unlink = 0; | |
2955 | return trans; | |
2956 | } | |
4df27c4d | 2957 | |
a22285a6 YZ |
2958 | path->skip_locking = 1; |
2959 | path->search_commit_root = 1; | |
4df27c4d | 2960 | |
a22285a6 YZ |
2961 | ret = btrfs_lookup_inode(trans, root, path, |
2962 | &BTRFS_I(dir)->location, 0); | |
2963 | if (ret < 0) { | |
2964 | err = ret; | |
2965 | goto out; | |
2966 | } | |
2967 | if (ret == 0) { | |
2968 | if (check_path_shared(root, path)) | |
2969 | goto out; | |
2970 | } else { | |
2971 | check_link = 0; | |
5df6a9f6 | 2972 | } |
b3b4aa74 | 2973 | btrfs_release_path(path); |
a22285a6 YZ |
2974 | |
2975 | ret = btrfs_lookup_inode(trans, root, path, | |
2976 | &BTRFS_I(inode)->location, 0); | |
2977 | if (ret < 0) { | |
2978 | err = ret; | |
2979 | goto out; | |
2980 | } | |
2981 | if (ret == 0) { | |
2982 | if (check_path_shared(root, path)) | |
2983 | goto out; | |
2984 | } else { | |
2985 | check_link = 0; | |
2986 | } | |
b3b4aa74 | 2987 | btrfs_release_path(path); |
a22285a6 YZ |
2988 | |
2989 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2990 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2991 | ino, (u64)-1, 0); |
a22285a6 YZ |
2992 | if (ret < 0) { |
2993 | err = ret; | |
2994 | goto out; | |
2995 | } | |
79787eaa | 2996 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
2997 | if (check_path_shared(root, path)) |
2998 | goto out; | |
b3b4aa74 | 2999 | btrfs_release_path(path); |
a22285a6 YZ |
3000 | } |
3001 | ||
3002 | if (!check_link) { | |
3003 | err = 0; | |
3004 | goto out; | |
3005 | } | |
3006 | ||
33345d01 | 3007 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
3008 | dentry->d_name.name, dentry->d_name.len, 0); |
3009 | if (IS_ERR(di)) { | |
3010 | err = PTR_ERR(di); | |
3011 | goto out; | |
3012 | } | |
3013 | if (di) { | |
3014 | if (check_path_shared(root, path)) | |
3015 | goto out; | |
3016 | } else { | |
3017 | err = 0; | |
3018 | goto out; | |
3019 | } | |
b3b4aa74 | 3020 | btrfs_release_path(path); |
a22285a6 YZ |
3021 | |
3022 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
3023 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 3024 | ino, dir_ino, 0); |
a22285a6 YZ |
3025 | if (IS_ERR(ref)) { |
3026 | err = PTR_ERR(ref); | |
3027 | goto out; | |
3028 | } | |
79787eaa | 3029 | BUG_ON(!ref); /* Logic error */ |
a22285a6 YZ |
3030 | if (check_path_shared(root, path)) |
3031 | goto out; | |
3032 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 3033 | btrfs_release_path(path); |
a22285a6 | 3034 | |
16cdcec7 MX |
3035 | /* |
3036 | * This is a commit root search, if we can lookup inode item and other | |
3037 | * relative items in the commit root, it means the transaction of | |
3038 | * dir/file creation has been committed, and the dir index item that we | |
3039 | * delay to insert has also been inserted into the commit root. So | |
3040 | * we needn't worry about the delayed insertion of the dir index item | |
3041 | * here. | |
3042 | */ | |
33345d01 | 3043 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
3044 | dentry->d_name.name, dentry->d_name.len, 0); |
3045 | if (IS_ERR(di)) { | |
3046 | err = PTR_ERR(di); | |
3047 | goto out; | |
3048 | } | |
3049 | BUG_ON(ret == -ENOENT); | |
3050 | if (check_path_shared(root, path)) | |
3051 | goto out; | |
3052 | ||
3053 | err = 0; | |
3054 | out: | |
3055 | btrfs_free_path(path); | |
3880a1b4 JB |
3056 | /* Migrate the orphan reservation over */ |
3057 | if (!err) | |
3058 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3059 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3060 | trans->bytes_reserved); |
3880a1b4 | 3061 | |
a22285a6 YZ |
3062 | if (err) { |
3063 | btrfs_end_transaction(trans, root); | |
3064 | root->fs_info->enospc_unlink = 0; | |
3065 | return ERR_PTR(err); | |
3066 | } | |
3067 | ||
3068 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3069 | return trans; | |
3070 | } | |
3071 | ||
3072 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3073 | struct btrfs_root *root) | |
3074 | { | |
3075 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
5a77d76c JB |
3076 | btrfs_block_rsv_release(root, trans->block_rsv, |
3077 | trans->bytes_reserved); | |
3078 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3079 | BUG_ON(!root->fs_info->enospc_unlink); |
3080 | root->fs_info->enospc_unlink = 0; | |
3081 | } | |
7ad85bb7 | 3082 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3083 | } |
3084 | ||
3085 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3086 | { | |
3087 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3088 | struct btrfs_trans_handle *trans; | |
3089 | struct inode *inode = dentry->d_inode; | |
3090 | int ret; | |
3091 | unsigned long nr = 0; | |
3092 | ||
3093 | trans = __unlink_start_trans(dir, dentry); | |
3094 | if (IS_ERR(trans)) | |
3095 | return PTR_ERR(trans); | |
5f39d397 | 3096 | |
12fcfd22 CM |
3097 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3098 | ||
e02119d5 CM |
3099 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3100 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3101 | if (ret) |
3102 | goto out; | |
7b128766 | 3103 | |
a22285a6 | 3104 | if (inode->i_nlink == 0) { |
7b128766 | 3105 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3106 | if (ret) |
3107 | goto out; | |
a22285a6 | 3108 | } |
7b128766 | 3109 | |
b532402e | 3110 | out: |
d3c2fdcf | 3111 | nr = trans->blocks_used; |
a22285a6 | 3112 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3113 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3114 | return ret; |
3115 | } | |
3116 | ||
4df27c4d YZ |
3117 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3118 | struct btrfs_root *root, | |
3119 | struct inode *dir, u64 objectid, | |
3120 | const char *name, int name_len) | |
3121 | { | |
3122 | struct btrfs_path *path; | |
3123 | struct extent_buffer *leaf; | |
3124 | struct btrfs_dir_item *di; | |
3125 | struct btrfs_key key; | |
3126 | u64 index; | |
3127 | int ret; | |
33345d01 | 3128 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3129 | |
3130 | path = btrfs_alloc_path(); | |
3131 | if (!path) | |
3132 | return -ENOMEM; | |
3133 | ||
33345d01 | 3134 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3135 | name, name_len, -1); |
79787eaa JM |
3136 | if (IS_ERR_OR_NULL(di)) { |
3137 | if (!di) | |
3138 | ret = -ENOENT; | |
3139 | else | |
3140 | ret = PTR_ERR(di); | |
3141 | goto out; | |
3142 | } | |
4df27c4d YZ |
3143 | |
3144 | leaf = path->nodes[0]; | |
3145 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3146 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3147 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3148 | if (ret) { |
3149 | btrfs_abort_transaction(trans, root, ret); | |
3150 | goto out; | |
3151 | } | |
b3b4aa74 | 3152 | btrfs_release_path(path); |
4df27c4d YZ |
3153 | |
3154 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3155 | objectid, root->root_key.objectid, | |
33345d01 | 3156 | dir_ino, &index, name, name_len); |
4df27c4d | 3157 | if (ret < 0) { |
79787eaa JM |
3158 | if (ret != -ENOENT) { |
3159 | btrfs_abort_transaction(trans, root, ret); | |
3160 | goto out; | |
3161 | } | |
33345d01 | 3162 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3163 | name, name_len); |
79787eaa JM |
3164 | if (IS_ERR_OR_NULL(di)) { |
3165 | if (!di) | |
3166 | ret = -ENOENT; | |
3167 | else | |
3168 | ret = PTR_ERR(di); | |
3169 | btrfs_abort_transaction(trans, root, ret); | |
3170 | goto out; | |
3171 | } | |
4df27c4d YZ |
3172 | |
3173 | leaf = path->nodes[0]; | |
3174 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3175 | btrfs_release_path(path); |
4df27c4d YZ |
3176 | index = key.offset; |
3177 | } | |
945d8962 | 3178 | btrfs_release_path(path); |
4df27c4d | 3179 | |
16cdcec7 | 3180 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3181 | if (ret) { |
3182 | btrfs_abort_transaction(trans, root, ret); | |
3183 | goto out; | |
3184 | } | |
4df27c4d YZ |
3185 | |
3186 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3187 | inode_inc_iversion(dir); |
4df27c4d | 3188 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3189 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3190 | if (ret) |
3191 | btrfs_abort_transaction(trans, root, ret); | |
3192 | out: | |
71d7aed0 | 3193 | btrfs_free_path(path); |
79787eaa | 3194 | return ret; |
4df27c4d YZ |
3195 | } |
3196 | ||
39279cc3 CM |
3197 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3198 | { | |
3199 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3200 | int err = 0; |
39279cc3 | 3201 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3202 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3203 | unsigned long nr = 0; |
39279cc3 | 3204 | |
3394e160 | 3205 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 3206 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3207 | return -ENOTEMPTY; |
3208 | ||
a22285a6 YZ |
3209 | trans = __unlink_start_trans(dir, dentry); |
3210 | if (IS_ERR(trans)) | |
5df6a9f6 | 3211 | return PTR_ERR(trans); |
5df6a9f6 | 3212 | |
33345d01 | 3213 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3214 | err = btrfs_unlink_subvol(trans, root, dir, |
3215 | BTRFS_I(inode)->location.objectid, | |
3216 | dentry->d_name.name, | |
3217 | dentry->d_name.len); | |
3218 | goto out; | |
3219 | } | |
3220 | ||
7b128766 JB |
3221 | err = btrfs_orphan_add(trans, inode); |
3222 | if (err) | |
4df27c4d | 3223 | goto out; |
7b128766 | 3224 | |
39279cc3 | 3225 | /* now the directory is empty */ |
e02119d5 CM |
3226 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3227 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3228 | if (!err) |
dbe674a9 | 3229 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3230 | out: |
d3c2fdcf | 3231 | nr = trans->blocks_used; |
a22285a6 | 3232 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3233 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3234 | |
39279cc3 CM |
3235 | return err; |
3236 | } | |
3237 | ||
39279cc3 CM |
3238 | /* |
3239 | * this can truncate away extent items, csum items and directory items. | |
3240 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3241 | * any higher than new_size |
39279cc3 CM |
3242 | * |
3243 | * csum items that cross the new i_size are truncated to the new size | |
3244 | * as well. | |
7b128766 JB |
3245 | * |
3246 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3247 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3248 | */ |
8082510e YZ |
3249 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3250 | struct btrfs_root *root, | |
3251 | struct inode *inode, | |
3252 | u64 new_size, u32 min_type) | |
39279cc3 | 3253 | { |
39279cc3 | 3254 | struct btrfs_path *path; |
5f39d397 | 3255 | struct extent_buffer *leaf; |
39279cc3 | 3256 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3257 | struct btrfs_key key; |
3258 | struct btrfs_key found_key; | |
39279cc3 | 3259 | u64 extent_start = 0; |
db94535d | 3260 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3261 | u64 extent_offset = 0; |
39279cc3 | 3262 | u64 item_end = 0; |
8082510e YZ |
3263 | u64 mask = root->sectorsize - 1; |
3264 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3265 | int found_extent; |
3266 | int del_item; | |
85e21bac CM |
3267 | int pending_del_nr = 0; |
3268 | int pending_del_slot = 0; | |
179e29e4 | 3269 | int extent_type = -1; |
8082510e YZ |
3270 | int ret; |
3271 | int err = 0; | |
33345d01 | 3272 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3273 | |
3274 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3275 | |
0eb0e19c MF |
3276 | path = btrfs_alloc_path(); |
3277 | if (!path) | |
3278 | return -ENOMEM; | |
3279 | path->reada = -1; | |
3280 | ||
5dc562c5 JB |
3281 | /* |
3282 | * We want to drop from the next block forward in case this new size is | |
3283 | * not block aligned since we will be keeping the last block of the | |
3284 | * extent just the way it is. | |
3285 | */ | |
0af3d00b | 3286 | if (root->ref_cows || root == root->fs_info->tree_root) |
5dc562c5 | 3287 | btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0); |
8082510e | 3288 | |
16cdcec7 MX |
3289 | /* |
3290 | * This function is also used to drop the items in the log tree before | |
3291 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3292 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3293 | * items. | |
3294 | */ | |
3295 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3296 | btrfs_kill_delayed_inode_items(inode); | |
3297 | ||
33345d01 | 3298 | key.objectid = ino; |
39279cc3 | 3299 | key.offset = (u64)-1; |
5f39d397 CM |
3300 | key.type = (u8)-1; |
3301 | ||
85e21bac | 3302 | search_again: |
b9473439 | 3303 | path->leave_spinning = 1; |
85e21bac | 3304 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3305 | if (ret < 0) { |
3306 | err = ret; | |
3307 | goto out; | |
3308 | } | |
d397712b | 3309 | |
85e21bac | 3310 | if (ret > 0) { |
e02119d5 CM |
3311 | /* there are no items in the tree for us to truncate, we're |
3312 | * done | |
3313 | */ | |
8082510e YZ |
3314 | if (path->slots[0] == 0) |
3315 | goto out; | |
85e21bac CM |
3316 | path->slots[0]--; |
3317 | } | |
3318 | ||
d397712b | 3319 | while (1) { |
39279cc3 | 3320 | fi = NULL; |
5f39d397 CM |
3321 | leaf = path->nodes[0]; |
3322 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3323 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3324 | |
33345d01 | 3325 | if (found_key.objectid != ino) |
39279cc3 | 3326 | break; |
5f39d397 | 3327 | |
85e21bac | 3328 | if (found_type < min_type) |
39279cc3 CM |
3329 | break; |
3330 | ||
5f39d397 | 3331 | item_end = found_key.offset; |
39279cc3 | 3332 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3333 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3334 | struct btrfs_file_extent_item); |
179e29e4 CM |
3335 | extent_type = btrfs_file_extent_type(leaf, fi); |
3336 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3337 | item_end += |
db94535d | 3338 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3339 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3340 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3341 | fi); |
39279cc3 | 3342 | } |
008630c1 | 3343 | item_end--; |
39279cc3 | 3344 | } |
8082510e YZ |
3345 | if (found_type > min_type) { |
3346 | del_item = 1; | |
3347 | } else { | |
3348 | if (item_end < new_size) | |
b888db2b | 3349 | break; |
8082510e YZ |
3350 | if (found_key.offset >= new_size) |
3351 | del_item = 1; | |
3352 | else | |
3353 | del_item = 0; | |
39279cc3 | 3354 | } |
39279cc3 | 3355 | found_extent = 0; |
39279cc3 | 3356 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3357 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3358 | goto delete; | |
3359 | ||
3360 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3361 | u64 num_dec; |
db94535d | 3362 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3363 | if (!del_item) { |
db94535d CM |
3364 | u64 orig_num_bytes = |
3365 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3366 | extent_num_bytes = new_size - |
5f39d397 | 3367 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3368 | extent_num_bytes = extent_num_bytes & |
3369 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3370 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3371 | extent_num_bytes); | |
3372 | num_dec = (orig_num_bytes - | |
9069218d | 3373 | extent_num_bytes); |
e02119d5 | 3374 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3375 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3376 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3377 | } else { |
db94535d CM |
3378 | extent_num_bytes = |
3379 | btrfs_file_extent_disk_num_bytes(leaf, | |
3380 | fi); | |
5d4f98a2 YZ |
3381 | extent_offset = found_key.offset - |
3382 | btrfs_file_extent_offset(leaf, fi); | |
3383 | ||
39279cc3 | 3384 | /* FIXME blocksize != 4096 */ |
9069218d | 3385 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3386 | if (extent_start != 0) { |
3387 | found_extent = 1; | |
e02119d5 | 3388 | if (root->ref_cows) |
a76a3cd4 | 3389 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3390 | } |
39279cc3 | 3391 | } |
9069218d | 3392 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3393 | /* |
3394 | * we can't truncate inline items that have had | |
3395 | * special encodings | |
3396 | */ | |
3397 | if (!del_item && | |
3398 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3399 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3400 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3401 | u32 size = new_size - found_key.offset; |
3402 | ||
3403 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3404 | inode_sub_bytes(inode, item_end + 1 - |
3405 | new_size); | |
e02119d5 CM |
3406 | } |
3407 | size = | |
3408 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3409 | btrfs_truncate_item(trans, root, path, |
3410 | size, 1); | |
e02119d5 | 3411 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3412 | inode_sub_bytes(inode, item_end + 1 - |
3413 | found_key.offset); | |
9069218d | 3414 | } |
39279cc3 | 3415 | } |
179e29e4 | 3416 | delete: |
39279cc3 | 3417 | if (del_item) { |
85e21bac CM |
3418 | if (!pending_del_nr) { |
3419 | /* no pending yet, add ourselves */ | |
3420 | pending_del_slot = path->slots[0]; | |
3421 | pending_del_nr = 1; | |
3422 | } else if (pending_del_nr && | |
3423 | path->slots[0] + 1 == pending_del_slot) { | |
3424 | /* hop on the pending chunk */ | |
3425 | pending_del_nr++; | |
3426 | pending_del_slot = path->slots[0]; | |
3427 | } else { | |
d397712b | 3428 | BUG(); |
85e21bac | 3429 | } |
39279cc3 CM |
3430 | } else { |
3431 | break; | |
3432 | } | |
0af3d00b JB |
3433 | if (found_extent && (root->ref_cows || |
3434 | root == root->fs_info->tree_root)) { | |
b9473439 | 3435 | btrfs_set_path_blocking(path); |
39279cc3 | 3436 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3437 | extent_num_bytes, 0, |
3438 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3439 | ino, extent_offset, 0); |
39279cc3 CM |
3440 | BUG_ON(ret); |
3441 | } | |
85e21bac | 3442 | |
8082510e YZ |
3443 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3444 | break; | |
3445 | ||
3446 | if (path->slots[0] == 0 || | |
3447 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
3448 | if (pending_del_nr) { |
3449 | ret = btrfs_del_items(trans, root, path, | |
3450 | pending_del_slot, | |
3451 | pending_del_nr); | |
79787eaa JM |
3452 | if (ret) { |
3453 | btrfs_abort_transaction(trans, | |
3454 | root, ret); | |
3455 | goto error; | |
3456 | } | |
8082510e YZ |
3457 | pending_del_nr = 0; |
3458 | } | |
b3b4aa74 | 3459 | btrfs_release_path(path); |
85e21bac | 3460 | goto search_again; |
8082510e YZ |
3461 | } else { |
3462 | path->slots[0]--; | |
85e21bac | 3463 | } |
39279cc3 | 3464 | } |
8082510e | 3465 | out: |
85e21bac CM |
3466 | if (pending_del_nr) { |
3467 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3468 | pending_del_nr); | |
79787eaa JM |
3469 | if (ret) |
3470 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3471 | } |
79787eaa | 3472 | error: |
39279cc3 | 3473 | btrfs_free_path(path); |
8082510e | 3474 | return err; |
39279cc3 CM |
3475 | } |
3476 | ||
3477 | /* | |
3478 | * taken from block_truncate_page, but does cow as it zeros out | |
3479 | * any bytes left in the last page in the file. | |
3480 | */ | |
3481 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3482 | { | |
3483 | struct inode *inode = mapping->host; | |
db94535d | 3484 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3485 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3486 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3487 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3488 | char *kaddr; |
db94535d | 3489 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3490 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3491 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3492 | struct page *page; | |
3b16a4e3 | 3493 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3494 | int ret = 0; |
a52d9a80 | 3495 | u64 page_start; |
e6dcd2dc | 3496 | u64 page_end; |
39279cc3 CM |
3497 | |
3498 | if ((offset & (blocksize - 1)) == 0) | |
3499 | goto out; | |
0ca1f7ce | 3500 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3501 | if (ret) |
3502 | goto out; | |
39279cc3 CM |
3503 | |
3504 | ret = -ENOMEM; | |
211c17f5 | 3505 | again: |
3b16a4e3 | 3506 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3507 | if (!page) { |
0ca1f7ce | 3508 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3509 | goto out; |
5d5e103a | 3510 | } |
e6dcd2dc CM |
3511 | |
3512 | page_start = page_offset(page); | |
3513 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3514 | ||
39279cc3 | 3515 | if (!PageUptodate(page)) { |
9ebefb18 | 3516 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3517 | lock_page(page); |
211c17f5 CM |
3518 | if (page->mapping != mapping) { |
3519 | unlock_page(page); | |
3520 | page_cache_release(page); | |
3521 | goto again; | |
3522 | } | |
39279cc3 CM |
3523 | if (!PageUptodate(page)) { |
3524 | ret = -EIO; | |
89642229 | 3525 | goto out_unlock; |
39279cc3 CM |
3526 | } |
3527 | } | |
211c17f5 | 3528 | wait_on_page_writeback(page); |
e6dcd2dc | 3529 | |
d0082371 | 3530 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3531 | set_page_extent_mapped(page); |
3532 | ||
3533 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3534 | if (ordered) { | |
2ac55d41 JB |
3535 | unlock_extent_cached(io_tree, page_start, page_end, |
3536 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3537 | unlock_page(page); |
3538 | page_cache_release(page); | |
eb84ae03 | 3539 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3540 | btrfs_put_ordered_extent(ordered); |
3541 | goto again; | |
3542 | } | |
3543 | ||
2ac55d41 | 3544 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3545 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3546 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3547 | |
2ac55d41 JB |
3548 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3549 | &cached_state); | |
9ed74f2d | 3550 | if (ret) { |
2ac55d41 JB |
3551 | unlock_extent_cached(io_tree, page_start, page_end, |
3552 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3553 | goto out_unlock; |
3554 | } | |
3555 | ||
e6dcd2dc CM |
3556 | ret = 0; |
3557 | if (offset != PAGE_CACHE_SIZE) { | |
3558 | kaddr = kmap(page); | |
3559 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3560 | flush_dcache_page(page); | |
3561 | kunmap(page); | |
3562 | } | |
247e743c | 3563 | ClearPageChecked(page); |
e6dcd2dc | 3564 | set_page_dirty(page); |
2ac55d41 JB |
3565 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3566 | GFP_NOFS); | |
39279cc3 | 3567 | |
89642229 | 3568 | out_unlock: |
5d5e103a | 3569 | if (ret) |
0ca1f7ce | 3570 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3571 | unlock_page(page); |
3572 | page_cache_release(page); | |
3573 | out: | |
3574 | return ret; | |
3575 | } | |
3576 | ||
695a0d0d JB |
3577 | /* |
3578 | * This function puts in dummy file extents for the area we're creating a hole | |
3579 | * for. So if we are truncating this file to a larger size we need to insert | |
3580 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3581 | * the range between oldsize and size | |
3582 | */ | |
a41ad394 | 3583 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3584 | { |
9036c102 YZ |
3585 | struct btrfs_trans_handle *trans; |
3586 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3587 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3588 | struct extent_map *em = NULL; |
2ac55d41 | 3589 | struct extent_state *cached_state = NULL; |
5dc562c5 | 3590 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9036c102 | 3591 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3592 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3593 | u64 block_end = (size + mask) & ~mask; |
3594 | u64 last_byte; | |
3595 | u64 cur_offset; | |
3596 | u64 hole_size; | |
9ed74f2d | 3597 | int err = 0; |
39279cc3 | 3598 | |
9036c102 YZ |
3599 | if (size <= hole_start) |
3600 | return 0; | |
3601 | ||
9036c102 YZ |
3602 | while (1) { |
3603 | struct btrfs_ordered_extent *ordered; | |
3604 | btrfs_wait_ordered_range(inode, hole_start, | |
3605 | block_end - hole_start); | |
2ac55d41 | 3606 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3607 | &cached_state); |
9036c102 YZ |
3608 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3609 | if (!ordered) | |
3610 | break; | |
2ac55d41 JB |
3611 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3612 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3613 | btrfs_put_ordered_extent(ordered); |
3614 | } | |
39279cc3 | 3615 | |
9036c102 YZ |
3616 | cur_offset = hole_start; |
3617 | while (1) { | |
3618 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3619 | block_end - cur_offset, 0); | |
79787eaa JM |
3620 | if (IS_ERR(em)) { |
3621 | err = PTR_ERR(em); | |
3622 | break; | |
3623 | } | |
9036c102 YZ |
3624 | last_byte = min(extent_map_end(em), block_end); |
3625 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3626 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 3627 | struct extent_map *hole_em; |
9036c102 | 3628 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3629 | |
3642320e | 3630 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3631 | if (IS_ERR(trans)) { |
3632 | err = PTR_ERR(trans); | |
9ed74f2d | 3633 | break; |
a22285a6 | 3634 | } |
8082510e | 3635 | |
5dc562c5 JB |
3636 | err = btrfs_drop_extents(trans, root, inode, |
3637 | cur_offset, | |
2671485d | 3638 | cur_offset + hole_size, 1); |
5b397377 | 3639 | if (err) { |
79787eaa | 3640 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3641 | btrfs_end_transaction(trans, root); |
3893e33b | 3642 | break; |
5b397377 | 3643 | } |
8082510e | 3644 | |
9036c102 | 3645 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3646 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3647 | 0, hole_size, 0, hole_size, |
3648 | 0, 0, 0); | |
5b397377 | 3649 | if (err) { |
79787eaa | 3650 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3651 | btrfs_end_transaction(trans, root); |
3893e33b | 3652 | break; |
5b397377 | 3653 | } |
8082510e | 3654 | |
5dc562c5 JB |
3655 | btrfs_drop_extent_cache(inode, cur_offset, |
3656 | cur_offset + hole_size - 1, 0); | |
3657 | hole_em = alloc_extent_map(); | |
3658 | if (!hole_em) { | |
3659 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3660 | &BTRFS_I(inode)->runtime_flags); | |
3661 | goto next; | |
3662 | } | |
3663 | hole_em->start = cur_offset; | |
3664 | hole_em->len = hole_size; | |
3665 | hole_em->orig_start = cur_offset; | |
3666 | ||
3667 | hole_em->block_start = EXTENT_MAP_HOLE; | |
3668 | hole_em->block_len = 0; | |
3669 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; | |
3670 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
3671 | hole_em->generation = trans->transid; | |
8082510e | 3672 | |
5dc562c5 JB |
3673 | while (1) { |
3674 | write_lock(&em_tree->lock); | |
3675 | err = add_extent_mapping(em_tree, hole_em); | |
3676 | if (!err) | |
3677 | list_move(&hole_em->list, | |
3678 | &em_tree->modified_extents); | |
3679 | write_unlock(&em_tree->lock); | |
3680 | if (err != -EEXIST) | |
3681 | break; | |
3682 | btrfs_drop_extent_cache(inode, cur_offset, | |
3683 | cur_offset + | |
3684 | hole_size - 1, 0); | |
3685 | } | |
3686 | free_extent_map(hole_em); | |
3687 | next: | |
3642320e | 3688 | btrfs_update_inode(trans, root, inode); |
8082510e | 3689 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3690 | } |
3691 | free_extent_map(em); | |
a22285a6 | 3692 | em = NULL; |
9036c102 | 3693 | cur_offset = last_byte; |
8082510e | 3694 | if (cur_offset >= block_end) |
9036c102 YZ |
3695 | break; |
3696 | } | |
1832a6d5 | 3697 | |
a22285a6 | 3698 | free_extent_map(em); |
2ac55d41 JB |
3699 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3700 | GFP_NOFS); | |
9036c102 YZ |
3701 | return err; |
3702 | } | |
39279cc3 | 3703 | |
a41ad394 | 3704 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3705 | { |
f4a2f4c5 MX |
3706 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3707 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3708 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3709 | int ret; |
3710 | ||
a41ad394 | 3711 | if (newsize == oldsize) |
8082510e YZ |
3712 | return 0; |
3713 | ||
a41ad394 | 3714 | if (newsize > oldsize) { |
a41ad394 JB |
3715 | truncate_pagecache(inode, oldsize, newsize); |
3716 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3717 | if (ret) |
8082510e | 3718 | return ret; |
8082510e | 3719 | |
f4a2f4c5 MX |
3720 | trans = btrfs_start_transaction(root, 1); |
3721 | if (IS_ERR(trans)) | |
3722 | return PTR_ERR(trans); | |
3723 | ||
3724 | i_size_write(inode, newsize); | |
3725 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3726 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3727 | btrfs_end_transaction(trans, root); |
a41ad394 | 3728 | } else { |
8082510e | 3729 | |
a41ad394 JB |
3730 | /* |
3731 | * We're truncating a file that used to have good data down to | |
3732 | * zero. Make sure it gets into the ordered flush list so that | |
3733 | * any new writes get down to disk quickly. | |
3734 | */ | |
3735 | if (newsize == 0) | |
72ac3c0d JB |
3736 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3737 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3738 | |
a41ad394 JB |
3739 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3740 | truncate_setsize(inode, newsize); | |
3741 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3742 | } |
3743 | ||
a41ad394 | 3744 | return ret; |
8082510e YZ |
3745 | } |
3746 | ||
9036c102 YZ |
3747 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3748 | { | |
3749 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3750 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3751 | int err; |
39279cc3 | 3752 | |
b83cc969 LZ |
3753 | if (btrfs_root_readonly(root)) |
3754 | return -EROFS; | |
3755 | ||
9036c102 YZ |
3756 | err = inode_change_ok(inode, attr); |
3757 | if (err) | |
3758 | return err; | |
2bf5a725 | 3759 | |
5a3f23d5 | 3760 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3761 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3762 | if (err) |
3763 | return err; | |
39279cc3 | 3764 | } |
9036c102 | 3765 | |
1025774c CH |
3766 | if (attr->ia_valid) { |
3767 | setattr_copy(inode, attr); | |
0c4d2d95 | 3768 | inode_inc_iversion(inode); |
22c44fe6 | 3769 | err = btrfs_dirty_inode(inode); |
1025774c | 3770 | |
22c44fe6 | 3771 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3772 | err = btrfs_acl_chmod(inode); |
3773 | } | |
33268eaf | 3774 | |
39279cc3 CM |
3775 | return err; |
3776 | } | |
61295eb8 | 3777 | |
bd555975 | 3778 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3779 | { |
3780 | struct btrfs_trans_handle *trans; | |
3781 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3782 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3783 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3784 | unsigned long nr; |
39279cc3 CM |
3785 | int ret; |
3786 | ||
1abe9b8a | 3787 | trace_btrfs_inode_evict(inode); |
3788 | ||
39279cc3 | 3789 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3790 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 3791 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
3792 | goto no_delete; |
3793 | ||
39279cc3 | 3794 | if (is_bad_inode(inode)) { |
7b128766 | 3795 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3796 | goto no_delete; |
3797 | } | |
bd555975 | 3798 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3799 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3800 | |
c71bf099 | 3801 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 3802 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 3803 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
3804 | goto no_delete; |
3805 | } | |
3806 | ||
76dda93c YZ |
3807 | if (inode->i_nlink > 0) { |
3808 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3809 | goto no_delete; | |
3810 | } | |
3811 | ||
4289a667 JB |
3812 | rsv = btrfs_alloc_block_rsv(root); |
3813 | if (!rsv) { | |
3814 | btrfs_orphan_del(NULL, inode); | |
3815 | goto no_delete; | |
3816 | } | |
4a338542 | 3817 | rsv->size = min_size; |
ca7e70f5 | 3818 | rsv->failfast = 1; |
726c35fa | 3819 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3820 | |
dbe674a9 | 3821 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3822 | |
4289a667 JB |
3823 | /* |
3824 | * This is a bit simpler than btrfs_truncate since | |
3825 | * | |
3826 | * 1) We've already reserved our space for our orphan item in the | |
3827 | * unlink. | |
3828 | * 2) We're going to delete the inode item, so we don't need to update | |
3829 | * it at all. | |
3830 | * | |
3831 | * So we just need to reserve some slack space in case we add bytes when | |
3832 | * doing the truncate. | |
3833 | */ | |
8082510e | 3834 | while (1) { |
aa38a711 | 3835 | ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size); |
726c35fa JB |
3836 | |
3837 | /* | |
3838 | * Try and steal from the global reserve since we will | |
3839 | * likely not use this space anyway, we want to try as | |
3840 | * hard as possible to get this to work. | |
3841 | */ | |
3842 | if (ret) | |
3843 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3844 | |
d68fc57b | 3845 | if (ret) { |
4289a667 | 3846 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3847 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3848 | btrfs_orphan_del(NULL, inode); |
3849 | btrfs_free_block_rsv(root, rsv); | |
3850 | goto no_delete; | |
d68fc57b | 3851 | } |
7b128766 | 3852 | |
4289a667 JB |
3853 | trans = btrfs_start_transaction(root, 0); |
3854 | if (IS_ERR(trans)) { | |
3855 | btrfs_orphan_del(NULL, inode); | |
3856 | btrfs_free_block_rsv(root, rsv); | |
3857 | goto no_delete; | |
d68fc57b | 3858 | } |
7b128766 | 3859 | |
4289a667 JB |
3860 | trans->block_rsv = rsv; |
3861 | ||
d68fc57b | 3862 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 3863 | if (ret != -ENOSPC) |
8082510e | 3864 | break; |
85e21bac | 3865 | |
8082510e YZ |
3866 | nr = trans->blocks_used; |
3867 | btrfs_end_transaction(trans, root); | |
3868 | trans = NULL; | |
3869 | btrfs_btree_balance_dirty(root, nr); | |
3870 | } | |
5f39d397 | 3871 | |
4289a667 JB |
3872 | btrfs_free_block_rsv(root, rsv); |
3873 | ||
8082510e | 3874 | if (ret == 0) { |
4289a667 | 3875 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3876 | ret = btrfs_orphan_del(trans, inode); |
3877 | BUG_ON(ret); | |
3878 | } | |
54aa1f4d | 3879 | |
4289a667 | 3880 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3881 | if (!(root == root->fs_info->tree_root || |
3882 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3883 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3884 | |
d3c2fdcf | 3885 | nr = trans->blocks_used; |
54aa1f4d | 3886 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3887 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3888 | no_delete: |
dbd5768f | 3889 | clear_inode(inode); |
8082510e | 3890 | return; |
39279cc3 CM |
3891 | } |
3892 | ||
3893 | /* | |
3894 | * this returns the key found in the dir entry in the location pointer. | |
3895 | * If no dir entries were found, location->objectid is 0. | |
3896 | */ | |
3897 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3898 | struct btrfs_key *location) | |
3899 | { | |
3900 | const char *name = dentry->d_name.name; | |
3901 | int namelen = dentry->d_name.len; | |
3902 | struct btrfs_dir_item *di; | |
3903 | struct btrfs_path *path; | |
3904 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3905 | int ret = 0; |
39279cc3 CM |
3906 | |
3907 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3908 | if (!path) |
3909 | return -ENOMEM; | |
3954401f | 3910 | |
33345d01 | 3911 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3912 | namelen, 0); |
0d9f7f3e Y |
3913 | if (IS_ERR(di)) |
3914 | ret = PTR_ERR(di); | |
d397712b | 3915 | |
c704005d | 3916 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3917 | goto out_err; |
d397712b | 3918 | |
5f39d397 | 3919 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3920 | out: |
39279cc3 CM |
3921 | btrfs_free_path(path); |
3922 | return ret; | |
3954401f CM |
3923 | out_err: |
3924 | location->objectid = 0; | |
3925 | goto out; | |
39279cc3 CM |
3926 | } |
3927 | ||
3928 | /* | |
3929 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3930 | * needs to be changed to reflect the root directory of the tree root. This | |
3931 | * is kind of like crossing a mount point. | |
3932 | */ | |
3933 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3934 | struct inode *dir, |
3935 | struct dentry *dentry, | |
3936 | struct btrfs_key *location, | |
3937 | struct btrfs_root **sub_root) | |
39279cc3 | 3938 | { |
4df27c4d YZ |
3939 | struct btrfs_path *path; |
3940 | struct btrfs_root *new_root; | |
3941 | struct btrfs_root_ref *ref; | |
3942 | struct extent_buffer *leaf; | |
3943 | int ret; | |
3944 | int err = 0; | |
39279cc3 | 3945 | |
4df27c4d YZ |
3946 | path = btrfs_alloc_path(); |
3947 | if (!path) { | |
3948 | err = -ENOMEM; | |
3949 | goto out; | |
3950 | } | |
39279cc3 | 3951 | |
4df27c4d YZ |
3952 | err = -ENOENT; |
3953 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3954 | BTRFS_I(dir)->root->root_key.objectid, | |
3955 | location->objectid); | |
3956 | if (ret) { | |
3957 | if (ret < 0) | |
3958 | err = ret; | |
3959 | goto out; | |
3960 | } | |
39279cc3 | 3961 | |
4df27c4d YZ |
3962 | leaf = path->nodes[0]; |
3963 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3964 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3965 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3966 | goto out; | |
39279cc3 | 3967 | |
4df27c4d YZ |
3968 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3969 | (unsigned long)(ref + 1), | |
3970 | dentry->d_name.len); | |
3971 | if (ret) | |
3972 | goto out; | |
3973 | ||
b3b4aa74 | 3974 | btrfs_release_path(path); |
4df27c4d YZ |
3975 | |
3976 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3977 | if (IS_ERR(new_root)) { | |
3978 | err = PTR_ERR(new_root); | |
3979 | goto out; | |
3980 | } | |
3981 | ||
3982 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3983 | err = -ENOENT; | |
3984 | goto out; | |
3985 | } | |
3986 | ||
3987 | *sub_root = new_root; | |
3988 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3989 | location->type = BTRFS_INODE_ITEM_KEY; | |
3990 | location->offset = 0; | |
3991 | err = 0; | |
3992 | out: | |
3993 | btrfs_free_path(path); | |
3994 | return err; | |
39279cc3 CM |
3995 | } |
3996 | ||
5d4f98a2 YZ |
3997 | static void inode_tree_add(struct inode *inode) |
3998 | { | |
3999 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4000 | struct btrfs_inode *entry; | |
03e860bd FNP |
4001 | struct rb_node **p; |
4002 | struct rb_node *parent; | |
33345d01 | 4003 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
4004 | again: |
4005 | p = &root->inode_tree.rb_node; | |
4006 | parent = NULL; | |
5d4f98a2 | 4007 | |
1d3382cb | 4008 | if (inode_unhashed(inode)) |
76dda93c YZ |
4009 | return; |
4010 | ||
5d4f98a2 YZ |
4011 | spin_lock(&root->inode_lock); |
4012 | while (*p) { | |
4013 | parent = *p; | |
4014 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4015 | ||
33345d01 | 4016 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4017 | p = &parent->rb_left; |
33345d01 | 4018 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4019 | p = &parent->rb_right; |
5d4f98a2 YZ |
4020 | else { |
4021 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4022 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4023 | rb_erase(parent, &root->inode_tree); |
4024 | RB_CLEAR_NODE(parent); | |
4025 | spin_unlock(&root->inode_lock); | |
4026 | goto again; | |
5d4f98a2 YZ |
4027 | } |
4028 | } | |
4029 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4030 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4031 | spin_unlock(&root->inode_lock); | |
4032 | } | |
4033 | ||
4034 | static void inode_tree_del(struct inode *inode) | |
4035 | { | |
4036 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4037 | int empty = 0; |
5d4f98a2 | 4038 | |
03e860bd | 4039 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4040 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4041 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4042 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4043 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4044 | } |
03e860bd | 4045 | spin_unlock(&root->inode_lock); |
76dda93c | 4046 | |
0af3d00b JB |
4047 | /* |
4048 | * Free space cache has inodes in the tree root, but the tree root has a | |
4049 | * root_refs of 0, so this could end up dropping the tree root as a | |
4050 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4051 | * make sure we don't drop it. | |
4052 | */ | |
4053 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4054 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4055 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4056 | spin_lock(&root->inode_lock); | |
4057 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4058 | spin_unlock(&root->inode_lock); | |
4059 | if (empty) | |
4060 | btrfs_add_dead_root(root); | |
4061 | } | |
4062 | } | |
4063 | ||
143bede5 | 4064 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4065 | { |
4066 | struct rb_node *node; | |
4067 | struct rb_node *prev; | |
4068 | struct btrfs_inode *entry; | |
4069 | struct inode *inode; | |
4070 | u64 objectid = 0; | |
4071 | ||
4072 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4073 | ||
4074 | spin_lock(&root->inode_lock); | |
4075 | again: | |
4076 | node = root->inode_tree.rb_node; | |
4077 | prev = NULL; | |
4078 | while (node) { | |
4079 | prev = node; | |
4080 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4081 | ||
33345d01 | 4082 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4083 | node = node->rb_left; |
33345d01 | 4084 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4085 | node = node->rb_right; |
4086 | else | |
4087 | break; | |
4088 | } | |
4089 | if (!node) { | |
4090 | while (prev) { | |
4091 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4092 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4093 | node = prev; |
4094 | break; | |
4095 | } | |
4096 | prev = rb_next(prev); | |
4097 | } | |
4098 | } | |
4099 | while (node) { | |
4100 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4101 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4102 | inode = igrab(&entry->vfs_inode); |
4103 | if (inode) { | |
4104 | spin_unlock(&root->inode_lock); | |
4105 | if (atomic_read(&inode->i_count) > 1) | |
4106 | d_prune_aliases(inode); | |
4107 | /* | |
45321ac5 | 4108 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4109 | * the inode cache when its usage count |
4110 | * hits zero. | |
4111 | */ | |
4112 | iput(inode); | |
4113 | cond_resched(); | |
4114 | spin_lock(&root->inode_lock); | |
4115 | goto again; | |
4116 | } | |
4117 | ||
4118 | if (cond_resched_lock(&root->inode_lock)) | |
4119 | goto again; | |
4120 | ||
4121 | node = rb_next(node); | |
4122 | } | |
4123 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4124 | } |
4125 | ||
e02119d5 CM |
4126 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4127 | { | |
4128 | struct btrfs_iget_args *args = p; | |
4129 | inode->i_ino = args->ino; | |
e02119d5 | 4130 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4131 | return 0; |
4132 | } | |
4133 | ||
4134 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4135 | { | |
4136 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4137 | return args->ino == btrfs_ino(inode) && |
d397712b | 4138 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4139 | } |
4140 | ||
5d4f98a2 YZ |
4141 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4142 | u64 objectid, | |
4143 | struct btrfs_root *root) | |
39279cc3 CM |
4144 | { |
4145 | struct inode *inode; | |
4146 | struct btrfs_iget_args args; | |
4147 | args.ino = objectid; | |
4148 | args.root = root; | |
4149 | ||
4150 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4151 | btrfs_init_locked_inode, | |
4152 | (void *)&args); | |
4153 | return inode; | |
4154 | } | |
4155 | ||
1a54ef8c BR |
4156 | /* Get an inode object given its location and corresponding root. |
4157 | * Returns in *is_new if the inode was read from disk | |
4158 | */ | |
4159 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4160 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4161 | { |
4162 | struct inode *inode; | |
4163 | ||
4164 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4165 | if (!inode) | |
5d4f98a2 | 4166 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4167 | |
4168 | if (inode->i_state & I_NEW) { | |
4169 | BTRFS_I(inode)->root = root; | |
4170 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4171 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4172 | if (!is_bad_inode(inode)) { |
4173 | inode_tree_add(inode); | |
4174 | unlock_new_inode(inode); | |
4175 | if (new) | |
4176 | *new = 1; | |
4177 | } else { | |
e0b6d65b ST |
4178 | unlock_new_inode(inode); |
4179 | iput(inode); | |
4180 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4181 | } |
4182 | } | |
4183 | ||
1a54ef8c BR |
4184 | return inode; |
4185 | } | |
4186 | ||
4df27c4d YZ |
4187 | static struct inode *new_simple_dir(struct super_block *s, |
4188 | struct btrfs_key *key, | |
4189 | struct btrfs_root *root) | |
4190 | { | |
4191 | struct inode *inode = new_inode(s); | |
4192 | ||
4193 | if (!inode) | |
4194 | return ERR_PTR(-ENOMEM); | |
4195 | ||
4df27c4d YZ |
4196 | BTRFS_I(inode)->root = root; |
4197 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4198 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4199 | |
4200 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4201 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4202 | inode->i_fop = &simple_dir_operations; |
4203 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4204 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4205 | ||
4206 | return inode; | |
4207 | } | |
4208 | ||
3de4586c | 4209 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4210 | { |
d397712b | 4211 | struct inode *inode; |
4df27c4d | 4212 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4213 | struct btrfs_root *sub_root = root; |
4214 | struct btrfs_key location; | |
76dda93c | 4215 | int index; |
b4aff1f8 | 4216 | int ret = 0; |
39279cc3 CM |
4217 | |
4218 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4219 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4220 | |
b4aff1f8 JB |
4221 | if (unlikely(d_need_lookup(dentry))) { |
4222 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4223 | kfree(dentry->d_fsdata); | |
4224 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4225 | /* This thing is hashed, drop it for now */ |
4226 | d_drop(dentry); | |
b4aff1f8 JB |
4227 | } else { |
4228 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4229 | } | |
5f39d397 | 4230 | |
39279cc3 CM |
4231 | if (ret < 0) |
4232 | return ERR_PTR(ret); | |
5f39d397 | 4233 | |
4df27c4d YZ |
4234 | if (location.objectid == 0) |
4235 | return NULL; | |
4236 | ||
4237 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4238 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4239 | return inode; |
4240 | } | |
4241 | ||
4242 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4243 | ||
76dda93c | 4244 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4245 | ret = fixup_tree_root_location(root, dir, dentry, |
4246 | &location, &sub_root); | |
4247 | if (ret < 0) { | |
4248 | if (ret != -ENOENT) | |
4249 | inode = ERR_PTR(ret); | |
4250 | else | |
4251 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4252 | } else { | |
73f73415 | 4253 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4254 | } |
76dda93c YZ |
4255 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4256 | ||
34d19bad | 4257 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4258 | down_read(&root->fs_info->cleanup_work_sem); |
4259 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4260 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4261 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4262 | if (ret) |
4263 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4264 | } |
4265 | ||
3de4586c CM |
4266 | return inode; |
4267 | } | |
4268 | ||
fe15ce44 | 4269 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4270 | { |
4271 | struct btrfs_root *root; | |
848cce0d | 4272 | struct inode *inode = dentry->d_inode; |
76dda93c | 4273 | |
848cce0d LZ |
4274 | if (!inode && !IS_ROOT(dentry)) |
4275 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4276 | |
848cce0d LZ |
4277 | if (inode) { |
4278 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4279 | if (btrfs_root_refs(&root->root_item) == 0) |
4280 | return 1; | |
848cce0d LZ |
4281 | |
4282 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4283 | return 1; | |
efefb143 | 4284 | } |
76dda93c YZ |
4285 | return 0; |
4286 | } | |
4287 | ||
b4aff1f8 JB |
4288 | static void btrfs_dentry_release(struct dentry *dentry) |
4289 | { | |
4290 | if (dentry->d_fsdata) | |
4291 | kfree(dentry->d_fsdata); | |
4292 | } | |
4293 | ||
3de4586c | 4294 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4295 | unsigned int flags) |
3de4586c | 4296 | { |
a66e7cc6 JB |
4297 | struct dentry *ret; |
4298 | ||
4299 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4300 | if (unlikely(d_need_lookup(dentry))) { | |
4301 | spin_lock(&dentry->d_lock); | |
4302 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4303 | spin_unlock(&dentry->d_lock); | |
4304 | } | |
4305 | return ret; | |
39279cc3 CM |
4306 | } |
4307 | ||
16cdcec7 | 4308 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4309 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4310 | }; | |
4311 | ||
cbdf5a24 DW |
4312 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4313 | filldir_t filldir) | |
39279cc3 | 4314 | { |
6da6abae | 4315 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4316 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4317 | struct btrfs_item *item; | |
4318 | struct btrfs_dir_item *di; | |
4319 | struct btrfs_key key; | |
5f39d397 | 4320 | struct btrfs_key found_key; |
39279cc3 | 4321 | struct btrfs_path *path; |
16cdcec7 MX |
4322 | struct list_head ins_list; |
4323 | struct list_head del_list; | |
39279cc3 | 4324 | int ret; |
5f39d397 | 4325 | struct extent_buffer *leaf; |
39279cc3 | 4326 | int slot; |
39279cc3 CM |
4327 | unsigned char d_type; |
4328 | int over = 0; | |
4329 | u32 di_cur; | |
4330 | u32 di_total; | |
4331 | u32 di_len; | |
4332 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4333 | char tmp_name[32]; |
4334 | char *name_ptr; | |
4335 | int name_len; | |
16cdcec7 | 4336 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4337 | |
4338 | /* FIXME, use a real flag for deciding about the key type */ | |
4339 | if (root->fs_info->tree_root == root) | |
4340 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4341 | |
3954401f CM |
4342 | /* special case for "." */ |
4343 | if (filp->f_pos == 0) { | |
3765fefa HS |
4344 | over = filldir(dirent, ".", 1, |
4345 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4346 | if (over) |
4347 | return 0; | |
4348 | filp->f_pos = 1; | |
4349 | } | |
3954401f CM |
4350 | /* special case for .., just use the back ref */ |
4351 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4352 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4353 | over = filldir(dirent, "..", 2, |
3765fefa | 4354 | filp->f_pos, pino, DT_DIR); |
3954401f | 4355 | if (over) |
49593bfa | 4356 | return 0; |
3954401f CM |
4357 | filp->f_pos = 2; |
4358 | } | |
49593bfa | 4359 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4360 | if (!path) |
4361 | return -ENOMEM; | |
ff5714cc | 4362 | |
026fd317 | 4363 | path->reada = 1; |
49593bfa | 4364 | |
16cdcec7 MX |
4365 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4366 | INIT_LIST_HEAD(&ins_list); | |
4367 | INIT_LIST_HEAD(&del_list); | |
4368 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4369 | } | |
4370 | ||
39279cc3 CM |
4371 | btrfs_set_key_type(&key, key_type); |
4372 | key.offset = filp->f_pos; | |
33345d01 | 4373 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4374 | |
39279cc3 CM |
4375 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4376 | if (ret < 0) | |
4377 | goto err; | |
49593bfa DW |
4378 | |
4379 | while (1) { | |
5f39d397 | 4380 | leaf = path->nodes[0]; |
39279cc3 | 4381 | slot = path->slots[0]; |
b9e03af0 LZ |
4382 | if (slot >= btrfs_header_nritems(leaf)) { |
4383 | ret = btrfs_next_leaf(root, path); | |
4384 | if (ret < 0) | |
4385 | goto err; | |
4386 | else if (ret > 0) | |
4387 | break; | |
4388 | continue; | |
39279cc3 | 4389 | } |
3de4586c | 4390 | |
5f39d397 CM |
4391 | item = btrfs_item_nr(leaf, slot); |
4392 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4393 | ||
4394 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4395 | break; |
5f39d397 | 4396 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4397 | break; |
5f39d397 | 4398 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4399 | goto next; |
16cdcec7 MX |
4400 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4401 | btrfs_should_delete_dir_index(&del_list, | |
4402 | found_key.offset)) | |
4403 | goto next; | |
5f39d397 CM |
4404 | |
4405 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4406 | is_curr = 1; |
49593bfa | 4407 | |
39279cc3 CM |
4408 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4409 | di_cur = 0; | |
5f39d397 | 4410 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4411 | |
4412 | while (di_cur < di_total) { | |
5f39d397 CM |
4413 | struct btrfs_key location; |
4414 | ||
22a94d44 JB |
4415 | if (verify_dir_item(root, leaf, di)) |
4416 | break; | |
4417 | ||
5f39d397 | 4418 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4419 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4420 | name_ptr = tmp_name; |
4421 | } else { | |
4422 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4423 | if (!name_ptr) { |
4424 | ret = -ENOMEM; | |
4425 | goto err; | |
4426 | } | |
5f39d397 CM |
4427 | } |
4428 | read_extent_buffer(leaf, name_ptr, | |
4429 | (unsigned long)(di + 1), name_len); | |
4430 | ||
4431 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4432 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4433 | |
fede766f | 4434 | |
3de4586c | 4435 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4436 | * skip it. |
4437 | * | |
4438 | * In contrast to old kernels, we insert the snapshot's | |
4439 | * dir item and dir index after it has been created, so | |
4440 | * we won't find a reference to our own snapshot. We | |
4441 | * still keep the following code for backward | |
4442 | * compatibility. | |
3de4586c CM |
4443 | */ |
4444 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4445 | location.objectid == root->root_key.objectid) { | |
4446 | over = 0; | |
4447 | goto skip; | |
4448 | } | |
5f39d397 | 4449 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4450 | found_key.offset, location.objectid, |
39279cc3 | 4451 | d_type); |
5f39d397 | 4452 | |
3de4586c | 4453 | skip: |
5f39d397 CM |
4454 | if (name_ptr != tmp_name) |
4455 | kfree(name_ptr); | |
4456 | ||
39279cc3 CM |
4457 | if (over) |
4458 | goto nopos; | |
5103e947 | 4459 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4460 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4461 | di_cur += di_len; |
4462 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4463 | } | |
b9e03af0 LZ |
4464 | next: |
4465 | path->slots[0]++; | |
39279cc3 | 4466 | } |
49593bfa | 4467 | |
16cdcec7 MX |
4468 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4469 | if (is_curr) | |
4470 | filp->f_pos++; | |
4471 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4472 | &ins_list); | |
4473 | if (ret) | |
4474 | goto nopos; | |
4475 | } | |
4476 | ||
49593bfa | 4477 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4478 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4479 | /* |
4480 | * 32-bit glibc will use getdents64, but then strtol - | |
4481 | * so the last number we can serve is this. | |
4482 | */ | |
4483 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4484 | else |
4485 | filp->f_pos++; | |
39279cc3 CM |
4486 | nopos: |
4487 | ret = 0; | |
4488 | err: | |
16cdcec7 MX |
4489 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4490 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4491 | btrfs_free_path(path); |
39279cc3 CM |
4492 | return ret; |
4493 | } | |
4494 | ||
a9185b41 | 4495 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4496 | { |
4497 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4498 | struct btrfs_trans_handle *trans; | |
4499 | int ret = 0; | |
0af3d00b | 4500 | bool nolock = false; |
39279cc3 | 4501 | |
72ac3c0d | 4502 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4503 | return 0; |
4504 | ||
83eea1f1 | 4505 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 4506 | nolock = true; |
0af3d00b | 4507 | |
a9185b41 | 4508 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4509 | if (nolock) |
7a7eaa40 | 4510 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4511 | else |
7a7eaa40 | 4512 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4513 | if (IS_ERR(trans)) |
4514 | return PTR_ERR(trans); | |
0af3d00b JB |
4515 | if (nolock) |
4516 | ret = btrfs_end_transaction_nolock(trans, root); | |
4517 | else | |
4518 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4519 | } |
4520 | return ret; | |
4521 | } | |
4522 | ||
4523 | /* | |
54aa1f4d | 4524 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4525 | * inode changes. But, it is most likely to find the inode in cache. |
4526 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4527 | * to keep or drop this code. | |
4528 | */ | |
22c44fe6 | 4529 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4530 | { |
4531 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4532 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4533 | int ret; |
4534 | ||
72ac3c0d | 4535 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4536 | return 0; |
39279cc3 | 4537 | |
7a7eaa40 | 4538 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4539 | if (IS_ERR(trans)) |
4540 | return PTR_ERR(trans); | |
8929ecfa YZ |
4541 | |
4542 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4543 | if (ret && ret == -ENOSPC) { |
4544 | /* whoops, lets try again with the full transaction */ | |
4545 | btrfs_end_transaction(trans, root); | |
4546 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4547 | if (IS_ERR(trans)) |
4548 | return PTR_ERR(trans); | |
8929ecfa | 4549 | |
94b60442 | 4550 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4551 | } |
39279cc3 | 4552 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4553 | if (BTRFS_I(inode)->delayed_node) |
4554 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4555 | |
4556 | return ret; | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * This is a copy of file_update_time. We need this so we can return error on | |
4561 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4562 | */ | |
e41f941a JB |
4563 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
4564 | int flags) | |
22c44fe6 | 4565 | { |
2bc55652 AB |
4566 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4567 | ||
4568 | if (btrfs_root_readonly(root)) | |
4569 | return -EROFS; | |
4570 | ||
e41f941a | 4571 | if (flags & S_VERSION) |
22c44fe6 | 4572 | inode_inc_iversion(inode); |
e41f941a JB |
4573 | if (flags & S_CTIME) |
4574 | inode->i_ctime = *now; | |
4575 | if (flags & S_MTIME) | |
4576 | inode->i_mtime = *now; | |
4577 | if (flags & S_ATIME) | |
4578 | inode->i_atime = *now; | |
4579 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
4580 | } |
4581 | ||
d352ac68 CM |
4582 | /* |
4583 | * find the highest existing sequence number in a directory | |
4584 | * and then set the in-memory index_cnt variable to reflect | |
4585 | * free sequence numbers | |
4586 | */ | |
aec7477b JB |
4587 | static int btrfs_set_inode_index_count(struct inode *inode) |
4588 | { | |
4589 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4590 | struct btrfs_key key, found_key; | |
4591 | struct btrfs_path *path; | |
4592 | struct extent_buffer *leaf; | |
4593 | int ret; | |
4594 | ||
33345d01 | 4595 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4596 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4597 | key.offset = (u64)-1; | |
4598 | ||
4599 | path = btrfs_alloc_path(); | |
4600 | if (!path) | |
4601 | return -ENOMEM; | |
4602 | ||
4603 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4604 | if (ret < 0) | |
4605 | goto out; | |
4606 | /* FIXME: we should be able to handle this */ | |
4607 | if (ret == 0) | |
4608 | goto out; | |
4609 | ret = 0; | |
4610 | ||
4611 | /* | |
4612 | * MAGIC NUMBER EXPLANATION: | |
4613 | * since we search a directory based on f_pos we have to start at 2 | |
4614 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4615 | * else has to start at 2 | |
4616 | */ | |
4617 | if (path->slots[0] == 0) { | |
4618 | BTRFS_I(inode)->index_cnt = 2; | |
4619 | goto out; | |
4620 | } | |
4621 | ||
4622 | path->slots[0]--; | |
4623 | ||
4624 | leaf = path->nodes[0]; | |
4625 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4626 | ||
33345d01 | 4627 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4628 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4629 | BTRFS_I(inode)->index_cnt = 2; | |
4630 | goto out; | |
4631 | } | |
4632 | ||
4633 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4634 | out: | |
4635 | btrfs_free_path(path); | |
4636 | return ret; | |
4637 | } | |
4638 | ||
d352ac68 CM |
4639 | /* |
4640 | * helper to find a free sequence number in a given directory. This current | |
4641 | * code is very simple, later versions will do smarter things in the btree | |
4642 | */ | |
3de4586c | 4643 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4644 | { |
4645 | int ret = 0; | |
4646 | ||
4647 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4648 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4649 | if (ret) { | |
4650 | ret = btrfs_set_inode_index_count(dir); | |
4651 | if (ret) | |
4652 | return ret; | |
4653 | } | |
aec7477b JB |
4654 | } |
4655 | ||
00e4e6b3 | 4656 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4657 | BTRFS_I(dir)->index_cnt++; |
4658 | ||
4659 | return ret; | |
4660 | } | |
4661 | ||
39279cc3 CM |
4662 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4663 | struct btrfs_root *root, | |
aec7477b | 4664 | struct inode *dir, |
9c58309d | 4665 | const char *name, int name_len, |
175a4eb7 AV |
4666 | u64 ref_objectid, u64 objectid, |
4667 | umode_t mode, u64 *index) | |
39279cc3 CM |
4668 | { |
4669 | struct inode *inode; | |
5f39d397 | 4670 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4671 | struct btrfs_key *location; |
5f39d397 | 4672 | struct btrfs_path *path; |
9c58309d CM |
4673 | struct btrfs_inode_ref *ref; |
4674 | struct btrfs_key key[2]; | |
4675 | u32 sizes[2]; | |
4676 | unsigned long ptr; | |
39279cc3 CM |
4677 | int ret; |
4678 | int owner; | |
4679 | ||
5f39d397 | 4680 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4681 | if (!path) |
4682 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4683 | |
39279cc3 | 4684 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4685 | if (!inode) { |
4686 | btrfs_free_path(path); | |
39279cc3 | 4687 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4688 | } |
39279cc3 | 4689 | |
581bb050 LZ |
4690 | /* |
4691 | * we have to initialize this early, so we can reclaim the inode | |
4692 | * number if we fail afterwards in this function. | |
4693 | */ | |
4694 | inode->i_ino = objectid; | |
4695 | ||
aec7477b | 4696 | if (dir) { |
1abe9b8a | 4697 | trace_btrfs_inode_request(dir); |
4698 | ||
3de4586c | 4699 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4700 | if (ret) { |
8fb27640 | 4701 | btrfs_free_path(path); |
09771430 | 4702 | iput(inode); |
aec7477b | 4703 | return ERR_PTR(ret); |
09771430 | 4704 | } |
aec7477b JB |
4705 | } |
4706 | /* | |
4707 | * index_cnt is ignored for everything but a dir, | |
4708 | * btrfs_get_inode_index_count has an explanation for the magic | |
4709 | * number | |
4710 | */ | |
4711 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4712 | BTRFS_I(inode)->root = root; |
e02119d5 | 4713 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4714 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 4715 | |
5dc562c5 JB |
4716 | /* |
4717 | * We could have gotten an inode number from somebody who was fsynced | |
4718 | * and then removed in this same transaction, so let's just set full | |
4719 | * sync since it will be a full sync anyway and this will blow away the | |
4720 | * old info in the log. | |
4721 | */ | |
4722 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
4723 | ||
569254b0 | 4724 | if (S_ISDIR(mode)) |
39279cc3 CM |
4725 | owner = 0; |
4726 | else | |
4727 | owner = 1; | |
9c58309d CM |
4728 | |
4729 | key[0].objectid = objectid; | |
4730 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4731 | key[0].offset = 0; | |
4732 | ||
4733 | key[1].objectid = objectid; | |
4734 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4735 | key[1].offset = ref_objectid; | |
4736 | ||
4737 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4738 | sizes[1] = name_len + sizeof(*ref); | |
4739 | ||
b9473439 | 4740 | path->leave_spinning = 1; |
9c58309d CM |
4741 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4742 | if (ret != 0) | |
5f39d397 CM |
4743 | goto fail; |
4744 | ||
ecc11fab | 4745 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4746 | inode_set_bytes(inode, 0); |
39279cc3 | 4747 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4748 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4749 | struct btrfs_inode_item); | |
293f7e07 LZ |
4750 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
4751 | sizeof(*inode_item)); | |
e02119d5 | 4752 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4753 | |
4754 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4755 | struct btrfs_inode_ref); | |
4756 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4757 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4758 | ptr = (unsigned long)(ref + 1); |
4759 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4760 | ||
5f39d397 CM |
4761 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4762 | btrfs_free_path(path); | |
4763 | ||
39279cc3 CM |
4764 | location = &BTRFS_I(inode)->location; |
4765 | location->objectid = objectid; | |
39279cc3 CM |
4766 | location->offset = 0; |
4767 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4768 | ||
6cbff00f CH |
4769 | btrfs_inherit_iflags(inode, dir); |
4770 | ||
569254b0 | 4771 | if (S_ISREG(mode)) { |
94272164 CM |
4772 | if (btrfs_test_opt(root, NODATASUM)) |
4773 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4774 | if (btrfs_test_opt(root, NODATACOW) || |
4775 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4776 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4777 | } | |
4778 | ||
39279cc3 | 4779 | insert_inode_hash(inode); |
5d4f98a2 | 4780 | inode_tree_add(inode); |
1abe9b8a | 4781 | |
4782 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4783 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4784 | |
8ea05e3a AB |
4785 | btrfs_update_root_times(trans, root); |
4786 | ||
39279cc3 | 4787 | return inode; |
5f39d397 | 4788 | fail: |
aec7477b JB |
4789 | if (dir) |
4790 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4791 | btrfs_free_path(path); |
09771430 | 4792 | iput(inode); |
5f39d397 | 4793 | return ERR_PTR(ret); |
39279cc3 CM |
4794 | } |
4795 | ||
4796 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4797 | { | |
4798 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4799 | } | |
4800 | ||
d352ac68 CM |
4801 | /* |
4802 | * utility function to add 'inode' into 'parent_inode' with | |
4803 | * a give name and a given sequence number. | |
4804 | * if 'add_backref' is true, also insert a backref from the | |
4805 | * inode to the parent directory. | |
4806 | */ | |
e02119d5 CM |
4807 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4808 | struct inode *parent_inode, struct inode *inode, | |
4809 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4810 | { |
4df27c4d | 4811 | int ret = 0; |
39279cc3 | 4812 | struct btrfs_key key; |
e02119d5 | 4813 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4814 | u64 ino = btrfs_ino(inode); |
4815 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4816 | |
33345d01 | 4817 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4818 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4819 | } else { | |
33345d01 | 4820 | key.objectid = ino; |
4df27c4d YZ |
4821 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4822 | key.offset = 0; | |
4823 | } | |
4824 | ||
33345d01 | 4825 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4826 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4827 | key.objectid, root->root_key.objectid, | |
33345d01 | 4828 | parent_ino, index, name, name_len); |
4df27c4d | 4829 | } else if (add_backref) { |
33345d01 LZ |
4830 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4831 | parent_ino, index); | |
4df27c4d | 4832 | } |
39279cc3 | 4833 | |
79787eaa JM |
4834 | /* Nothing to clean up yet */ |
4835 | if (ret) | |
4836 | return ret; | |
4df27c4d | 4837 | |
79787eaa JM |
4838 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4839 | parent_inode, &key, | |
4840 | btrfs_inode_type(inode), index); | |
4841 | if (ret == -EEXIST) | |
4842 | goto fail_dir_item; | |
4843 | else if (ret) { | |
4844 | btrfs_abort_transaction(trans, root, ret); | |
4845 | return ret; | |
39279cc3 | 4846 | } |
79787eaa JM |
4847 | |
4848 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4849 | name_len * 2); | |
0c4d2d95 | 4850 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4851 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4852 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4853 | if (ret) | |
4854 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4855 | return ret; |
fe66a05a CM |
4856 | |
4857 | fail_dir_item: | |
4858 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4859 | u64 local_index; | |
4860 | int err; | |
4861 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4862 | key.objectid, root->root_key.objectid, | |
4863 | parent_ino, &local_index, name, name_len); | |
4864 | ||
4865 | } else if (add_backref) { | |
4866 | u64 local_index; | |
4867 | int err; | |
4868 | ||
4869 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
4870 | ino, parent_ino, &local_index); | |
4871 | } | |
4872 | return ret; | |
39279cc3 CM |
4873 | } |
4874 | ||
4875 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4876 | struct inode *dir, struct dentry *dentry, |
4877 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4878 | { |
a1b075d2 JB |
4879 | int err = btrfs_add_link(trans, dir, inode, |
4880 | dentry->d_name.name, dentry->d_name.len, | |
4881 | backref, index); | |
39279cc3 CM |
4882 | if (err > 0) |
4883 | err = -EEXIST; | |
4884 | return err; | |
4885 | } | |
4886 | ||
618e21d5 | 4887 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 4888 | umode_t mode, dev_t rdev) |
618e21d5 JB |
4889 | { |
4890 | struct btrfs_trans_handle *trans; | |
4891 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4892 | struct inode *inode = NULL; |
618e21d5 JB |
4893 | int err; |
4894 | int drop_inode = 0; | |
4895 | u64 objectid; | |
1832a6d5 | 4896 | unsigned long nr = 0; |
00e4e6b3 | 4897 | u64 index = 0; |
618e21d5 JB |
4898 | |
4899 | if (!new_valid_dev(rdev)) | |
4900 | return -EINVAL; | |
4901 | ||
9ed74f2d JB |
4902 | /* |
4903 | * 2 for inode item and ref | |
4904 | * 2 for dir items | |
4905 | * 1 for xattr if selinux is on | |
4906 | */ | |
a22285a6 YZ |
4907 | trans = btrfs_start_transaction(root, 5); |
4908 | if (IS_ERR(trans)) | |
4909 | return PTR_ERR(trans); | |
1832a6d5 | 4910 | |
581bb050 LZ |
4911 | err = btrfs_find_free_ino(root, &objectid); |
4912 | if (err) | |
4913 | goto out_unlock; | |
4914 | ||
aec7477b | 4915 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4916 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4917 | mode, &index); |
7cf96da3 TI |
4918 | if (IS_ERR(inode)) { |
4919 | err = PTR_ERR(inode); | |
618e21d5 | 4920 | goto out_unlock; |
7cf96da3 | 4921 | } |
618e21d5 | 4922 | |
2a7dba39 | 4923 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4924 | if (err) { |
4925 | drop_inode = 1; | |
4926 | goto out_unlock; | |
4927 | } | |
4928 | ||
ad19db71 CS |
4929 | /* |
4930 | * If the active LSM wants to access the inode during | |
4931 | * d_instantiate it needs these. Smack checks to see | |
4932 | * if the filesystem supports xattrs by looking at the | |
4933 | * ops vector. | |
4934 | */ | |
4935 | ||
4936 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4937 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4938 | if (err) |
4939 | drop_inode = 1; | |
4940 | else { | |
618e21d5 | 4941 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4942 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4943 | d_instantiate(dentry, inode); |
618e21d5 | 4944 | } |
618e21d5 | 4945 | out_unlock: |
d3c2fdcf | 4946 | nr = trans->blocks_used; |
7ad85bb7 | 4947 | btrfs_end_transaction(trans, root); |
a22285a6 | 4948 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4949 | if (drop_inode) { |
4950 | inode_dec_link_count(inode); | |
4951 | iput(inode); | |
4952 | } | |
618e21d5 JB |
4953 | return err; |
4954 | } | |
4955 | ||
39279cc3 | 4956 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 4957 | umode_t mode, bool excl) |
39279cc3 CM |
4958 | { |
4959 | struct btrfs_trans_handle *trans; | |
4960 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4961 | struct inode *inode = NULL; |
39279cc3 | 4962 | int drop_inode = 0; |
a22285a6 | 4963 | int err; |
1832a6d5 | 4964 | unsigned long nr = 0; |
39279cc3 | 4965 | u64 objectid; |
00e4e6b3 | 4966 | u64 index = 0; |
39279cc3 | 4967 | |
9ed74f2d JB |
4968 | /* |
4969 | * 2 for inode item and ref | |
4970 | * 2 for dir items | |
4971 | * 1 for xattr if selinux is on | |
4972 | */ | |
a22285a6 YZ |
4973 | trans = btrfs_start_transaction(root, 5); |
4974 | if (IS_ERR(trans)) | |
4975 | return PTR_ERR(trans); | |
9ed74f2d | 4976 | |
581bb050 LZ |
4977 | err = btrfs_find_free_ino(root, &objectid); |
4978 | if (err) | |
4979 | goto out_unlock; | |
4980 | ||
aec7477b | 4981 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4982 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4983 | mode, &index); |
7cf96da3 TI |
4984 | if (IS_ERR(inode)) { |
4985 | err = PTR_ERR(inode); | |
39279cc3 | 4986 | goto out_unlock; |
7cf96da3 | 4987 | } |
39279cc3 | 4988 | |
2a7dba39 | 4989 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4990 | if (err) { |
4991 | drop_inode = 1; | |
4992 | goto out_unlock; | |
4993 | } | |
4994 | ||
ad19db71 CS |
4995 | /* |
4996 | * If the active LSM wants to access the inode during | |
4997 | * d_instantiate it needs these. Smack checks to see | |
4998 | * if the filesystem supports xattrs by looking at the | |
4999 | * ops vector. | |
5000 | */ | |
5001 | inode->i_fop = &btrfs_file_operations; | |
5002 | inode->i_op = &btrfs_file_inode_operations; | |
5003 | ||
a1b075d2 | 5004 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
5005 | if (err) |
5006 | drop_inode = 1; | |
5007 | else { | |
5008 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5009 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 5010 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
08c422c2 | 5011 | d_instantiate(dentry, inode); |
39279cc3 | 5012 | } |
39279cc3 | 5013 | out_unlock: |
d3c2fdcf | 5014 | nr = trans->blocks_used; |
7ad85bb7 | 5015 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5016 | if (drop_inode) { |
5017 | inode_dec_link_count(inode); | |
5018 | iput(inode); | |
5019 | } | |
d3c2fdcf | 5020 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5021 | return err; |
5022 | } | |
5023 | ||
5024 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5025 | struct dentry *dentry) | |
5026 | { | |
5027 | struct btrfs_trans_handle *trans; | |
5028 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5029 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5030 | u64 index; |
1832a6d5 | 5031 | unsigned long nr = 0; |
39279cc3 CM |
5032 | int err; |
5033 | int drop_inode = 0; | |
5034 | ||
4a8be425 TH |
5035 | /* do not allow sys_link's with other subvols of the same device */ |
5036 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5037 | return -EXDEV; |
4a8be425 | 5038 | |
c055e99e AV |
5039 | if (inode->i_nlink == ~0U) |
5040 | return -EMLINK; | |
4a8be425 | 5041 | |
3de4586c | 5042 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5043 | if (err) |
5044 | goto fail; | |
5045 | ||
a22285a6 | 5046 | /* |
7e6b6465 | 5047 | * 2 items for inode and inode ref |
a22285a6 | 5048 | * 2 items for dir items |
7e6b6465 | 5049 | * 1 item for parent inode |
a22285a6 | 5050 | */ |
7e6b6465 | 5051 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5052 | if (IS_ERR(trans)) { |
5053 | err = PTR_ERR(trans); | |
5054 | goto fail; | |
5055 | } | |
5f39d397 | 5056 | |
3153495d | 5057 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5058 | inode_inc_iversion(inode); |
3153495d | 5059 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5060 | ihold(inode); |
aec7477b | 5061 | |
a1b075d2 | 5062 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5063 | |
a5719521 | 5064 | if (err) { |
54aa1f4d | 5065 | drop_inode = 1; |
a5719521 | 5066 | } else { |
10d9f309 | 5067 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5068 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5069 | if (err) |
5070 | goto fail; | |
08c422c2 | 5071 | d_instantiate(dentry, inode); |
6a912213 | 5072 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5073 | } |
39279cc3 | 5074 | |
d3c2fdcf | 5075 | nr = trans->blocks_used; |
7ad85bb7 | 5076 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5077 | fail: |
39279cc3 CM |
5078 | if (drop_inode) { |
5079 | inode_dec_link_count(inode); | |
5080 | iput(inode); | |
5081 | } | |
d3c2fdcf | 5082 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5083 | return err; |
5084 | } | |
5085 | ||
18bb1db3 | 5086 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5087 | { |
b9d86667 | 5088 | struct inode *inode = NULL; |
39279cc3 CM |
5089 | struct btrfs_trans_handle *trans; |
5090 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5091 | int err = 0; | |
5092 | int drop_on_err = 0; | |
b9d86667 | 5093 | u64 objectid = 0; |
00e4e6b3 | 5094 | u64 index = 0; |
d3c2fdcf | 5095 | unsigned long nr = 1; |
39279cc3 | 5096 | |
9ed74f2d JB |
5097 | /* |
5098 | * 2 items for inode and ref | |
5099 | * 2 items for dir items | |
5100 | * 1 for xattr if selinux is on | |
5101 | */ | |
a22285a6 YZ |
5102 | trans = btrfs_start_transaction(root, 5); |
5103 | if (IS_ERR(trans)) | |
5104 | return PTR_ERR(trans); | |
39279cc3 | 5105 | |
581bb050 LZ |
5106 | err = btrfs_find_free_ino(root, &objectid); |
5107 | if (err) | |
5108 | goto out_fail; | |
5109 | ||
aec7477b | 5110 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5111 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5112 | S_IFDIR | mode, &index); |
39279cc3 CM |
5113 | if (IS_ERR(inode)) { |
5114 | err = PTR_ERR(inode); | |
5115 | goto out_fail; | |
5116 | } | |
5f39d397 | 5117 | |
39279cc3 | 5118 | drop_on_err = 1; |
33268eaf | 5119 | |
2a7dba39 | 5120 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5121 | if (err) |
5122 | goto out_fail; | |
5123 | ||
39279cc3 CM |
5124 | inode->i_op = &btrfs_dir_inode_operations; |
5125 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5126 | |
dbe674a9 | 5127 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5128 | err = btrfs_update_inode(trans, root, inode); |
5129 | if (err) | |
5130 | goto out_fail; | |
5f39d397 | 5131 | |
a1b075d2 JB |
5132 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5133 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5134 | if (err) |
5135 | goto out_fail; | |
5f39d397 | 5136 | |
39279cc3 CM |
5137 | d_instantiate(dentry, inode); |
5138 | drop_on_err = 0; | |
39279cc3 CM |
5139 | |
5140 | out_fail: | |
d3c2fdcf | 5141 | nr = trans->blocks_used; |
7ad85bb7 | 5142 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5143 | if (drop_on_err) |
5144 | iput(inode); | |
d3c2fdcf | 5145 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5146 | return err; |
5147 | } | |
5148 | ||
d352ac68 CM |
5149 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5150 | * and an extent that you want to insert, deal with overlap and insert | |
5151 | * the new extent into the tree. | |
5152 | */ | |
3b951516 CM |
5153 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5154 | struct extent_map *existing, | |
e6dcd2dc CM |
5155 | struct extent_map *em, |
5156 | u64 map_start, u64 map_len) | |
3b951516 CM |
5157 | { |
5158 | u64 start_diff; | |
3b951516 | 5159 | |
e6dcd2dc CM |
5160 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5161 | start_diff = map_start - em->start; | |
5162 | em->start = map_start; | |
5163 | em->len = map_len; | |
c8b97818 CM |
5164 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5165 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5166 | em->block_start += start_diff; |
c8b97818 CM |
5167 | em->block_len -= start_diff; |
5168 | } | |
e6dcd2dc | 5169 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5170 | } |
5171 | ||
c8b97818 CM |
5172 | static noinline int uncompress_inline(struct btrfs_path *path, |
5173 | struct inode *inode, struct page *page, | |
5174 | size_t pg_offset, u64 extent_offset, | |
5175 | struct btrfs_file_extent_item *item) | |
5176 | { | |
5177 | int ret; | |
5178 | struct extent_buffer *leaf = path->nodes[0]; | |
5179 | char *tmp; | |
5180 | size_t max_size; | |
5181 | unsigned long inline_size; | |
5182 | unsigned long ptr; | |
261507a0 | 5183 | int compress_type; |
c8b97818 CM |
5184 | |
5185 | WARN_ON(pg_offset != 0); | |
261507a0 | 5186 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5187 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5188 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5189 | btrfs_item_nr(leaf, path->slots[0])); | |
5190 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5191 | if (!tmp) |
5192 | return -ENOMEM; | |
c8b97818 CM |
5193 | ptr = btrfs_file_extent_inline_start(item); |
5194 | ||
5195 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5196 | ||
5b050f04 | 5197 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5198 | ret = btrfs_decompress(compress_type, tmp, page, |
5199 | extent_offset, inline_size, max_size); | |
c8b97818 | 5200 | if (ret) { |
7ac687d9 | 5201 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5202 | unsigned long copy_size = min_t(u64, |
5203 | PAGE_CACHE_SIZE - pg_offset, | |
5204 | max_size - extent_offset); | |
5205 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5206 | kunmap_atomic(kaddr); |
c8b97818 CM |
5207 | } |
5208 | kfree(tmp); | |
5209 | return 0; | |
5210 | } | |
5211 | ||
d352ac68 CM |
5212 | /* |
5213 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5214 | * the ugly parts come from merging extents from the disk with the in-ram |
5215 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5216 | * where the in-ram extents might be locked pending data=ordered completion. |
5217 | * | |
5218 | * This also copies inline extents directly into the page. | |
5219 | */ | |
d397712b | 5220 | |
a52d9a80 | 5221 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5222 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5223 | int create) |
5224 | { | |
5225 | int ret; | |
5226 | int err = 0; | |
db94535d | 5227 | u64 bytenr; |
a52d9a80 CM |
5228 | u64 extent_start = 0; |
5229 | u64 extent_end = 0; | |
33345d01 | 5230 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5231 | u32 found_type; |
f421950f | 5232 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5233 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5234 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5235 | struct extent_buffer *leaf; |
5236 | struct btrfs_key found_key; | |
a52d9a80 CM |
5237 | struct extent_map *em = NULL; |
5238 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5239 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5240 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5241 | int compress_type; |
a52d9a80 | 5242 | |
a52d9a80 | 5243 | again: |
890871be | 5244 | read_lock(&em_tree->lock); |
d1310b2e | 5245 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5246 | if (em) |
5247 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5248 | read_unlock(&em_tree->lock); |
d1310b2e | 5249 | |
a52d9a80 | 5250 | if (em) { |
e1c4b745 CM |
5251 | if (em->start > start || em->start + em->len <= start) |
5252 | free_extent_map(em); | |
5253 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5254 | free_extent_map(em); |
5255 | else | |
5256 | goto out; | |
a52d9a80 | 5257 | } |
172ddd60 | 5258 | em = alloc_extent_map(); |
a52d9a80 | 5259 | if (!em) { |
d1310b2e CM |
5260 | err = -ENOMEM; |
5261 | goto out; | |
a52d9a80 | 5262 | } |
e6dcd2dc | 5263 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5264 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5265 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5266 | em->len = (u64)-1; |
c8b97818 | 5267 | em->block_len = (u64)-1; |
f421950f CM |
5268 | |
5269 | if (!path) { | |
5270 | path = btrfs_alloc_path(); | |
026fd317 JB |
5271 | if (!path) { |
5272 | err = -ENOMEM; | |
5273 | goto out; | |
5274 | } | |
5275 | /* | |
5276 | * Chances are we'll be called again, so go ahead and do | |
5277 | * readahead | |
5278 | */ | |
5279 | path->reada = 1; | |
f421950f CM |
5280 | } |
5281 | ||
179e29e4 CM |
5282 | ret = btrfs_lookup_file_extent(trans, root, path, |
5283 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5284 | if (ret < 0) { |
5285 | err = ret; | |
5286 | goto out; | |
5287 | } | |
5288 | ||
5289 | if (ret != 0) { | |
5290 | if (path->slots[0] == 0) | |
5291 | goto not_found; | |
5292 | path->slots[0]--; | |
5293 | } | |
5294 | ||
5f39d397 CM |
5295 | leaf = path->nodes[0]; |
5296 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5297 | struct btrfs_file_extent_item); |
a52d9a80 | 5298 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5299 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5300 | found_type = btrfs_key_type(&found_key); | |
5301 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5302 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5303 | goto not_found; | |
5304 | } | |
5305 | ||
5f39d397 CM |
5306 | found_type = btrfs_file_extent_type(leaf, item); |
5307 | extent_start = found_key.offset; | |
261507a0 | 5308 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5309 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5310 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5311 | extent_end = extent_start + |
db94535d | 5312 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5313 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5314 | size_t size; | |
5315 | size = btrfs_file_extent_inline_len(leaf, item); | |
5316 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5317 | ~((u64)root->sectorsize - 1); | |
5318 | } | |
5319 | ||
5320 | if (start >= extent_end) { | |
5321 | path->slots[0]++; | |
5322 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5323 | ret = btrfs_next_leaf(root, path); | |
5324 | if (ret < 0) { | |
5325 | err = ret; | |
5326 | goto out; | |
a52d9a80 | 5327 | } |
9036c102 YZ |
5328 | if (ret > 0) |
5329 | goto not_found; | |
5330 | leaf = path->nodes[0]; | |
a52d9a80 | 5331 | } |
9036c102 YZ |
5332 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5333 | if (found_key.objectid != objectid || | |
5334 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5335 | goto not_found; | |
5336 | if (start + len <= found_key.offset) | |
5337 | goto not_found; | |
5338 | em->start = start; | |
5339 | em->len = found_key.offset - start; | |
5340 | goto not_found_em; | |
5341 | } | |
5342 | ||
d899e052 YZ |
5343 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5344 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5345 | em->start = extent_start; |
5346 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5347 | em->orig_start = extent_start - |
5348 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5349 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5350 | if (bytenr == 0) { | |
5f39d397 | 5351 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5352 | goto insert; |
5353 | } | |
261507a0 | 5354 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5355 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5356 | em->compress_type = compress_type; |
c8b97818 CM |
5357 | em->block_start = bytenr; |
5358 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5359 | item); | |
5360 | } else { | |
5361 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5362 | em->block_start = bytenr; | |
5363 | em->block_len = em->len; | |
d899e052 YZ |
5364 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5365 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5366 | } |
a52d9a80 CM |
5367 | goto insert; |
5368 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5369 | unsigned long ptr; |
a52d9a80 | 5370 | char *map; |
3326d1b0 CM |
5371 | size_t size; |
5372 | size_t extent_offset; | |
5373 | size_t copy_size; | |
a52d9a80 | 5374 | |
689f9346 | 5375 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5376 | if (!page || create) { |
689f9346 | 5377 | em->start = extent_start; |
9036c102 | 5378 | em->len = extent_end - extent_start; |
689f9346 Y |
5379 | goto out; |
5380 | } | |
5f39d397 | 5381 | |
9036c102 YZ |
5382 | size = btrfs_file_extent_inline_len(leaf, item); |
5383 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5384 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5385 | size - extent_offset); |
3326d1b0 | 5386 | em->start = extent_start + extent_offset; |
70dec807 CM |
5387 | em->len = (copy_size + root->sectorsize - 1) & |
5388 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5389 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5390 | if (compress_type) { |
c8b97818 | 5391 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5392 | em->compress_type = compress_type; |
5393 | } | |
689f9346 | 5394 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5395 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5396 | if (btrfs_file_extent_compression(leaf, item) != |
5397 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5398 | ret = uncompress_inline(path, inode, page, |
5399 | pg_offset, | |
5400 | extent_offset, item); | |
79787eaa | 5401 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5402 | } else { |
5403 | map = kmap(page); | |
5404 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5405 | copy_size); | |
93c82d57 CM |
5406 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5407 | memset(map + pg_offset + copy_size, 0, | |
5408 | PAGE_CACHE_SIZE - pg_offset - | |
5409 | copy_size); | |
5410 | } | |
c8b97818 CM |
5411 | kunmap(page); |
5412 | } | |
179e29e4 CM |
5413 | flush_dcache_page(page); |
5414 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5415 | BUG(); |
179e29e4 CM |
5416 | if (!trans) { |
5417 | kunmap(page); | |
5418 | free_extent_map(em); | |
5419 | em = NULL; | |
ff5714cc | 5420 | |
b3b4aa74 | 5421 | btrfs_release_path(path); |
7a7eaa40 | 5422 | trans = btrfs_join_transaction(root); |
ff5714cc | 5423 | |
3612b495 TI |
5424 | if (IS_ERR(trans)) |
5425 | return ERR_CAST(trans); | |
179e29e4 CM |
5426 | goto again; |
5427 | } | |
c8b97818 | 5428 | map = kmap(page); |
70dec807 | 5429 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5430 | copy_size); |
c8b97818 | 5431 | kunmap(page); |
179e29e4 | 5432 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5433 | } |
d1310b2e | 5434 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5435 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5436 | goto insert; |
5437 | } else { | |
d397712b | 5438 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5439 | WARN_ON(1); |
5440 | } | |
5441 | not_found: | |
5442 | em->start = start; | |
d1310b2e | 5443 | em->len = len; |
a52d9a80 | 5444 | not_found_em: |
5f39d397 | 5445 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5446 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5447 | insert: |
b3b4aa74 | 5448 | btrfs_release_path(path); |
d1310b2e | 5449 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5450 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5451 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5452 | (unsigned long long)em->len, | |
5453 | (unsigned long long)start, | |
5454 | (unsigned long long)len); | |
a52d9a80 CM |
5455 | err = -EIO; |
5456 | goto out; | |
5457 | } | |
d1310b2e CM |
5458 | |
5459 | err = 0; | |
890871be | 5460 | write_lock(&em_tree->lock); |
a52d9a80 | 5461 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5462 | /* it is possible that someone inserted the extent into the tree |
5463 | * while we had the lock dropped. It is also possible that | |
5464 | * an overlapping map exists in the tree | |
5465 | */ | |
a52d9a80 | 5466 | if (ret == -EEXIST) { |
3b951516 | 5467 | struct extent_map *existing; |
e6dcd2dc CM |
5468 | |
5469 | ret = 0; | |
5470 | ||
3b951516 | 5471 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5472 | if (existing && (existing->start > start || |
5473 | existing->start + existing->len <= start)) { | |
5474 | free_extent_map(existing); | |
5475 | existing = NULL; | |
5476 | } | |
3b951516 CM |
5477 | if (!existing) { |
5478 | existing = lookup_extent_mapping(em_tree, em->start, | |
5479 | em->len); | |
5480 | if (existing) { | |
5481 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5482 | em, start, |
5483 | root->sectorsize); | |
3b951516 CM |
5484 | free_extent_map(existing); |
5485 | if (err) { | |
5486 | free_extent_map(em); | |
5487 | em = NULL; | |
5488 | } | |
5489 | } else { | |
5490 | err = -EIO; | |
3b951516 CM |
5491 | free_extent_map(em); |
5492 | em = NULL; | |
5493 | } | |
5494 | } else { | |
5495 | free_extent_map(em); | |
5496 | em = existing; | |
e6dcd2dc | 5497 | err = 0; |
a52d9a80 | 5498 | } |
a52d9a80 | 5499 | } |
890871be | 5500 | write_unlock(&em_tree->lock); |
a52d9a80 | 5501 | out: |
1abe9b8a | 5502 | |
5503 | trace_btrfs_get_extent(root, em); | |
5504 | ||
f421950f CM |
5505 | if (path) |
5506 | btrfs_free_path(path); | |
a52d9a80 CM |
5507 | if (trans) { |
5508 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5509 | if (!err) |
a52d9a80 CM |
5510 | err = ret; |
5511 | } | |
a52d9a80 CM |
5512 | if (err) { |
5513 | free_extent_map(em); | |
a52d9a80 CM |
5514 | return ERR_PTR(err); |
5515 | } | |
79787eaa | 5516 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5517 | return em; |
5518 | } | |
5519 | ||
ec29ed5b CM |
5520 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5521 | size_t pg_offset, u64 start, u64 len, | |
5522 | int create) | |
5523 | { | |
5524 | struct extent_map *em; | |
5525 | struct extent_map *hole_em = NULL; | |
5526 | u64 range_start = start; | |
5527 | u64 end; | |
5528 | u64 found; | |
5529 | u64 found_end; | |
5530 | int err = 0; | |
5531 | ||
5532 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5533 | if (IS_ERR(em)) | |
5534 | return em; | |
5535 | if (em) { | |
5536 | /* | |
5537 | * if our em maps to a hole, there might | |
5538 | * actually be delalloc bytes behind it | |
5539 | */ | |
5540 | if (em->block_start != EXTENT_MAP_HOLE) | |
5541 | return em; | |
5542 | else | |
5543 | hole_em = em; | |
5544 | } | |
5545 | ||
5546 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5547 | end = start + len; | |
5548 | if (end < start) | |
5549 | end = (u64)-1; | |
5550 | else | |
5551 | end -= 1; | |
5552 | ||
5553 | em = NULL; | |
5554 | ||
5555 | /* ok, we didn't find anything, lets look for delalloc */ | |
5556 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5557 | end, len, EXTENT_DELALLOC, 1); | |
5558 | found_end = range_start + found; | |
5559 | if (found_end < range_start) | |
5560 | found_end = (u64)-1; | |
5561 | ||
5562 | /* | |
5563 | * we didn't find anything useful, return | |
5564 | * the original results from get_extent() | |
5565 | */ | |
5566 | if (range_start > end || found_end <= start) { | |
5567 | em = hole_em; | |
5568 | hole_em = NULL; | |
5569 | goto out; | |
5570 | } | |
5571 | ||
5572 | /* adjust the range_start to make sure it doesn't | |
5573 | * go backwards from the start they passed in | |
5574 | */ | |
5575 | range_start = max(start,range_start); | |
5576 | found = found_end - range_start; | |
5577 | ||
5578 | if (found > 0) { | |
5579 | u64 hole_start = start; | |
5580 | u64 hole_len = len; | |
5581 | ||
172ddd60 | 5582 | em = alloc_extent_map(); |
ec29ed5b CM |
5583 | if (!em) { |
5584 | err = -ENOMEM; | |
5585 | goto out; | |
5586 | } | |
5587 | /* | |
5588 | * when btrfs_get_extent can't find anything it | |
5589 | * returns one huge hole | |
5590 | * | |
5591 | * make sure what it found really fits our range, and | |
5592 | * adjust to make sure it is based on the start from | |
5593 | * the caller | |
5594 | */ | |
5595 | if (hole_em) { | |
5596 | u64 calc_end = extent_map_end(hole_em); | |
5597 | ||
5598 | if (calc_end <= start || (hole_em->start > end)) { | |
5599 | free_extent_map(hole_em); | |
5600 | hole_em = NULL; | |
5601 | } else { | |
5602 | hole_start = max(hole_em->start, start); | |
5603 | hole_len = calc_end - hole_start; | |
5604 | } | |
5605 | } | |
5606 | em->bdev = NULL; | |
5607 | if (hole_em && range_start > hole_start) { | |
5608 | /* our hole starts before our delalloc, so we | |
5609 | * have to return just the parts of the hole | |
5610 | * that go until the delalloc starts | |
5611 | */ | |
5612 | em->len = min(hole_len, | |
5613 | range_start - hole_start); | |
5614 | em->start = hole_start; | |
5615 | em->orig_start = hole_start; | |
5616 | /* | |
5617 | * don't adjust block start at all, | |
5618 | * it is fixed at EXTENT_MAP_HOLE | |
5619 | */ | |
5620 | em->block_start = hole_em->block_start; | |
5621 | em->block_len = hole_len; | |
5622 | } else { | |
5623 | em->start = range_start; | |
5624 | em->len = found; | |
5625 | em->orig_start = range_start; | |
5626 | em->block_start = EXTENT_MAP_DELALLOC; | |
5627 | em->block_len = found; | |
5628 | } | |
5629 | } else if (hole_em) { | |
5630 | return hole_em; | |
5631 | } | |
5632 | out: | |
5633 | ||
5634 | free_extent_map(hole_em); | |
5635 | if (err) { | |
5636 | free_extent_map(em); | |
5637 | return ERR_PTR(err); | |
5638 | } | |
5639 | return em; | |
5640 | } | |
5641 | ||
4b46fce2 | 5642 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5643 | struct extent_map *em, |
4b46fce2 JB |
5644 | u64 start, u64 len) |
5645 | { | |
5646 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5647 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5648 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5649 | struct btrfs_key ins; | |
5650 | u64 alloc_hint; | |
5651 | int ret; | |
16d299ac | 5652 | bool insert = false; |
4b46fce2 | 5653 | |
16d299ac JB |
5654 | /* |
5655 | * Ok if the extent map we looked up is a hole and is for the exact | |
5656 | * range we want, there is no reason to allocate a new one, however if | |
5657 | * it is not right then we need to free this one and drop the cache for | |
5658 | * our range. | |
5659 | */ | |
5660 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5661 | em->len != len) { | |
5662 | free_extent_map(em); | |
5663 | em = NULL; | |
5664 | insert = true; | |
5665 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5666 | } | |
4b46fce2 | 5667 | |
7a7eaa40 | 5668 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5669 | if (IS_ERR(trans)) |
5670 | return ERR_CAST(trans); | |
4b46fce2 | 5671 | |
4cb5300b CM |
5672 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5673 | btrfs_add_inode_defrag(trans, inode); | |
5674 | ||
4b46fce2 JB |
5675 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5676 | ||
5677 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5678 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5679 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5680 | if (ret) { |
5681 | em = ERR_PTR(ret); | |
5682 | goto out; | |
5683 | } | |
5684 | ||
4b46fce2 | 5685 | if (!em) { |
172ddd60 | 5686 | em = alloc_extent_map(); |
16d299ac JB |
5687 | if (!em) { |
5688 | em = ERR_PTR(-ENOMEM); | |
5689 | goto out; | |
5690 | } | |
4b46fce2 JB |
5691 | } |
5692 | ||
5693 | em->start = start; | |
5694 | em->orig_start = em->start; | |
5695 | em->len = ins.offset; | |
5696 | ||
5697 | em->block_start = ins.objectid; | |
5698 | em->block_len = ins.offset; | |
5699 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5700 | |
5701 | /* | |
5702 | * We need to do this because if we're using the original em we searched | |
5703 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5704 | */ | |
5705 | em->flags = 0; | |
4b46fce2 JB |
5706 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5707 | ||
16d299ac | 5708 | while (insert) { |
4b46fce2 JB |
5709 | write_lock(&em_tree->lock); |
5710 | ret = add_extent_mapping(em_tree, em); | |
5711 | write_unlock(&em_tree->lock); | |
5712 | if (ret != -EEXIST) | |
5713 | break; | |
5714 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5715 | } | |
5716 | ||
5717 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5718 | ins.offset, ins.offset, 0); | |
5719 | if (ret) { | |
5720 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5721 | em = ERR_PTR(ret); | |
5722 | } | |
5723 | out: | |
5724 | btrfs_end_transaction(trans, root); | |
5725 | return em; | |
5726 | } | |
5727 | ||
46bfbb5c CM |
5728 | /* |
5729 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5730 | * block must be cow'd | |
5731 | */ | |
5732 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5733 | struct inode *inode, u64 offset, u64 len) | |
5734 | { | |
5735 | struct btrfs_path *path; | |
5736 | int ret; | |
5737 | struct extent_buffer *leaf; | |
5738 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5739 | struct btrfs_file_extent_item *fi; | |
5740 | struct btrfs_key key; | |
5741 | u64 disk_bytenr; | |
5742 | u64 backref_offset; | |
5743 | u64 extent_end; | |
5744 | u64 num_bytes; | |
5745 | int slot; | |
5746 | int found_type; | |
5747 | ||
5748 | path = btrfs_alloc_path(); | |
5749 | if (!path) | |
5750 | return -ENOMEM; | |
5751 | ||
33345d01 | 5752 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5753 | offset, 0); |
5754 | if (ret < 0) | |
5755 | goto out; | |
5756 | ||
5757 | slot = path->slots[0]; | |
5758 | if (ret == 1) { | |
5759 | if (slot == 0) { | |
5760 | /* can't find the item, must cow */ | |
5761 | ret = 0; | |
5762 | goto out; | |
5763 | } | |
5764 | slot--; | |
5765 | } | |
5766 | ret = 0; | |
5767 | leaf = path->nodes[0]; | |
5768 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5769 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5770 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5771 | /* not our file or wrong item type, must cow */ | |
5772 | goto out; | |
5773 | } | |
5774 | ||
5775 | if (key.offset > offset) { | |
5776 | /* Wrong offset, must cow */ | |
5777 | goto out; | |
5778 | } | |
5779 | ||
5780 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5781 | found_type = btrfs_file_extent_type(leaf, fi); | |
5782 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5783 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5784 | /* not a regular extent, must cow */ | |
5785 | goto out; | |
5786 | } | |
5787 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5788 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5789 | ||
5790 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5791 | if (extent_end < offset + len) { | |
5792 | /* extent doesn't include our full range, must cow */ | |
5793 | goto out; | |
5794 | } | |
5795 | ||
5796 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5797 | goto out; | |
5798 | ||
5799 | /* | |
5800 | * look for other files referencing this extent, if we | |
5801 | * find any we must cow | |
5802 | */ | |
33345d01 | 5803 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5804 | key.offset - backref_offset, disk_bytenr)) |
5805 | goto out; | |
5806 | ||
5807 | /* | |
5808 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5809 | * in this extent we are about to write. If there | |
5810 | * are any csums in that range we have to cow in order | |
5811 | * to keep the csums correct | |
5812 | */ | |
5813 | disk_bytenr += backref_offset; | |
5814 | disk_bytenr += offset - key.offset; | |
5815 | num_bytes = min(offset + len, extent_end) - offset; | |
5816 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5817 | goto out; | |
5818 | /* | |
5819 | * all of the above have passed, it is safe to overwrite this extent | |
5820 | * without cow | |
5821 | */ | |
5822 | ret = 1; | |
5823 | out: | |
5824 | btrfs_free_path(path); | |
5825 | return ret; | |
5826 | } | |
5827 | ||
eb838e73 JB |
5828 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
5829 | struct extent_state **cached_state, int writing) | |
5830 | { | |
5831 | struct btrfs_ordered_extent *ordered; | |
5832 | int ret = 0; | |
5833 | ||
5834 | while (1) { | |
5835 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5836 | 0, cached_state); | |
5837 | /* | |
5838 | * We're concerned with the entire range that we're going to be | |
5839 | * doing DIO to, so we need to make sure theres no ordered | |
5840 | * extents in this range. | |
5841 | */ | |
5842 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
5843 | lockend - lockstart + 1); | |
5844 | ||
5845 | /* | |
5846 | * We need to make sure there are no buffered pages in this | |
5847 | * range either, we could have raced between the invalidate in | |
5848 | * generic_file_direct_write and locking the extent. The | |
5849 | * invalidate needs to happen so that reads after a write do not | |
5850 | * get stale data. | |
5851 | */ | |
5852 | if (!ordered && (!writing || | |
5853 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
5854 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
5855 | *cached_state))) | |
5856 | break; | |
5857 | ||
5858 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5859 | cached_state, GFP_NOFS); | |
5860 | ||
5861 | if (ordered) { | |
5862 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5863 | btrfs_put_ordered_extent(ordered); | |
5864 | } else { | |
5865 | /* Screw you mmap */ | |
5866 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
5867 | lockstart, | |
5868 | lockend); | |
5869 | if (ret) | |
5870 | break; | |
5871 | ||
5872 | /* | |
5873 | * If we found a page that couldn't be invalidated just | |
5874 | * fall back to buffered. | |
5875 | */ | |
5876 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5877 | lockstart >> PAGE_CACHE_SHIFT, | |
5878 | lockend >> PAGE_CACHE_SHIFT); | |
5879 | if (ret) | |
5880 | break; | |
5881 | } | |
5882 | ||
5883 | cond_resched(); | |
5884 | } | |
5885 | ||
5886 | return ret; | |
5887 | } | |
5888 | ||
4b46fce2 JB |
5889 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5890 | struct buffer_head *bh_result, int create) | |
5891 | { | |
5892 | struct extent_map *em; | |
5893 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 5894 | struct extent_state *cached_state = NULL; |
4b46fce2 | 5895 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 5896 | u64 lockstart, lockend; |
4b46fce2 | 5897 | u64 len = bh_result->b_size; |
46bfbb5c | 5898 | struct btrfs_trans_handle *trans; |
eb838e73 JB |
5899 | int unlock_bits = EXTENT_LOCKED; |
5900 | int ret; | |
5901 | ||
eb838e73 JB |
5902 | if (create) { |
5903 | ret = btrfs_delalloc_reserve_space(inode, len); | |
5904 | if (ret) | |
5905 | return ret; | |
5906 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; | |
c329861d JB |
5907 | } else { |
5908 | len = min_t(u64, len, root->sectorsize); | |
eb838e73 JB |
5909 | } |
5910 | ||
c329861d JB |
5911 | lockstart = start; |
5912 | lockend = start + len - 1; | |
5913 | ||
eb838e73 JB |
5914 | /* |
5915 | * If this errors out it's because we couldn't invalidate pagecache for | |
5916 | * this range and we need to fallback to buffered. | |
5917 | */ | |
5918 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
5919 | return -ENOTBLK; | |
5920 | ||
5921 | if (create) { | |
5922 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
5923 | lockend, EXTENT_DELALLOC, NULL, | |
5924 | &cached_state, GFP_NOFS); | |
5925 | if (ret) | |
5926 | goto unlock_err; | |
5927 | } | |
4b46fce2 JB |
5928 | |
5929 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
eb838e73 JB |
5930 | if (IS_ERR(em)) { |
5931 | ret = PTR_ERR(em); | |
5932 | goto unlock_err; | |
5933 | } | |
4b46fce2 JB |
5934 | |
5935 | /* | |
5936 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5937 | * io. INLINE is special, and we could probably kludge it in here, but | |
5938 | * it's still buffered so for safety lets just fall back to the generic | |
5939 | * buffered path. | |
5940 | * | |
5941 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5942 | * decompress it, so there will be buffering required no matter what we | |
5943 | * do, so go ahead and fallback to buffered. | |
5944 | * | |
5945 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5946 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5947 | * the generic code. | |
5948 | */ | |
5949 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5950 | em->block_start == EXTENT_MAP_INLINE) { | |
5951 | free_extent_map(em); | |
eb838e73 JB |
5952 | ret = -ENOTBLK; |
5953 | goto unlock_err; | |
4b46fce2 JB |
5954 | } |
5955 | ||
5956 | /* Just a good old fashioned hole, return */ | |
5957 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5958 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5959 | free_extent_map(em); | |
eb838e73 JB |
5960 | ret = 0; |
5961 | goto unlock_err; | |
4b46fce2 JB |
5962 | } |
5963 | ||
5964 | /* | |
5965 | * We don't allocate a new extent in the following cases | |
5966 | * | |
5967 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5968 | * existing extent. | |
5969 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5970 | * just use the extent. | |
5971 | * | |
5972 | */ | |
46bfbb5c | 5973 | if (!create) { |
eb838e73 JB |
5974 | len = min(len, em->len - (start - em->start)); |
5975 | lockstart = start + len; | |
5976 | goto unlock; | |
46bfbb5c | 5977 | } |
4b46fce2 JB |
5978 | |
5979 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5980 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5981 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5982 | int type; |
5983 | int ret; | |
46bfbb5c | 5984 | u64 block_start; |
4b46fce2 JB |
5985 | |
5986 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5987 | type = BTRFS_ORDERED_PREALLOC; | |
5988 | else | |
5989 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5990 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5991 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5992 | |
5993 | /* | |
5994 | * we're not going to log anything, but we do need | |
5995 | * to make sure the current transaction stays open | |
5996 | * while we look for nocow cross refs | |
5997 | */ | |
7a7eaa40 | 5998 | trans = btrfs_join_transaction(root); |
3612b495 | 5999 | if (IS_ERR(trans)) |
46bfbb5c CM |
6000 | goto must_cow; |
6001 | ||
6002 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
6003 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
6004 | block_start, len, len, type); | |
6005 | btrfs_end_transaction(trans, root); | |
6006 | if (ret) { | |
6007 | free_extent_map(em); | |
eb838e73 | 6008 | goto unlock_err; |
46bfbb5c CM |
6009 | } |
6010 | goto unlock; | |
4b46fce2 | 6011 | } |
46bfbb5c | 6012 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6013 | } |
46bfbb5c CM |
6014 | must_cow: |
6015 | /* | |
6016 | * this will cow the extent, reset the len in case we changed | |
6017 | * it above | |
6018 | */ | |
6019 | len = bh_result->b_size; | |
16d299ac | 6020 | em = btrfs_new_extent_direct(inode, em, start, len); |
eb838e73 JB |
6021 | if (IS_ERR(em)) { |
6022 | ret = PTR_ERR(em); | |
6023 | goto unlock_err; | |
6024 | } | |
46bfbb5c CM |
6025 | len = min(len, em->len - (start - em->start)); |
6026 | unlock: | |
4b46fce2 JB |
6027 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6028 | inode->i_blkbits; | |
46bfbb5c | 6029 | bh_result->b_size = len; |
4b46fce2 JB |
6030 | bh_result->b_bdev = em->bdev; |
6031 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6032 | if (create) { |
6033 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6034 | set_buffer_new(bh_result); | |
6035 | ||
6036 | /* | |
6037 | * Need to update the i_size under the extent lock so buffered | |
6038 | * readers will get the updated i_size when we unlock. | |
6039 | */ | |
6040 | if (start + len > i_size_read(inode)) | |
6041 | i_size_write(inode, start + len); | |
6042 | } | |
4b46fce2 | 6043 | |
eb838e73 JB |
6044 | /* |
6045 | * In the case of write we need to clear and unlock the entire range, | |
6046 | * in the case of read we need to unlock only the end area that we | |
6047 | * aren't using if there is any left over space. | |
6048 | */ | |
24c03fa5 LB |
6049 | if (lockstart < lockend) { |
6050 | if (create && len < lockend - lockstart) { | |
6051 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6052 | lockstart + len - 1, unlock_bits, 1, 0, | |
6053 | &cached_state, GFP_NOFS); | |
6054 | /* | |
6055 | * Beside unlock, we also need to cleanup reserved space | |
6056 | * for the left range by attaching EXTENT_DO_ACCOUNTING. | |
6057 | */ | |
6058 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
6059 | lockstart + len, lockend, | |
6060 | unlock_bits | EXTENT_DO_ACCOUNTING, | |
6061 | 1, 0, NULL, GFP_NOFS); | |
6062 | } else { | |
6063 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6064 | lockend, unlock_bits, 1, 0, | |
6065 | &cached_state, GFP_NOFS); | |
6066 | } | |
6067 | } else { | |
eb838e73 | 6068 | free_extent_state(cached_state); |
24c03fa5 | 6069 | } |
eb838e73 | 6070 | |
4b46fce2 JB |
6071 | free_extent_map(em); |
6072 | ||
6073 | return 0; | |
eb838e73 JB |
6074 | |
6075 | unlock_err: | |
6076 | if (create) | |
6077 | unlock_bits |= EXTENT_DO_ACCOUNTING; | |
6078 | ||
6079 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6080 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6081 | return ret; | |
4b46fce2 JB |
6082 | } |
6083 | ||
6084 | struct btrfs_dio_private { | |
6085 | struct inode *inode; | |
6086 | u64 logical_offset; | |
6087 | u64 disk_bytenr; | |
6088 | u64 bytes; | |
4b46fce2 | 6089 | void *private; |
e65e1535 MX |
6090 | |
6091 | /* number of bios pending for this dio */ | |
6092 | atomic_t pending_bios; | |
6093 | ||
6094 | /* IO errors */ | |
6095 | int errors; | |
6096 | ||
6097 | struct bio *orig_bio; | |
4b46fce2 JB |
6098 | }; |
6099 | ||
6100 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6101 | { | |
e65e1535 | 6102 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6103 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6104 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6105 | struct inode *inode = dip->inode; |
6106 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6107 | u64 start; | |
4b46fce2 JB |
6108 | |
6109 | start = dip->logical_offset; | |
6110 | do { | |
6111 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6112 | struct page *page = bvec->bv_page; | |
6113 | char *kaddr; | |
6114 | u32 csum = ~(u32)0; | |
c329861d | 6115 | u64 private = ~(u32)0; |
4b46fce2 JB |
6116 | unsigned long flags; |
6117 | ||
c329861d JB |
6118 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6119 | start, &private)) | |
6120 | goto failed; | |
4b46fce2 | 6121 | local_irq_save(flags); |
7ac687d9 | 6122 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
6123 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
6124 | csum, bvec->bv_len); | |
6125 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6126 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6127 | local_irq_restore(flags); |
6128 | ||
6129 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6130 | if (csum != private) { |
6131 | failed: | |
33345d01 | 6132 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 6133 | " %llu csum %u private %u\n", |
33345d01 LZ |
6134 | (unsigned long long)btrfs_ino(inode), |
6135 | (unsigned long long)start, | |
c329861d | 6136 | csum, (unsigned)private); |
4b46fce2 JB |
6137 | err = -EIO; |
6138 | } | |
6139 | } | |
6140 | ||
6141 | start += bvec->bv_len; | |
4b46fce2 JB |
6142 | bvec++; |
6143 | } while (bvec <= bvec_end); | |
6144 | ||
6145 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6146 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
6147 | bio->bi_private = dip->private; |
6148 | ||
4b46fce2 | 6149 | kfree(dip); |
c0da7aa1 JB |
6150 | |
6151 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6152 | if (err) | |
6153 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6154 | dio_end_io(bio, err); |
6155 | } | |
6156 | ||
6157 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6158 | { | |
6159 | struct btrfs_dio_private *dip = bio->bi_private; | |
6160 | struct inode *inode = dip->inode; | |
6161 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6162 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6163 | u64 ordered_offset = dip->logical_offset; |
6164 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
6165 | int ret; |
6166 | ||
6167 | if (err) | |
6168 | goto out_done; | |
163cf09c CM |
6169 | again: |
6170 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6171 | &ordered_offset, | |
5fd02043 | 6172 | ordered_bytes, !err); |
4b46fce2 | 6173 | if (!ret) |
163cf09c | 6174 | goto out_test; |
4b46fce2 | 6175 | |
5fd02043 JB |
6176 | ordered->work.func = finish_ordered_fn; |
6177 | ordered->work.flags = 0; | |
6178 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6179 | &ordered->work); | |
163cf09c CM |
6180 | out_test: |
6181 | /* | |
6182 | * our bio might span multiple ordered extents. If we haven't | |
6183 | * completed the accounting for the whole dio, go back and try again | |
6184 | */ | |
6185 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6186 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6187 | ordered_offset; | |
5fd02043 | 6188 | ordered = NULL; |
163cf09c CM |
6189 | goto again; |
6190 | } | |
4b46fce2 JB |
6191 | out_done: |
6192 | bio->bi_private = dip->private; | |
6193 | ||
4b46fce2 | 6194 | kfree(dip); |
c0da7aa1 JB |
6195 | |
6196 | /* If we had an error make sure to clear the uptodate flag */ | |
6197 | if (err) | |
6198 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6199 | dio_end_io(bio, err); |
6200 | } | |
6201 | ||
eaf25d93 CM |
6202 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6203 | struct bio *bio, int mirror_num, | |
6204 | unsigned long bio_flags, u64 offset) | |
6205 | { | |
6206 | int ret; | |
6207 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6208 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6209 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6210 | return 0; |
6211 | } | |
6212 | ||
e65e1535 MX |
6213 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6214 | { | |
6215 | struct btrfs_dio_private *dip = bio->bi_private; | |
6216 | ||
6217 | if (err) { | |
33345d01 | 6218 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6219 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6220 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6221 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6222 | dip->errors = 1; |
6223 | ||
6224 | /* | |
6225 | * before atomic variable goto zero, we must make sure | |
6226 | * dip->errors is perceived to be set. | |
6227 | */ | |
6228 | smp_mb__before_atomic_dec(); | |
6229 | } | |
6230 | ||
6231 | /* if there are more bios still pending for this dio, just exit */ | |
6232 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6233 | goto out; | |
6234 | ||
6235 | if (dip->errors) | |
6236 | bio_io_error(dip->orig_bio); | |
6237 | else { | |
6238 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6239 | bio_endio(dip->orig_bio, 0); | |
6240 | } | |
6241 | out: | |
6242 | bio_put(bio); | |
6243 | } | |
6244 | ||
6245 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6246 | u64 first_sector, gfp_t gfp_flags) | |
6247 | { | |
6248 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6249 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6250 | } | |
6251 | ||
6252 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6253 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6254 | int async_submit) |
e65e1535 MX |
6255 | { |
6256 | int write = rw & REQ_WRITE; | |
6257 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6258 | int ret; | |
6259 | ||
6260 | bio_get(bio); | |
5fd02043 JB |
6261 | |
6262 | if (!write) { | |
6263 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6264 | if (ret) | |
6265 | goto err; | |
6266 | } | |
e65e1535 | 6267 | |
1ae39938 JB |
6268 | if (skip_sum) |
6269 | goto map; | |
6270 | ||
6271 | if (write && async_submit) { | |
e65e1535 MX |
6272 | ret = btrfs_wq_submit_bio(root->fs_info, |
6273 | inode, rw, bio, 0, 0, | |
6274 | file_offset, | |
6275 | __btrfs_submit_bio_start_direct_io, | |
6276 | __btrfs_submit_bio_done); | |
6277 | goto err; | |
1ae39938 JB |
6278 | } else if (write) { |
6279 | /* | |
6280 | * If we aren't doing async submit, calculate the csum of the | |
6281 | * bio now. | |
6282 | */ | |
6283 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6284 | if (ret) | |
6285 | goto err; | |
c2db1073 | 6286 | } else if (!skip_sum) { |
c329861d | 6287 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6288 | if (ret) |
6289 | goto err; | |
6290 | } | |
e65e1535 | 6291 | |
1ae39938 JB |
6292 | map: |
6293 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6294 | err: |
6295 | bio_put(bio); | |
6296 | return ret; | |
6297 | } | |
6298 | ||
6299 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6300 | int skip_sum) | |
6301 | { | |
6302 | struct inode *inode = dip->inode; | |
6303 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6304 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6305 | struct bio *bio; | |
6306 | struct bio *orig_bio = dip->orig_bio; | |
6307 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6308 | u64 start_sector = orig_bio->bi_sector; | |
6309 | u64 file_offset = dip->logical_offset; | |
6310 | u64 submit_len = 0; | |
6311 | u64 map_length; | |
6312 | int nr_pages = 0; | |
e65e1535 | 6313 | int ret = 0; |
1ae39938 | 6314 | int async_submit = 0; |
e65e1535 | 6315 | |
e65e1535 MX |
6316 | map_length = orig_bio->bi_size; |
6317 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6318 | &map_length, NULL, 0); | |
6319 | if (ret) { | |
64728bbb | 6320 | bio_put(orig_bio); |
e65e1535 MX |
6321 | return -EIO; |
6322 | } | |
6323 | ||
02f57c7a JB |
6324 | if (map_length >= orig_bio->bi_size) { |
6325 | bio = orig_bio; | |
6326 | goto submit; | |
6327 | } | |
6328 | ||
1ae39938 | 6329 | async_submit = 1; |
02f57c7a JB |
6330 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6331 | if (!bio) | |
6332 | return -ENOMEM; | |
6333 | bio->bi_private = dip; | |
6334 | bio->bi_end_io = btrfs_end_dio_bio; | |
6335 | atomic_inc(&dip->pending_bios); | |
6336 | ||
e65e1535 MX |
6337 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6338 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6339 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6340 | bvec->bv_offset) < bvec->bv_len)) { | |
6341 | /* | |
6342 | * inc the count before we submit the bio so | |
6343 | * we know the end IO handler won't happen before | |
6344 | * we inc the count. Otherwise, the dip might get freed | |
6345 | * before we're done setting it up | |
6346 | */ | |
6347 | atomic_inc(&dip->pending_bios); | |
6348 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6349 | file_offset, skip_sum, | |
c329861d | 6350 | async_submit); |
e65e1535 MX |
6351 | if (ret) { |
6352 | bio_put(bio); | |
6353 | atomic_dec(&dip->pending_bios); | |
6354 | goto out_err; | |
6355 | } | |
6356 | ||
e65e1535 MX |
6357 | start_sector += submit_len >> 9; |
6358 | file_offset += submit_len; | |
6359 | ||
6360 | submit_len = 0; | |
6361 | nr_pages = 0; | |
6362 | ||
6363 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6364 | start_sector, GFP_NOFS); | |
6365 | if (!bio) | |
6366 | goto out_err; | |
6367 | bio->bi_private = dip; | |
6368 | bio->bi_end_io = btrfs_end_dio_bio; | |
6369 | ||
6370 | map_length = orig_bio->bi_size; | |
6371 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6372 | &map_length, NULL, 0); | |
6373 | if (ret) { | |
6374 | bio_put(bio); | |
6375 | goto out_err; | |
6376 | } | |
6377 | } else { | |
6378 | submit_len += bvec->bv_len; | |
6379 | nr_pages ++; | |
6380 | bvec++; | |
6381 | } | |
6382 | } | |
6383 | ||
02f57c7a | 6384 | submit: |
e65e1535 | 6385 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 6386 | async_submit); |
e65e1535 MX |
6387 | if (!ret) |
6388 | return 0; | |
6389 | ||
6390 | bio_put(bio); | |
6391 | out_err: | |
6392 | dip->errors = 1; | |
6393 | /* | |
6394 | * before atomic variable goto zero, we must | |
6395 | * make sure dip->errors is perceived to be set. | |
6396 | */ | |
6397 | smp_mb__before_atomic_dec(); | |
6398 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6399 | bio_io_error(dip->orig_bio); | |
6400 | ||
6401 | /* bio_end_io() will handle error, so we needn't return it */ | |
6402 | return 0; | |
6403 | } | |
6404 | ||
4b46fce2 JB |
6405 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6406 | loff_t file_offset) | |
6407 | { | |
6408 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6409 | struct btrfs_dio_private *dip; | |
6410 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6411 | int skip_sum; |
7b6d91da | 6412 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6413 | int ret = 0; |
6414 | ||
6415 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6416 | ||
6417 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6418 | if (!dip) { | |
6419 | ret = -ENOMEM; | |
6420 | goto free_ordered; | |
6421 | } | |
4b46fce2 JB |
6422 | |
6423 | dip->private = bio->bi_private; | |
6424 | dip->inode = inode; | |
6425 | dip->logical_offset = file_offset; | |
6426 | ||
4b46fce2 JB |
6427 | dip->bytes = 0; |
6428 | do { | |
6429 | dip->bytes += bvec->bv_len; | |
6430 | bvec++; | |
6431 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6432 | ||
46bfbb5c | 6433 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6434 | bio->bi_private = dip; |
e65e1535 MX |
6435 | dip->errors = 0; |
6436 | dip->orig_bio = bio; | |
6437 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6438 | |
6439 | if (write) | |
6440 | bio->bi_end_io = btrfs_endio_direct_write; | |
6441 | else | |
6442 | bio->bi_end_io = btrfs_endio_direct_read; | |
6443 | ||
e65e1535 MX |
6444 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6445 | if (!ret) | |
eaf25d93 | 6446 | return; |
4b46fce2 JB |
6447 | free_ordered: |
6448 | /* | |
6449 | * If this is a write, we need to clean up the reserved space and kill | |
6450 | * the ordered extent. | |
6451 | */ | |
6452 | if (write) { | |
6453 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6454 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6455 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6456 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6457 | btrfs_free_reserved_extent(root, ordered->start, | |
6458 | ordered->disk_len); | |
6459 | btrfs_put_ordered_extent(ordered); | |
6460 | btrfs_put_ordered_extent(ordered); | |
6461 | } | |
6462 | bio_endio(bio, ret); | |
6463 | } | |
6464 | ||
5a5f79b5 CM |
6465 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6466 | const struct iovec *iov, loff_t offset, | |
6467 | unsigned long nr_segs) | |
6468 | { | |
6469 | int seg; | |
a1b75f7d | 6470 | int i; |
5a5f79b5 CM |
6471 | size_t size; |
6472 | unsigned long addr; | |
6473 | unsigned blocksize_mask = root->sectorsize - 1; | |
6474 | ssize_t retval = -EINVAL; | |
6475 | loff_t end = offset; | |
6476 | ||
6477 | if (offset & blocksize_mask) | |
6478 | goto out; | |
6479 | ||
6480 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6481 | for (seg = 0; seg < nr_segs; seg++) { | |
6482 | addr = (unsigned long)iov[seg].iov_base; | |
6483 | size = iov[seg].iov_len; | |
6484 | end += size; | |
a1b75f7d | 6485 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6486 | goto out; |
a1b75f7d JB |
6487 | |
6488 | /* If this is a write we don't need to check anymore */ | |
6489 | if (rw & WRITE) | |
6490 | continue; | |
6491 | ||
6492 | /* | |
6493 | * Check to make sure we don't have duplicate iov_base's in this | |
6494 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6495 | * when reading back. | |
6496 | */ | |
6497 | for (i = seg + 1; i < nr_segs; i++) { | |
6498 | if (iov[seg].iov_base == iov[i].iov_base) | |
6499 | goto out; | |
6500 | } | |
5a5f79b5 CM |
6501 | } |
6502 | retval = 0; | |
6503 | out: | |
6504 | return retval; | |
6505 | } | |
eb838e73 | 6506 | |
16432985 CM |
6507 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6508 | const struct iovec *iov, loff_t offset, | |
6509 | unsigned long nr_segs) | |
6510 | { | |
4b46fce2 JB |
6511 | struct file *file = iocb->ki_filp; |
6512 | struct inode *inode = file->f_mapping->host; | |
4b46fce2 | 6513 | |
5a5f79b5 | 6514 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 6515 | offset, nr_segs)) |
5a5f79b5 | 6516 | return 0; |
3f7c579c | 6517 | |
eb838e73 | 6518 | return __blockdev_direct_IO(rw, iocb, inode, |
5a5f79b5 CM |
6519 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, |
6520 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6521 | btrfs_submit_direct, 0); | |
16432985 CM |
6522 | } |
6523 | ||
1506fcc8 YS |
6524 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6525 | __u64 start, __u64 len) | |
6526 | { | |
ec29ed5b | 6527 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6528 | } |
6529 | ||
a52d9a80 | 6530 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6531 | { |
d1310b2e CM |
6532 | struct extent_io_tree *tree; |
6533 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6534 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6535 | } |
1832a6d5 | 6536 | |
a52d9a80 | 6537 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6538 | { |
d1310b2e | 6539 | struct extent_io_tree *tree; |
b888db2b CM |
6540 | |
6541 | ||
6542 | if (current->flags & PF_MEMALLOC) { | |
6543 | redirty_page_for_writepage(wbc, page); | |
6544 | unlock_page(page); | |
6545 | return 0; | |
6546 | } | |
d1310b2e | 6547 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6548 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6549 | } |
6550 | ||
f421950f CM |
6551 | int btrfs_writepages(struct address_space *mapping, |
6552 | struct writeback_control *wbc) | |
b293f02e | 6553 | { |
d1310b2e | 6554 | struct extent_io_tree *tree; |
771ed689 | 6555 | |
d1310b2e | 6556 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6557 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6558 | } | |
6559 | ||
3ab2fb5a CM |
6560 | static int |
6561 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6562 | struct list_head *pages, unsigned nr_pages) | |
6563 | { | |
d1310b2e CM |
6564 | struct extent_io_tree *tree; |
6565 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6566 | return extent_readpages(tree, mapping, pages, nr_pages, |
6567 | btrfs_get_extent); | |
6568 | } | |
e6dcd2dc | 6569 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6570 | { |
d1310b2e CM |
6571 | struct extent_io_tree *tree; |
6572 | struct extent_map_tree *map; | |
a52d9a80 | 6573 | int ret; |
8c2383c3 | 6574 | |
d1310b2e CM |
6575 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6576 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6577 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6578 | if (ret == 1) { |
6579 | ClearPagePrivate(page); | |
6580 | set_page_private(page, 0); | |
6581 | page_cache_release(page); | |
39279cc3 | 6582 | } |
a52d9a80 | 6583 | return ret; |
39279cc3 CM |
6584 | } |
6585 | ||
e6dcd2dc CM |
6586 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6587 | { | |
98509cfc CM |
6588 | if (PageWriteback(page) || PageDirty(page)) |
6589 | return 0; | |
b335b003 | 6590 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6591 | } |
6592 | ||
a52d9a80 | 6593 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6594 | { |
5fd02043 | 6595 | struct inode *inode = page->mapping->host; |
d1310b2e | 6596 | struct extent_io_tree *tree; |
e6dcd2dc | 6597 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6598 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6599 | u64 page_start = page_offset(page); |
6600 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6601 | |
8b62b72b CM |
6602 | /* |
6603 | * we have the page locked, so new writeback can't start, | |
6604 | * and the dirty bit won't be cleared while we are here. | |
6605 | * | |
6606 | * Wait for IO on this page so that we can safely clear | |
6607 | * the PagePrivate2 bit and do ordered accounting | |
6608 | */ | |
e6dcd2dc | 6609 | wait_on_page_writeback(page); |
8b62b72b | 6610 | |
5fd02043 | 6611 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6612 | if (offset) { |
6613 | btrfs_releasepage(page, GFP_NOFS); | |
6614 | return; | |
6615 | } | |
d0082371 | 6616 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
5fd02043 | 6617 | ordered = btrfs_lookup_ordered_extent(inode, |
e6dcd2dc CM |
6618 | page_offset(page)); |
6619 | if (ordered) { | |
eb84ae03 CM |
6620 | /* |
6621 | * IO on this page will never be started, so we need | |
6622 | * to account for any ordered extents now | |
6623 | */ | |
e6dcd2dc CM |
6624 | clear_extent_bit(tree, page_start, page_end, |
6625 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6626 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6627 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6628 | /* |
6629 | * whoever cleared the private bit is responsible | |
6630 | * for the finish_ordered_io | |
6631 | */ | |
5fd02043 JB |
6632 | if (TestClearPagePrivate2(page) && |
6633 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6634 | PAGE_CACHE_SIZE, 1)) { | |
6635 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6636 | } |
e6dcd2dc | 6637 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6638 | cached_state = NULL; |
d0082371 | 6639 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6640 | } |
6641 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6642 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6643 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6644 | __btrfs_releasepage(page, GFP_NOFS); |
6645 | ||
4a096752 | 6646 | ClearPageChecked(page); |
9ad6b7bc | 6647 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6648 | ClearPagePrivate(page); |
6649 | set_page_private(page, 0); | |
6650 | page_cache_release(page); | |
6651 | } | |
39279cc3 CM |
6652 | } |
6653 | ||
9ebefb18 CM |
6654 | /* |
6655 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6656 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6657 | * be careful to check for EOF conditions here. We set the page up correctly | |
6658 | * for a written page which means we get ENOSPC checking when writing into | |
6659 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6660 | * support these features. | |
6661 | * | |
6662 | * We are not allowed to take the i_mutex here so we have to play games to | |
6663 | * protect against truncate races as the page could now be beyond EOF. Because | |
6664 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6665 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6666 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6667 | * unlock the page. | |
6668 | */ | |
c2ec175c | 6669 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6670 | { |
c2ec175c | 6671 | struct page *page = vmf->page; |
6da6abae | 6672 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6673 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6674 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6675 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6676 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6677 | char *kaddr; |
6678 | unsigned long zero_start; | |
9ebefb18 | 6679 | loff_t size; |
1832a6d5 | 6680 | int ret; |
9998eb70 | 6681 | int reserved = 0; |
a52d9a80 | 6682 | u64 page_start; |
e6dcd2dc | 6683 | u64 page_end; |
9ebefb18 | 6684 | |
b2b5ef5c | 6685 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 6686 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6687 | if (!ret) { |
e41f941a | 6688 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
6689 | reserved = 1; |
6690 | } | |
56a76f82 NP |
6691 | if (ret) { |
6692 | if (ret == -ENOMEM) | |
6693 | ret = VM_FAULT_OOM; | |
6694 | else /* -ENOSPC, -EIO, etc */ | |
6695 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6696 | if (reserved) |
6697 | goto out; | |
6698 | goto out_noreserve; | |
56a76f82 | 6699 | } |
1832a6d5 | 6700 | |
56a76f82 | 6701 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6702 | again: |
9ebefb18 | 6703 | lock_page(page); |
9ebefb18 | 6704 | size = i_size_read(inode); |
e6dcd2dc CM |
6705 | page_start = page_offset(page); |
6706 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6707 | |
9ebefb18 | 6708 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6709 | (page_start >= size)) { |
9ebefb18 CM |
6710 | /* page got truncated out from underneath us */ |
6711 | goto out_unlock; | |
6712 | } | |
e6dcd2dc CM |
6713 | wait_on_page_writeback(page); |
6714 | ||
d0082371 | 6715 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6716 | set_page_extent_mapped(page); |
6717 | ||
eb84ae03 CM |
6718 | /* |
6719 | * we can't set the delalloc bits if there are pending ordered | |
6720 | * extents. Drop our locks and wait for them to finish | |
6721 | */ | |
e6dcd2dc CM |
6722 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6723 | if (ordered) { | |
2ac55d41 JB |
6724 | unlock_extent_cached(io_tree, page_start, page_end, |
6725 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6726 | unlock_page(page); |
eb84ae03 | 6727 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6728 | btrfs_put_ordered_extent(ordered); |
6729 | goto again; | |
6730 | } | |
6731 | ||
fbf19087 JB |
6732 | /* |
6733 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6734 | * if it was already dirty, so for space accounting reasons we need to | |
6735 | * clear any delalloc bits for the range we are fixing to save. There | |
6736 | * is probably a better way to do this, but for now keep consistent with | |
6737 | * prepare_pages in the normal write path. | |
6738 | */ | |
2ac55d41 | 6739 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6740 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6741 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6742 | |
2ac55d41 JB |
6743 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6744 | &cached_state); | |
9ed74f2d | 6745 | if (ret) { |
2ac55d41 JB |
6746 | unlock_extent_cached(io_tree, page_start, page_end, |
6747 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6748 | ret = VM_FAULT_SIGBUS; |
6749 | goto out_unlock; | |
6750 | } | |
e6dcd2dc | 6751 | ret = 0; |
9ebefb18 CM |
6752 | |
6753 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6754 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6755 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6756 | else |
e6dcd2dc | 6757 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6758 | |
e6dcd2dc CM |
6759 | if (zero_start != PAGE_CACHE_SIZE) { |
6760 | kaddr = kmap(page); | |
6761 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6762 | flush_dcache_page(page); | |
6763 | kunmap(page); | |
6764 | } | |
247e743c | 6765 | ClearPageChecked(page); |
e6dcd2dc | 6766 | set_page_dirty(page); |
50a9b214 | 6767 | SetPageUptodate(page); |
5a3f23d5 | 6768 | |
257c62e1 CM |
6769 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6770 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 6771 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 6772 | |
2ac55d41 | 6773 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6774 | |
6775 | out_unlock: | |
b2b5ef5c JK |
6776 | if (!ret) { |
6777 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 6778 | return VM_FAULT_LOCKED; |
b2b5ef5c | 6779 | } |
9ebefb18 | 6780 | unlock_page(page); |
1832a6d5 | 6781 | out: |
ec39e180 | 6782 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6783 | out_noreserve: |
b2b5ef5c | 6784 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
6785 | return ret; |
6786 | } | |
6787 | ||
a41ad394 | 6788 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6789 | { |
6790 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6791 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6792 | int ret; |
3893e33b | 6793 | int err = 0; |
39279cc3 | 6794 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6795 | unsigned long nr; |
dbe674a9 | 6796 | u64 mask = root->sectorsize - 1; |
07127184 | 6797 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6798 | |
5d5e103a JB |
6799 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6800 | if (ret) | |
a41ad394 | 6801 | return ret; |
8082510e | 6802 | |
4a096752 | 6803 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6804 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6805 | |
fcb80c2a JB |
6806 | /* |
6807 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6808 | * 3 things going on here | |
6809 | * | |
6810 | * 1) We need to reserve space for our orphan item and the space to | |
6811 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6812 | * orphan item because we didn't reserve space to remove it. | |
6813 | * | |
6814 | * 2) We need to reserve space to update our inode. | |
6815 | * | |
6816 | * 3) We need to have something to cache all the space that is going to | |
6817 | * be free'd up by the truncate operation, but also have some slack | |
6818 | * space reserved in case it uses space during the truncate (thank you | |
6819 | * very much snapshotting). | |
6820 | * | |
6821 | * And we need these to all be seperate. The fact is we can use alot of | |
6822 | * space doing the truncate, and we have no earthly idea how much space | |
6823 | * we will use, so we need the truncate reservation to be seperate so it | |
6824 | * doesn't end up using space reserved for updating the inode or | |
6825 | * removing the orphan item. We also need to be able to stop the | |
6826 | * transaction and start a new one, which means we need to be able to | |
6827 | * update the inode several times, and we have no idea of knowing how | |
6828 | * many times that will be, so we can't just reserve 1 item for the | |
6829 | * entirety of the opration, so that has to be done seperately as well. | |
6830 | * Then there is the orphan item, which does indeed need to be held on | |
6831 | * to for the whole operation, and we need nobody to touch this reserved | |
6832 | * space except the orphan code. | |
6833 | * | |
6834 | * So that leaves us with | |
6835 | * | |
6836 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6837 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6838 | * transaction reservation. | |
6839 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6840 | * updating the inode. | |
6841 | */ | |
6842 | rsv = btrfs_alloc_block_rsv(root); | |
6843 | if (!rsv) | |
6844 | return -ENOMEM; | |
4a338542 | 6845 | rsv->size = min_size; |
ca7e70f5 | 6846 | rsv->failfast = 1; |
f0cd846e | 6847 | |
907cbceb | 6848 | /* |
07127184 | 6849 | * 1 for the truncate slack space |
907cbceb JB |
6850 | * 1 for the orphan item we're going to add |
6851 | * 1 for the orphan item deletion | |
6852 | * 1 for updating the inode. | |
6853 | */ | |
fcb80c2a JB |
6854 | trans = btrfs_start_transaction(root, 4); |
6855 | if (IS_ERR(trans)) { | |
6856 | err = PTR_ERR(trans); | |
6857 | goto out; | |
6858 | } | |
f0cd846e | 6859 | |
907cbceb JB |
6860 | /* Migrate the slack space for the truncate to our reserve */ |
6861 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6862 | min_size); | |
fcb80c2a | 6863 | BUG_ON(ret); |
f0cd846e JB |
6864 | |
6865 | ret = btrfs_orphan_add(trans, inode); | |
6866 | if (ret) { | |
6867 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6868 | goto out; |
f0cd846e JB |
6869 | } |
6870 | ||
5a3f23d5 CM |
6871 | /* |
6872 | * setattr is responsible for setting the ordered_data_close flag, | |
6873 | * but that is only tested during the last file release. That | |
6874 | * could happen well after the next commit, leaving a great big | |
6875 | * window where new writes may get lost if someone chooses to write | |
6876 | * to this file after truncating to zero | |
6877 | * | |
6878 | * The inode doesn't have any dirty data here, and so if we commit | |
6879 | * this is a noop. If someone immediately starts writing to the inode | |
6880 | * it is very likely we'll catch some of their writes in this | |
6881 | * transaction, and the commit will find this file on the ordered | |
6882 | * data list with good things to send down. | |
6883 | * | |
6884 | * This is a best effort solution, there is still a window where | |
6885 | * using truncate to replace the contents of the file will | |
6886 | * end up with a zero length file after a crash. | |
6887 | */ | |
72ac3c0d JB |
6888 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
6889 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
6890 | btrfs_add_ordered_operation(trans, root, inode); |
6891 | ||
5dc562c5 JB |
6892 | /* |
6893 | * So if we truncate and then write and fsync we normally would just | |
6894 | * write the extents that changed, which is a problem if we need to | |
6895 | * first truncate that entire inode. So set this flag so we write out | |
6896 | * all of the extents in the inode to the sync log so we're completely | |
6897 | * safe. | |
6898 | */ | |
6899 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 6900 | trans->block_rsv = rsv; |
5dc562c5 | 6901 | |
8082510e YZ |
6902 | while (1) { |
6903 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
6904 | inode->i_size, | |
6905 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 6906 | if (ret != -ENOSPC) { |
3893e33b | 6907 | err = ret; |
8082510e | 6908 | break; |
3893e33b | 6909 | } |
39279cc3 | 6910 | |
fcb80c2a | 6911 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6912 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6913 | if (ret) { |
6914 | err = ret; | |
6915 | break; | |
6916 | } | |
ca7e70f5 | 6917 | |
8082510e YZ |
6918 | nr = trans->blocks_used; |
6919 | btrfs_end_transaction(trans, root); | |
6920 | btrfs_btree_balance_dirty(root, nr); | |
ca7e70f5 JB |
6921 | |
6922 | trans = btrfs_start_transaction(root, 2); | |
6923 | if (IS_ERR(trans)) { | |
6924 | ret = err = PTR_ERR(trans); | |
6925 | trans = NULL; | |
6926 | break; | |
6927 | } | |
6928 | ||
6929 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
6930 | rsv, min_size); | |
6931 | BUG_ON(ret); /* shouldn't happen */ | |
6932 | trans->block_rsv = rsv; | |
8082510e YZ |
6933 | } |
6934 | ||
6935 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6936 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6937 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6938 | if (ret) |
6939 | err = ret; | |
ded5db9d JB |
6940 | } else if (ret && inode->i_nlink > 0) { |
6941 | /* | |
6942 | * Failed to do the truncate, remove us from the in memory | |
6943 | * orphan list. | |
6944 | */ | |
6945 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6946 | } |
6947 | ||
917c16b2 CM |
6948 | if (trans) { |
6949 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
6950 | ret = btrfs_update_inode(trans, root, inode); | |
6951 | if (ret && !err) | |
6952 | err = ret; | |
7b128766 | 6953 | |
917c16b2 | 6954 | nr = trans->blocks_used; |
7ad85bb7 | 6955 | ret = btrfs_end_transaction(trans, root); |
917c16b2 CM |
6956 | btrfs_btree_balance_dirty(root, nr); |
6957 | } | |
fcb80c2a JB |
6958 | |
6959 | out: | |
6960 | btrfs_free_block_rsv(root, rsv); | |
6961 | ||
3893e33b JB |
6962 | if (ret && !err) |
6963 | err = ret; | |
a41ad394 | 6964 | |
3893e33b | 6965 | return err; |
39279cc3 CM |
6966 | } |
6967 | ||
d352ac68 CM |
6968 | /* |
6969 | * create a new subvolume directory/inode (helper for the ioctl). | |
6970 | */ | |
d2fb3437 | 6971 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6972 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6973 | { |
39279cc3 | 6974 | struct inode *inode; |
76dda93c | 6975 | int err; |
00e4e6b3 | 6976 | u64 index = 0; |
39279cc3 | 6977 | |
12fc9d09 FA |
6978 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
6979 | new_dirid, new_dirid, | |
6980 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
6981 | &index); | |
54aa1f4d | 6982 | if (IS_ERR(inode)) |
f46b5a66 | 6983 | return PTR_ERR(inode); |
39279cc3 CM |
6984 | inode->i_op = &btrfs_dir_inode_operations; |
6985 | inode->i_fop = &btrfs_dir_file_operations; | |
6986 | ||
bfe86848 | 6987 | set_nlink(inode, 1); |
dbe674a9 | 6988 | btrfs_i_size_write(inode, 0); |
3b96362c | 6989 | |
76dda93c | 6990 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 6991 | |
76dda93c | 6992 | iput(inode); |
ce598979 | 6993 | return err; |
39279cc3 CM |
6994 | } |
6995 | ||
39279cc3 CM |
6996 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6997 | { | |
6998 | struct btrfs_inode *ei; | |
2ead6ae7 | 6999 | struct inode *inode; |
39279cc3 CM |
7000 | |
7001 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7002 | if (!ei) | |
7003 | return NULL; | |
2ead6ae7 YZ |
7004 | |
7005 | ei->root = NULL; | |
2ead6ae7 | 7006 | ei->generation = 0; |
15ee9bc7 | 7007 | ei->last_trans = 0; |
257c62e1 | 7008 | ei->last_sub_trans = 0; |
e02119d5 | 7009 | ei->logged_trans = 0; |
2ead6ae7 | 7010 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7011 | ei->disk_i_size = 0; |
7012 | ei->flags = 0; | |
7709cde3 | 7013 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7014 | ei->index_cnt = (u64)-1; |
7015 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7016 | ei->last_log_commit = 0; |
2ead6ae7 | 7017 | |
9e0baf60 JB |
7018 | spin_lock_init(&ei->lock); |
7019 | ei->outstanding_extents = 0; | |
7020 | ei->reserved_extents = 0; | |
2ead6ae7 | 7021 | |
72ac3c0d | 7022 | ei->runtime_flags = 0; |
261507a0 | 7023 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7024 | |
16cdcec7 MX |
7025 | ei->delayed_node = NULL; |
7026 | ||
2ead6ae7 | 7027 | inode = &ei->vfs_inode; |
a8067e02 | 7028 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7029 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7030 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7031 | ei->io_tree.track_uptodate = 1; |
7032 | ei->io_failure_tree.track_uptodate = 1; | |
2ead6ae7 | 7033 | mutex_init(&ei->log_mutex); |
f248679e | 7034 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7035 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7036 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7037 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7038 | RB_CLEAR_NODE(&ei->rb_node); |
7039 | ||
7040 | return inode; | |
39279cc3 CM |
7041 | } |
7042 | ||
fa0d7e3d NP |
7043 | static void btrfs_i_callback(struct rcu_head *head) |
7044 | { | |
7045 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7046 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7047 | } | |
7048 | ||
39279cc3 CM |
7049 | void btrfs_destroy_inode(struct inode *inode) |
7050 | { | |
e6dcd2dc | 7051 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7052 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7053 | ||
b3d9b7a3 | 7054 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7055 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7056 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7057 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7058 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7059 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7060 | |
a6dbd429 JB |
7061 | /* |
7062 | * This can happen where we create an inode, but somebody else also | |
7063 | * created the same inode and we need to destroy the one we already | |
7064 | * created. | |
7065 | */ | |
7066 | if (!root) | |
7067 | goto free; | |
7068 | ||
5a3f23d5 CM |
7069 | /* |
7070 | * Make sure we're properly removed from the ordered operation | |
7071 | * lists. | |
7072 | */ | |
7073 | smp_mb(); | |
7074 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
7075 | spin_lock(&root->fs_info->ordered_extent_lock); | |
7076 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
7077 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
7078 | } | |
7079 | ||
8a35d95f JB |
7080 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7081 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
7082 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
7083 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7084 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7085 | } |
7b128766 | 7086 | |
d397712b | 7087 | while (1) { |
e6dcd2dc CM |
7088 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7089 | if (!ordered) | |
7090 | break; | |
7091 | else { | |
d397712b CM |
7092 | printk(KERN_ERR "btrfs found ordered " |
7093 | "extent %llu %llu on inode cleanup\n", | |
7094 | (unsigned long long)ordered->file_offset, | |
7095 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7096 | btrfs_remove_ordered_extent(inode, ordered); |
7097 | btrfs_put_ordered_extent(ordered); | |
7098 | btrfs_put_ordered_extent(ordered); | |
7099 | } | |
7100 | } | |
5d4f98a2 | 7101 | inode_tree_del(inode); |
5b21f2ed | 7102 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7103 | free: |
16cdcec7 | 7104 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 7105 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7106 | } |
7107 | ||
45321ac5 | 7108 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7109 | { |
7110 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7111 | |
0af3d00b | 7112 | if (btrfs_root_refs(&root->root_item) == 0 && |
83eea1f1 | 7113 | !btrfs_is_free_space_inode(inode)) |
45321ac5 | 7114 | return 1; |
76dda93c | 7115 | else |
45321ac5 | 7116 | return generic_drop_inode(inode); |
76dda93c YZ |
7117 | } |
7118 | ||
0ee0fda0 | 7119 | static void init_once(void *foo) |
39279cc3 CM |
7120 | { |
7121 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7122 | ||
7123 | inode_init_once(&ei->vfs_inode); | |
7124 | } | |
7125 | ||
7126 | void btrfs_destroy_cachep(void) | |
7127 | { | |
7128 | if (btrfs_inode_cachep) | |
7129 | kmem_cache_destroy(btrfs_inode_cachep); | |
7130 | if (btrfs_trans_handle_cachep) | |
7131 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7132 | if (btrfs_transaction_cachep) | |
7133 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7134 | if (btrfs_path_cachep) |
7135 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7136 | if (btrfs_free_space_cachep) |
7137 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
7138 | } |
7139 | ||
7140 | int btrfs_init_cachep(void) | |
7141 | { | |
9601e3f6 CH |
7142 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
7143 | sizeof(struct btrfs_inode), 0, | |
7144 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7145 | if (!btrfs_inode_cachep) |
7146 | goto fail; | |
9601e3f6 CH |
7147 | |
7148 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
7149 | sizeof(struct btrfs_trans_handle), 0, | |
7150 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7151 | if (!btrfs_trans_handle_cachep) |
7152 | goto fail; | |
9601e3f6 CH |
7153 | |
7154 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
7155 | sizeof(struct btrfs_transaction), 0, | |
7156 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7157 | if (!btrfs_transaction_cachep) |
7158 | goto fail; | |
9601e3f6 CH |
7159 | |
7160 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
7161 | sizeof(struct btrfs_path), 0, | |
7162 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7163 | if (!btrfs_path_cachep) |
7164 | goto fail; | |
9601e3f6 | 7165 | |
dc89e982 JB |
7166 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
7167 | sizeof(struct btrfs_free_space), 0, | |
7168 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7169 | if (!btrfs_free_space_cachep) | |
7170 | goto fail; | |
7171 | ||
39279cc3 CM |
7172 | return 0; |
7173 | fail: | |
7174 | btrfs_destroy_cachep(); | |
7175 | return -ENOMEM; | |
7176 | } | |
7177 | ||
7178 | static int btrfs_getattr(struct vfsmount *mnt, | |
7179 | struct dentry *dentry, struct kstat *stat) | |
7180 | { | |
7181 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
7182 | u32 blocksize = inode->i_sb->s_blocksize; |
7183 | ||
39279cc3 | 7184 | generic_fillattr(inode, stat); |
0ee5dc67 | 7185 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7186 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
7187 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
7188 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
7189 | return 0; |
7190 | } | |
7191 | ||
75e7cb7f LB |
7192 | /* |
7193 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7194 | * directory. | |
7195 | */ | |
7196 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7197 | { | |
7198 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7199 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7200 | ||
7201 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7202 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7203 | else | |
7204 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7205 | ||
bc178237 | 7206 | if (b_dir->flags & BTRFS_INODE_COMPRESS) { |
75e7cb7f | 7207 | b_inode->flags |= BTRFS_INODE_COMPRESS; |
bc178237 LB |
7208 | b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
7209 | } else { | |
7210 | b_inode->flags &= ~(BTRFS_INODE_COMPRESS | | |
7211 | BTRFS_INODE_NOCOMPRESS); | |
7212 | } | |
75e7cb7f LB |
7213 | } |
7214 | ||
d397712b CM |
7215 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7216 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7217 | { |
7218 | struct btrfs_trans_handle *trans; | |
7219 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7220 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7221 | struct inode *new_inode = new_dentry->d_inode; |
7222 | struct inode *old_inode = old_dentry->d_inode; | |
7223 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7224 | u64 index = 0; |
4df27c4d | 7225 | u64 root_objectid; |
39279cc3 | 7226 | int ret; |
33345d01 | 7227 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7228 | |
33345d01 | 7229 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7230 | return -EPERM; |
7231 | ||
4df27c4d | 7232 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7233 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7234 | return -EXDEV; |
7235 | ||
33345d01 LZ |
7236 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7237 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7238 | return -ENOTEMPTY; |
5f39d397 | 7239 | |
4df27c4d YZ |
7240 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7241 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7242 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7243 | /* |
7244 | * we're using rename to replace one file with another. | |
7245 | * and the replacement file is large. Start IO on it now so | |
7246 | * we don't add too much work to the end of the transaction | |
7247 | */ | |
4baf8c92 | 7248 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7249 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7250 | filemap_flush(old_inode->i_mapping); | |
7251 | ||
76dda93c | 7252 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7253 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7254 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7255 | /* |
7256 | * We want to reserve the absolute worst case amount of items. So if | |
7257 | * both inodes are subvols and we need to unlink them then that would | |
7258 | * require 4 item modifications, but if they are both normal inodes it | |
7259 | * would require 5 item modifications, so we'll assume their normal | |
7260 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7261 | * should cover the worst case number of items we'll modify. | |
7262 | */ | |
7263 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7264 | if (IS_ERR(trans)) { |
7265 | ret = PTR_ERR(trans); | |
7266 | goto out_notrans; | |
7267 | } | |
76dda93c | 7268 | |
4df27c4d YZ |
7269 | if (dest != root) |
7270 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7271 | |
a5719521 YZ |
7272 | ret = btrfs_set_inode_index(new_dir, &index); |
7273 | if (ret) | |
7274 | goto out_fail; | |
5a3f23d5 | 7275 | |
33345d01 | 7276 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7277 | /* force full log commit if subvolume involved. */ |
7278 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7279 | } else { | |
a5719521 YZ |
7280 | ret = btrfs_insert_inode_ref(trans, dest, |
7281 | new_dentry->d_name.name, | |
7282 | new_dentry->d_name.len, | |
33345d01 LZ |
7283 | old_ino, |
7284 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7285 | if (ret) |
7286 | goto out_fail; | |
4df27c4d YZ |
7287 | /* |
7288 | * this is an ugly little race, but the rename is required | |
7289 | * to make sure that if we crash, the inode is either at the | |
7290 | * old name or the new one. pinning the log transaction lets | |
7291 | * us make sure we don't allow a log commit to come in after | |
7292 | * we unlink the name but before we add the new name back in. | |
7293 | */ | |
7294 | btrfs_pin_log_trans(root); | |
7295 | } | |
5a3f23d5 CM |
7296 | /* |
7297 | * make sure the inode gets flushed if it is replacing | |
7298 | * something. | |
7299 | */ | |
33345d01 | 7300 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7301 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7302 | |
0c4d2d95 JB |
7303 | inode_inc_iversion(old_dir); |
7304 | inode_inc_iversion(new_dir); | |
7305 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7306 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7307 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7308 | old_inode->i_ctime = ctime; | |
5f39d397 | 7309 | |
12fcfd22 CM |
7310 | if (old_dentry->d_parent != new_dentry->d_parent) |
7311 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7312 | ||
33345d01 | 7313 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7314 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7315 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7316 | old_dentry->d_name.name, | |
7317 | old_dentry->d_name.len); | |
7318 | } else { | |
92986796 AV |
7319 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7320 | old_dentry->d_inode, | |
7321 | old_dentry->d_name.name, | |
7322 | old_dentry->d_name.len); | |
7323 | if (!ret) | |
7324 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7325 | } |
79787eaa JM |
7326 | if (ret) { |
7327 | btrfs_abort_transaction(trans, root, ret); | |
7328 | goto out_fail; | |
7329 | } | |
39279cc3 CM |
7330 | |
7331 | if (new_inode) { | |
0c4d2d95 | 7332 | inode_inc_iversion(new_inode); |
39279cc3 | 7333 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7334 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7335 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7336 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7337 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7338 | root_objectid, | |
7339 | new_dentry->d_name.name, | |
7340 | new_dentry->d_name.len); | |
7341 | BUG_ON(new_inode->i_nlink == 0); | |
7342 | } else { | |
7343 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7344 | new_dentry->d_inode, | |
7345 | new_dentry->d_name.name, | |
7346 | new_dentry->d_name.len); | |
7347 | } | |
79787eaa | 7348 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7349 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7350 | BUG_ON(ret); |
7b128766 | 7351 | } |
79787eaa JM |
7352 | if (ret) { |
7353 | btrfs_abort_transaction(trans, root, ret); | |
7354 | goto out_fail; | |
7355 | } | |
39279cc3 | 7356 | } |
aec7477b | 7357 | |
75e7cb7f LB |
7358 | fixup_inode_flags(new_dir, old_inode); |
7359 | ||
4df27c4d YZ |
7360 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7361 | new_dentry->d_name.name, | |
a5719521 | 7362 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7363 | if (ret) { |
7364 | btrfs_abort_transaction(trans, root, ret); | |
7365 | goto out_fail; | |
7366 | } | |
39279cc3 | 7367 | |
33345d01 | 7368 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7369 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7370 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7371 | btrfs_end_log_trans(root); |
7372 | } | |
39279cc3 | 7373 | out_fail: |
7ad85bb7 | 7374 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7375 | out_notrans: |
33345d01 | 7376 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7377 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7378 | |
39279cc3 CM |
7379 | return ret; |
7380 | } | |
7381 | ||
d352ac68 CM |
7382 | /* |
7383 | * some fairly slow code that needs optimization. This walks the list | |
7384 | * of all the inodes with pending delalloc and forces them to disk. | |
7385 | */ | |
24bbcf04 | 7386 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7387 | { |
7388 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7389 | struct btrfs_inode *binode; | |
5b21f2ed | 7390 | struct inode *inode; |
ea8c2819 | 7391 | |
c146afad YZ |
7392 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7393 | return -EROFS; | |
7394 | ||
75eff68e | 7395 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7396 | while (!list_empty(head)) { |
ea8c2819 CM |
7397 | binode = list_entry(head->next, struct btrfs_inode, |
7398 | delalloc_inodes); | |
5b21f2ed ZY |
7399 | inode = igrab(&binode->vfs_inode); |
7400 | if (!inode) | |
7401 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7402 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7403 | if (inode) { |
8c8bee1d | 7404 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7405 | if (delay_iput) |
7406 | btrfs_add_delayed_iput(inode); | |
7407 | else | |
7408 | iput(inode); | |
5b21f2ed ZY |
7409 | } |
7410 | cond_resched(); | |
75eff68e | 7411 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7412 | } |
75eff68e | 7413 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7414 | |
7415 | /* the filemap_flush will queue IO into the worker threads, but | |
7416 | * we have to make sure the IO is actually started and that | |
7417 | * ordered extents get created before we return | |
7418 | */ | |
7419 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7420 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7421 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7422 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7423 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7424 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7425 | } |
7426 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7427 | return 0; |
7428 | } | |
7429 | ||
39279cc3 CM |
7430 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7431 | const char *symname) | |
7432 | { | |
7433 | struct btrfs_trans_handle *trans; | |
7434 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7435 | struct btrfs_path *path; | |
7436 | struct btrfs_key key; | |
1832a6d5 | 7437 | struct inode *inode = NULL; |
39279cc3 CM |
7438 | int err; |
7439 | int drop_inode = 0; | |
7440 | u64 objectid; | |
00e4e6b3 | 7441 | u64 index = 0 ; |
39279cc3 CM |
7442 | int name_len; |
7443 | int datasize; | |
5f39d397 | 7444 | unsigned long ptr; |
39279cc3 | 7445 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7446 | struct extent_buffer *leaf; |
1832a6d5 | 7447 | unsigned long nr = 0; |
39279cc3 CM |
7448 | |
7449 | name_len = strlen(symname) + 1; | |
7450 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7451 | return -ENAMETOOLONG; | |
1832a6d5 | 7452 | |
9ed74f2d JB |
7453 | /* |
7454 | * 2 items for inode item and ref | |
7455 | * 2 items for dir items | |
7456 | * 1 item for xattr if selinux is on | |
7457 | */ | |
a22285a6 YZ |
7458 | trans = btrfs_start_transaction(root, 5); |
7459 | if (IS_ERR(trans)) | |
7460 | return PTR_ERR(trans); | |
1832a6d5 | 7461 | |
581bb050 LZ |
7462 | err = btrfs_find_free_ino(root, &objectid); |
7463 | if (err) | |
7464 | goto out_unlock; | |
7465 | ||
aec7477b | 7466 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7467 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7468 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7469 | if (IS_ERR(inode)) { |
7470 | err = PTR_ERR(inode); | |
39279cc3 | 7471 | goto out_unlock; |
7cf96da3 | 7472 | } |
39279cc3 | 7473 | |
2a7dba39 | 7474 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7475 | if (err) { |
7476 | drop_inode = 1; | |
7477 | goto out_unlock; | |
7478 | } | |
7479 | ||
ad19db71 CS |
7480 | /* |
7481 | * If the active LSM wants to access the inode during | |
7482 | * d_instantiate it needs these. Smack checks to see | |
7483 | * if the filesystem supports xattrs by looking at the | |
7484 | * ops vector. | |
7485 | */ | |
7486 | inode->i_fop = &btrfs_file_operations; | |
7487 | inode->i_op = &btrfs_file_inode_operations; | |
7488 | ||
a1b075d2 | 7489 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7490 | if (err) |
7491 | drop_inode = 1; | |
7492 | else { | |
7493 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7494 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7495 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7496 | } |
39279cc3 CM |
7497 | if (drop_inode) |
7498 | goto out_unlock; | |
7499 | ||
7500 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7501 | if (!path) { |
7502 | err = -ENOMEM; | |
7503 | drop_inode = 1; | |
7504 | goto out_unlock; | |
7505 | } | |
33345d01 | 7506 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7507 | key.offset = 0; |
39279cc3 CM |
7508 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7509 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7510 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7511 | datasize); | |
54aa1f4d CM |
7512 | if (err) { |
7513 | drop_inode = 1; | |
b0839166 | 7514 | btrfs_free_path(path); |
54aa1f4d CM |
7515 | goto out_unlock; |
7516 | } | |
5f39d397 CM |
7517 | leaf = path->nodes[0]; |
7518 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7519 | struct btrfs_file_extent_item); | |
7520 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7521 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7522 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7523 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7524 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7525 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7526 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7527 | ||
39279cc3 | 7528 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7529 | write_extent_buffer(leaf, symname, ptr, name_len); |
7530 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7531 | btrfs_free_path(path); |
5f39d397 | 7532 | |
39279cc3 CM |
7533 | inode->i_op = &btrfs_symlink_inode_operations; |
7534 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7535 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7536 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7537 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7538 | err = btrfs_update_inode(trans, root, inode); |
7539 | if (err) | |
7540 | drop_inode = 1; | |
39279cc3 CM |
7541 | |
7542 | out_unlock: | |
08c422c2 AV |
7543 | if (!err) |
7544 | d_instantiate(dentry, inode); | |
d3c2fdcf | 7545 | nr = trans->blocks_used; |
7ad85bb7 | 7546 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7547 | if (drop_inode) { |
7548 | inode_dec_link_count(inode); | |
7549 | iput(inode); | |
7550 | } | |
d3c2fdcf | 7551 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7552 | return err; |
7553 | } | |
16432985 | 7554 | |
0af3d00b JB |
7555 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7556 | u64 start, u64 num_bytes, u64 min_size, | |
7557 | loff_t actual_len, u64 *alloc_hint, | |
7558 | struct btrfs_trans_handle *trans) | |
d899e052 | 7559 | { |
5dc562c5 JB |
7560 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7561 | struct extent_map *em; | |
d899e052 YZ |
7562 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7563 | struct btrfs_key ins; | |
d899e052 | 7564 | u64 cur_offset = start; |
55a61d1d | 7565 | u64 i_size; |
d899e052 | 7566 | int ret = 0; |
0af3d00b | 7567 | bool own_trans = true; |
d899e052 | 7568 | |
0af3d00b JB |
7569 | if (trans) |
7570 | own_trans = false; | |
d899e052 | 7571 | while (num_bytes > 0) { |
0af3d00b JB |
7572 | if (own_trans) { |
7573 | trans = btrfs_start_transaction(root, 3); | |
7574 | if (IS_ERR(trans)) { | |
7575 | ret = PTR_ERR(trans); | |
7576 | break; | |
7577 | } | |
5a303d5d YZ |
7578 | } |
7579 | ||
efa56464 | 7580 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7581 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7582 | if (ret) { |
0af3d00b JB |
7583 | if (own_trans) |
7584 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7585 | break; |
d899e052 | 7586 | } |
5a303d5d | 7587 | |
d899e052 YZ |
7588 | ret = insert_reserved_file_extent(trans, inode, |
7589 | cur_offset, ins.objectid, | |
7590 | ins.offset, ins.offset, | |
920bbbfb | 7591 | ins.offset, 0, 0, 0, |
d899e052 | 7592 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7593 | if (ret) { |
7594 | btrfs_abort_transaction(trans, root, ret); | |
7595 | if (own_trans) | |
7596 | btrfs_end_transaction(trans, root); | |
7597 | break; | |
7598 | } | |
a1ed835e CM |
7599 | btrfs_drop_extent_cache(inode, cur_offset, |
7600 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7601 | |
5dc562c5 JB |
7602 | em = alloc_extent_map(); |
7603 | if (!em) { | |
7604 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
7605 | &BTRFS_I(inode)->runtime_flags); | |
7606 | goto next; | |
7607 | } | |
7608 | ||
7609 | em->start = cur_offset; | |
7610 | em->orig_start = cur_offset; | |
7611 | em->len = ins.offset; | |
7612 | em->block_start = ins.objectid; | |
7613 | em->block_len = ins.offset; | |
7614 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
7615 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
7616 | em->generation = trans->transid; | |
7617 | ||
7618 | while (1) { | |
7619 | write_lock(&em_tree->lock); | |
7620 | ret = add_extent_mapping(em_tree, em); | |
7621 | if (!ret) | |
7622 | list_move(&em->list, | |
7623 | &em_tree->modified_extents); | |
7624 | write_unlock(&em_tree->lock); | |
7625 | if (ret != -EEXIST) | |
7626 | break; | |
7627 | btrfs_drop_extent_cache(inode, cur_offset, | |
7628 | cur_offset + ins.offset - 1, | |
7629 | 0); | |
7630 | } | |
7631 | free_extent_map(em); | |
7632 | next: | |
d899e052 YZ |
7633 | num_bytes -= ins.offset; |
7634 | cur_offset += ins.offset; | |
efa56464 | 7635 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7636 | |
0c4d2d95 | 7637 | inode_inc_iversion(inode); |
d899e052 | 7638 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7639 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7640 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7641 | (actual_len > inode->i_size) && |
7642 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7643 | if (cur_offset > actual_len) |
55a61d1d | 7644 | i_size = actual_len; |
d1ea6a61 | 7645 | else |
55a61d1d JB |
7646 | i_size = cur_offset; |
7647 | i_size_write(inode, i_size); | |
7648 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7649 | } |
7650 | ||
d899e052 | 7651 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7652 | |
7653 | if (ret) { | |
7654 | btrfs_abort_transaction(trans, root, ret); | |
7655 | if (own_trans) | |
7656 | btrfs_end_transaction(trans, root); | |
7657 | break; | |
7658 | } | |
d899e052 | 7659 | |
0af3d00b JB |
7660 | if (own_trans) |
7661 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7662 | } |
d899e052 YZ |
7663 | return ret; |
7664 | } | |
7665 | ||
0af3d00b JB |
7666 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7667 | u64 start, u64 num_bytes, u64 min_size, | |
7668 | loff_t actual_len, u64 *alloc_hint) | |
7669 | { | |
7670 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7671 | min_size, actual_len, alloc_hint, | |
7672 | NULL); | |
7673 | } | |
7674 | ||
7675 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7676 | struct btrfs_trans_handle *trans, int mode, | |
7677 | u64 start, u64 num_bytes, u64 min_size, | |
7678 | loff_t actual_len, u64 *alloc_hint) | |
7679 | { | |
7680 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7681 | min_size, actual_len, alloc_hint, trans); | |
7682 | } | |
7683 | ||
e6dcd2dc CM |
7684 | static int btrfs_set_page_dirty(struct page *page) |
7685 | { | |
e6dcd2dc CM |
7686 | return __set_page_dirty_nobuffers(page); |
7687 | } | |
7688 | ||
10556cb2 | 7689 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7690 | { |
b83cc969 | 7691 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7692 | umode_t mode = inode->i_mode; |
b83cc969 | 7693 | |
cb6db4e5 JM |
7694 | if (mask & MAY_WRITE && |
7695 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7696 | if (btrfs_root_readonly(root)) | |
7697 | return -EROFS; | |
7698 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7699 | return -EACCES; | |
7700 | } | |
2830ba7f | 7701 | return generic_permission(inode, mask); |
fdebe2bd | 7702 | } |
39279cc3 | 7703 | |
6e1d5dcc | 7704 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7705 | .getattr = btrfs_getattr, |
39279cc3 CM |
7706 | .lookup = btrfs_lookup, |
7707 | .create = btrfs_create, | |
7708 | .unlink = btrfs_unlink, | |
7709 | .link = btrfs_link, | |
7710 | .mkdir = btrfs_mkdir, | |
7711 | .rmdir = btrfs_rmdir, | |
7712 | .rename = btrfs_rename, | |
7713 | .symlink = btrfs_symlink, | |
7714 | .setattr = btrfs_setattr, | |
618e21d5 | 7715 | .mknod = btrfs_mknod, |
95819c05 CH |
7716 | .setxattr = btrfs_setxattr, |
7717 | .getxattr = btrfs_getxattr, | |
5103e947 | 7718 | .listxattr = btrfs_listxattr, |
95819c05 | 7719 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7720 | .permission = btrfs_permission, |
4e34e719 | 7721 | .get_acl = btrfs_get_acl, |
39279cc3 | 7722 | }; |
6e1d5dcc | 7723 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7724 | .lookup = btrfs_lookup, |
fdebe2bd | 7725 | .permission = btrfs_permission, |
4e34e719 | 7726 | .get_acl = btrfs_get_acl, |
39279cc3 | 7727 | }; |
76dda93c | 7728 | |
828c0950 | 7729 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7730 | .llseek = generic_file_llseek, |
7731 | .read = generic_read_dir, | |
cbdf5a24 | 7732 | .readdir = btrfs_real_readdir, |
34287aa3 | 7733 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7734 | #ifdef CONFIG_COMPAT |
34287aa3 | 7735 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7736 | #endif |
6bf13c0c | 7737 | .release = btrfs_release_file, |
e02119d5 | 7738 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7739 | }; |
7740 | ||
d1310b2e | 7741 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7742 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7743 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7744 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7745 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7746 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7747 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7748 | .set_bit_hook = btrfs_set_bit_hook, |
7749 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7750 | .merge_extent_hook = btrfs_merge_extent_hook, |
7751 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7752 | }; |
7753 | ||
35054394 CM |
7754 | /* |
7755 | * btrfs doesn't support the bmap operation because swapfiles | |
7756 | * use bmap to make a mapping of extents in the file. They assume | |
7757 | * these extents won't change over the life of the file and they | |
7758 | * use the bmap result to do IO directly to the drive. | |
7759 | * | |
7760 | * the btrfs bmap call would return logical addresses that aren't | |
7761 | * suitable for IO and they also will change frequently as COW | |
7762 | * operations happen. So, swapfile + btrfs == corruption. | |
7763 | * | |
7764 | * For now we're avoiding this by dropping bmap. | |
7765 | */ | |
7f09410b | 7766 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7767 | .readpage = btrfs_readpage, |
7768 | .writepage = btrfs_writepage, | |
b293f02e | 7769 | .writepages = btrfs_writepages, |
3ab2fb5a | 7770 | .readpages = btrfs_readpages, |
16432985 | 7771 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7772 | .invalidatepage = btrfs_invalidatepage, |
7773 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7774 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7775 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7776 | }; |
7777 | ||
7f09410b | 7778 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7779 | .readpage = btrfs_readpage, |
7780 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7781 | .invalidatepage = btrfs_invalidatepage, |
7782 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7783 | }; |
7784 | ||
6e1d5dcc | 7785 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7786 | .getattr = btrfs_getattr, |
7787 | .setattr = btrfs_setattr, | |
95819c05 CH |
7788 | .setxattr = btrfs_setxattr, |
7789 | .getxattr = btrfs_getxattr, | |
5103e947 | 7790 | .listxattr = btrfs_listxattr, |
95819c05 | 7791 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7792 | .permission = btrfs_permission, |
1506fcc8 | 7793 | .fiemap = btrfs_fiemap, |
4e34e719 | 7794 | .get_acl = btrfs_get_acl, |
e41f941a | 7795 | .update_time = btrfs_update_time, |
39279cc3 | 7796 | }; |
6e1d5dcc | 7797 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7798 | .getattr = btrfs_getattr, |
7799 | .setattr = btrfs_setattr, | |
fdebe2bd | 7800 | .permission = btrfs_permission, |
95819c05 CH |
7801 | .setxattr = btrfs_setxattr, |
7802 | .getxattr = btrfs_getxattr, | |
33268eaf | 7803 | .listxattr = btrfs_listxattr, |
95819c05 | 7804 | .removexattr = btrfs_removexattr, |
4e34e719 | 7805 | .get_acl = btrfs_get_acl, |
e41f941a | 7806 | .update_time = btrfs_update_time, |
618e21d5 | 7807 | }; |
6e1d5dcc | 7808 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7809 | .readlink = generic_readlink, |
7810 | .follow_link = page_follow_link_light, | |
7811 | .put_link = page_put_link, | |
f209561a | 7812 | .getattr = btrfs_getattr, |
22c44fe6 | 7813 | .setattr = btrfs_setattr, |
fdebe2bd | 7814 | .permission = btrfs_permission, |
0279b4cd JO |
7815 | .setxattr = btrfs_setxattr, |
7816 | .getxattr = btrfs_getxattr, | |
7817 | .listxattr = btrfs_listxattr, | |
7818 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7819 | .get_acl = btrfs_get_acl, |
e41f941a | 7820 | .update_time = btrfs_update_time, |
39279cc3 | 7821 | }; |
76dda93c | 7822 | |
82d339d9 | 7823 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7824 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7825 | .d_release = btrfs_dentry_release, |
76dda93c | 7826 | }; |