Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License version 2 as published by | |
8 | * the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., 51 | |
17 | * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | * | |
19 | * Authors: Artem Bityutskiy (Битюцкий Артём) | |
20 | * Adrian Hunter | |
21 | */ | |
22 | ||
23 | /* | |
873a64c7 | 24 | * This file implements VFS file and inode operations for regular files, device |
1e51764a AB |
25 | * nodes and symlinks as well as address space operations. |
26 | * | |
873a64c7 AB |
27 | * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if |
28 | * the page is dirty and is used for optimization purposes - dirty pages are | |
29 | * not budgeted so the flag shows that 'ubifs_write_end()' should not release | |
30 | * the budget for this page. The @PG_checked flag is set if full budgeting is | |
31 | * required for the page e.g., when it corresponds to a file hole or it is | |
32 | * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because | |
33 | * it is OK to fail in this function, and the budget is released in | |
34 | * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry | |
35 | * information about how the page was budgeted, to make it possible to release | |
36 | * the budget properly. | |
1e51764a | 37 | * |
873a64c7 AB |
38 | * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we |
39 | * implement. However, this is not true for 'ubifs_writepage()', which may be | |
40 | * called with @i_mutex unlocked. For example, when pdflush is doing background | |
41 | * write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex. At "normal" | |
42 | * work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g. in the | |
43 | * "sys_write -> alloc_pages -> direct reclaim path". So, in 'ubifs_writepage()' | |
44 | * we are only guaranteed that the page is locked. | |
1e51764a | 45 | * |
873a64c7 AB |
46 | * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the |
47 | * read-ahead path does not lock it ("sys_read -> generic_file_aio_read -> | |
eaff8079 | 48 | * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not |
873a64c7 | 49 | * set as well. However, UBIFS disables readahead. |
1e51764a AB |
50 | */ |
51 | ||
52 | #include "ubifs.h" | |
53 | #include <linux/mount.h> | |
3f8206d4 | 54 | #include <linux/namei.h> |
5a0e3ad6 | 55 | #include <linux/slab.h> |
1e51764a AB |
56 | |
57 | static int read_block(struct inode *inode, void *addr, unsigned int block, | |
58 | struct ubifs_data_node *dn) | |
59 | { | |
60 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
61 | int err, len, out_len; | |
62 | union ubifs_key key; | |
63 | unsigned int dlen; | |
64 | ||
65 | data_key_init(c, &key, inode->i_ino, block); | |
66 | err = ubifs_tnc_lookup(c, &key, dn); | |
67 | if (err) { | |
68 | if (err == -ENOENT) | |
69 | /* Not found, so it must be a hole */ | |
70 | memset(addr, 0, UBIFS_BLOCK_SIZE); | |
71 | return err; | |
72 | } | |
73 | ||
f92b9826 AB |
74 | ubifs_assert(le64_to_cpu(dn->ch.sqnum) > |
75 | ubifs_inode(inode)->creat_sqnum); | |
1e51764a AB |
76 | len = le32_to_cpu(dn->size); |
77 | if (len <= 0 || len > UBIFS_BLOCK_SIZE) | |
78 | goto dump; | |
79 | ||
80 | dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ; | |
81 | out_len = UBIFS_BLOCK_SIZE; | |
82 | err = ubifs_decompress(&dn->data, dlen, addr, &out_len, | |
83 | le16_to_cpu(dn->compr_type)); | |
84 | if (err || len != out_len) | |
85 | goto dump; | |
86 | ||
87 | /* | |
88 | * Data length can be less than a full block, even for blocks that are | |
89 | * not the last in the file (e.g., as a result of making a hole and | |
90 | * appending data). Ensure that the remainder is zeroed out. | |
91 | */ | |
92 | if (len < UBIFS_BLOCK_SIZE) | |
93 | memset(addr + len, 0, UBIFS_BLOCK_SIZE - len); | |
94 | ||
95 | return 0; | |
96 | ||
97 | dump: | |
98 | ubifs_err("bad data node (block %u, inode %lu)", | |
99 | block, inode->i_ino); | |
100 | dbg_dump_node(c, dn); | |
101 | return -EINVAL; | |
102 | } | |
103 | ||
104 | static int do_readpage(struct page *page) | |
105 | { | |
106 | void *addr; | |
107 | int err = 0, i; | |
108 | unsigned int block, beyond; | |
109 | struct ubifs_data_node *dn; | |
110 | struct inode *inode = page->mapping->host; | |
111 | loff_t i_size = i_size_read(inode); | |
112 | ||
113 | dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx", | |
114 | inode->i_ino, page->index, i_size, page->flags); | |
115 | ubifs_assert(!PageChecked(page)); | |
116 | ubifs_assert(!PagePrivate(page)); | |
117 | ||
118 | addr = kmap(page); | |
119 | ||
120 | block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT; | |
121 | beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT; | |
122 | if (block >= beyond) { | |
123 | /* Reading beyond inode */ | |
124 | SetPageChecked(page); | |
125 | memset(addr, 0, PAGE_CACHE_SIZE); | |
126 | goto out; | |
127 | } | |
128 | ||
129 | dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS); | |
130 | if (!dn) { | |
131 | err = -ENOMEM; | |
132 | goto error; | |
133 | } | |
134 | ||
135 | i = 0; | |
136 | while (1) { | |
137 | int ret; | |
138 | ||
139 | if (block >= beyond) { | |
140 | /* Reading beyond inode */ | |
141 | err = -ENOENT; | |
142 | memset(addr, 0, UBIFS_BLOCK_SIZE); | |
143 | } else { | |
144 | ret = read_block(inode, addr, block, dn); | |
145 | if (ret) { | |
146 | err = ret; | |
147 | if (err != -ENOENT) | |
148 | break; | |
ed382d58 AH |
149 | } else if (block + 1 == beyond) { |
150 | int dlen = le32_to_cpu(dn->size); | |
151 | int ilen = i_size & (UBIFS_BLOCK_SIZE - 1); | |
152 | ||
153 | if (ilen && ilen < dlen) | |
154 | memset(addr + ilen, 0, dlen - ilen); | |
1e51764a AB |
155 | } |
156 | } | |
157 | if (++i >= UBIFS_BLOCKS_PER_PAGE) | |
158 | break; | |
159 | block += 1; | |
160 | addr += UBIFS_BLOCK_SIZE; | |
161 | } | |
162 | if (err) { | |
163 | if (err == -ENOENT) { | |
164 | /* Not found, so it must be a hole */ | |
165 | SetPageChecked(page); | |
166 | dbg_gen("hole"); | |
167 | goto out_free; | |
168 | } | |
169 | ubifs_err("cannot read page %lu of inode %lu, error %d", | |
170 | page->index, inode->i_ino, err); | |
171 | goto error; | |
172 | } | |
173 | ||
174 | out_free: | |
175 | kfree(dn); | |
176 | out: | |
177 | SetPageUptodate(page); | |
178 | ClearPageError(page); | |
179 | flush_dcache_page(page); | |
180 | kunmap(page); | |
181 | return 0; | |
182 | ||
183 | error: | |
184 | kfree(dn); | |
185 | ClearPageUptodate(page); | |
186 | SetPageError(page); | |
187 | flush_dcache_page(page); | |
188 | kunmap(page); | |
189 | return err; | |
190 | } | |
191 | ||
192 | /** | |
193 | * release_new_page_budget - release budget of a new page. | |
194 | * @c: UBIFS file-system description object | |
195 | * | |
196 | * This is a helper function which releases budget corresponding to the budget | |
197 | * of one new page of data. | |
198 | */ | |
199 | static void release_new_page_budget(struct ubifs_info *c) | |
200 | { | |
201 | struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 }; | |
202 | ||
203 | ubifs_release_budget(c, &req); | |
204 | } | |
205 | ||
206 | /** | |
207 | * release_existing_page_budget - release budget of an existing page. | |
208 | * @c: UBIFS file-system description object | |
209 | * | |
210 | * This is a helper function which releases budget corresponding to the budget | |
211 | * of changing one one page of data which already exists on the flash media. | |
212 | */ | |
213 | static void release_existing_page_budget(struct ubifs_info *c) | |
214 | { | |
215 | struct ubifs_budget_req req = { .dd_growth = c->page_budget}; | |
216 | ||
217 | ubifs_release_budget(c, &req); | |
218 | } | |
219 | ||
220 | static int write_begin_slow(struct address_space *mapping, | |
54566b2c NP |
221 | loff_t pos, unsigned len, struct page **pagep, |
222 | unsigned flags) | |
1e51764a AB |
223 | { |
224 | struct inode *inode = mapping->host; | |
225 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
226 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
227 | struct ubifs_budget_req req = { .new_page = 1 }; | |
228 | int uninitialized_var(err), appending = !!(pos + len > inode->i_size); | |
229 | struct page *page; | |
230 | ||
231 | dbg_gen("ino %lu, pos %llu, len %u, i_size %lld", | |
232 | inode->i_ino, pos, len, inode->i_size); | |
233 | ||
234 | /* | |
235 | * At the slow path we have to budget before locking the page, because | |
236 | * budgeting may force write-back, which would wait on locked pages and | |
237 | * deadlock if we had the page locked. At this point we do not know | |
238 | * anything about the page, so assume that this is a new page which is | |
239 | * written to a hole. This corresponds to largest budget. Later the | |
240 | * budget will be amended if this is not true. | |
241 | */ | |
242 | if (appending) | |
243 | /* We are appending data, budget for inode change */ | |
244 | req.dirtied_ino = 1; | |
245 | ||
246 | err = ubifs_budget_space(c, &req); | |
247 | if (unlikely(err)) | |
248 | return err; | |
249 | ||
54566b2c | 250 | page = grab_cache_page_write_begin(mapping, index, flags); |
1e51764a AB |
251 | if (unlikely(!page)) { |
252 | ubifs_release_budget(c, &req); | |
253 | return -ENOMEM; | |
254 | } | |
255 | ||
256 | if (!PageUptodate(page)) { | |
7bbe5b5a | 257 | if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) |
1e51764a AB |
258 | SetPageChecked(page); |
259 | else { | |
260 | err = do_readpage(page); | |
261 | if (err) { | |
262 | unlock_page(page); | |
263 | page_cache_release(page); | |
264 | return err; | |
265 | } | |
266 | } | |
267 | ||
268 | SetPageUptodate(page); | |
269 | ClearPageError(page); | |
270 | } | |
271 | ||
272 | if (PagePrivate(page)) | |
273 | /* | |
274 | * The page is dirty, which means it was budgeted twice: | |
275 | * o first time the budget was allocated by the task which | |
276 | * made the page dirty and set the PG_private flag; | |
277 | * o and then we budgeted for it for the second time at the | |
278 | * very beginning of this function. | |
279 | * | |
280 | * So what we have to do is to release the page budget we | |
281 | * allocated. | |
282 | */ | |
283 | release_new_page_budget(c); | |
284 | else if (!PageChecked(page)) | |
285 | /* | |
286 | * We are changing a page which already exists on the media. | |
287 | * This means that changing the page does not make the amount | |
288 | * of indexing information larger, and this part of the budget | |
289 | * which we have already acquired may be released. | |
290 | */ | |
291 | ubifs_convert_page_budget(c); | |
292 | ||
293 | if (appending) { | |
294 | struct ubifs_inode *ui = ubifs_inode(inode); | |
295 | ||
296 | /* | |
297 | * 'ubifs_write_end()' is optimized from the fast-path part of | |
298 | * 'ubifs_write_begin()' and expects the @ui_mutex to be locked | |
299 | * if data is appended. | |
300 | */ | |
301 | mutex_lock(&ui->ui_mutex); | |
302 | if (ui->dirty) | |
303 | /* | |
304 | * The inode is dirty already, so we may free the | |
305 | * budget we allocated. | |
306 | */ | |
307 | ubifs_release_dirty_inode_budget(c, ui); | |
308 | } | |
309 | ||
310 | *pagep = page; | |
311 | return 0; | |
312 | } | |
313 | ||
314 | /** | |
315 | * allocate_budget - allocate budget for 'ubifs_write_begin()'. | |
316 | * @c: UBIFS file-system description object | |
317 | * @page: page to allocate budget for | |
318 | * @ui: UBIFS inode object the page belongs to | |
319 | * @appending: non-zero if the page is appended | |
320 | * | |
321 | * This is a helper function for 'ubifs_write_begin()' which allocates budget | |
322 | * for the operation. The budget is allocated differently depending on whether | |
323 | * this is appending, whether the page is dirty or not, and so on. This | |
324 | * function leaves the @ui->ui_mutex locked in case of appending. Returns zero | |
325 | * in case of success and %-ENOSPC in case of failure. | |
326 | */ | |
327 | static int allocate_budget(struct ubifs_info *c, struct page *page, | |
328 | struct ubifs_inode *ui, int appending) | |
329 | { | |
330 | struct ubifs_budget_req req = { .fast = 1 }; | |
331 | ||
332 | if (PagePrivate(page)) { | |
333 | if (!appending) | |
334 | /* | |
335 | * The page is dirty and we are not appending, which | |
336 | * means no budget is needed at all. | |
337 | */ | |
338 | return 0; | |
339 | ||
340 | mutex_lock(&ui->ui_mutex); | |
341 | if (ui->dirty) | |
342 | /* | |
343 | * The page is dirty and we are appending, so the inode | |
344 | * has to be marked as dirty. However, it is already | |
345 | * dirty, so we do not need any budget. We may return, | |
346 | * but @ui->ui_mutex hast to be left locked because we | |
347 | * should prevent write-back from flushing the inode | |
348 | * and freeing the budget. The lock will be released in | |
349 | * 'ubifs_write_end()'. | |
350 | */ | |
351 | return 0; | |
352 | ||
353 | /* | |
354 | * The page is dirty, we are appending, the inode is clean, so | |
355 | * we need to budget the inode change. | |
356 | */ | |
357 | req.dirtied_ino = 1; | |
358 | } else { | |
359 | if (PageChecked(page)) | |
360 | /* | |
361 | * The page corresponds to a hole and does not | |
362 | * exist on the media. So changing it makes | |
363 | * make the amount of indexing information | |
364 | * larger, and we have to budget for a new | |
365 | * page. | |
366 | */ | |
367 | req.new_page = 1; | |
368 | else | |
369 | /* | |
370 | * Not a hole, the change will not add any new | |
371 | * indexing information, budget for page | |
372 | * change. | |
373 | */ | |
374 | req.dirtied_page = 1; | |
375 | ||
376 | if (appending) { | |
377 | mutex_lock(&ui->ui_mutex); | |
378 | if (!ui->dirty) | |
379 | /* | |
380 | * The inode is clean but we will have to mark | |
381 | * it as dirty because we are appending. This | |
382 | * needs a budget. | |
383 | */ | |
384 | req.dirtied_ino = 1; | |
385 | } | |
386 | } | |
387 | ||
388 | return ubifs_budget_space(c, &req); | |
389 | } | |
390 | ||
391 | /* | |
392 | * This function is called when a page of data is going to be written. Since | |
393 | * the page of data will not necessarily go to the flash straight away, UBIFS | |
394 | * has to reserve space on the media for it, which is done by means of | |
395 | * budgeting. | |
396 | * | |
397 | * This is the hot-path of the file-system and we are trying to optimize it as | |
398 | * much as possible. For this reasons it is split on 2 parts - slow and fast. | |
399 | * | |
400 | * There many budgeting cases: | |
401 | * o a new page is appended - we have to budget for a new page and for | |
402 | * changing the inode; however, if the inode is already dirty, there is | |
403 | * no need to budget for it; | |
404 | * o an existing clean page is changed - we have budget for it; if the page | |
405 | * does not exist on the media (a hole), we have to budget for a new | |
406 | * page; otherwise, we may budget for changing an existing page; the | |
407 | * difference between these cases is that changing an existing page does | |
408 | * not introduce anything new to the FS indexing information, so it does | |
409 | * not grow, and smaller budget is acquired in this case; | |
410 | * o an existing dirty page is changed - no need to budget at all, because | |
411 | * the page budget has been acquired by earlier, when the page has been | |
412 | * marked dirty. | |
413 | * | |
414 | * UBIFS budgeting sub-system may force write-back if it thinks there is no | |
415 | * space to reserve. This imposes some locking restrictions and makes it | |
416 | * impossible to take into account the above cases, and makes it impossible to | |
417 | * optimize budgeting. | |
418 | * | |
419 | * The solution for this is that the fast path of 'ubifs_write_begin()' assumes | |
420 | * there is a plenty of flash space and the budget will be acquired quickly, | |
421 | * without forcing write-back. The slow path does not make this assumption. | |
422 | */ | |
423 | static int ubifs_write_begin(struct file *file, struct address_space *mapping, | |
424 | loff_t pos, unsigned len, unsigned flags, | |
425 | struct page **pagep, void **fsdata) | |
426 | { | |
427 | struct inode *inode = mapping->host; | |
428 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
429 | struct ubifs_inode *ui = ubifs_inode(inode); | |
430 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
431 | int uninitialized_var(err), appending = !!(pos + len > inode->i_size); | |
f55aa591 | 432 | int skipped_read = 0; |
1e51764a AB |
433 | struct page *page; |
434 | ||
1e51764a AB |
435 | ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size); |
436 | ||
437 | if (unlikely(c->ro_media)) | |
438 | return -EROFS; | |
439 | ||
440 | /* Try out the fast-path part first */ | |
54566b2c | 441 | page = grab_cache_page_write_begin(mapping, index, flags); |
1e51764a AB |
442 | if (unlikely(!page)) |
443 | return -ENOMEM; | |
444 | ||
445 | if (!PageUptodate(page)) { | |
446 | /* The page is not loaded from the flash */ | |
f55aa591 | 447 | if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) { |
1e51764a AB |
448 | /* |
449 | * We change whole page so no need to load it. But we | |
450 | * have to set the @PG_checked flag to make the further | |
873a64c7 AB |
451 | * code know that the page is new. This might be not |
452 | * true, but it is better to budget more than to read | |
453 | * the page from the media. | |
1e51764a AB |
454 | */ |
455 | SetPageChecked(page); | |
f55aa591 AH |
456 | skipped_read = 1; |
457 | } else { | |
1e51764a AB |
458 | err = do_readpage(page); |
459 | if (err) { | |
460 | unlock_page(page); | |
461 | page_cache_release(page); | |
462 | return err; | |
463 | } | |
464 | } | |
465 | ||
466 | SetPageUptodate(page); | |
467 | ClearPageError(page); | |
468 | } | |
469 | ||
470 | err = allocate_budget(c, page, ui, appending); | |
471 | if (unlikely(err)) { | |
472 | ubifs_assert(err == -ENOSPC); | |
f55aa591 AH |
473 | /* |
474 | * If we skipped reading the page because we were going to | |
475 | * write all of it, then it is not up to date. | |
476 | */ | |
477 | if (skipped_read) { | |
478 | ClearPageChecked(page); | |
479 | ClearPageUptodate(page); | |
480 | } | |
1e51764a AB |
481 | /* |
482 | * Budgeting failed which means it would have to force | |
483 | * write-back but didn't, because we set the @fast flag in the | |
484 | * request. Write-back cannot be done now, while we have the | |
485 | * page locked, because it would deadlock. Unlock and free | |
486 | * everything and fall-back to slow-path. | |
487 | */ | |
488 | if (appending) { | |
489 | ubifs_assert(mutex_is_locked(&ui->ui_mutex)); | |
490 | mutex_unlock(&ui->ui_mutex); | |
491 | } | |
492 | unlock_page(page); | |
493 | page_cache_release(page); | |
494 | ||
54566b2c | 495 | return write_begin_slow(mapping, pos, len, pagep, flags); |
1e51764a AB |
496 | } |
497 | ||
498 | /* | |
873a64c7 AB |
499 | * Whee, we acquired budgeting quickly - without involving |
500 | * garbage-collection, committing or forcing write-back. We return | |
1e51764a AB |
501 | * with @ui->ui_mutex locked if we are appending pages, and unlocked |
502 | * otherwise. This is an optimization (slightly hacky though). | |
503 | */ | |
504 | *pagep = page; | |
505 | return 0; | |
506 | ||
507 | } | |
508 | ||
509 | /** | |
510 | * cancel_budget - cancel budget. | |
511 | * @c: UBIFS file-system description object | |
512 | * @page: page to cancel budget for | |
513 | * @ui: UBIFS inode object the page belongs to | |
514 | * @appending: non-zero if the page is appended | |
515 | * | |
516 | * This is a helper function for a page write operation. It unlocks the | |
517 | * @ui->ui_mutex in case of appending. | |
518 | */ | |
519 | static void cancel_budget(struct ubifs_info *c, struct page *page, | |
520 | struct ubifs_inode *ui, int appending) | |
521 | { | |
522 | if (appending) { | |
523 | if (!ui->dirty) | |
524 | ubifs_release_dirty_inode_budget(c, ui); | |
525 | mutex_unlock(&ui->ui_mutex); | |
526 | } | |
527 | if (!PagePrivate(page)) { | |
528 | if (PageChecked(page)) | |
529 | release_new_page_budget(c); | |
530 | else | |
531 | release_existing_page_budget(c); | |
532 | } | |
533 | } | |
534 | ||
535 | static int ubifs_write_end(struct file *file, struct address_space *mapping, | |
536 | loff_t pos, unsigned len, unsigned copied, | |
537 | struct page *page, void *fsdata) | |
538 | { | |
539 | struct inode *inode = mapping->host; | |
540 | struct ubifs_inode *ui = ubifs_inode(inode); | |
541 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
542 | loff_t end_pos = pos + len; | |
543 | int appending = !!(end_pos > inode->i_size); | |
544 | ||
545 | dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld", | |
546 | inode->i_ino, pos, page->index, len, copied, inode->i_size); | |
547 | ||
548 | if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) { | |
549 | /* | |
550 | * VFS copied less data to the page that it intended and | |
551 | * declared in its '->write_begin()' call via the @len | |
552 | * argument. If the page was not up-to-date, and @len was | |
553 | * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did | |
554 | * not load it from the media (for optimization reasons). This | |
555 | * means that part of the page contains garbage. So read the | |
556 | * page now. | |
557 | */ | |
558 | dbg_gen("copied %d instead of %d, read page and repeat", | |
559 | copied, len); | |
560 | cancel_budget(c, page, ui, appending); | |
561 | ||
562 | /* | |
563 | * Return 0 to force VFS to repeat the whole operation, or the | |
873a64c7 | 564 | * error code if 'do_readpage()' fails. |
1e51764a AB |
565 | */ |
566 | copied = do_readpage(page); | |
567 | goto out; | |
568 | } | |
569 | ||
570 | if (!PagePrivate(page)) { | |
571 | SetPagePrivate(page); | |
572 | atomic_long_inc(&c->dirty_pg_cnt); | |
573 | __set_page_dirty_nobuffers(page); | |
574 | } | |
575 | ||
576 | if (appending) { | |
577 | i_size_write(inode, end_pos); | |
578 | ui->ui_size = end_pos; | |
579 | /* | |
580 | * Note, we do not set @I_DIRTY_PAGES (which means that the | |
581 | * inode has dirty pages), this has been done in | |
582 | * '__set_page_dirty_nobuffers()'. | |
583 | */ | |
584 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
585 | ubifs_assert(mutex_is_locked(&ui->ui_mutex)); | |
586 | mutex_unlock(&ui->ui_mutex); | |
587 | } | |
588 | ||
589 | out: | |
590 | unlock_page(page); | |
591 | page_cache_release(page); | |
592 | return copied; | |
593 | } | |
594 | ||
4793e7c5 AH |
595 | /** |
596 | * populate_page - copy data nodes into a page for bulk-read. | |
597 | * @c: UBIFS file-system description object | |
598 | * @page: page | |
599 | * @bu: bulk-read information | |
600 | * @n: next zbranch slot | |
601 | * | |
602 | * This function returns %0 on success and a negative error code on failure. | |
603 | */ | |
604 | static int populate_page(struct ubifs_info *c, struct page *page, | |
605 | struct bu_info *bu, int *n) | |
606 | { | |
5c0013c1 | 607 | int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0; |
4793e7c5 AH |
608 | struct inode *inode = page->mapping->host; |
609 | loff_t i_size = i_size_read(inode); | |
610 | unsigned int page_block; | |
611 | void *addr, *zaddr; | |
612 | pgoff_t end_index; | |
613 | ||
614 | dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx", | |
615 | inode->i_ino, page->index, i_size, page->flags); | |
616 | ||
617 | addr = zaddr = kmap(page); | |
618 | ||
ed382d58 | 619 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; |
4793e7c5 | 620 | if (!i_size || page->index > end_index) { |
5c0013c1 | 621 | hole = 1; |
4793e7c5 AH |
622 | memset(addr, 0, PAGE_CACHE_SIZE); |
623 | goto out_hole; | |
624 | } | |
625 | ||
626 | page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT; | |
627 | while (1) { | |
628 | int err, len, out_len, dlen; | |
629 | ||
5c0013c1 AH |
630 | if (nn >= bu->cnt) { |
631 | hole = 1; | |
4793e7c5 | 632 | memset(addr, 0, UBIFS_BLOCK_SIZE); |
5c0013c1 | 633 | } else if (key_block(c, &bu->zbranch[nn].key) == page_block) { |
4793e7c5 AH |
634 | struct ubifs_data_node *dn; |
635 | ||
636 | dn = bu->buf + (bu->zbranch[nn].offs - offs); | |
637 | ||
0ecb9529 | 638 | ubifs_assert(le64_to_cpu(dn->ch.sqnum) > |
4793e7c5 AH |
639 | ubifs_inode(inode)->creat_sqnum); |
640 | ||
641 | len = le32_to_cpu(dn->size); | |
642 | if (len <= 0 || len > UBIFS_BLOCK_SIZE) | |
643 | goto out_err; | |
644 | ||
645 | dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ; | |
646 | out_len = UBIFS_BLOCK_SIZE; | |
647 | err = ubifs_decompress(&dn->data, dlen, addr, &out_len, | |
648 | le16_to_cpu(dn->compr_type)); | |
649 | if (err || len != out_len) | |
650 | goto out_err; | |
651 | ||
652 | if (len < UBIFS_BLOCK_SIZE) | |
653 | memset(addr + len, 0, UBIFS_BLOCK_SIZE - len); | |
654 | ||
655 | nn += 1; | |
4793e7c5 | 656 | read = (i << UBIFS_BLOCK_SHIFT) + len; |
5c0013c1 AH |
657 | } else if (key_block(c, &bu->zbranch[nn].key) < page_block) { |
658 | nn += 1; | |
659 | continue; | |
660 | } else { | |
661 | hole = 1; | |
662 | memset(addr, 0, UBIFS_BLOCK_SIZE); | |
4793e7c5 AH |
663 | } |
664 | if (++i >= UBIFS_BLOCKS_PER_PAGE) | |
665 | break; | |
666 | addr += UBIFS_BLOCK_SIZE; | |
667 | page_block += 1; | |
668 | } | |
669 | ||
670 | if (end_index == page->index) { | |
671 | int len = i_size & (PAGE_CACHE_SIZE - 1); | |
672 | ||
ed382d58 | 673 | if (len && len < read) |
4793e7c5 AH |
674 | memset(zaddr + len, 0, read - len); |
675 | } | |
676 | ||
677 | out_hole: | |
678 | if (hole) { | |
679 | SetPageChecked(page); | |
680 | dbg_gen("hole"); | |
681 | } | |
682 | ||
683 | SetPageUptodate(page); | |
684 | ClearPageError(page); | |
685 | flush_dcache_page(page); | |
686 | kunmap(page); | |
687 | *n = nn; | |
688 | return 0; | |
689 | ||
690 | out_err: | |
691 | ClearPageUptodate(page); | |
692 | SetPageError(page); | |
693 | flush_dcache_page(page); | |
694 | kunmap(page); | |
695 | ubifs_err("bad data node (block %u, inode %lu)", | |
696 | page_block, inode->i_ino); | |
697 | return -EINVAL; | |
698 | } | |
699 | ||
700 | /** | |
701 | * ubifs_do_bulk_read - do bulk-read. | |
702 | * @c: UBIFS file-system description object | |
6c0c42cd AB |
703 | * @bu: bulk-read information |
704 | * @page1: first page to read | |
4793e7c5 AH |
705 | * |
706 | * This function returns %1 if the bulk-read is done, otherwise %0 is returned. | |
707 | */ | |
6c0c42cd AB |
708 | static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu, |
709 | struct page *page1) | |
4793e7c5 AH |
710 | { |
711 | pgoff_t offset = page1->index, end_index; | |
712 | struct address_space *mapping = page1->mapping; | |
713 | struct inode *inode = mapping->host; | |
714 | struct ubifs_inode *ui = ubifs_inode(inode); | |
4793e7c5 | 715 | int err, page_idx, page_cnt, ret = 0, n = 0; |
6c0c42cd | 716 | int allocate = bu->buf ? 0 : 1; |
4793e7c5 AH |
717 | loff_t isize; |
718 | ||
4793e7c5 AH |
719 | err = ubifs_tnc_get_bu_keys(c, bu); |
720 | if (err) | |
721 | goto out_warn; | |
722 | ||
723 | if (bu->eof) { | |
724 | /* Turn off bulk-read at the end of the file */ | |
725 | ui->read_in_a_row = 1; | |
726 | ui->bulk_read = 0; | |
727 | } | |
728 | ||
729 | page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT; | |
730 | if (!page_cnt) { | |
731 | /* | |
732 | * This happens when there are multiple blocks per page and the | |
733 | * blocks for the first page we are looking for, are not | |
734 | * together. If all the pages were like this, bulk-read would | |
735 | * reduce performance, so we turn it off for a while. | |
736 | */ | |
6c0c42cd | 737 | goto out_bu_off; |
4793e7c5 AH |
738 | } |
739 | ||
740 | if (bu->cnt) { | |
6c0c42cd AB |
741 | if (allocate) { |
742 | /* | |
743 | * Allocate bulk-read buffer depending on how many data | |
744 | * nodes we are going to read. | |
745 | */ | |
746 | bu->buf_len = bu->zbranch[bu->cnt - 1].offs + | |
747 | bu->zbranch[bu->cnt - 1].len - | |
748 | bu->zbranch[0].offs; | |
749 | ubifs_assert(bu->buf_len > 0); | |
750 | ubifs_assert(bu->buf_len <= c->leb_size); | |
751 | bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN); | |
752 | if (!bu->buf) | |
753 | goto out_bu_off; | |
754 | } | |
755 | ||
4793e7c5 AH |
756 | err = ubifs_tnc_bulk_read(c, bu); |
757 | if (err) | |
758 | goto out_warn; | |
759 | } | |
760 | ||
761 | err = populate_page(c, page1, bu, &n); | |
762 | if (err) | |
763 | goto out_warn; | |
764 | ||
765 | unlock_page(page1); | |
766 | ret = 1; | |
767 | ||
768 | isize = i_size_read(inode); | |
769 | if (isize == 0) | |
770 | goto out_free; | |
771 | end_index = ((isize - 1) >> PAGE_CACHE_SHIFT); | |
772 | ||
773 | for (page_idx = 1; page_idx < page_cnt; page_idx++) { | |
774 | pgoff_t page_offset = offset + page_idx; | |
775 | struct page *page; | |
776 | ||
777 | if (page_offset > end_index) | |
778 | break; | |
779 | page = find_or_create_page(mapping, page_offset, | |
780 | GFP_NOFS | __GFP_COLD); | |
781 | if (!page) | |
782 | break; | |
783 | if (!PageUptodate(page)) | |
784 | err = populate_page(c, page, bu, &n); | |
785 | unlock_page(page); | |
786 | page_cache_release(page); | |
787 | if (err) | |
788 | break; | |
789 | } | |
790 | ||
791 | ui->last_page_read = offset + page_idx - 1; | |
792 | ||
793 | out_free: | |
6c0c42cd AB |
794 | if (allocate) |
795 | kfree(bu->buf); | |
4793e7c5 AH |
796 | return ret; |
797 | ||
798 | out_warn: | |
799 | ubifs_warn("ignoring error %d and skipping bulk-read", err); | |
800 | goto out_free; | |
6c0c42cd AB |
801 | |
802 | out_bu_off: | |
803 | ui->read_in_a_row = ui->bulk_read = 0; | |
804 | goto out_free; | |
4793e7c5 AH |
805 | } |
806 | ||
807 | /** | |
808 | * ubifs_bulk_read - determine whether to bulk-read and, if so, do it. | |
809 | * @page: page from which to start bulk-read. | |
810 | * | |
811 | * Some flash media are capable of reading sequentially at faster rates. UBIFS | |
812 | * bulk-read facility is designed to take advantage of that, by reading in one | |
813 | * go consecutive data nodes that are also located consecutively in the same | |
814 | * LEB. This function returns %1 if a bulk-read is done and %0 otherwise. | |
815 | */ | |
816 | static int ubifs_bulk_read(struct page *page) | |
817 | { | |
818 | struct inode *inode = page->mapping->host; | |
819 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
820 | struct ubifs_inode *ui = ubifs_inode(inode); | |
821 | pgoff_t index = page->index, last_page_read = ui->last_page_read; | |
6c0c42cd | 822 | struct bu_info *bu; |
3477d204 | 823 | int err = 0, allocated = 0; |
4793e7c5 AH |
824 | |
825 | ui->last_page_read = index; | |
4793e7c5 AH |
826 | if (!c->bulk_read) |
827 | return 0; | |
6c0c42cd | 828 | |
4793e7c5 | 829 | /* |
3477d204 AB |
830 | * Bulk-read is protected by @ui->ui_mutex, but it is an optimization, |
831 | * so don't bother if we cannot lock the mutex. | |
4793e7c5 AH |
832 | */ |
833 | if (!mutex_trylock(&ui->ui_mutex)) | |
834 | return 0; | |
6c0c42cd | 835 | |
4793e7c5 AH |
836 | if (index != last_page_read + 1) { |
837 | /* Turn off bulk-read if we stop reading sequentially */ | |
838 | ui->read_in_a_row = 1; | |
839 | if (ui->bulk_read) | |
840 | ui->bulk_read = 0; | |
841 | goto out_unlock; | |
842 | } | |
6c0c42cd | 843 | |
4793e7c5 AH |
844 | if (!ui->bulk_read) { |
845 | ui->read_in_a_row += 1; | |
846 | if (ui->read_in_a_row < 3) | |
847 | goto out_unlock; | |
848 | /* Three reads in a row, so switch on bulk-read */ | |
849 | ui->bulk_read = 1; | |
850 | } | |
6c0c42cd | 851 | |
3477d204 AB |
852 | /* |
853 | * If possible, try to use pre-allocated bulk-read information, which | |
854 | * is protected by @c->bu_mutex. | |
855 | */ | |
856 | if (mutex_trylock(&c->bu_mutex)) | |
857 | bu = &c->bu; | |
858 | else { | |
859 | bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN); | |
860 | if (!bu) | |
861 | goto out_unlock; | |
862 | ||
863 | bu->buf = NULL; | |
864 | allocated = 1; | |
865 | } | |
6c0c42cd | 866 | |
6c0c42cd AB |
867 | bu->buf_len = c->max_bu_buf_len; |
868 | data_key_init(c, &bu->key, inode->i_ino, | |
869 | page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT); | |
6c0c42cd | 870 | err = ubifs_do_bulk_read(c, bu, page); |
3477d204 AB |
871 | |
872 | if (!allocated) | |
873 | mutex_unlock(&c->bu_mutex); | |
874 | else | |
875 | kfree(bu); | |
6c0c42cd | 876 | |
4793e7c5 AH |
877 | out_unlock: |
878 | mutex_unlock(&ui->ui_mutex); | |
6c0c42cd | 879 | return err; |
4793e7c5 AH |
880 | } |
881 | ||
1e51764a AB |
882 | static int ubifs_readpage(struct file *file, struct page *page) |
883 | { | |
4793e7c5 AH |
884 | if (ubifs_bulk_read(page)) |
885 | return 0; | |
1e51764a AB |
886 | do_readpage(page); |
887 | unlock_page(page); | |
888 | return 0; | |
889 | } | |
890 | ||
891 | static int do_writepage(struct page *page, int len) | |
892 | { | |
893 | int err = 0, i, blen; | |
894 | unsigned int block; | |
895 | void *addr; | |
896 | union ubifs_key key; | |
897 | struct inode *inode = page->mapping->host; | |
898 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
899 | ||
900 | #ifdef UBIFS_DEBUG | |
901 | spin_lock(&ui->ui_lock); | |
902 | ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE); | |
903 | spin_unlock(&ui->ui_lock); | |
904 | #endif | |
905 | ||
906 | /* Update radix tree tags */ | |
907 | set_page_writeback(page); | |
908 | ||
909 | addr = kmap(page); | |
910 | block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT; | |
911 | i = 0; | |
912 | while (len) { | |
913 | blen = min_t(int, len, UBIFS_BLOCK_SIZE); | |
914 | data_key_init(c, &key, inode->i_ino, block); | |
915 | err = ubifs_jnl_write_data(c, inode, &key, addr, blen); | |
916 | if (err) | |
917 | break; | |
918 | if (++i >= UBIFS_BLOCKS_PER_PAGE) | |
919 | break; | |
920 | block += 1; | |
921 | addr += blen; | |
922 | len -= blen; | |
923 | } | |
924 | if (err) { | |
925 | SetPageError(page); | |
926 | ubifs_err("cannot write page %lu of inode %lu, error %d", | |
927 | page->index, inode->i_ino, err); | |
928 | ubifs_ro_mode(c, err); | |
929 | } | |
930 | ||
931 | ubifs_assert(PagePrivate(page)); | |
932 | if (PageChecked(page)) | |
933 | release_new_page_budget(c); | |
934 | else | |
935 | release_existing_page_budget(c); | |
936 | ||
937 | atomic_long_dec(&c->dirty_pg_cnt); | |
938 | ClearPagePrivate(page); | |
939 | ClearPageChecked(page); | |
940 | ||
941 | kunmap(page); | |
942 | unlock_page(page); | |
943 | end_page_writeback(page); | |
944 | return err; | |
945 | } | |
946 | ||
947 | /* | |
948 | * When writing-back dirty inodes, VFS first writes-back pages belonging to the | |
949 | * inode, then the inode itself. For UBIFS this may cause a problem. Consider a | |
950 | * situation when a we have an inode with size 0, then a megabyte of data is | |
951 | * appended to the inode, then write-back starts and flushes some amount of the | |
952 | * dirty pages, the journal becomes full, commit happens and finishes, and then | |
953 | * an unclean reboot happens. When the file system is mounted next time, the | |
954 | * inode size would still be 0, but there would be many pages which are beyond | |
955 | * the inode size, they would be indexed and consume flash space. Because the | |
956 | * journal has been committed, the replay would not be able to detect this | |
957 | * situation and correct the inode size. This means UBIFS would have to scan | |
958 | * whole index and correct all inode sizes, which is long an unacceptable. | |
959 | * | |
960 | * To prevent situations like this, UBIFS writes pages back only if they are | |
7d4e9ccb | 961 | * within the last synchronized inode size, i.e. the size which has been |
1e51764a AB |
962 | * written to the flash media last time. Otherwise, UBIFS forces inode |
963 | * write-back, thus making sure the on-flash inode contains current inode size, | |
964 | * and then keeps writing pages back. | |
965 | * | |
966 | * Some locking issues explanation. 'ubifs_writepage()' first is called with | |
967 | * the page locked, and it locks @ui_mutex. However, write-back does take inode | |
968 | * @i_mutex, which means other VFS operations may be run on this inode at the | |
969 | * same time. And the problematic one is truncation to smaller size, from where | |
2c27c65e | 970 | * we have to call 'truncate_setsize()', which first changes @inode->i_size, then |
1e51764a | 971 | * drops the truncated pages. And while dropping the pages, it takes the page |
2c27c65e | 972 | * lock. This means that 'do_truncation()' cannot call 'truncate_setsize()' with |
1e51764a AB |
973 | * @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'. This |
974 | * means that @inode->i_size is changed while @ui_mutex is unlocked. | |
975 | * | |
2c27c65e CH |
976 | * XXX(truncate): with the new truncate sequence this is not true anymore, |
977 | * and the calls to truncate_setsize can be move around freely. They should | |
978 | * be moved to the very end of the truncate sequence. | |
15c6fd97 | 979 | * |
1e51764a AB |
980 | * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond |
981 | * inode size. How do we do this if @inode->i_size may became smaller while we | |
982 | * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the | |
983 | * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size | |
984 | * internally and updates it under @ui_mutex. | |
985 | * | |
986 | * Q: why we do not worry that if we race with truncation, we may end up with a | |
987 | * situation when the inode is truncated while we are in the middle of | |
988 | * 'do_writepage()', so we do write beyond inode size? | |
989 | * A: If we are in the middle of 'do_writepage()', truncation would be locked | |
990 | * on the page lock and it would not write the truncated inode node to the | |
991 | * journal before we have finished. | |
992 | */ | |
993 | static int ubifs_writepage(struct page *page, struct writeback_control *wbc) | |
994 | { | |
995 | struct inode *inode = page->mapping->host; | |
996 | struct ubifs_inode *ui = ubifs_inode(inode); | |
997 | loff_t i_size = i_size_read(inode), synced_i_size; | |
998 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
999 | int err, len = i_size & (PAGE_CACHE_SIZE - 1); | |
1000 | void *kaddr; | |
1001 | ||
1002 | dbg_gen("ino %lu, pg %lu, pg flags %#lx", | |
1003 | inode->i_ino, page->index, page->flags); | |
1004 | ubifs_assert(PagePrivate(page)); | |
1005 | ||
1006 | /* Is the page fully outside @i_size? (truncate in progress) */ | |
1007 | if (page->index > end_index || (page->index == end_index && !len)) { | |
1008 | err = 0; | |
1009 | goto out_unlock; | |
1010 | } | |
1011 | ||
1012 | spin_lock(&ui->ui_lock); | |
1013 | synced_i_size = ui->synced_i_size; | |
1014 | spin_unlock(&ui->ui_lock); | |
1015 | ||
1016 | /* Is the page fully inside @i_size? */ | |
1017 | if (page->index < end_index) { | |
1018 | if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) { | |
a9185b41 | 1019 | err = inode->i_sb->s_op->write_inode(inode, NULL); |
1e51764a AB |
1020 | if (err) |
1021 | goto out_unlock; | |
1022 | /* | |
1023 | * The inode has been written, but the write-buffer has | |
1024 | * not been synchronized, so in case of an unclean | |
1025 | * reboot we may end up with some pages beyond inode | |
1026 | * size, but they would be in the journal (because | |
1027 | * commit flushes write buffers) and recovery would deal | |
1028 | * with this. | |
1029 | */ | |
1030 | } | |
1031 | return do_writepage(page, PAGE_CACHE_SIZE); | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * The page straddles @i_size. It must be zeroed out on each and every | |
1036 | * writepage invocation because it may be mmapped. "A file is mapped | |
1037 | * in multiples of the page size. For a file that is not a multiple of | |
1038 | * the page size, the remaining memory is zeroed when mapped, and | |
1039 | * writes to that region are not written out to the file." | |
1040 | */ | |
1041 | kaddr = kmap_atomic(page, KM_USER0); | |
1042 | memset(kaddr + len, 0, PAGE_CACHE_SIZE - len); | |
1043 | flush_dcache_page(page); | |
1044 | kunmap_atomic(kaddr, KM_USER0); | |
1045 | ||
1046 | if (i_size > synced_i_size) { | |
a9185b41 | 1047 | err = inode->i_sb->s_op->write_inode(inode, NULL); |
1e51764a AB |
1048 | if (err) |
1049 | goto out_unlock; | |
1050 | } | |
1051 | ||
1052 | return do_writepage(page, len); | |
1053 | ||
1054 | out_unlock: | |
1055 | unlock_page(page); | |
1056 | return err; | |
1057 | } | |
1058 | ||
1059 | /** | |
1060 | * do_attr_changes - change inode attributes. | |
1061 | * @inode: inode to change attributes for | |
1062 | * @attr: describes attributes to change | |
1063 | */ | |
1064 | static void do_attr_changes(struct inode *inode, const struct iattr *attr) | |
1065 | { | |
1066 | if (attr->ia_valid & ATTR_UID) | |
1067 | inode->i_uid = attr->ia_uid; | |
1068 | if (attr->ia_valid & ATTR_GID) | |
1069 | inode->i_gid = attr->ia_gid; | |
1070 | if (attr->ia_valid & ATTR_ATIME) | |
1071 | inode->i_atime = timespec_trunc(attr->ia_atime, | |
1072 | inode->i_sb->s_time_gran); | |
1073 | if (attr->ia_valid & ATTR_MTIME) | |
1074 | inode->i_mtime = timespec_trunc(attr->ia_mtime, | |
1075 | inode->i_sb->s_time_gran); | |
1076 | if (attr->ia_valid & ATTR_CTIME) | |
1077 | inode->i_ctime = timespec_trunc(attr->ia_ctime, | |
1078 | inode->i_sb->s_time_gran); | |
1079 | if (attr->ia_valid & ATTR_MODE) { | |
1080 | umode_t mode = attr->ia_mode; | |
1081 | ||
1082 | if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) | |
1083 | mode &= ~S_ISGID; | |
1084 | inode->i_mode = mode; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | /** | |
1089 | * do_truncation - truncate an inode. | |
1090 | * @c: UBIFS file-system description object | |
1091 | * @inode: inode to truncate | |
1092 | * @attr: inode attribute changes description | |
1093 | * | |
1094 | * This function implements VFS '->setattr()' call when the inode is truncated | |
1095 | * to a smaller size. Returns zero in case of success and a negative error code | |
1096 | * in case of failure. | |
1097 | */ | |
1098 | static int do_truncation(struct ubifs_info *c, struct inode *inode, | |
1099 | const struct iattr *attr) | |
1100 | { | |
1101 | int err; | |
1102 | struct ubifs_budget_req req; | |
1103 | loff_t old_size = inode->i_size, new_size = attr->ia_size; | |
04da11bf | 1104 | int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1; |
1e51764a AB |
1105 | struct ubifs_inode *ui = ubifs_inode(inode); |
1106 | ||
1107 | dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size); | |
1108 | memset(&req, 0, sizeof(struct ubifs_budget_req)); | |
1109 | ||
1110 | /* | |
1111 | * If this is truncation to a smaller size, and we do not truncate on a | |
1112 | * block boundary, budget for changing one data block, because the last | |
1113 | * block will be re-written. | |
1114 | */ | |
1115 | if (new_size & (UBIFS_BLOCK_SIZE - 1)) | |
1116 | req.dirtied_page = 1; | |
1117 | ||
1118 | req.dirtied_ino = 1; | |
1119 | /* A funny way to budget for truncation node */ | |
1120 | req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ; | |
1121 | err = ubifs_budget_space(c, &req); | |
04da11bf AB |
1122 | if (err) { |
1123 | /* | |
1124 | * Treat truncations to zero as deletion and always allow them, | |
1125 | * just like we do for '->unlink()'. | |
1126 | */ | |
1127 | if (new_size || err != -ENOSPC) | |
1128 | return err; | |
1129 | budgeted = 0; | |
1130 | } | |
1e51764a | 1131 | |
2c27c65e | 1132 | truncate_setsize(inode, new_size); |
1e51764a AB |
1133 | |
1134 | if (offset) { | |
1135 | pgoff_t index = new_size >> PAGE_CACHE_SHIFT; | |
1136 | struct page *page; | |
1137 | ||
1138 | page = find_lock_page(inode->i_mapping, index); | |
1139 | if (page) { | |
1140 | if (PageDirty(page)) { | |
1141 | /* | |
1142 | * 'ubifs_jnl_truncate()' will try to truncate | |
1143 | * the last data node, but it contains | |
1144 | * out-of-date data because the page is dirty. | |
1145 | * Write the page now, so that | |
1146 | * 'ubifs_jnl_truncate()' will see an already | |
1147 | * truncated (and up to date) data node. | |
1148 | */ | |
1149 | ubifs_assert(PagePrivate(page)); | |
1150 | ||
1151 | clear_page_dirty_for_io(page); | |
1152 | if (UBIFS_BLOCKS_PER_PAGE_SHIFT) | |
1153 | offset = new_size & | |
1154 | (PAGE_CACHE_SIZE - 1); | |
1155 | err = do_writepage(page, offset); | |
1156 | page_cache_release(page); | |
1157 | if (err) | |
1158 | goto out_budg; | |
1159 | /* | |
1160 | * We could now tell 'ubifs_jnl_truncate()' not | |
1161 | * to read the last block. | |
1162 | */ | |
1163 | } else { | |
1164 | /* | |
1165 | * We could 'kmap()' the page and pass the data | |
1166 | * to 'ubifs_jnl_truncate()' to save it from | |
1167 | * having to read it. | |
1168 | */ | |
1169 | unlock_page(page); | |
1170 | page_cache_release(page); | |
1171 | } | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | mutex_lock(&ui->ui_mutex); | |
1176 | ui->ui_size = inode->i_size; | |
1177 | /* Truncation changes inode [mc]time */ | |
1178 | inode->i_mtime = inode->i_ctime = ubifs_current_time(inode); | |
873a64c7 | 1179 | /* Other attributes may be changed at the same time as well */ |
1e51764a | 1180 | do_attr_changes(inode, attr); |
1e51764a AB |
1181 | err = ubifs_jnl_truncate(c, inode, old_size, new_size); |
1182 | mutex_unlock(&ui->ui_mutex); | |
873a64c7 | 1183 | |
1e51764a | 1184 | out_budg: |
04da11bf AB |
1185 | if (budgeted) |
1186 | ubifs_release_budget(c, &req); | |
1187 | else { | |
1188 | c->nospace = c->nospace_rp = 0; | |
1189 | smp_wmb(); | |
1190 | } | |
1e51764a AB |
1191 | return err; |
1192 | } | |
1193 | ||
1194 | /** | |
1195 | * do_setattr - change inode attributes. | |
1196 | * @c: UBIFS file-system description object | |
1197 | * @inode: inode to change attributes for | |
1198 | * @attr: inode attribute changes description | |
1199 | * | |
1200 | * This function implements VFS '->setattr()' call for all cases except | |
1201 | * truncations to smaller size. Returns zero in case of success and a negative | |
1202 | * error code in case of failure. | |
1203 | */ | |
1204 | static int do_setattr(struct ubifs_info *c, struct inode *inode, | |
1205 | const struct iattr *attr) | |
1206 | { | |
1207 | int err, release; | |
1208 | loff_t new_size = attr->ia_size; | |
1209 | struct ubifs_inode *ui = ubifs_inode(inode); | |
1210 | struct ubifs_budget_req req = { .dirtied_ino = 1, | |
dab4b4d2 | 1211 | .dirtied_ino_d = ALIGN(ui->data_len, 8) }; |
1e51764a AB |
1212 | |
1213 | err = ubifs_budget_space(c, &req); | |
1214 | if (err) | |
1215 | return err; | |
1216 | ||
1217 | if (attr->ia_valid & ATTR_SIZE) { | |
1218 | dbg_gen("size %lld -> %lld", inode->i_size, new_size); | |
2c27c65e | 1219 | truncate_setsize(inode, new_size); |
1e51764a AB |
1220 | } |
1221 | ||
1222 | mutex_lock(&ui->ui_mutex); | |
1223 | if (attr->ia_valid & ATTR_SIZE) { | |
1224 | /* Truncation changes inode [mc]time */ | |
1225 | inode->i_mtime = inode->i_ctime = ubifs_current_time(inode); | |
2c27c65e | 1226 | /* 'truncate_setsize()' changed @i_size, update @ui_size */ |
1e51764a AB |
1227 | ui->ui_size = inode->i_size; |
1228 | } | |
1229 | ||
1230 | do_attr_changes(inode, attr); | |
1231 | ||
1232 | release = ui->dirty; | |
1233 | if (attr->ia_valid & ATTR_SIZE) | |
1234 | /* | |
1235 | * Inode length changed, so we have to make sure | |
1236 | * @I_DIRTY_DATASYNC is set. | |
1237 | */ | |
1238 | __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
1239 | else | |
1240 | mark_inode_dirty_sync(inode); | |
1241 | mutex_unlock(&ui->ui_mutex); | |
1242 | ||
1243 | if (release) | |
1244 | ubifs_release_budget(c, &req); | |
1245 | if (IS_SYNC(inode)) | |
a9185b41 | 1246 | err = inode->i_sb->s_op->write_inode(inode, NULL); |
1e51764a | 1247 | return err; |
1e51764a AB |
1248 | } |
1249 | ||
1250 | int ubifs_setattr(struct dentry *dentry, struct iattr *attr) | |
1251 | { | |
1252 | int err; | |
1253 | struct inode *inode = dentry->d_inode; | |
1254 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
1255 | ||
7d32c2bb AB |
1256 | dbg_gen("ino %lu, mode %#x, ia_valid %#x", |
1257 | inode->i_ino, inode->i_mode, attr->ia_valid); | |
1e51764a AB |
1258 | err = inode_change_ok(inode, attr); |
1259 | if (err) | |
1260 | return err; | |
1261 | ||
1262 | err = dbg_check_synced_i_size(inode); | |
1263 | if (err) | |
1264 | return err; | |
1265 | ||
1266 | if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size) | |
1267 | /* Truncation to a smaller size */ | |
1268 | err = do_truncation(c, inode, attr); | |
1269 | else | |
1270 | err = do_setattr(c, inode, attr); | |
1271 | ||
1272 | return err; | |
1273 | } | |
1274 | ||
1275 | static void ubifs_invalidatepage(struct page *page, unsigned long offset) | |
1276 | { | |
1277 | struct inode *inode = page->mapping->host; | |
1278 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
1279 | ||
1280 | ubifs_assert(PagePrivate(page)); | |
1281 | if (offset) | |
1282 | /* Partial page remains dirty */ | |
1283 | return; | |
1284 | ||
1285 | if (PageChecked(page)) | |
1286 | release_new_page_budget(c); | |
1287 | else | |
1288 | release_existing_page_budget(c); | |
1289 | ||
1290 | atomic_long_dec(&c->dirty_pg_cnt); | |
1291 | ClearPagePrivate(page); | |
1292 | ClearPageChecked(page); | |
1293 | } | |
1294 | ||
1295 | static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd) | |
1296 | { | |
1297 | struct ubifs_inode *ui = ubifs_inode(dentry->d_inode); | |
1298 | ||
1299 | nd_set_link(nd, ui->data); | |
1300 | return NULL; | |
1301 | } | |
1302 | ||
7ea80859 | 1303 | int ubifs_fsync(struct file *file, int datasync) |
1e51764a | 1304 | { |
7ea80859 | 1305 | struct inode *inode = file->f_mapping->host; |
1e51764a AB |
1306 | struct ubifs_info *c = inode->i_sb->s_fs_info; |
1307 | int err; | |
1308 | ||
1309 | dbg_gen("syncing inode %lu", inode->i_ino); | |
1310 | ||
1311 | /* | |
1312 | * VFS has already synchronized dirty pages for this inode. Synchronize | |
1313 | * the inode unless this is a 'datasync()' call. | |
1314 | */ | |
1315 | if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) { | |
a9185b41 | 1316 | err = inode->i_sb->s_op->write_inode(inode, NULL); |
1e51764a AB |
1317 | if (err) |
1318 | return err; | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Nodes related to this inode may still sit in a write-buffer. Flush | |
1323 | * them. | |
1324 | */ | |
1325 | err = ubifs_sync_wbufs_by_inode(c, inode); | |
1326 | if (err) | |
1327 | return err; | |
1328 | ||
1329 | return 0; | |
1330 | } | |
1331 | ||
1332 | /** | |
1333 | * mctime_update_needed - check if mtime or ctime update is needed. | |
1334 | * @inode: the inode to do the check for | |
1335 | * @now: current time | |
1336 | * | |
1337 | * This helper function checks if the inode mtime/ctime should be updated or | |
1338 | * not. If current values of the time-stamps are within the UBIFS inode time | |
1339 | * granularity, they are not updated. This is an optimization. | |
1340 | */ | |
1341 | static inline int mctime_update_needed(const struct inode *inode, | |
1342 | const struct timespec *now) | |
1343 | { | |
1344 | if (!timespec_equal(&inode->i_mtime, now) || | |
1345 | !timespec_equal(&inode->i_ctime, now)) | |
1346 | return 1; | |
1347 | return 0; | |
1348 | } | |
1349 | ||
1350 | /** | |
1351 | * update_ctime - update mtime and ctime of an inode. | |
1352 | * @c: UBIFS file-system description object | |
1353 | * @inode: inode to update | |
1354 | * | |
1355 | * This function updates mtime and ctime of the inode if it is not equivalent to | |
1356 | * current time. Returns zero in case of success and a negative error code in | |
1357 | * case of failure. | |
1358 | */ | |
1359 | static int update_mctime(struct ubifs_info *c, struct inode *inode) | |
1360 | { | |
1361 | struct timespec now = ubifs_current_time(inode); | |
1362 | struct ubifs_inode *ui = ubifs_inode(inode); | |
1363 | ||
1364 | if (mctime_update_needed(inode, &now)) { | |
1365 | int err, release; | |
1366 | struct ubifs_budget_req req = { .dirtied_ino = 1, | |
dab4b4d2 | 1367 | .dirtied_ino_d = ALIGN(ui->data_len, 8) }; |
1e51764a AB |
1368 | |
1369 | err = ubifs_budget_space(c, &req); | |
1370 | if (err) | |
1371 | return err; | |
1372 | ||
1373 | mutex_lock(&ui->ui_mutex); | |
1374 | inode->i_mtime = inode->i_ctime = ubifs_current_time(inode); | |
1375 | release = ui->dirty; | |
1376 | mark_inode_dirty_sync(inode); | |
1377 | mutex_unlock(&ui->ui_mutex); | |
1378 | if (release) | |
1379 | ubifs_release_budget(c, &req); | |
1380 | } | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1386 | unsigned long nr_segs, loff_t pos) | |
1387 | { | |
1388 | int err; | |
1e51764a AB |
1389 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1390 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
1391 | ||
1392 | err = update_mctime(c, inode); | |
1393 | if (err) | |
1394 | return err; | |
1395 | ||
774888bc | 1396 | return generic_file_aio_write(iocb, iov, nr_segs, pos); |
1e51764a AB |
1397 | } |
1398 | ||
1399 | static int ubifs_set_page_dirty(struct page *page) | |
1400 | { | |
1401 | int ret; | |
1402 | ||
1403 | ret = __set_page_dirty_nobuffers(page); | |
1404 | /* | |
1405 | * An attempt to dirty a page without budgeting for it - should not | |
1406 | * happen. | |
1407 | */ | |
1408 | ubifs_assert(ret == 0); | |
1409 | return ret; | |
1410 | } | |
1411 | ||
1412 | static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags) | |
1413 | { | |
1414 | /* | |
1415 | * An attempt to release a dirty page without budgeting for it - should | |
1416 | * not happen. | |
1417 | */ | |
1418 | if (PageWriteback(page)) | |
1419 | return 0; | |
1420 | ubifs_assert(PagePrivate(page)); | |
1421 | ubifs_assert(0); | |
1422 | ClearPagePrivate(page); | |
1423 | ClearPageChecked(page); | |
1424 | return 1; | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * mmap()d file has taken write protection fault and is being made | |
1429 | * writable. UBIFS must ensure page is budgeted for. | |
1430 | */ | |
c2ec175c | 1431 | static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
1e51764a | 1432 | { |
c2ec175c | 1433 | struct page *page = vmf->page; |
1e51764a AB |
1434 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1435 | struct ubifs_info *c = inode->i_sb->s_fs_info; | |
1436 | struct timespec now = ubifs_current_time(inode); | |
1437 | struct ubifs_budget_req req = { .new_page = 1 }; | |
1438 | int err, update_time; | |
1439 | ||
1440 | dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index, | |
1441 | i_size_read(inode)); | |
1442 | ubifs_assert(!(inode->i_sb->s_flags & MS_RDONLY)); | |
1443 | ||
1444 | if (unlikely(c->ro_media)) | |
c2ec175c | 1445 | return VM_FAULT_SIGBUS; /* -EROFS */ |
1e51764a AB |
1446 | |
1447 | /* | |
1448 | * We have not locked @page so far so we may budget for changing the | |
1449 | * page. Note, we cannot do this after we locked the page, because | |
1450 | * budgeting may cause write-back which would cause deadlock. | |
1451 | * | |
1452 | * At the moment we do not know whether the page is dirty or not, so we | |
1453 | * assume that it is not and budget for a new page. We could look at | |
1454 | * the @PG_private flag and figure this out, but we may race with write | |
1455 | * back and the page state may change by the time we lock it, so this | |
1456 | * would need additional care. We do not bother with this at the | |
1457 | * moment, although it might be good idea to do. Instead, we allocate | |
1458 | * budget for a new page and amend it later on if the page was in fact | |
1459 | * dirty. | |
1460 | * | |
1461 | * The budgeting-related logic of this function is similar to what we | |
1462 | * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there | |
1463 | * for more comments. | |
1464 | */ | |
1465 | update_time = mctime_update_needed(inode, &now); | |
1466 | if (update_time) | |
1467 | /* | |
1468 | * We have to change inode time stamp which requires extra | |
1469 | * budgeting. | |
1470 | */ | |
1471 | req.dirtied_ino = 1; | |
1472 | ||
1473 | err = ubifs_budget_space(c, &req); | |
1474 | if (unlikely(err)) { | |
1475 | if (err == -ENOSPC) | |
1476 | ubifs_warn("out of space for mmapped file " | |
1477 | "(inode number %lu)", inode->i_ino); | |
c2ec175c | 1478 | return VM_FAULT_SIGBUS; |
1e51764a AB |
1479 | } |
1480 | ||
1481 | lock_page(page); | |
1482 | if (unlikely(page->mapping != inode->i_mapping || | |
1483 | page_offset(page) > i_size_read(inode))) { | |
1484 | /* Page got truncated out from underneath us */ | |
1485 | err = -EINVAL; | |
1486 | goto out_unlock; | |
1487 | } | |
1488 | ||
1489 | if (PagePrivate(page)) | |
1490 | release_new_page_budget(c); | |
1491 | else { | |
1492 | if (!PageChecked(page)) | |
1493 | ubifs_convert_page_budget(c); | |
1494 | SetPagePrivate(page); | |
1495 | atomic_long_inc(&c->dirty_pg_cnt); | |
1496 | __set_page_dirty_nobuffers(page); | |
1497 | } | |
1498 | ||
1499 | if (update_time) { | |
1500 | int release; | |
1501 | struct ubifs_inode *ui = ubifs_inode(inode); | |
1502 | ||
1503 | mutex_lock(&ui->ui_mutex); | |
1504 | inode->i_mtime = inode->i_ctime = ubifs_current_time(inode); | |
1505 | release = ui->dirty; | |
1506 | mark_inode_dirty_sync(inode); | |
1507 | mutex_unlock(&ui->ui_mutex); | |
1508 | if (release) | |
1509 | ubifs_release_dirty_inode_budget(c, ui); | |
1510 | } | |
1511 | ||
1512 | unlock_page(page); | |
1513 | return 0; | |
1514 | ||
1515 | out_unlock: | |
1516 | unlock_page(page); | |
1517 | ubifs_release_budget(c, &req); | |
c2ec175c NP |
1518 | if (err) |
1519 | err = VM_FAULT_SIGBUS; | |
1e51764a AB |
1520 | return err; |
1521 | } | |
1522 | ||
f0f37e2f | 1523 | static const struct vm_operations_struct ubifs_file_vm_ops = { |
1e51764a AB |
1524 | .fault = filemap_fault, |
1525 | .page_mkwrite = ubifs_vm_page_mkwrite, | |
1526 | }; | |
1527 | ||
1528 | static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma) | |
1529 | { | |
1530 | int err; | |
1531 | ||
1532 | /* 'generic_file_mmap()' takes care of NOMMU case */ | |
1533 | err = generic_file_mmap(file, vma); | |
1534 | if (err) | |
1535 | return err; | |
1536 | vma->vm_ops = &ubifs_file_vm_ops; | |
1537 | return 0; | |
1538 | } | |
1539 | ||
e8b81566 | 1540 | const struct address_space_operations ubifs_file_address_operations = { |
1e51764a AB |
1541 | .readpage = ubifs_readpage, |
1542 | .writepage = ubifs_writepage, | |
1543 | .write_begin = ubifs_write_begin, | |
1544 | .write_end = ubifs_write_end, | |
1545 | .invalidatepage = ubifs_invalidatepage, | |
1546 | .set_page_dirty = ubifs_set_page_dirty, | |
1547 | .releasepage = ubifs_releasepage, | |
1548 | }; | |
1549 | ||
e8b81566 | 1550 | const struct inode_operations ubifs_file_inode_operations = { |
1e51764a AB |
1551 | .setattr = ubifs_setattr, |
1552 | .getattr = ubifs_getattr, | |
1553 | #ifdef CONFIG_UBIFS_FS_XATTR | |
1554 | .setxattr = ubifs_setxattr, | |
1555 | .getxattr = ubifs_getxattr, | |
1556 | .listxattr = ubifs_listxattr, | |
1557 | .removexattr = ubifs_removexattr, | |
1558 | #endif | |
1559 | }; | |
1560 | ||
e8b81566 | 1561 | const struct inode_operations ubifs_symlink_inode_operations = { |
1e51764a AB |
1562 | .readlink = generic_readlink, |
1563 | .follow_link = ubifs_follow_link, | |
1564 | .setattr = ubifs_setattr, | |
1565 | .getattr = ubifs_getattr, | |
1566 | }; | |
1567 | ||
e8b81566 | 1568 | const struct file_operations ubifs_file_operations = { |
1e51764a AB |
1569 | .llseek = generic_file_llseek, |
1570 | .read = do_sync_read, | |
1571 | .write = do_sync_write, | |
1572 | .aio_read = generic_file_aio_read, | |
1573 | .aio_write = ubifs_aio_write, | |
1574 | .mmap = ubifs_file_mmap, | |
1575 | .fsync = ubifs_fsync, | |
1576 | .unlocked_ioctl = ubifs_ioctl, | |
1577 | .splice_read = generic_file_splice_read, | |
22bc7fa8 | 1578 | .splice_write = generic_file_splice_write, |
1e51764a AB |
1579 | #ifdef CONFIG_COMPAT |
1580 | .compat_ioctl = ubifs_compat_ioctl, | |
1581 | #endif | |
1582 | }; |