blob: 6d34dc7e33e10c87adb4df4f25ab54c6eb5587fc [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
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/*
24 * This file implements VFS file and inode operations of regular files, device
25 * nodes and symlinks as well as address space operations.
26 *
27 * UBIFS uses 2 page flags: PG_private and PG_checked. PG_private is set if the
28 * page is dirty and is used for budgeting purposes - dirty pages should not be
29 * budgeted. The PG_checked flag is set if full budgeting is required for the
30 * page e.g., when it corresponds to a file hole or it is just beyond the file
31 * size. The budgeting is done in 'ubifs_write_begin()', because it is OK to
32 * fail in this function, and the budget is released in 'ubifs_write_end()'. So
33 * the PG_private and PG_checked flags carry the information about how the page
34 * was budgeted, to make it possible to release the budget properly.
35 *
36 * A thing to keep in mind: inode's 'i_mutex' is locked in most VFS operations
37 * we implement. However, this is not true for '->writepage()', which might be
38 * called with 'i_mutex' unlocked. For example, when pdflush is performing
39 * write-back, it calls 'writepage()' with unlocked 'i_mutex', although the
40 * inode has 'I_LOCK' flag in this case. At "normal" work-paths 'i_mutex' is
41 * locked in '->writepage', e.g. in "sys_write -> alloc_pages -> direct reclaim
42 * path'. So, in '->writepage()' we are only guaranteed that the page is
43 * locked.
44 *
45 * Similarly, 'i_mutex' does not have to be locked in readpage(), e.g.,
46 * readahead path does not have it locked ("sys_read -> generic_file_aio_read
47 * -> ondemand_readahead -> readpage"). In case of readahead, 'I_LOCK' flag is
48 * not set as well. However, UBIFS disables readahead.
49 *
50 * This, for example means that there might be 2 concurrent '->writepage()'
51 * calls for the same inode, but different inode dirty pages.
52 */
53
54#include "ubifs.h"
55#include <linux/mount.h>
Al Viro3f8206d2008-07-26 03:46:43 -040056#include <linux/namei.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030057
58static int read_block(struct inode *inode, void *addr, unsigned int block,
59 struct ubifs_data_node *dn)
60{
61 struct ubifs_info *c = inode->i_sb->s_fs_info;
62 int err, len, out_len;
63 union ubifs_key key;
64 unsigned int dlen;
65
66 data_key_init(c, &key, inode->i_ino, block);
67 err = ubifs_tnc_lookup(c, &key, dn);
68 if (err) {
69 if (err == -ENOENT)
70 /* Not found, so it must be a hole */
71 memset(addr, 0, UBIFS_BLOCK_SIZE);
72 return err;
73 }
74
Artem Bityutskiyf92b9822008-12-28 11:34:26 +020075 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
76 ubifs_inode(inode)->creat_sqnum);
Artem Bityutskiy1e517642008-07-14 19:08:37 +030077 len = le32_to_cpu(dn->size);
78 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
79 goto dump;
80
81 dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
82 out_len = UBIFS_BLOCK_SIZE;
83 err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
84 le16_to_cpu(dn->compr_type));
85 if (err || len != out_len)
86 goto dump;
87
88 /*
89 * Data length can be less than a full block, even for blocks that are
90 * not the last in the file (e.g., as a result of making a hole and
91 * appending data). Ensure that the remainder is zeroed out.
92 */
93 if (len < UBIFS_BLOCK_SIZE)
94 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
95
96 return 0;
97
98dump:
99 ubifs_err("bad data node (block %u, inode %lu)",
100 block, inode->i_ino);
101 dbg_dump_node(c, dn);
102 return -EINVAL;
103}
104
105static int do_readpage(struct page *page)
106{
107 void *addr;
108 int err = 0, i;
109 unsigned int block, beyond;
110 struct ubifs_data_node *dn;
111 struct inode *inode = page->mapping->host;
112 loff_t i_size = i_size_read(inode);
113
114 dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
115 inode->i_ino, page->index, i_size, page->flags);
116 ubifs_assert(!PageChecked(page));
117 ubifs_assert(!PagePrivate(page));
118
119 addr = kmap(page);
120
121 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
122 beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
123 if (block >= beyond) {
124 /* Reading beyond inode */
125 SetPageChecked(page);
126 memset(addr, 0, PAGE_CACHE_SIZE);
127 goto out;
128 }
129
130 dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
131 if (!dn) {
132 err = -ENOMEM;
133 goto error;
134 }
135
136 i = 0;
137 while (1) {
138 int ret;
139
140 if (block >= beyond) {
141 /* Reading beyond inode */
142 err = -ENOENT;
143 memset(addr, 0, UBIFS_BLOCK_SIZE);
144 } else {
145 ret = read_block(inode, addr, block, dn);
146 if (ret) {
147 err = ret;
148 if (err != -ENOENT)
149 break;
Adrian Huntered382d52008-09-05 16:17:42 +0300150 } else if (block + 1 == beyond) {
151 int dlen = le32_to_cpu(dn->size);
152 int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
153
154 if (ilen && ilen < dlen)
155 memset(addr + ilen, 0, dlen - ilen);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300156 }
157 }
158 if (++i >= UBIFS_BLOCKS_PER_PAGE)
159 break;
160 block += 1;
161 addr += UBIFS_BLOCK_SIZE;
162 }
163 if (err) {
164 if (err == -ENOENT) {
165 /* Not found, so it must be a hole */
166 SetPageChecked(page);
167 dbg_gen("hole");
168 goto out_free;
169 }
170 ubifs_err("cannot read page %lu of inode %lu, error %d",
171 page->index, inode->i_ino, err);
172 goto error;
173 }
174
175out_free:
176 kfree(dn);
177out:
178 SetPageUptodate(page);
179 ClearPageError(page);
180 flush_dcache_page(page);
181 kunmap(page);
182 return 0;
183
184error:
185 kfree(dn);
186 ClearPageUptodate(page);
187 SetPageError(page);
188 flush_dcache_page(page);
189 kunmap(page);
190 return err;
191}
192
193/**
194 * release_new_page_budget - release budget of a new page.
195 * @c: UBIFS file-system description object
196 *
197 * This is a helper function which releases budget corresponding to the budget
198 * of one new page of data.
199 */
200static void release_new_page_budget(struct ubifs_info *c)
201{
202 struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
203
204 ubifs_release_budget(c, &req);
205}
206
207/**
208 * release_existing_page_budget - release budget of an existing page.
209 * @c: UBIFS file-system description object
210 *
211 * This is a helper function which releases budget corresponding to the budget
212 * of changing one one page of data which already exists on the flash media.
213 */
214static void release_existing_page_budget(struct ubifs_info *c)
215{
216 struct ubifs_budget_req req = { .dd_growth = c->page_budget};
217
218 ubifs_release_budget(c, &req);
219}
220
221static int write_begin_slow(struct address_space *mapping,
Nick Piggin54566b22009-01-04 12:00:53 -0800222 loff_t pos, unsigned len, struct page **pagep,
223 unsigned flags)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300224{
225 struct inode *inode = mapping->host;
226 struct ubifs_info *c = inode->i_sb->s_fs_info;
227 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
228 struct ubifs_budget_req req = { .new_page = 1 };
229 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
230 struct page *page;
231
232 dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
233 inode->i_ino, pos, len, inode->i_size);
234
235 /*
236 * At the slow path we have to budget before locking the page, because
237 * budgeting may force write-back, which would wait on locked pages and
238 * deadlock if we had the page locked. At this point we do not know
239 * anything about the page, so assume that this is a new page which is
240 * written to a hole. This corresponds to largest budget. Later the
241 * budget will be amended if this is not true.
242 */
243 if (appending)
244 /* We are appending data, budget for inode change */
245 req.dirtied_ino = 1;
246
247 err = ubifs_budget_space(c, &req);
248 if (unlikely(err))
249 return err;
250
Nick Piggin54566b22009-01-04 12:00:53 -0800251 page = grab_cache_page_write_begin(mapping, index, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300252 if (unlikely(!page)) {
253 ubifs_release_budget(c, &req);
254 return -ENOMEM;
255 }
256
257 if (!PageUptodate(page)) {
Artem Bityutskiy7bbe5b52008-12-09 11:02:51 -0500258 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300259 SetPageChecked(page);
260 else {
261 err = do_readpage(page);
262 if (err) {
263 unlock_page(page);
264 page_cache_release(page);
265 return err;
266 }
267 }
268
269 SetPageUptodate(page);
270 ClearPageError(page);
271 }
272
273 if (PagePrivate(page))
274 /*
275 * The page is dirty, which means it was budgeted twice:
276 * o first time the budget was allocated by the task which
277 * made the page dirty and set the PG_private flag;
278 * o and then we budgeted for it for the second time at the
279 * very beginning of this function.
280 *
281 * So what we have to do is to release the page budget we
282 * allocated.
283 */
284 release_new_page_budget(c);
285 else if (!PageChecked(page))
286 /*
287 * We are changing a page which already exists on the media.
288 * This means that changing the page does not make the amount
289 * of indexing information larger, and this part of the budget
290 * which we have already acquired may be released.
291 */
292 ubifs_convert_page_budget(c);
293
294 if (appending) {
295 struct ubifs_inode *ui = ubifs_inode(inode);
296
297 /*
298 * 'ubifs_write_end()' is optimized from the fast-path part of
299 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
300 * if data is appended.
301 */
302 mutex_lock(&ui->ui_mutex);
303 if (ui->dirty)
304 /*
305 * The inode is dirty already, so we may free the
306 * budget we allocated.
307 */
308 ubifs_release_dirty_inode_budget(c, ui);
309 }
310
311 *pagep = page;
312 return 0;
313}
314
315/**
316 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
317 * @c: UBIFS file-system description object
318 * @page: page to allocate budget for
319 * @ui: UBIFS inode object the page belongs to
320 * @appending: non-zero if the page is appended
321 *
322 * This is a helper function for 'ubifs_write_begin()' which allocates budget
323 * for the operation. The budget is allocated differently depending on whether
324 * this is appending, whether the page is dirty or not, and so on. This
325 * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
326 * in case of success and %-ENOSPC in case of failure.
327 */
328static int allocate_budget(struct ubifs_info *c, struct page *page,
329 struct ubifs_inode *ui, int appending)
330{
331 struct ubifs_budget_req req = { .fast = 1 };
332
333 if (PagePrivate(page)) {
334 if (!appending)
335 /*
336 * The page is dirty and we are not appending, which
337 * means no budget is needed at all.
338 */
339 return 0;
340
341 mutex_lock(&ui->ui_mutex);
342 if (ui->dirty)
343 /*
344 * The page is dirty and we are appending, so the inode
345 * has to be marked as dirty. However, it is already
346 * dirty, so we do not need any budget. We may return,
347 * but @ui->ui_mutex hast to be left locked because we
348 * should prevent write-back from flushing the inode
349 * and freeing the budget. The lock will be released in
350 * 'ubifs_write_end()'.
351 */
352 return 0;
353
354 /*
355 * The page is dirty, we are appending, the inode is clean, so
356 * we need to budget the inode change.
357 */
358 req.dirtied_ino = 1;
359 } else {
360 if (PageChecked(page))
361 /*
362 * The page corresponds to a hole and does not
363 * exist on the media. So changing it makes
364 * make the amount of indexing information
365 * larger, and we have to budget for a new
366 * page.
367 */
368 req.new_page = 1;
369 else
370 /*
371 * Not a hole, the change will not add any new
372 * indexing information, budget for page
373 * change.
374 */
375 req.dirtied_page = 1;
376
377 if (appending) {
378 mutex_lock(&ui->ui_mutex);
379 if (!ui->dirty)
380 /*
381 * The inode is clean but we will have to mark
382 * it as dirty because we are appending. This
383 * needs a budget.
384 */
385 req.dirtied_ino = 1;
386 }
387 }
388
389 return ubifs_budget_space(c, &req);
390}
391
392/*
393 * This function is called when a page of data is going to be written. Since
394 * the page of data will not necessarily go to the flash straight away, UBIFS
395 * has to reserve space on the media for it, which is done by means of
396 * budgeting.
397 *
398 * This is the hot-path of the file-system and we are trying to optimize it as
399 * much as possible. For this reasons it is split on 2 parts - slow and fast.
400 *
401 * There many budgeting cases:
402 * o a new page is appended - we have to budget for a new page and for
403 * changing the inode; however, if the inode is already dirty, there is
404 * no need to budget for it;
405 * o an existing clean page is changed - we have budget for it; if the page
406 * does not exist on the media (a hole), we have to budget for a new
407 * page; otherwise, we may budget for changing an existing page; the
408 * difference between these cases is that changing an existing page does
409 * not introduce anything new to the FS indexing information, so it does
410 * not grow, and smaller budget is acquired in this case;
411 * o an existing dirty page is changed - no need to budget at all, because
412 * the page budget has been acquired by earlier, when the page has been
413 * marked dirty.
414 *
415 * UBIFS budgeting sub-system may force write-back if it thinks there is no
416 * space to reserve. This imposes some locking restrictions and makes it
417 * impossible to take into account the above cases, and makes it impossible to
418 * optimize budgeting.
419 *
420 * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
421 * there is a plenty of flash space and the budget will be acquired quickly,
422 * without forcing write-back. The slow path does not make this assumption.
423 */
424static int ubifs_write_begin(struct file *file, struct address_space *mapping,
425 loff_t pos, unsigned len, unsigned flags,
426 struct page **pagep, void **fsdata)
427{
428 struct inode *inode = mapping->host;
429 struct ubifs_info *c = inode->i_sb->s_fs_info;
430 struct ubifs_inode *ui = ubifs_inode(inode);
431 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
432 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
Adrian Hunterf55aa592009-02-23 12:47:25 +0200433 int skipped_read = 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300434 struct page *page;
435
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300436 ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
437
438 if (unlikely(c->ro_media))
439 return -EROFS;
440
441 /* Try out the fast-path part first */
Nick Piggin54566b22009-01-04 12:00:53 -0800442 page = grab_cache_page_write_begin(mapping, index, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300443 if (unlikely(!page))
444 return -ENOMEM;
445
446 if (!PageUptodate(page)) {
447 /* The page is not loaded from the flash */
Adrian Hunterf55aa592009-02-23 12:47:25 +0200448 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300449 /*
450 * We change whole page so no need to load it. But we
451 * have to set the @PG_checked flag to make the further
452 * code the page is new. This might be not true, but it
453 * is better to budget more that to read the page from
454 * the media.
455 */
456 SetPageChecked(page);
Adrian Hunterf55aa592009-02-23 12:47:25 +0200457 skipped_read = 1;
458 } else {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300459 err = do_readpage(page);
460 if (err) {
461 unlock_page(page);
462 page_cache_release(page);
463 return err;
464 }
465 }
466
467 SetPageUptodate(page);
468 ClearPageError(page);
469 }
470
471 err = allocate_budget(c, page, ui, appending);
472 if (unlikely(err)) {
473 ubifs_assert(err == -ENOSPC);
474 /*
Adrian Hunterf55aa592009-02-23 12:47:25 +0200475 * If we skipped reading the page because we were going to
476 * write all of it, then it is not up to date.
477 */
478 if (skipped_read) {
479 ClearPageChecked(page);
480 ClearPageUptodate(page);
481 }
482 /*
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300483 * Budgeting failed which means it would have to force
484 * write-back but didn't, because we set the @fast flag in the
485 * request. Write-back cannot be done now, while we have the
486 * page locked, because it would deadlock. Unlock and free
487 * everything and fall-back to slow-path.
488 */
489 if (appending) {
490 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
491 mutex_unlock(&ui->ui_mutex);
492 }
493 unlock_page(page);
494 page_cache_release(page);
495
Nick Piggin54566b22009-01-04 12:00:53 -0800496 return write_begin_slow(mapping, pos, len, pagep, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300497 }
498
499 /*
500 * Whee, we aquired budgeting quickly - without involving
501 * garbage-collection, committing or forceing write-back. We return
502 * with @ui->ui_mutex locked if we are appending pages, and unlocked
503 * otherwise. This is an optimization (slightly hacky though).
504 */
505 *pagep = page;
506 return 0;
507
508}
509
510/**
511 * cancel_budget - cancel budget.
512 * @c: UBIFS file-system description object
513 * @page: page to cancel budget for
514 * @ui: UBIFS inode object the page belongs to
515 * @appending: non-zero if the page is appended
516 *
517 * This is a helper function for a page write operation. It unlocks the
518 * @ui->ui_mutex in case of appending.
519 */
520static void cancel_budget(struct ubifs_info *c, struct page *page,
521 struct ubifs_inode *ui, int appending)
522{
523 if (appending) {
524 if (!ui->dirty)
525 ubifs_release_dirty_inode_budget(c, ui);
526 mutex_unlock(&ui->ui_mutex);
527 }
528 if (!PagePrivate(page)) {
529 if (PageChecked(page))
530 release_new_page_budget(c);
531 else
532 release_existing_page_budget(c);
533 }
534}
535
536static int ubifs_write_end(struct file *file, struct address_space *mapping,
537 loff_t pos, unsigned len, unsigned copied,
538 struct page *page, void *fsdata)
539{
540 struct inode *inode = mapping->host;
541 struct ubifs_inode *ui = ubifs_inode(inode);
542 struct ubifs_info *c = inode->i_sb->s_fs_info;
543 loff_t end_pos = pos + len;
544 int appending = !!(end_pos > inode->i_size);
545
546 dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
547 inode->i_ino, pos, page->index, len, copied, inode->i_size);
548
549 if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
550 /*
551 * VFS copied less data to the page that it intended and
552 * declared in its '->write_begin()' call via the @len
553 * argument. If the page was not up-to-date, and @len was
554 * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
555 * not load it from the media (for optimization reasons). This
556 * means that part of the page contains garbage. So read the
557 * page now.
558 */
559 dbg_gen("copied %d instead of %d, read page and repeat",
560 copied, len);
561 cancel_budget(c, page, ui, appending);
562
563 /*
564 * Return 0 to force VFS to repeat the whole operation, or the
565 * error code if 'do_readpage()' failes.
566 */
567 copied = do_readpage(page);
568 goto out;
569 }
570
571 if (!PagePrivate(page)) {
572 SetPagePrivate(page);
573 atomic_long_inc(&c->dirty_pg_cnt);
574 __set_page_dirty_nobuffers(page);
575 }
576
577 if (appending) {
578 i_size_write(inode, end_pos);
579 ui->ui_size = end_pos;
580 /*
581 * Note, we do not set @I_DIRTY_PAGES (which means that the
582 * inode has dirty pages), this has been done in
583 * '__set_page_dirty_nobuffers()'.
584 */
585 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
586 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
587 mutex_unlock(&ui->ui_mutex);
588 }
589
590out:
591 unlock_page(page);
592 page_cache_release(page);
593 return copied;
594}
595
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300596/**
597 * populate_page - copy data nodes into a page for bulk-read.
598 * @c: UBIFS file-system description object
599 * @page: page
600 * @bu: bulk-read information
601 * @n: next zbranch slot
602 *
603 * This function returns %0 on success and a negative error code on failure.
604 */
605static int populate_page(struct ubifs_info *c, struct page *page,
606 struct bu_info *bu, int *n)
607{
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300608 int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300609 struct inode *inode = page->mapping->host;
610 loff_t i_size = i_size_read(inode);
611 unsigned int page_block;
612 void *addr, *zaddr;
613 pgoff_t end_index;
614
615 dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
616 inode->i_ino, page->index, i_size, page->flags);
617
618 addr = zaddr = kmap(page);
619
Adrian Huntered382d52008-09-05 16:17:42 +0300620 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300621 if (!i_size || page->index > end_index) {
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300622 hole = 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300623 memset(addr, 0, PAGE_CACHE_SIZE);
624 goto out_hole;
625 }
626
627 page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
628 while (1) {
629 int err, len, out_len, dlen;
630
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300631 if (nn >= bu->cnt) {
632 hole = 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300633 memset(addr, 0, UBIFS_BLOCK_SIZE);
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300634 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300635 struct ubifs_data_node *dn;
636
637 dn = bu->buf + (bu->zbranch[nn].offs - offs);
638
Harvey Harrison0ecb9522008-10-24 10:52:57 -0700639 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300640 ubifs_inode(inode)->creat_sqnum);
641
642 len = le32_to_cpu(dn->size);
643 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
644 goto out_err;
645
646 dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
647 out_len = UBIFS_BLOCK_SIZE;
648 err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
649 le16_to_cpu(dn->compr_type));
650 if (err || len != out_len)
651 goto out_err;
652
653 if (len < UBIFS_BLOCK_SIZE)
654 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
655
656 nn += 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300657 read = (i << UBIFS_BLOCK_SHIFT) + len;
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300658 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
659 nn += 1;
660 continue;
661 } else {
662 hole = 1;
663 memset(addr, 0, UBIFS_BLOCK_SIZE);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300664 }
665 if (++i >= UBIFS_BLOCKS_PER_PAGE)
666 break;
667 addr += UBIFS_BLOCK_SIZE;
668 page_block += 1;
669 }
670
671 if (end_index == page->index) {
672 int len = i_size & (PAGE_CACHE_SIZE - 1);
673
Adrian Huntered382d52008-09-05 16:17:42 +0300674 if (len && len < read)
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300675 memset(zaddr + len, 0, read - len);
676 }
677
678out_hole:
679 if (hole) {
680 SetPageChecked(page);
681 dbg_gen("hole");
682 }
683
684 SetPageUptodate(page);
685 ClearPageError(page);
686 flush_dcache_page(page);
687 kunmap(page);
688 *n = nn;
689 return 0;
690
691out_err:
692 ClearPageUptodate(page);
693 SetPageError(page);
694 flush_dcache_page(page);
695 kunmap(page);
696 ubifs_err("bad data node (block %u, inode %lu)",
697 page_block, inode->i_ino);
698 return -EINVAL;
699}
700
701/**
702 * ubifs_do_bulk_read - do bulk-read.
703 * @c: UBIFS file-system description object
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200704 * @bu: bulk-read information
705 * @page1: first page to read
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300706 *
707 * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
708 */
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200709static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
710 struct page *page1)
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300711{
712 pgoff_t offset = page1->index, end_index;
713 struct address_space *mapping = page1->mapping;
714 struct inode *inode = mapping->host;
715 struct ubifs_inode *ui = ubifs_inode(inode);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300716 int err, page_idx, page_cnt, ret = 0, n = 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200717 int allocate = bu->buf ? 0 : 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300718 loff_t isize;
719
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300720 err = ubifs_tnc_get_bu_keys(c, bu);
721 if (err)
722 goto out_warn;
723
724 if (bu->eof) {
725 /* Turn off bulk-read at the end of the file */
726 ui->read_in_a_row = 1;
727 ui->bulk_read = 0;
728 }
729
730 page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
731 if (!page_cnt) {
732 /*
733 * This happens when there are multiple blocks per page and the
734 * blocks for the first page we are looking for, are not
735 * together. If all the pages were like this, bulk-read would
736 * reduce performance, so we turn it off for a while.
737 */
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200738 goto out_bu_off;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300739 }
740
741 if (bu->cnt) {
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200742 if (allocate) {
743 /*
744 * Allocate bulk-read buffer depending on how many data
745 * nodes we are going to read.
746 */
747 bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
748 bu->zbranch[bu->cnt - 1].len -
749 bu->zbranch[0].offs;
750 ubifs_assert(bu->buf_len > 0);
751 ubifs_assert(bu->buf_len <= c->leb_size);
752 bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
753 if (!bu->buf)
754 goto out_bu_off;
755 }
756
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300757 err = ubifs_tnc_bulk_read(c, bu);
758 if (err)
759 goto out_warn;
760 }
761
762 err = populate_page(c, page1, bu, &n);
763 if (err)
764 goto out_warn;
765
766 unlock_page(page1);
767 ret = 1;
768
769 isize = i_size_read(inode);
770 if (isize == 0)
771 goto out_free;
772 end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
773
774 for (page_idx = 1; page_idx < page_cnt; page_idx++) {
775 pgoff_t page_offset = offset + page_idx;
776 struct page *page;
777
778 if (page_offset > end_index)
779 break;
780 page = find_or_create_page(mapping, page_offset,
781 GFP_NOFS | __GFP_COLD);
782 if (!page)
783 break;
784 if (!PageUptodate(page))
785 err = populate_page(c, page, bu, &n);
786 unlock_page(page);
787 page_cache_release(page);
788 if (err)
789 break;
790 }
791
792 ui->last_page_read = offset + page_idx - 1;
793
794out_free:
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200795 if (allocate)
796 kfree(bu->buf);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300797 return ret;
798
799out_warn:
800 ubifs_warn("ignoring error %d and skipping bulk-read", err);
801 goto out_free;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200802
803out_bu_off:
804 ui->read_in_a_row = ui->bulk_read = 0;
805 goto out_free;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300806}
807
808/**
809 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
810 * @page: page from which to start bulk-read.
811 *
812 * Some flash media are capable of reading sequentially at faster rates. UBIFS
813 * bulk-read facility is designed to take advantage of that, by reading in one
814 * go consecutive data nodes that are also located consecutively in the same
815 * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
816 */
817static int ubifs_bulk_read(struct page *page)
818{
819 struct inode *inode = page->mapping->host;
820 struct ubifs_info *c = inode->i_sb->s_fs_info;
821 struct ubifs_inode *ui = ubifs_inode(inode);
822 pgoff_t index = page->index, last_page_read = ui->last_page_read;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200823 struct bu_info *bu;
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200824 int err = 0, allocated = 0;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300825
826 ui->last_page_read = index;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300827 if (!c->bulk_read)
828 return 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200829
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300830 /*
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200831 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
832 * so don't bother if we cannot lock the mutex.
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300833 */
834 if (!mutex_trylock(&ui->ui_mutex))
835 return 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200836
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300837 if (index != last_page_read + 1) {
838 /* Turn off bulk-read if we stop reading sequentially */
839 ui->read_in_a_row = 1;
840 if (ui->bulk_read)
841 ui->bulk_read = 0;
842 goto out_unlock;
843 }
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200844
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300845 if (!ui->bulk_read) {
846 ui->read_in_a_row += 1;
847 if (ui->read_in_a_row < 3)
848 goto out_unlock;
849 /* Three reads in a row, so switch on bulk-read */
850 ui->bulk_read = 1;
851 }
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200852
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200853 /*
854 * If possible, try to use pre-allocated bulk-read information, which
855 * is protected by @c->bu_mutex.
856 */
857 if (mutex_trylock(&c->bu_mutex))
858 bu = &c->bu;
859 else {
860 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
861 if (!bu)
862 goto out_unlock;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200863
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200864 bu->buf = NULL;
865 allocated = 1;
866 }
867
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200868 bu->buf_len = c->max_bu_buf_len;
869 data_key_init(c, &bu->key, inode->i_ino,
870 page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200871 err = ubifs_do_bulk_read(c, bu, page);
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200872
873 if (!allocated)
874 mutex_unlock(&c->bu_mutex);
875 else
876 kfree(bu);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200877
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300878out_unlock:
879 mutex_unlock(&ui->ui_mutex);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200880 return err;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300881}
882
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300883static int ubifs_readpage(struct file *file, struct page *page)
884{
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300885 if (ubifs_bulk_read(page))
886 return 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300887 do_readpage(page);
888 unlock_page(page);
889 return 0;
890}
891
892static int do_writepage(struct page *page, int len)
893{
894 int err = 0, i, blen;
895 unsigned int block;
896 void *addr;
897 union ubifs_key key;
898 struct inode *inode = page->mapping->host;
899 struct ubifs_info *c = inode->i_sb->s_fs_info;
900
901#ifdef UBIFS_DEBUG
902 spin_lock(&ui->ui_lock);
903 ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
904 spin_unlock(&ui->ui_lock);
905#endif
906
907 /* Update radix tree tags */
908 set_page_writeback(page);
909
910 addr = kmap(page);
911 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
912 i = 0;
913 while (len) {
914 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
915 data_key_init(c, &key, inode->i_ino, block);
916 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
917 if (err)
918 break;
919 if (++i >= UBIFS_BLOCKS_PER_PAGE)
920 break;
921 block += 1;
922 addr += blen;
923 len -= blen;
924 }
925 if (err) {
926 SetPageError(page);
927 ubifs_err("cannot write page %lu of inode %lu, error %d",
928 page->index, inode->i_ino, err);
929 ubifs_ro_mode(c, err);
930 }
931
932 ubifs_assert(PagePrivate(page));
933 if (PageChecked(page))
934 release_new_page_budget(c);
935 else
936 release_existing_page_budget(c);
937
938 atomic_long_dec(&c->dirty_pg_cnt);
939 ClearPagePrivate(page);
940 ClearPageChecked(page);
941
942 kunmap(page);
943 unlock_page(page);
944 end_page_writeback(page);
945 return err;
946}
947
948/*
949 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
950 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
951 * situation when a we have an inode with size 0, then a megabyte of data is
952 * appended to the inode, then write-back starts and flushes some amount of the
953 * dirty pages, the journal becomes full, commit happens and finishes, and then
954 * an unclean reboot happens. When the file system is mounted next time, the
955 * inode size would still be 0, but there would be many pages which are beyond
956 * the inode size, they would be indexed and consume flash space. Because the
957 * journal has been committed, the replay would not be able to detect this
958 * situation and correct the inode size. This means UBIFS would have to scan
959 * whole index and correct all inode sizes, which is long an unacceptable.
960 *
961 * To prevent situations like this, UBIFS writes pages back only if they are
Artem Bityutskiy7d4e9cc2009-03-20 19:11:12 +0200962 * within the last synchronized inode size, i.e. the size which has been
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300963 * written to the flash media last time. Otherwise, UBIFS forces inode
964 * write-back, thus making sure the on-flash inode contains current inode size,
965 * and then keeps writing pages back.
966 *
967 * Some locking issues explanation. 'ubifs_writepage()' first is called with
968 * the page locked, and it locks @ui_mutex. However, write-back does take inode
969 * @i_mutex, which means other VFS operations may be run on this inode at the
970 * same time. And the problematic one is truncation to smaller size, from where
971 * we have to call 'vmtruncate()', which first changes @inode->i_size, then
972 * drops the truncated pages. And while dropping the pages, it takes the page
973 * lock. This means that 'do_truncation()' cannot call 'vmtruncate()' with
974 * @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'. This
975 * means that @inode->i_size is changed while @ui_mutex is unlocked.
976 *
977 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
978 * inode size. How do we do this if @inode->i_size may became smaller while we
979 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
980 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
981 * internally and updates it under @ui_mutex.
982 *
983 * Q: why we do not worry that if we race with truncation, we may end up with a
984 * situation when the inode is truncated while we are in the middle of
985 * 'do_writepage()', so we do write beyond inode size?
986 * A: If we are in the middle of 'do_writepage()', truncation would be locked
987 * on the page lock and it would not write the truncated inode node to the
988 * journal before we have finished.
989 */
990static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
991{
992 struct inode *inode = page->mapping->host;
993 struct ubifs_inode *ui = ubifs_inode(inode);
994 loff_t i_size = i_size_read(inode), synced_i_size;
995 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
996 int err, len = i_size & (PAGE_CACHE_SIZE - 1);
997 void *kaddr;
998
999 dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1000 inode->i_ino, page->index, page->flags);
1001 ubifs_assert(PagePrivate(page));
1002
1003 /* Is the page fully outside @i_size? (truncate in progress) */
1004 if (page->index > end_index || (page->index == end_index && !len)) {
1005 err = 0;
1006 goto out_unlock;
1007 }
1008
1009 spin_lock(&ui->ui_lock);
1010 synced_i_size = ui->synced_i_size;
1011 spin_unlock(&ui->ui_lock);
1012
1013 /* Is the page fully inside @i_size? */
1014 if (page->index < end_index) {
1015 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
1016 err = inode->i_sb->s_op->write_inode(inode, 1);
1017 if (err)
1018 goto out_unlock;
1019 /*
1020 * The inode has been written, but the write-buffer has
1021 * not been synchronized, so in case of an unclean
1022 * reboot we may end up with some pages beyond inode
1023 * size, but they would be in the journal (because
1024 * commit flushes write buffers) and recovery would deal
1025 * with this.
1026 */
1027 }
1028 return do_writepage(page, PAGE_CACHE_SIZE);
1029 }
1030
1031 /*
1032 * The page straddles @i_size. It must be zeroed out on each and every
1033 * writepage invocation because it may be mmapped. "A file is mapped
1034 * in multiples of the page size. For a file that is not a multiple of
1035 * the page size, the remaining memory is zeroed when mapped, and
1036 * writes to that region are not written out to the file."
1037 */
1038 kaddr = kmap_atomic(page, KM_USER0);
1039 memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1040 flush_dcache_page(page);
1041 kunmap_atomic(kaddr, KM_USER0);
1042
1043 if (i_size > synced_i_size) {
1044 err = inode->i_sb->s_op->write_inode(inode, 1);
1045 if (err)
1046 goto out_unlock;
1047 }
1048
1049 return do_writepage(page, len);
1050
1051out_unlock:
1052 unlock_page(page);
1053 return err;
1054}
1055
1056/**
1057 * do_attr_changes - change inode attributes.
1058 * @inode: inode to change attributes for
1059 * @attr: describes attributes to change
1060 */
1061static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1062{
1063 if (attr->ia_valid & ATTR_UID)
1064 inode->i_uid = attr->ia_uid;
1065 if (attr->ia_valid & ATTR_GID)
1066 inode->i_gid = attr->ia_gid;
1067 if (attr->ia_valid & ATTR_ATIME)
1068 inode->i_atime = timespec_trunc(attr->ia_atime,
1069 inode->i_sb->s_time_gran);
1070 if (attr->ia_valid & ATTR_MTIME)
1071 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1072 inode->i_sb->s_time_gran);
1073 if (attr->ia_valid & ATTR_CTIME)
1074 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1075 inode->i_sb->s_time_gran);
1076 if (attr->ia_valid & ATTR_MODE) {
1077 umode_t mode = attr->ia_mode;
1078
1079 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1080 mode &= ~S_ISGID;
1081 inode->i_mode = mode;
1082 }
1083}
1084
1085/**
1086 * do_truncation - truncate an inode.
1087 * @c: UBIFS file-system description object
1088 * @inode: inode to truncate
1089 * @attr: inode attribute changes description
1090 *
1091 * This function implements VFS '->setattr()' call when the inode is truncated
1092 * to a smaller size. Returns zero in case of success and a negative error code
1093 * in case of failure.
1094 */
1095static int do_truncation(struct ubifs_info *c, struct inode *inode,
1096 const struct iattr *attr)
1097{
1098 int err;
1099 struct ubifs_budget_req req;
1100 loff_t old_size = inode->i_size, new_size = attr->ia_size;
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001101 int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001102 struct ubifs_inode *ui = ubifs_inode(inode);
1103
1104 dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1105 memset(&req, 0, sizeof(struct ubifs_budget_req));
1106
1107 /*
1108 * If this is truncation to a smaller size, and we do not truncate on a
1109 * block boundary, budget for changing one data block, because the last
1110 * block will be re-written.
1111 */
1112 if (new_size & (UBIFS_BLOCK_SIZE - 1))
1113 req.dirtied_page = 1;
1114
1115 req.dirtied_ino = 1;
1116 /* A funny way to budget for truncation node */
1117 req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1118 err = ubifs_budget_space(c, &req);
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001119 if (err) {
1120 /*
1121 * Treat truncations to zero as deletion and always allow them,
1122 * just like we do for '->unlink()'.
1123 */
1124 if (new_size || err != -ENOSPC)
1125 return err;
1126 budgeted = 0;
1127 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001128
1129 err = vmtruncate(inode, new_size);
1130 if (err)
1131 goto out_budg;
1132
1133 if (offset) {
1134 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1135 struct page *page;
1136
1137 page = find_lock_page(inode->i_mapping, index);
1138 if (page) {
1139 if (PageDirty(page)) {
1140 /*
1141 * 'ubifs_jnl_truncate()' will try to truncate
1142 * the last data node, but it contains
1143 * out-of-date data because the page is dirty.
1144 * Write the page now, so that
1145 * 'ubifs_jnl_truncate()' will see an already
1146 * truncated (and up to date) data node.
1147 */
1148 ubifs_assert(PagePrivate(page));
1149
1150 clear_page_dirty_for_io(page);
1151 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1152 offset = new_size &
1153 (PAGE_CACHE_SIZE - 1);
1154 err = do_writepage(page, offset);
1155 page_cache_release(page);
1156 if (err)
1157 goto out_budg;
1158 /*
1159 * We could now tell 'ubifs_jnl_truncate()' not
1160 * to read the last block.
1161 */
1162 } else {
1163 /*
1164 * We could 'kmap()' the page and pass the data
1165 * to 'ubifs_jnl_truncate()' to save it from
1166 * having to read it.
1167 */
1168 unlock_page(page);
1169 page_cache_release(page);
1170 }
1171 }
1172 }
1173
1174 mutex_lock(&ui->ui_mutex);
1175 ui->ui_size = inode->i_size;
1176 /* Truncation changes inode [mc]time */
1177 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1178 /* The other attributes may be changed at the same time as well */
1179 do_attr_changes(inode, attr);
1180
1181 err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1182 mutex_unlock(&ui->ui_mutex);
1183out_budg:
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001184 if (budgeted)
1185 ubifs_release_budget(c, &req);
1186 else {
1187 c->nospace = c->nospace_rp = 0;
1188 smp_wmb();
1189 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001190 return err;
1191}
1192
1193/**
1194 * do_setattr - change inode attributes.
1195 * @c: UBIFS file-system description object
1196 * @inode: inode to change attributes for
1197 * @attr: inode attribute changes description
1198 *
1199 * This function implements VFS '->setattr()' call for all cases except
1200 * truncations to smaller size. Returns zero in case of success and a negative
1201 * error code in case of failure.
1202 */
1203static int do_setattr(struct ubifs_info *c, struct inode *inode,
1204 const struct iattr *attr)
1205{
1206 int err, release;
1207 loff_t new_size = attr->ia_size;
1208 struct ubifs_inode *ui = ubifs_inode(inode);
1209 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001210 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001211
1212 err = ubifs_budget_space(c, &req);
1213 if (err)
1214 return err;
1215
1216 if (attr->ia_valid & ATTR_SIZE) {
1217 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1218 err = vmtruncate(inode, new_size);
1219 if (err)
1220 goto out;
1221 }
1222
1223 mutex_lock(&ui->ui_mutex);
1224 if (attr->ia_valid & ATTR_SIZE) {
1225 /* Truncation changes inode [mc]time */
1226 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1227 /* 'vmtruncate()' changed @i_size, update @ui_size */
1228 ui->ui_size = inode->i_size;
1229 }
1230
1231 do_attr_changes(inode, attr);
1232
1233 release = ui->dirty;
1234 if (attr->ia_valid & ATTR_SIZE)
1235 /*
1236 * Inode length changed, so we have to make sure
1237 * @I_DIRTY_DATASYNC is set.
1238 */
1239 __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1240 else
1241 mark_inode_dirty_sync(inode);
1242 mutex_unlock(&ui->ui_mutex);
1243
1244 if (release)
1245 ubifs_release_budget(c, &req);
1246 if (IS_SYNC(inode))
1247 err = inode->i_sb->s_op->write_inode(inode, 1);
1248 return err;
1249
1250out:
1251 ubifs_release_budget(c, &req);
1252 return err;
1253}
1254
1255int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1256{
1257 int err;
1258 struct inode *inode = dentry->d_inode;
1259 struct ubifs_info *c = inode->i_sb->s_fs_info;
1260
Artem Bityutskiy7d32c2b2008-07-18 18:54:29 +03001261 dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1262 inode->i_ino, inode->i_mode, attr->ia_valid);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001263 err = inode_change_ok(inode, attr);
1264 if (err)
1265 return err;
1266
1267 err = dbg_check_synced_i_size(inode);
1268 if (err)
1269 return err;
1270
1271 if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1272 /* Truncation to a smaller size */
1273 err = do_truncation(c, inode, attr);
1274 else
1275 err = do_setattr(c, inode, attr);
1276
1277 return err;
1278}
1279
1280static void ubifs_invalidatepage(struct page *page, unsigned long offset)
1281{
1282 struct inode *inode = page->mapping->host;
1283 struct ubifs_info *c = inode->i_sb->s_fs_info;
1284
1285 ubifs_assert(PagePrivate(page));
1286 if (offset)
1287 /* Partial page remains dirty */
1288 return;
1289
1290 if (PageChecked(page))
1291 release_new_page_budget(c);
1292 else
1293 release_existing_page_budget(c);
1294
1295 atomic_long_dec(&c->dirty_pg_cnt);
1296 ClearPagePrivate(page);
1297 ClearPageChecked(page);
1298}
1299
1300static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
1301{
1302 struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
1303
1304 nd_set_link(nd, ui->data);
1305 return NULL;
1306}
1307
1308int ubifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1309{
1310 struct inode *inode = dentry->d_inode;
1311 struct ubifs_info *c = inode->i_sb->s_fs_info;
1312 int err;
1313
1314 dbg_gen("syncing inode %lu", inode->i_ino);
1315
1316 /*
1317 * VFS has already synchronized dirty pages for this inode. Synchronize
1318 * the inode unless this is a 'datasync()' call.
1319 */
1320 if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1321 err = inode->i_sb->s_op->write_inode(inode, 1);
1322 if (err)
1323 return err;
1324 }
1325
1326 /*
1327 * Nodes related to this inode may still sit in a write-buffer. Flush
1328 * them.
1329 */
1330 err = ubifs_sync_wbufs_by_inode(c, inode);
1331 if (err)
1332 return err;
1333
1334 return 0;
1335}
1336
1337/**
1338 * mctime_update_needed - check if mtime or ctime update is needed.
1339 * @inode: the inode to do the check for
1340 * @now: current time
1341 *
1342 * This helper function checks if the inode mtime/ctime should be updated or
1343 * not. If current values of the time-stamps are within the UBIFS inode time
1344 * granularity, they are not updated. This is an optimization.
1345 */
1346static inline int mctime_update_needed(const struct inode *inode,
1347 const struct timespec *now)
1348{
1349 if (!timespec_equal(&inode->i_mtime, now) ||
1350 !timespec_equal(&inode->i_ctime, now))
1351 return 1;
1352 return 0;
1353}
1354
1355/**
1356 * update_ctime - update mtime and ctime of an inode.
1357 * @c: UBIFS file-system description object
1358 * @inode: inode to update
1359 *
1360 * This function updates mtime and ctime of the inode if it is not equivalent to
1361 * current time. Returns zero in case of success and a negative error code in
1362 * case of failure.
1363 */
1364static int update_mctime(struct ubifs_info *c, struct inode *inode)
1365{
1366 struct timespec now = ubifs_current_time(inode);
1367 struct ubifs_inode *ui = ubifs_inode(inode);
1368
1369 if (mctime_update_needed(inode, &now)) {
1370 int err, release;
1371 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001372 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001373
1374 err = ubifs_budget_space(c, &req);
1375 if (err)
1376 return err;
1377
1378 mutex_lock(&ui->ui_mutex);
1379 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1380 release = ui->dirty;
1381 mark_inode_dirty_sync(inode);
1382 mutex_unlock(&ui->ui_mutex);
1383 if (release)
1384 ubifs_release_budget(c, &req);
1385 }
1386
1387 return 0;
1388}
1389
1390static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1391 unsigned long nr_segs, loff_t pos)
1392{
1393 int err;
1394 ssize_t ret;
1395 struct inode *inode = iocb->ki_filp->f_mapping->host;
1396 struct ubifs_info *c = inode->i_sb->s_fs_info;
1397
1398 err = update_mctime(c, inode);
1399 if (err)
1400 return err;
1401
1402 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
1403 if (ret < 0)
1404 return ret;
1405
1406 if (ret > 0 && (IS_SYNC(inode) || iocb->ki_filp->f_flags & O_SYNC)) {
1407 err = ubifs_sync_wbufs_by_inode(c, inode);
1408 if (err)
1409 return err;
1410 }
1411
1412 return ret;
1413}
1414
1415static int ubifs_set_page_dirty(struct page *page)
1416{
1417 int ret;
1418
1419 ret = __set_page_dirty_nobuffers(page);
1420 /*
1421 * An attempt to dirty a page without budgeting for it - should not
1422 * happen.
1423 */
1424 ubifs_assert(ret == 0);
1425 return ret;
1426}
1427
1428static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1429{
1430 /*
1431 * An attempt to release a dirty page without budgeting for it - should
1432 * not happen.
1433 */
1434 if (PageWriteback(page))
1435 return 0;
1436 ubifs_assert(PagePrivate(page));
1437 ubifs_assert(0);
1438 ClearPagePrivate(page);
1439 ClearPageChecked(page);
1440 return 1;
1441}
1442
1443/*
1444 * mmap()d file has taken write protection fault and is being made
1445 * writable. UBIFS must ensure page is budgeted for.
1446 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07001447static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001448{
Nick Pigginc2ec1752009-03-31 15:23:21 -07001449 struct page *page = vmf->page;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001450 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1451 struct ubifs_info *c = inode->i_sb->s_fs_info;
1452 struct timespec now = ubifs_current_time(inode);
1453 struct ubifs_budget_req req = { .new_page = 1 };
1454 int err, update_time;
1455
1456 dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1457 i_size_read(inode));
1458 ubifs_assert(!(inode->i_sb->s_flags & MS_RDONLY));
1459
1460 if (unlikely(c->ro_media))
Nick Pigginc2ec1752009-03-31 15:23:21 -07001461 return VM_FAULT_SIGBUS; /* -EROFS */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001462
1463 /*
1464 * We have not locked @page so far so we may budget for changing the
1465 * page. Note, we cannot do this after we locked the page, because
1466 * budgeting may cause write-back which would cause deadlock.
1467 *
1468 * At the moment we do not know whether the page is dirty or not, so we
1469 * assume that it is not and budget for a new page. We could look at
1470 * the @PG_private flag and figure this out, but we may race with write
1471 * back and the page state may change by the time we lock it, so this
1472 * would need additional care. We do not bother with this at the
1473 * moment, although it might be good idea to do. Instead, we allocate
1474 * budget for a new page and amend it later on if the page was in fact
1475 * dirty.
1476 *
1477 * The budgeting-related logic of this function is similar to what we
1478 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1479 * for more comments.
1480 */
1481 update_time = mctime_update_needed(inode, &now);
1482 if (update_time)
1483 /*
1484 * We have to change inode time stamp which requires extra
1485 * budgeting.
1486 */
1487 req.dirtied_ino = 1;
1488
1489 err = ubifs_budget_space(c, &req);
1490 if (unlikely(err)) {
1491 if (err == -ENOSPC)
1492 ubifs_warn("out of space for mmapped file "
1493 "(inode number %lu)", inode->i_ino);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001494 return VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001495 }
1496
1497 lock_page(page);
1498 if (unlikely(page->mapping != inode->i_mapping ||
1499 page_offset(page) > i_size_read(inode))) {
1500 /* Page got truncated out from underneath us */
1501 err = -EINVAL;
1502 goto out_unlock;
1503 }
1504
1505 if (PagePrivate(page))
1506 release_new_page_budget(c);
1507 else {
1508 if (!PageChecked(page))
1509 ubifs_convert_page_budget(c);
1510 SetPagePrivate(page);
1511 atomic_long_inc(&c->dirty_pg_cnt);
1512 __set_page_dirty_nobuffers(page);
1513 }
1514
1515 if (update_time) {
1516 int release;
1517 struct ubifs_inode *ui = ubifs_inode(inode);
1518
1519 mutex_lock(&ui->ui_mutex);
1520 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1521 release = ui->dirty;
1522 mark_inode_dirty_sync(inode);
1523 mutex_unlock(&ui->ui_mutex);
1524 if (release)
1525 ubifs_release_dirty_inode_budget(c, ui);
1526 }
1527
1528 unlock_page(page);
1529 return 0;
1530
1531out_unlock:
1532 unlock_page(page);
1533 ubifs_release_budget(c, &req);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001534 if (err)
1535 err = VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001536 return err;
1537}
1538
1539static struct vm_operations_struct ubifs_file_vm_ops = {
1540 .fault = filemap_fault,
1541 .page_mkwrite = ubifs_vm_page_mkwrite,
1542};
1543
1544static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1545{
1546 int err;
1547
1548 /* 'generic_file_mmap()' takes care of NOMMU case */
1549 err = generic_file_mmap(file, vma);
1550 if (err)
1551 return err;
1552 vma->vm_ops = &ubifs_file_vm_ops;
1553 return 0;
1554}
1555
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001556const struct address_space_operations ubifs_file_address_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001557 .readpage = ubifs_readpage,
1558 .writepage = ubifs_writepage,
1559 .write_begin = ubifs_write_begin,
1560 .write_end = ubifs_write_end,
1561 .invalidatepage = ubifs_invalidatepage,
1562 .set_page_dirty = ubifs_set_page_dirty,
1563 .releasepage = ubifs_releasepage,
1564};
1565
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001566const struct inode_operations ubifs_file_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001567 .setattr = ubifs_setattr,
1568 .getattr = ubifs_getattr,
1569#ifdef CONFIG_UBIFS_FS_XATTR
1570 .setxattr = ubifs_setxattr,
1571 .getxattr = ubifs_getxattr,
1572 .listxattr = ubifs_listxattr,
1573 .removexattr = ubifs_removexattr,
1574#endif
1575};
1576
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001577const struct inode_operations ubifs_symlink_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001578 .readlink = generic_readlink,
1579 .follow_link = ubifs_follow_link,
1580 .setattr = ubifs_setattr,
1581 .getattr = ubifs_getattr,
1582};
1583
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001584const struct file_operations ubifs_file_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001585 .llseek = generic_file_llseek,
1586 .read = do_sync_read,
1587 .write = do_sync_write,
1588 .aio_read = generic_file_aio_read,
1589 .aio_write = ubifs_aio_write,
1590 .mmap = ubifs_file_mmap,
1591 .fsync = ubifs_fsync,
1592 .unlocked_ioctl = ubifs_ioctl,
1593 .splice_read = generic_file_splice_read,
Zoltan Sogor22bc7fa2008-07-28 16:28:49 +02001594 .splice_write = generic_file_splice_write,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001595#ifdef CONFIG_COMPAT
1596 .compat_ioctl = ubifs_compat_ioctl,
1597#endif
1598};