blob: d77db7e36484e3f79bece27ee4a11f3a6b80df6f [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/*
Artem Bityutskiy873a64c2009-09-01 17:06:43 +030024 * This file implements VFS file and inode operations for regular files, device
Artem Bityutskiy1e517642008-07-14 19:08:37 +030025 * nodes and symlinks as well as address space operations.
26 *
Artem Bityutskiy873a64c2009-09-01 17:06:43 +030027 * 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.
Artem Bityutskiy1e517642008-07-14 19:08:37 +030037 *
Artem Bityutskiy873a64c2009-09-01 17:06:43 +030038 * 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.
Artem Bityutskiy1e517642008-07-14 19:08:37 +030045 *
Artem Bityutskiy873a64c2009-09-01 17:06:43 +030046 * 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 ->
Christoph Hellwigeaff8072009-12-17 14:25:01 +010048 * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
Artem Bityutskiy873a64c2009-09-01 17:06:43 +030049 * set as well. However, UBIFS disables readahead.
Artem Bityutskiy1e517642008-07-14 19:08:37 +030050 */
51
52#include "ubifs.h"
53#include <linux/mount.h>
Al Viro3f8206d2008-07-26 03:46:43 -040054#include <linux/namei.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090055#include <linux/slab.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030056
57static 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
Artem Bityutskiyf92b9822008-12-28 11:34:26 +020074 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
75 ubifs_inode(inode)->creat_sqnum);
Artem Bityutskiy1e517642008-07-14 19:08:37 +030076 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
97dump:
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
104static 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;
Adrian Huntered382d52008-09-05 16:17:42 +0300149 } 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);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300155 }
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
174out_free:
175 kfree(dn);
176out:
177 SetPageUptodate(page);
178 ClearPageError(page);
179 flush_dcache_page(page);
180 kunmap(page);
181 return 0;
182
183error:
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 */
199static 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 */
213static 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
220static int write_begin_slow(struct address_space *mapping,
Nick Piggin54566b22009-01-04 12:00:53 -0800221 loff_t pos, unsigned len, struct page **pagep,
222 unsigned flags)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300223{
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
Nick Piggin54566b22009-01-04 12:00:53 -0800250 page = grab_cache_page_write_begin(mapping, index, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300251 if (unlikely(!page)) {
252 ubifs_release_budget(c, &req);
253 return -ENOMEM;
254 }
255
256 if (!PageUptodate(page)) {
Artem Bityutskiy7bbe5b52008-12-09 11:02:51 -0500257 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300258 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 */
327static 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 */
423static 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);
Adrian Hunterf55aa592009-02-23 12:47:25 +0200432 int skipped_read = 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300433 struct page *page;
434
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300435 ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
Artem Bityutskiy2ef13292010-09-19 18:34:26 +0300436 ubifs_assert(!c->ro_media && !c->ro_mount);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300437
Artem Bityutskiy2680d722010-09-17 16:44:28 +0300438 if (unlikely(c->ro_error))
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300439 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
Artem Bityutskiy873a64c2009-09-01 17:06:43 +0300452 * code know that the page is new. This might be not
453 * true, but it is better to budget more than to read
454 * the page from the media.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300455 */
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 /*
Artem Bityutskiy873a64c2009-09-01 17:06:43 +0300500 * Whee, we acquired budgeting quickly - without involving
501 * garbage-collection, committing or forcing write-back. We return
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300502 * 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
Artem Bityutskiy873a64c2009-09-01 17:06:43 +0300565 * error code if 'do_readpage()' fails.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300566 */
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
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200971 * we have to call 'truncate_setsize()', which first changes @inode->i_size, then
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300972 * drops the truncated pages. And while dropping the pages, it takes the page
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200973 * lock. This means that 'do_truncation()' cannot call 'truncate_setsize()' with
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300974 * @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 *
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200977 * XXX(truncate): with the new truncate sequence this is not true anymore,
978 * and the calls to truncate_setsize can be move around freely. They should
979 * be moved to the very end of the truncate sequence.
npiggin@suse.de15c6fd92010-05-27 01:05:34 +1000980 *
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300981 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
982 * inode size. How do we do this if @inode->i_size may became smaller while we
983 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
984 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
985 * internally and updates it under @ui_mutex.
986 *
987 * Q: why we do not worry that if we race with truncation, we may end up with a
988 * situation when the inode is truncated while we are in the middle of
989 * 'do_writepage()', so we do write beyond inode size?
990 * A: If we are in the middle of 'do_writepage()', truncation would be locked
991 * on the page lock and it would not write the truncated inode node to the
992 * journal before we have finished.
993 */
994static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
995{
996 struct inode *inode = page->mapping->host;
997 struct ubifs_inode *ui = ubifs_inode(inode);
998 loff_t i_size = i_size_read(inode), synced_i_size;
999 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1000 int err, len = i_size & (PAGE_CACHE_SIZE - 1);
1001 void *kaddr;
1002
1003 dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1004 inode->i_ino, page->index, page->flags);
1005 ubifs_assert(PagePrivate(page));
1006
1007 /* Is the page fully outside @i_size? (truncate in progress) */
1008 if (page->index > end_index || (page->index == end_index && !len)) {
1009 err = 0;
1010 goto out_unlock;
1011 }
1012
1013 spin_lock(&ui->ui_lock);
1014 synced_i_size = ui->synced_i_size;
1015 spin_unlock(&ui->ui_lock);
1016
1017 /* Is the page fully inside @i_size? */
1018 if (page->index < end_index) {
1019 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001020 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001021 if (err)
1022 goto out_unlock;
1023 /*
1024 * The inode has been written, but the write-buffer has
1025 * not been synchronized, so in case of an unclean
1026 * reboot we may end up with some pages beyond inode
1027 * size, but they would be in the journal (because
1028 * commit flushes write buffers) and recovery would deal
1029 * with this.
1030 */
1031 }
1032 return do_writepage(page, PAGE_CACHE_SIZE);
1033 }
1034
1035 /*
1036 * The page straddles @i_size. It must be zeroed out on each and every
1037 * writepage invocation because it may be mmapped. "A file is mapped
1038 * in multiples of the page size. For a file that is not a multiple of
1039 * the page size, the remaining memory is zeroed when mapped, and
1040 * writes to that region are not written out to the file."
1041 */
1042 kaddr = kmap_atomic(page, KM_USER0);
1043 memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1044 flush_dcache_page(page);
1045 kunmap_atomic(kaddr, KM_USER0);
1046
1047 if (i_size > synced_i_size) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001048 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001049 if (err)
1050 goto out_unlock;
1051 }
1052
1053 return do_writepage(page, len);
1054
1055out_unlock:
1056 unlock_page(page);
1057 return err;
1058}
1059
1060/**
1061 * do_attr_changes - change inode attributes.
1062 * @inode: inode to change attributes for
1063 * @attr: describes attributes to change
1064 */
1065static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1066{
1067 if (attr->ia_valid & ATTR_UID)
1068 inode->i_uid = attr->ia_uid;
1069 if (attr->ia_valid & ATTR_GID)
1070 inode->i_gid = attr->ia_gid;
1071 if (attr->ia_valid & ATTR_ATIME)
1072 inode->i_atime = timespec_trunc(attr->ia_atime,
1073 inode->i_sb->s_time_gran);
1074 if (attr->ia_valid & ATTR_MTIME)
1075 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1076 inode->i_sb->s_time_gran);
1077 if (attr->ia_valid & ATTR_CTIME)
1078 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1079 inode->i_sb->s_time_gran);
1080 if (attr->ia_valid & ATTR_MODE) {
1081 umode_t mode = attr->ia_mode;
1082
1083 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1084 mode &= ~S_ISGID;
1085 inode->i_mode = mode;
1086 }
1087}
1088
1089/**
1090 * do_truncation - truncate an inode.
1091 * @c: UBIFS file-system description object
1092 * @inode: inode to truncate
1093 * @attr: inode attribute changes description
1094 *
1095 * This function implements VFS '->setattr()' call when the inode is truncated
1096 * to a smaller size. Returns zero in case of success and a negative error code
1097 * in case of failure.
1098 */
1099static int do_truncation(struct ubifs_info *c, struct inode *inode,
1100 const struct iattr *attr)
1101{
1102 int err;
1103 struct ubifs_budget_req req;
1104 loff_t old_size = inode->i_size, new_size = attr->ia_size;
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001105 int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001106 struct ubifs_inode *ui = ubifs_inode(inode);
1107
1108 dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1109 memset(&req, 0, sizeof(struct ubifs_budget_req));
1110
1111 /*
1112 * If this is truncation to a smaller size, and we do not truncate on a
1113 * block boundary, budget for changing one data block, because the last
1114 * block will be re-written.
1115 */
1116 if (new_size & (UBIFS_BLOCK_SIZE - 1))
1117 req.dirtied_page = 1;
1118
1119 req.dirtied_ino = 1;
1120 /* A funny way to budget for truncation node */
1121 req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1122 err = ubifs_budget_space(c, &req);
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001123 if (err) {
1124 /*
1125 * Treat truncations to zero as deletion and always allow them,
1126 * just like we do for '->unlink()'.
1127 */
1128 if (new_size || err != -ENOSPC)
1129 return err;
1130 budgeted = 0;
1131 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001132
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001133 truncate_setsize(inode, new_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001134
1135 if (offset) {
1136 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1137 struct page *page;
1138
1139 page = find_lock_page(inode->i_mapping, index);
1140 if (page) {
1141 if (PageDirty(page)) {
1142 /*
1143 * 'ubifs_jnl_truncate()' will try to truncate
1144 * the last data node, but it contains
1145 * out-of-date data because the page is dirty.
1146 * Write the page now, so that
1147 * 'ubifs_jnl_truncate()' will see an already
1148 * truncated (and up to date) data node.
1149 */
1150 ubifs_assert(PagePrivate(page));
1151
1152 clear_page_dirty_for_io(page);
1153 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1154 offset = new_size &
1155 (PAGE_CACHE_SIZE - 1);
1156 err = do_writepage(page, offset);
1157 page_cache_release(page);
1158 if (err)
1159 goto out_budg;
1160 /*
1161 * We could now tell 'ubifs_jnl_truncate()' not
1162 * to read the last block.
1163 */
1164 } else {
1165 /*
1166 * We could 'kmap()' the page and pass the data
1167 * to 'ubifs_jnl_truncate()' to save it from
1168 * having to read it.
1169 */
1170 unlock_page(page);
1171 page_cache_release(page);
1172 }
1173 }
1174 }
1175
1176 mutex_lock(&ui->ui_mutex);
1177 ui->ui_size = inode->i_size;
1178 /* Truncation changes inode [mc]time */
1179 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
Artem Bityutskiy873a64c2009-09-01 17:06:43 +03001180 /* Other attributes may be changed at the same time as well */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001181 do_attr_changes(inode, attr);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001182 err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1183 mutex_unlock(&ui->ui_mutex);
Artem Bityutskiy873a64c2009-09-01 17:06:43 +03001184
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001185out_budg:
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001186 if (budgeted)
1187 ubifs_release_budget(c, &req);
1188 else {
1189 c->nospace = c->nospace_rp = 0;
1190 smp_wmb();
1191 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001192 return err;
1193}
1194
1195/**
1196 * do_setattr - change inode attributes.
1197 * @c: UBIFS file-system description object
1198 * @inode: inode to change attributes for
1199 * @attr: inode attribute changes description
1200 *
1201 * This function implements VFS '->setattr()' call for all cases except
1202 * truncations to smaller size. Returns zero in case of success and a negative
1203 * error code in case of failure.
1204 */
1205static int do_setattr(struct ubifs_info *c, struct inode *inode,
1206 const struct iattr *attr)
1207{
1208 int err, release;
1209 loff_t new_size = attr->ia_size;
1210 struct ubifs_inode *ui = ubifs_inode(inode);
1211 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001212 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001213
1214 err = ubifs_budget_space(c, &req);
1215 if (err)
1216 return err;
1217
1218 if (attr->ia_valid & ATTR_SIZE) {
1219 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001220 truncate_setsize(inode, new_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001221 }
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);
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001227 /* 'truncate_setsize()' changed @i_size, update @ui_size */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001228 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))
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001247 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001248 return err;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001249}
1250
1251int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1252{
1253 int err;
1254 struct inode *inode = dentry->d_inode;
1255 struct ubifs_info *c = inode->i_sb->s_fs_info;
1256
Artem Bityutskiy7d32c2b2008-07-18 18:54:29 +03001257 dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1258 inode->i_ino, inode->i_mode, attr->ia_valid);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001259 err = inode_change_ok(inode, attr);
1260 if (err)
1261 return err;
1262
1263 err = dbg_check_synced_i_size(inode);
1264 if (err)
1265 return err;
1266
1267 if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1268 /* Truncation to a smaller size */
1269 err = do_truncation(c, inode, attr);
1270 else
1271 err = do_setattr(c, inode, attr);
1272
1273 return err;
1274}
1275
1276static void ubifs_invalidatepage(struct page *page, unsigned long offset)
1277{
1278 struct inode *inode = page->mapping->host;
1279 struct ubifs_info *c = inode->i_sb->s_fs_info;
1280
1281 ubifs_assert(PagePrivate(page));
1282 if (offset)
1283 /* Partial page remains dirty */
1284 return;
1285
1286 if (PageChecked(page))
1287 release_new_page_budget(c);
1288 else
1289 release_existing_page_budget(c);
1290
1291 atomic_long_dec(&c->dirty_pg_cnt);
1292 ClearPagePrivate(page);
1293 ClearPageChecked(page);
1294}
1295
1296static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
1297{
1298 struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
1299
1300 nd_set_link(nd, ui->data);
1301 return NULL;
1302}
1303
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001304int ubifs_fsync(struct file *file, int datasync)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001305{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001306 struct inode *inode = file->f_mapping->host;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001307 struct ubifs_info *c = inode->i_sb->s_fs_info;
1308 int err;
1309
1310 dbg_gen("syncing inode %lu", inode->i_ino);
1311
1312 /*
1313 * VFS has already synchronized dirty pages for this inode. Synchronize
1314 * the inode unless this is a 'datasync()' call.
1315 */
1316 if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001317 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001318 if (err)
1319 return err;
1320 }
1321
1322 /*
1323 * Nodes related to this inode may still sit in a write-buffer. Flush
1324 * them.
1325 */
1326 err = ubifs_sync_wbufs_by_inode(c, inode);
1327 if (err)
1328 return err;
1329
1330 return 0;
1331}
1332
1333/**
1334 * mctime_update_needed - check if mtime or ctime update is needed.
1335 * @inode: the inode to do the check for
1336 * @now: current time
1337 *
1338 * This helper function checks if the inode mtime/ctime should be updated or
1339 * not. If current values of the time-stamps are within the UBIFS inode time
1340 * granularity, they are not updated. This is an optimization.
1341 */
1342static inline int mctime_update_needed(const struct inode *inode,
1343 const struct timespec *now)
1344{
1345 if (!timespec_equal(&inode->i_mtime, now) ||
1346 !timespec_equal(&inode->i_ctime, now))
1347 return 1;
1348 return 0;
1349}
1350
1351/**
1352 * update_ctime - update mtime and ctime of an inode.
1353 * @c: UBIFS file-system description object
1354 * @inode: inode to update
1355 *
1356 * This function updates mtime and ctime of the inode if it is not equivalent to
1357 * current time. Returns zero in case of success and a negative error code in
1358 * case of failure.
1359 */
1360static int update_mctime(struct ubifs_info *c, struct inode *inode)
1361{
1362 struct timespec now = ubifs_current_time(inode);
1363 struct ubifs_inode *ui = ubifs_inode(inode);
1364
1365 if (mctime_update_needed(inode, &now)) {
1366 int err, release;
1367 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001368 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001369
1370 err = ubifs_budget_space(c, &req);
1371 if (err)
1372 return err;
1373
1374 mutex_lock(&ui->ui_mutex);
1375 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1376 release = ui->dirty;
1377 mark_inode_dirty_sync(inode);
1378 mutex_unlock(&ui->ui_mutex);
1379 if (release)
1380 ubifs_release_budget(c, &req);
1381 }
1382
1383 return 0;
1384}
1385
1386static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1387 unsigned long nr_segs, loff_t pos)
1388{
1389 int err;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001390 struct inode *inode = iocb->ki_filp->f_mapping->host;
1391 struct ubifs_info *c = inode->i_sb->s_fs_info;
1392
1393 err = update_mctime(c, inode);
1394 if (err)
1395 return err;
1396
Christoph Hellwig774888b2009-09-30 22:17:16 +02001397 return generic_file_aio_write(iocb, iov, nr_segs, pos);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001398}
1399
1400static int ubifs_set_page_dirty(struct page *page)
1401{
1402 int ret;
1403
1404 ret = __set_page_dirty_nobuffers(page);
1405 /*
1406 * An attempt to dirty a page without budgeting for it - should not
1407 * happen.
1408 */
1409 ubifs_assert(ret == 0);
1410 return ret;
1411}
1412
1413static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1414{
1415 /*
1416 * An attempt to release a dirty page without budgeting for it - should
1417 * not happen.
1418 */
1419 if (PageWriteback(page))
1420 return 0;
1421 ubifs_assert(PagePrivate(page));
1422 ubifs_assert(0);
1423 ClearPagePrivate(page);
1424 ClearPageChecked(page);
1425 return 1;
1426}
1427
1428/*
1429 * mmap()d file has taken write protection fault and is being made
1430 * writable. UBIFS must ensure page is budgeted for.
1431 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07001432static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001433{
Nick Pigginc2ec1752009-03-31 15:23:21 -07001434 struct page *page = vmf->page;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001435 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1436 struct ubifs_info *c = inode->i_sb->s_fs_info;
1437 struct timespec now = ubifs_current_time(inode);
1438 struct ubifs_budget_req req = { .new_page = 1 };
1439 int err, update_time;
1440
1441 dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1442 i_size_read(inode));
Artem Bityutskiy2ef13292010-09-19 18:34:26 +03001443 ubifs_assert(!c->ro_media && !c->ro_mount);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001444
Artem Bityutskiy2680d722010-09-17 16:44:28 +03001445 if (unlikely(c->ro_error))
Nick Pigginc2ec1752009-03-31 15:23:21 -07001446 return VM_FAULT_SIGBUS; /* -EROFS */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001447
1448 /*
1449 * We have not locked @page so far so we may budget for changing the
1450 * page. Note, we cannot do this after we locked the page, because
1451 * budgeting may cause write-back which would cause deadlock.
1452 *
1453 * At the moment we do not know whether the page is dirty or not, so we
1454 * assume that it is not and budget for a new page. We could look at
1455 * the @PG_private flag and figure this out, but we may race with write
1456 * back and the page state may change by the time we lock it, so this
1457 * would need additional care. We do not bother with this at the
1458 * moment, although it might be good idea to do. Instead, we allocate
1459 * budget for a new page and amend it later on if the page was in fact
1460 * dirty.
1461 *
1462 * The budgeting-related logic of this function is similar to what we
1463 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1464 * for more comments.
1465 */
1466 update_time = mctime_update_needed(inode, &now);
1467 if (update_time)
1468 /*
1469 * We have to change inode time stamp which requires extra
1470 * budgeting.
1471 */
1472 req.dirtied_ino = 1;
1473
1474 err = ubifs_budget_space(c, &req);
1475 if (unlikely(err)) {
1476 if (err == -ENOSPC)
1477 ubifs_warn("out of space for mmapped file "
1478 "(inode number %lu)", inode->i_ino);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001479 return VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001480 }
1481
1482 lock_page(page);
1483 if (unlikely(page->mapping != inode->i_mapping ||
1484 page_offset(page) > i_size_read(inode))) {
1485 /* Page got truncated out from underneath us */
1486 err = -EINVAL;
1487 goto out_unlock;
1488 }
1489
1490 if (PagePrivate(page))
1491 release_new_page_budget(c);
1492 else {
1493 if (!PageChecked(page))
1494 ubifs_convert_page_budget(c);
1495 SetPagePrivate(page);
1496 atomic_long_inc(&c->dirty_pg_cnt);
1497 __set_page_dirty_nobuffers(page);
1498 }
1499
1500 if (update_time) {
1501 int release;
1502 struct ubifs_inode *ui = ubifs_inode(inode);
1503
1504 mutex_lock(&ui->ui_mutex);
1505 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1506 release = ui->dirty;
1507 mark_inode_dirty_sync(inode);
1508 mutex_unlock(&ui->ui_mutex);
1509 if (release)
1510 ubifs_release_dirty_inode_budget(c, ui);
1511 }
1512
1513 unlock_page(page);
1514 return 0;
1515
1516out_unlock:
1517 unlock_page(page);
1518 ubifs_release_budget(c, &req);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001519 if (err)
1520 err = VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001521 return err;
1522}
1523
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001524static const struct vm_operations_struct ubifs_file_vm_ops = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001525 .fault = filemap_fault,
1526 .page_mkwrite = ubifs_vm_page_mkwrite,
1527};
1528
1529static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1530{
1531 int err;
1532
1533 /* 'generic_file_mmap()' takes care of NOMMU case */
1534 err = generic_file_mmap(file, vma);
1535 if (err)
1536 return err;
1537 vma->vm_ops = &ubifs_file_vm_ops;
1538 return 0;
1539}
1540
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001541const struct address_space_operations ubifs_file_address_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001542 .readpage = ubifs_readpage,
1543 .writepage = ubifs_writepage,
1544 .write_begin = ubifs_write_begin,
1545 .write_end = ubifs_write_end,
1546 .invalidatepage = ubifs_invalidatepage,
1547 .set_page_dirty = ubifs_set_page_dirty,
1548 .releasepage = ubifs_releasepage,
1549};
1550
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001551const struct inode_operations ubifs_file_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001552 .setattr = ubifs_setattr,
1553 .getattr = ubifs_getattr,
1554#ifdef CONFIG_UBIFS_FS_XATTR
1555 .setxattr = ubifs_setxattr,
1556 .getxattr = ubifs_getxattr,
1557 .listxattr = ubifs_listxattr,
1558 .removexattr = ubifs_removexattr,
1559#endif
1560};
1561
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001562const struct inode_operations ubifs_symlink_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001563 .readlink = generic_readlink,
1564 .follow_link = ubifs_follow_link,
1565 .setattr = ubifs_setattr,
1566 .getattr = ubifs_getattr,
1567};
1568
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001569const struct file_operations ubifs_file_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001570 .llseek = generic_file_llseek,
1571 .read = do_sync_read,
1572 .write = do_sync_write,
1573 .aio_read = generic_file_aio_read,
1574 .aio_write = ubifs_aio_write,
1575 .mmap = ubifs_file_mmap,
1576 .fsync = ubifs_fsync,
1577 .unlocked_ioctl = ubifs_ioctl,
1578 .splice_read = generic_file_splice_read,
Zoltan Sogor22bc7fa2008-07-28 16:28:49 +02001579 .splice_write = generic_file_splice_write,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001580#ifdef CONFIG_COMPAT
1581 .compat_ioctl = ubifs_compat_ioctl,
1582#endif
1583};