blob: 14374530784c683f36a15e52b64d792578a8ec07 [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
Artem Bityutskiy5c57f202012-07-25 18:12:14 +030040 * called with @i_mutex unlocked. For example, when flusher thread is doing
41 * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
42 * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
43 * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
44 * 'ubifs_writepage()' 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"
Kent Overstreeta27bb332013-05-07 16:19:08 -070053#include <linux/aio.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030054#include <linux/mount.h>
Al Viro3f8206d2008-07-26 03:46:43 -040055#include <linux/namei.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090056#include <linux/slab.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);
Artem Bityutskiyedf6be22012-05-16 19:15:56 +0300101 ubifs_dump_node(c, dn);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300102 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{
Artem Bityutskiyb1375452011-03-29 18:04:05 +0300216 struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300217
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);
Artem Bityutskiy2ef13292010-09-19 18:34:26 +0300437 ubifs_assert(!c->ro_media && !c->ro_mount);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300438
Artem Bityutskiy2680d722010-09-17 16:44:28 +0300439 if (unlikely(c->ro_error))
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300440 return -EROFS;
441
442 /* Try out the fast-path part first */
Nick Piggin54566b22009-01-04 12:00:53 -0800443 page = grab_cache_page_write_begin(mapping, index, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300444 if (unlikely(!page))
445 return -ENOMEM;
446
447 if (!PageUptodate(page)) {
448 /* The page is not loaded from the flash */
Adrian Hunterf55aa592009-02-23 12:47:25 +0200449 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300450 /*
451 * We change whole page so no need to load it. But we
Artem Bityutskiy6ed09c32011-03-23 10:32:58 +0200452 * do not know whether this page exists on the media or
453 * not, so we assume the latter because it requires
454 * larger budget. The assumption is that it is better
455 * to budget a bit more than to read the page from the
456 * media. Thus, we are setting the @PG_checked flag
457 * here.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300458 */
459 SetPageChecked(page);
Adrian Hunterf55aa592009-02-23 12:47:25 +0200460 skipped_read = 1;
461 } else {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300462 err = do_readpage(page);
463 if (err) {
464 unlock_page(page);
465 page_cache_release(page);
466 return err;
467 }
468 }
469
470 SetPageUptodate(page);
471 ClearPageError(page);
472 }
473
474 err = allocate_budget(c, page, ui, appending);
475 if (unlikely(err)) {
476 ubifs_assert(err == -ENOSPC);
477 /*
Adrian Hunterf55aa592009-02-23 12:47:25 +0200478 * If we skipped reading the page because we were going to
479 * write all of it, then it is not up to date.
480 */
481 if (skipped_read) {
482 ClearPageChecked(page);
483 ClearPageUptodate(page);
484 }
485 /*
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300486 * Budgeting failed which means it would have to force
487 * write-back but didn't, because we set the @fast flag in the
488 * request. Write-back cannot be done now, while we have the
489 * page locked, because it would deadlock. Unlock and free
490 * everything and fall-back to slow-path.
491 */
492 if (appending) {
493 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
494 mutex_unlock(&ui->ui_mutex);
495 }
496 unlock_page(page);
497 page_cache_release(page);
498
Nick Piggin54566b22009-01-04 12:00:53 -0800499 return write_begin_slow(mapping, pos, len, pagep, flags);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300500 }
501
502 /*
Artem Bityutskiy873a64c2009-09-01 17:06:43 +0300503 * Whee, we acquired budgeting quickly - without involving
504 * garbage-collection, committing or forcing write-back. We return
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300505 * with @ui->ui_mutex locked if we are appending pages, and unlocked
506 * otherwise. This is an optimization (slightly hacky though).
507 */
508 *pagep = page;
509 return 0;
510
511}
512
513/**
514 * cancel_budget - cancel budget.
515 * @c: UBIFS file-system description object
516 * @page: page to cancel budget for
517 * @ui: UBIFS inode object the page belongs to
518 * @appending: non-zero if the page is appended
519 *
520 * This is a helper function for a page write operation. It unlocks the
521 * @ui->ui_mutex in case of appending.
522 */
523static void cancel_budget(struct ubifs_info *c, struct page *page,
524 struct ubifs_inode *ui, int appending)
525{
526 if (appending) {
527 if (!ui->dirty)
528 ubifs_release_dirty_inode_budget(c, ui);
529 mutex_unlock(&ui->ui_mutex);
530 }
531 if (!PagePrivate(page)) {
532 if (PageChecked(page))
533 release_new_page_budget(c);
534 else
535 release_existing_page_budget(c);
536 }
537}
538
539static int ubifs_write_end(struct file *file, struct address_space *mapping,
540 loff_t pos, unsigned len, unsigned copied,
541 struct page *page, void *fsdata)
542{
543 struct inode *inode = mapping->host;
544 struct ubifs_inode *ui = ubifs_inode(inode);
545 struct ubifs_info *c = inode->i_sb->s_fs_info;
546 loff_t end_pos = pos + len;
547 int appending = !!(end_pos > inode->i_size);
548
549 dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
550 inode->i_ino, pos, page->index, len, copied, inode->i_size);
551
552 if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
553 /*
554 * VFS copied less data to the page that it intended and
555 * declared in its '->write_begin()' call via the @len
556 * argument. If the page was not up-to-date, and @len was
557 * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
558 * not load it from the media (for optimization reasons). This
559 * means that part of the page contains garbage. So read the
560 * page now.
561 */
562 dbg_gen("copied %d instead of %d, read page and repeat",
563 copied, len);
564 cancel_budget(c, page, ui, appending);
Artem Bityutskiy6ed09c32011-03-23 10:32:58 +0200565 ClearPageChecked(page);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300566
567 /*
568 * Return 0 to force VFS to repeat the whole operation, or the
Artem Bityutskiy873a64c2009-09-01 17:06:43 +0300569 * error code if 'do_readpage()' fails.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300570 */
571 copied = do_readpage(page);
572 goto out;
573 }
574
575 if (!PagePrivate(page)) {
576 SetPagePrivate(page);
577 atomic_long_inc(&c->dirty_pg_cnt);
578 __set_page_dirty_nobuffers(page);
579 }
580
581 if (appending) {
582 i_size_write(inode, end_pos);
583 ui->ui_size = end_pos;
584 /*
585 * Note, we do not set @I_DIRTY_PAGES (which means that the
586 * inode has dirty pages), this has been done in
587 * '__set_page_dirty_nobuffers()'.
588 */
589 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
590 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
591 mutex_unlock(&ui->ui_mutex);
592 }
593
594out:
595 unlock_page(page);
596 page_cache_release(page);
597 return copied;
598}
599
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300600/**
601 * populate_page - copy data nodes into a page for bulk-read.
602 * @c: UBIFS file-system description object
603 * @page: page
604 * @bu: bulk-read information
605 * @n: next zbranch slot
606 *
607 * This function returns %0 on success and a negative error code on failure.
608 */
609static int populate_page(struct ubifs_info *c, struct page *page,
610 struct bu_info *bu, int *n)
611{
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300612 int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300613 struct inode *inode = page->mapping->host;
614 loff_t i_size = i_size_read(inode);
615 unsigned int page_block;
616 void *addr, *zaddr;
617 pgoff_t end_index;
618
619 dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
620 inode->i_ino, page->index, i_size, page->flags);
621
622 addr = zaddr = kmap(page);
623
Adrian Huntered382d52008-09-05 16:17:42 +0300624 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300625 if (!i_size || page->index > end_index) {
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300626 hole = 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300627 memset(addr, 0, PAGE_CACHE_SIZE);
628 goto out_hole;
629 }
630
631 page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
632 while (1) {
633 int err, len, out_len, dlen;
634
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300635 if (nn >= bu->cnt) {
636 hole = 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300637 memset(addr, 0, UBIFS_BLOCK_SIZE);
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300638 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300639 struct ubifs_data_node *dn;
640
641 dn = bu->buf + (bu->zbranch[nn].offs - offs);
642
Harvey Harrison0ecb9522008-10-24 10:52:57 -0700643 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300644 ubifs_inode(inode)->creat_sqnum);
645
646 len = le32_to_cpu(dn->size);
647 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
648 goto out_err;
649
650 dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
651 out_len = UBIFS_BLOCK_SIZE;
652 err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
653 le16_to_cpu(dn->compr_type));
654 if (err || len != out_len)
655 goto out_err;
656
657 if (len < UBIFS_BLOCK_SIZE)
658 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
659
660 nn += 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300661 read = (i << UBIFS_BLOCK_SHIFT) + len;
Adrian Hunter5c0013c2008-09-12 10:34:51 +0300662 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
663 nn += 1;
664 continue;
665 } else {
666 hole = 1;
667 memset(addr, 0, UBIFS_BLOCK_SIZE);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300668 }
669 if (++i >= UBIFS_BLOCKS_PER_PAGE)
670 break;
671 addr += UBIFS_BLOCK_SIZE;
672 page_block += 1;
673 }
674
675 if (end_index == page->index) {
676 int len = i_size & (PAGE_CACHE_SIZE - 1);
677
Adrian Huntered382d52008-09-05 16:17:42 +0300678 if (len && len < read)
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300679 memset(zaddr + len, 0, read - len);
680 }
681
682out_hole:
683 if (hole) {
684 SetPageChecked(page);
685 dbg_gen("hole");
686 }
687
688 SetPageUptodate(page);
689 ClearPageError(page);
690 flush_dcache_page(page);
691 kunmap(page);
692 *n = nn;
693 return 0;
694
695out_err:
696 ClearPageUptodate(page);
697 SetPageError(page);
698 flush_dcache_page(page);
699 kunmap(page);
700 ubifs_err("bad data node (block %u, inode %lu)",
701 page_block, inode->i_ino);
702 return -EINVAL;
703}
704
705/**
706 * ubifs_do_bulk_read - do bulk-read.
707 * @c: UBIFS file-system description object
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200708 * @bu: bulk-read information
709 * @page1: first page to read
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300710 *
711 * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
712 */
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200713static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
714 struct page *page1)
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300715{
716 pgoff_t offset = page1->index, end_index;
717 struct address_space *mapping = page1->mapping;
718 struct inode *inode = mapping->host;
719 struct ubifs_inode *ui = ubifs_inode(inode);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300720 int err, page_idx, page_cnt, ret = 0, n = 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200721 int allocate = bu->buf ? 0 : 1;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300722 loff_t isize;
723
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300724 err = ubifs_tnc_get_bu_keys(c, bu);
725 if (err)
726 goto out_warn;
727
728 if (bu->eof) {
729 /* Turn off bulk-read at the end of the file */
730 ui->read_in_a_row = 1;
731 ui->bulk_read = 0;
732 }
733
734 page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
735 if (!page_cnt) {
736 /*
737 * This happens when there are multiple blocks per page and the
738 * blocks for the first page we are looking for, are not
739 * together. If all the pages were like this, bulk-read would
740 * reduce performance, so we turn it off for a while.
741 */
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200742 goto out_bu_off;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300743 }
744
745 if (bu->cnt) {
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200746 if (allocate) {
747 /*
748 * Allocate bulk-read buffer depending on how many data
749 * nodes we are going to read.
750 */
751 bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
752 bu->zbranch[bu->cnt - 1].len -
753 bu->zbranch[0].offs;
754 ubifs_assert(bu->buf_len > 0);
755 ubifs_assert(bu->buf_len <= c->leb_size);
756 bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
757 if (!bu->buf)
758 goto out_bu_off;
759 }
760
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300761 err = ubifs_tnc_bulk_read(c, bu);
762 if (err)
763 goto out_warn;
764 }
765
766 err = populate_page(c, page1, bu, &n);
767 if (err)
768 goto out_warn;
769
770 unlock_page(page1);
771 ret = 1;
772
773 isize = i_size_read(inode);
774 if (isize == 0)
775 goto out_free;
776 end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
777
778 for (page_idx = 1; page_idx < page_cnt; page_idx++) {
779 pgoff_t page_offset = offset + page_idx;
780 struct page *page;
781
782 if (page_offset > end_index)
783 break;
784 page = find_or_create_page(mapping, page_offset,
785 GFP_NOFS | __GFP_COLD);
786 if (!page)
787 break;
788 if (!PageUptodate(page))
789 err = populate_page(c, page, bu, &n);
790 unlock_page(page);
791 page_cache_release(page);
792 if (err)
793 break;
794 }
795
796 ui->last_page_read = offset + page_idx - 1;
797
798out_free:
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200799 if (allocate)
800 kfree(bu->buf);
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300801 return ret;
802
803out_warn:
804 ubifs_warn("ignoring error %d and skipping bulk-read", err);
805 goto out_free;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200806
807out_bu_off:
808 ui->read_in_a_row = ui->bulk_read = 0;
809 goto out_free;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300810}
811
812/**
813 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
814 * @page: page from which to start bulk-read.
815 *
816 * Some flash media are capable of reading sequentially at faster rates. UBIFS
817 * bulk-read facility is designed to take advantage of that, by reading in one
818 * go consecutive data nodes that are also located consecutively in the same
819 * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
820 */
821static int ubifs_bulk_read(struct page *page)
822{
823 struct inode *inode = page->mapping->host;
824 struct ubifs_info *c = inode->i_sb->s_fs_info;
825 struct ubifs_inode *ui = ubifs_inode(inode);
826 pgoff_t index = page->index, last_page_read = ui->last_page_read;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200827 struct bu_info *bu;
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200828 int err = 0, allocated = 0;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300829
830 ui->last_page_read = index;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300831 if (!c->bulk_read)
832 return 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200833
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300834 /*
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200835 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
836 * so don't bother if we cannot lock the mutex.
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300837 */
838 if (!mutex_trylock(&ui->ui_mutex))
839 return 0;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200840
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300841 if (index != last_page_read + 1) {
842 /* Turn off bulk-read if we stop reading sequentially */
843 ui->read_in_a_row = 1;
844 if (ui->bulk_read)
845 ui->bulk_read = 0;
846 goto out_unlock;
847 }
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200848
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300849 if (!ui->bulk_read) {
850 ui->read_in_a_row += 1;
851 if (ui->read_in_a_row < 3)
852 goto out_unlock;
853 /* Three reads in a row, so switch on bulk-read */
854 ui->bulk_read = 1;
855 }
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200856
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200857 /*
858 * If possible, try to use pre-allocated bulk-read information, which
859 * is protected by @c->bu_mutex.
860 */
861 if (mutex_trylock(&c->bu_mutex))
862 bu = &c->bu;
863 else {
864 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
865 if (!bu)
866 goto out_unlock;
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200867
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200868 bu->buf = NULL;
869 allocated = 1;
870 }
871
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200872 bu->buf_len = c->max_bu_buf_len;
873 data_key_init(c, &bu->key, inode->i_ino,
874 page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200875 err = ubifs_do_bulk_read(c, bu, page);
Artem Bityutskiy3477d202008-11-19 11:53:15 +0200876
877 if (!allocated)
878 mutex_unlock(&c->bu_mutex);
879 else
880 kfree(bu);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200881
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300882out_unlock:
883 mutex_unlock(&ui->ui_mutex);
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +0200884 return err;
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300885}
886
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300887static int ubifs_readpage(struct file *file, struct page *page)
888{
Adrian Hunter4793e7c2008-09-02 16:29:46 +0300889 if (ubifs_bulk_read(page))
890 return 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300891 do_readpage(page);
892 unlock_page(page);
893 return 0;
894}
895
896static int do_writepage(struct page *page, int len)
897{
898 int err = 0, i, blen;
899 unsigned int block;
900 void *addr;
901 union ubifs_key key;
902 struct inode *inode = page->mapping->host;
903 struct ubifs_info *c = inode->i_sb->s_fs_info;
904
905#ifdef UBIFS_DEBUG
906 spin_lock(&ui->ui_lock);
907 ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
908 spin_unlock(&ui->ui_lock);
909#endif
910
911 /* Update radix tree tags */
912 set_page_writeback(page);
913
914 addr = kmap(page);
915 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
916 i = 0;
917 while (len) {
918 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
919 data_key_init(c, &key, inode->i_ino, block);
920 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
921 if (err)
922 break;
923 if (++i >= UBIFS_BLOCKS_PER_PAGE)
924 break;
925 block += 1;
926 addr += blen;
927 len -= blen;
928 }
929 if (err) {
930 SetPageError(page);
931 ubifs_err("cannot write page %lu of inode %lu, error %d",
932 page->index, inode->i_ino, err);
933 ubifs_ro_mode(c, err);
934 }
935
936 ubifs_assert(PagePrivate(page));
937 if (PageChecked(page))
938 release_new_page_budget(c);
939 else
940 release_existing_page_budget(c);
941
942 atomic_long_dec(&c->dirty_pg_cnt);
943 ClearPagePrivate(page);
944 ClearPageChecked(page);
945
946 kunmap(page);
947 unlock_page(page);
948 end_page_writeback(page);
949 return err;
950}
951
952/*
953 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
954 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
955 * situation when a we have an inode with size 0, then a megabyte of data is
956 * appended to the inode, then write-back starts and flushes some amount of the
957 * dirty pages, the journal becomes full, commit happens and finishes, and then
958 * an unclean reboot happens. When the file system is mounted next time, the
959 * inode size would still be 0, but there would be many pages which are beyond
960 * the inode size, they would be indexed and consume flash space. Because the
961 * journal has been committed, the replay would not be able to detect this
962 * situation and correct the inode size. This means UBIFS would have to scan
963 * whole index and correct all inode sizes, which is long an unacceptable.
964 *
965 * To prevent situations like this, UBIFS writes pages back only if they are
Artem Bityutskiy7d4e9cc2009-03-20 19:11:12 +0200966 * within the last synchronized inode size, i.e. the size which has been
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300967 * written to the flash media last time. Otherwise, UBIFS forces inode
968 * write-back, thus making sure the on-flash inode contains current inode size,
969 * and then keeps writing pages back.
970 *
971 * Some locking issues explanation. 'ubifs_writepage()' first is called with
972 * the page locked, and it locks @ui_mutex. However, write-back does take inode
973 * @i_mutex, which means other VFS operations may be run on this inode at the
974 * same time. And the problematic one is truncation to smaller size, from where
Artem Bityutskiyc4361572011-03-25 15:27:40 +0200975 * we have to call 'truncate_setsize()', which first changes @inode->i_size,
976 * then drops the truncated pages. And while dropping the pages, it takes the
977 * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
978 * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
979 * This means that @inode->i_size is changed while @ui_mutex is unlocked.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300980 *
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200981 * XXX(truncate): with the new truncate sequence this is not true anymore,
982 * and the calls to truncate_setsize can be move around freely. They should
983 * be moved to the very end of the truncate sequence.
npiggin@suse.de15c6fd92010-05-27 01:05:34 +1000984 *
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300985 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
986 * inode size. How do we do this if @inode->i_size may became smaller while we
987 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
988 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
989 * internally and updates it under @ui_mutex.
990 *
991 * Q: why we do not worry that if we race with truncation, we may end up with a
992 * situation when the inode is truncated while we are in the middle of
993 * 'do_writepage()', so we do write beyond inode size?
994 * A: If we are in the middle of 'do_writepage()', truncation would be locked
995 * on the page lock and it would not write the truncated inode node to the
996 * journal before we have finished.
997 */
998static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
999{
1000 struct inode *inode = page->mapping->host;
1001 struct ubifs_inode *ui = ubifs_inode(inode);
1002 loff_t i_size = i_size_read(inode), synced_i_size;
1003 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1004 int err, len = i_size & (PAGE_CACHE_SIZE - 1);
1005 void *kaddr;
1006
1007 dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1008 inode->i_ino, page->index, page->flags);
1009 ubifs_assert(PagePrivate(page));
1010
1011 /* Is the page fully outside @i_size? (truncate in progress) */
1012 if (page->index > end_index || (page->index == end_index && !len)) {
1013 err = 0;
1014 goto out_unlock;
1015 }
1016
1017 spin_lock(&ui->ui_lock);
1018 synced_i_size = ui->synced_i_size;
1019 spin_unlock(&ui->ui_lock);
1020
1021 /* Is the page fully inside @i_size? */
1022 if (page->index < end_index) {
1023 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001024 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001025 if (err)
1026 goto out_unlock;
1027 /*
1028 * The inode has been written, but the write-buffer has
1029 * not been synchronized, so in case of an unclean
1030 * reboot we may end up with some pages beyond inode
1031 * size, but they would be in the journal (because
1032 * commit flushes write buffers) and recovery would deal
1033 * with this.
1034 */
1035 }
1036 return do_writepage(page, PAGE_CACHE_SIZE);
1037 }
1038
1039 /*
1040 * The page straddles @i_size. It must be zeroed out on each and every
1041 * writepage invocation because it may be mmapped. "A file is mapped
1042 * in multiples of the page size. For a file that is not a multiple of
1043 * the page size, the remaining memory is zeroed when mapped, and
1044 * writes to that region are not written out to the file."
1045 */
Cong Wanga1c7c1372011-11-25 23:14:36 +08001046 kaddr = kmap_atomic(page);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001047 memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1048 flush_dcache_page(page);
Cong Wanga1c7c1372011-11-25 23:14:36 +08001049 kunmap_atomic(kaddr);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001050
1051 if (i_size > synced_i_size) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001052 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001053 if (err)
1054 goto out_unlock;
1055 }
1056
1057 return do_writepage(page, len);
1058
1059out_unlock:
1060 unlock_page(page);
1061 return err;
1062}
1063
1064/**
1065 * do_attr_changes - change inode attributes.
1066 * @inode: inode to change attributes for
1067 * @attr: describes attributes to change
1068 */
1069static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1070{
1071 if (attr->ia_valid & ATTR_UID)
1072 inode->i_uid = attr->ia_uid;
1073 if (attr->ia_valid & ATTR_GID)
1074 inode->i_gid = attr->ia_gid;
1075 if (attr->ia_valid & ATTR_ATIME)
1076 inode->i_atime = timespec_trunc(attr->ia_atime,
1077 inode->i_sb->s_time_gran);
1078 if (attr->ia_valid & ATTR_MTIME)
1079 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1080 inode->i_sb->s_time_gran);
1081 if (attr->ia_valid & ATTR_CTIME)
1082 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1083 inode->i_sb->s_time_gran);
1084 if (attr->ia_valid & ATTR_MODE) {
1085 umode_t mode = attr->ia_mode;
1086
1087 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1088 mode &= ~S_ISGID;
1089 inode->i_mode = mode;
1090 }
1091}
1092
1093/**
1094 * do_truncation - truncate an inode.
1095 * @c: UBIFS file-system description object
1096 * @inode: inode to truncate
1097 * @attr: inode attribute changes description
1098 *
1099 * This function implements VFS '->setattr()' call when the inode is truncated
1100 * to a smaller size. Returns zero in case of success and a negative error code
1101 * in case of failure.
1102 */
1103static int do_truncation(struct ubifs_info *c, struct inode *inode,
1104 const struct iattr *attr)
1105{
1106 int err;
1107 struct ubifs_budget_req req;
1108 loff_t old_size = inode->i_size, new_size = attr->ia_size;
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001109 int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001110 struct ubifs_inode *ui = ubifs_inode(inode);
1111
1112 dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1113 memset(&req, 0, sizeof(struct ubifs_budget_req));
1114
1115 /*
1116 * If this is truncation to a smaller size, and we do not truncate on a
1117 * block boundary, budget for changing one data block, because the last
1118 * block will be re-written.
1119 */
1120 if (new_size & (UBIFS_BLOCK_SIZE - 1))
1121 req.dirtied_page = 1;
1122
1123 req.dirtied_ino = 1;
1124 /* A funny way to budget for truncation node */
1125 req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1126 err = ubifs_budget_space(c, &req);
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001127 if (err) {
1128 /*
1129 * Treat truncations to zero as deletion and always allow them,
1130 * just like we do for '->unlink()'.
1131 */
1132 if (new_size || err != -ENOSPC)
1133 return err;
1134 budgeted = 0;
1135 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001136
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001137 truncate_setsize(inode, new_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001138
1139 if (offset) {
1140 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1141 struct page *page;
1142
1143 page = find_lock_page(inode->i_mapping, index);
1144 if (page) {
1145 if (PageDirty(page)) {
1146 /*
1147 * 'ubifs_jnl_truncate()' will try to truncate
1148 * the last data node, but it contains
1149 * out-of-date data because the page is dirty.
1150 * Write the page now, so that
1151 * 'ubifs_jnl_truncate()' will see an already
1152 * truncated (and up to date) data node.
1153 */
1154 ubifs_assert(PagePrivate(page));
1155
1156 clear_page_dirty_for_io(page);
1157 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1158 offset = new_size &
1159 (PAGE_CACHE_SIZE - 1);
1160 err = do_writepage(page, offset);
1161 page_cache_release(page);
1162 if (err)
1163 goto out_budg;
1164 /*
1165 * We could now tell 'ubifs_jnl_truncate()' not
1166 * to read the last block.
1167 */
1168 } else {
1169 /*
1170 * We could 'kmap()' the page and pass the data
1171 * to 'ubifs_jnl_truncate()' to save it from
1172 * having to read it.
1173 */
1174 unlock_page(page);
1175 page_cache_release(page);
1176 }
1177 }
1178 }
1179
1180 mutex_lock(&ui->ui_mutex);
1181 ui->ui_size = inode->i_size;
1182 /* Truncation changes inode [mc]time */
1183 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
Artem Bityutskiy873a64c2009-09-01 17:06:43 +03001184 /* Other attributes may be changed at the same time as well */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001185 do_attr_changes(inode, attr);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001186 err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1187 mutex_unlock(&ui->ui_mutex);
Artem Bityutskiy873a64c2009-09-01 17:06:43 +03001188
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001189out_budg:
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001190 if (budgeted)
1191 ubifs_release_budget(c, &req);
1192 else {
Artem Bityutskiyb1375452011-03-29 18:04:05 +03001193 c->bi.nospace = c->bi.nospace_rp = 0;
Artem Bityutskiy04da11b2008-08-20 17:16:34 +03001194 smp_wmb();
1195 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001196 return err;
1197}
1198
1199/**
1200 * do_setattr - change inode attributes.
1201 * @c: UBIFS file-system description object
1202 * @inode: inode to change attributes for
1203 * @attr: inode attribute changes description
1204 *
1205 * This function implements VFS '->setattr()' call for all cases except
1206 * truncations to smaller size. Returns zero in case of success and a negative
1207 * error code in case of failure.
1208 */
1209static int do_setattr(struct ubifs_info *c, struct inode *inode,
1210 const struct iattr *attr)
1211{
1212 int err, release;
1213 loff_t new_size = attr->ia_size;
1214 struct ubifs_inode *ui = ubifs_inode(inode);
1215 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001216 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001217
1218 err = ubifs_budget_space(c, &req);
1219 if (err)
1220 return err;
1221
1222 if (attr->ia_valid & ATTR_SIZE) {
1223 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001224 truncate_setsize(inode, new_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001225 }
1226
1227 mutex_lock(&ui->ui_mutex);
1228 if (attr->ia_valid & ATTR_SIZE) {
1229 /* Truncation changes inode [mc]time */
1230 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001231 /* 'truncate_setsize()' changed @i_size, update @ui_size */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001232 ui->ui_size = inode->i_size;
1233 }
1234
1235 do_attr_changes(inode, attr);
1236
1237 release = ui->dirty;
1238 if (attr->ia_valid & ATTR_SIZE)
1239 /*
1240 * Inode length changed, so we have to make sure
1241 * @I_DIRTY_DATASYNC is set.
1242 */
1243 __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1244 else
1245 mark_inode_dirty_sync(inode);
1246 mutex_unlock(&ui->ui_mutex);
1247
1248 if (release)
1249 ubifs_release_budget(c, &req);
1250 if (IS_SYNC(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001251 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001252 return err;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001253}
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
Artem Bityutskiyd808efb2011-05-31 18:14:38 +03001267 err = dbg_check_synced_i_size(c, inode);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001268 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
Josef Bacik02c24a82011-07-16 20:44:56 -04001308int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001309{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001310 struct inode *inode = file->f_mapping->host;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001311 struct ubifs_info *c = inode->i_sb->s_fs_info;
1312 int err;
1313
1314 dbg_gen("syncing inode %lu", inode->i_ino);
1315
Artem Bityutskiy3b2f9a02011-04-21 15:11:33 +03001316 if (c->ro_mount)
1317 /*
1318 * For some really strange reasons VFS does not filter out
1319 * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1320 */
Artem Bityutskiy78530bf2011-04-13 10:31:52 +03001321 return 0;
1322
Josef Bacik02c24a82011-07-16 20:44:56 -04001323 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
1324 if (err)
1325 return err;
1326 mutex_lock(&inode->i_mutex);
1327
1328 /* Synchronize the inode unless this is a 'datasync()' call. */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001329 if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001330 err = inode->i_sb->s_op->write_inode(inode, NULL);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001331 if (err)
Josef Bacik02c24a82011-07-16 20:44:56 -04001332 goto out;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001333 }
1334
1335 /*
1336 * Nodes related to this inode may still sit in a write-buffer. Flush
1337 * them.
1338 */
1339 err = ubifs_sync_wbufs_by_inode(c, inode);
Josef Bacik02c24a82011-07-16 20:44:56 -04001340out:
1341 mutex_unlock(&inode->i_mutex);
1342 return err;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001343}
1344
1345/**
1346 * mctime_update_needed - check if mtime or ctime update is needed.
1347 * @inode: the inode to do the check for
1348 * @now: current time
1349 *
1350 * This helper function checks if the inode mtime/ctime should be updated or
1351 * not. If current values of the time-stamps are within the UBIFS inode time
1352 * granularity, they are not updated. This is an optimization.
1353 */
1354static inline int mctime_update_needed(const struct inode *inode,
1355 const struct timespec *now)
1356{
1357 if (!timespec_equal(&inode->i_mtime, now) ||
1358 !timespec_equal(&inode->i_ctime, now))
1359 return 1;
1360 return 0;
1361}
1362
1363/**
1364 * update_ctime - update mtime and ctime of an inode.
1365 * @c: UBIFS file-system description object
1366 * @inode: inode to update
1367 *
1368 * This function updates mtime and ctime of the inode if it is not equivalent to
1369 * current time. Returns zero in case of success and a negative error code in
1370 * case of failure.
1371 */
1372static int update_mctime(struct ubifs_info *c, struct inode *inode)
1373{
1374 struct timespec now = ubifs_current_time(inode);
1375 struct ubifs_inode *ui = ubifs_inode(inode);
1376
1377 if (mctime_update_needed(inode, &now)) {
1378 int err, release;
1379 struct ubifs_budget_req req = { .dirtied_ino = 1,
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +03001380 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001381
1382 err = ubifs_budget_space(c, &req);
1383 if (err)
1384 return err;
1385
1386 mutex_lock(&ui->ui_mutex);
1387 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1388 release = ui->dirty;
1389 mark_inode_dirty_sync(inode);
1390 mutex_unlock(&ui->ui_mutex);
1391 if (release)
1392 ubifs_release_budget(c, &req);
1393 }
1394
1395 return 0;
1396}
1397
1398static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1399 unsigned long nr_segs, loff_t pos)
1400{
1401 int err;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001402 struct inode *inode = iocb->ki_filp->f_mapping->host;
1403 struct ubifs_info *c = inode->i_sb->s_fs_info;
1404
1405 err = update_mctime(c, inode);
1406 if (err)
1407 return err;
1408
Christoph Hellwig774888b2009-09-30 22:17:16 +02001409 return generic_file_aio_write(iocb, iov, nr_segs, pos);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001410}
1411
1412static int ubifs_set_page_dirty(struct page *page)
1413{
1414 int ret;
1415
1416 ret = __set_page_dirty_nobuffers(page);
1417 /*
1418 * An attempt to dirty a page without budgeting for it - should not
1419 * happen.
1420 */
1421 ubifs_assert(ret == 0);
1422 return ret;
1423}
1424
1425static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1426{
1427 /*
1428 * An attempt to release a dirty page without budgeting for it - should
1429 * not happen.
1430 */
1431 if (PageWriteback(page))
1432 return 0;
1433 ubifs_assert(PagePrivate(page));
1434 ubifs_assert(0);
1435 ClearPagePrivate(page);
1436 ClearPageChecked(page);
1437 return 1;
1438}
1439
1440/*
Artem Bityutskiyc4361572011-03-25 15:27:40 +02001441 * mmap()d file has taken write protection fault and is being made writable.
1442 * UBIFS must ensure page is budgeted for.
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001443 */
Artem Bityutskiyc4361572011-03-25 15:27:40 +02001444static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma,
1445 struct vm_fault *vmf)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001446{
Nick Pigginc2ec1752009-03-31 15:23:21 -07001447 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05001448 struct inode *inode = file_inode(vma->vm_file);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001449 struct ubifs_info *c = inode->i_sb->s_fs_info;
1450 struct timespec now = ubifs_current_time(inode);
1451 struct ubifs_budget_req req = { .new_page = 1 };
1452 int err, update_time;
1453
1454 dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1455 i_size_read(inode));
Artem Bityutskiy2ef13292010-09-19 18:34:26 +03001456 ubifs_assert(!c->ro_media && !c->ro_mount);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001457
Artem Bityutskiy2680d722010-09-17 16:44:28 +03001458 if (unlikely(c->ro_error))
Nick Pigginc2ec1752009-03-31 15:23:21 -07001459 return VM_FAULT_SIGBUS; /* -EROFS */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001460
1461 /*
1462 * We have not locked @page so far so we may budget for changing the
1463 * page. Note, we cannot do this after we locked the page, because
1464 * budgeting may cause write-back which would cause deadlock.
1465 *
1466 * At the moment we do not know whether the page is dirty or not, so we
1467 * assume that it is not and budget for a new page. We could look at
1468 * the @PG_private flag and figure this out, but we may race with write
1469 * back and the page state may change by the time we lock it, so this
1470 * would need additional care. We do not bother with this at the
1471 * moment, although it might be good idea to do. Instead, we allocate
1472 * budget for a new page and amend it later on if the page was in fact
1473 * dirty.
1474 *
1475 * The budgeting-related logic of this function is similar to what we
1476 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1477 * for more comments.
1478 */
1479 update_time = mctime_update_needed(inode, &now);
1480 if (update_time)
1481 /*
1482 * We have to change inode time stamp which requires extra
1483 * budgeting.
1484 */
1485 req.dirtied_ino = 1;
1486
1487 err = ubifs_budget_space(c, &req);
1488 if (unlikely(err)) {
1489 if (err == -ENOSPC)
Artem Bityutskiy79fda512012-08-27 13:34:09 +03001490 ubifs_warn("out of space for mmapped file (inode number %lu)",
1491 inode->i_ino);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001492 return VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001493 }
1494
1495 lock_page(page);
1496 if (unlikely(page->mapping != inode->i_mapping ||
1497 page_offset(page) > i_size_read(inode))) {
1498 /* Page got truncated out from underneath us */
1499 err = -EINVAL;
1500 goto out_unlock;
1501 }
1502
1503 if (PagePrivate(page))
1504 release_new_page_budget(c);
1505 else {
1506 if (!PageChecked(page))
1507 ubifs_convert_page_budget(c);
1508 SetPagePrivate(page);
1509 atomic_long_inc(&c->dirty_pg_cnt);
1510 __set_page_dirty_nobuffers(page);
1511 }
1512
1513 if (update_time) {
1514 int release;
1515 struct ubifs_inode *ui = ubifs_inode(inode);
1516
1517 mutex_lock(&ui->ui_mutex);
1518 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1519 release = ui->dirty;
1520 mark_inode_dirty_sync(inode);
1521 mutex_unlock(&ui->ui_mutex);
1522 if (release)
1523 ubifs_release_dirty_inode_budget(c, ui);
1524 }
1525
Jan Kara182dcfd2013-02-21 16:42:59 -08001526 wait_for_stable_page(page);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001527 unlock_page(page);
1528 return 0;
1529
1530out_unlock:
1531 unlock_page(page);
1532 ubifs_release_budget(c, &req);
Nick Pigginc2ec1752009-03-31 15:23:21 -07001533 if (err)
1534 err = VM_FAULT_SIGBUS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001535 return err;
1536}
1537
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001538static const struct vm_operations_struct ubifs_file_vm_ops = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001539 .fault = filemap_fault,
1540 .page_mkwrite = ubifs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001541 .remap_pages = generic_file_remap_pages,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001542};
1543
1544static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1545{
1546 int err;
1547
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001548 err = generic_file_mmap(file, vma);
1549 if (err)
1550 return err;
1551 vma->vm_ops = &ubifs_file_vm_ops;
1552 return 0;
1553}
1554
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001555const struct address_space_operations ubifs_file_address_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001556 .readpage = ubifs_readpage,
1557 .writepage = ubifs_writepage,
1558 .write_begin = ubifs_write_begin,
1559 .write_end = ubifs_write_end,
1560 .invalidatepage = ubifs_invalidatepage,
1561 .set_page_dirty = ubifs_set_page_dirty,
1562 .releasepage = ubifs_releasepage,
1563};
1564
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001565const struct inode_operations ubifs_file_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001566 .setattr = ubifs_setattr,
1567 .getattr = ubifs_getattr,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001568 .setxattr = ubifs_setxattr,
1569 .getxattr = ubifs_getxattr,
1570 .listxattr = ubifs_listxattr,
1571 .removexattr = ubifs_removexattr,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001572};
1573
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001574const struct inode_operations ubifs_symlink_inode_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001575 .readlink = generic_readlink,
1576 .follow_link = ubifs_follow_link,
1577 .setattr = ubifs_setattr,
1578 .getattr = ubifs_getattr,
1579};
1580
Artem Bityutskiye8b81562009-01-15 17:43:23 +02001581const struct file_operations ubifs_file_operations = {
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001582 .llseek = generic_file_llseek,
1583 .read = do_sync_read,
1584 .write = do_sync_write,
1585 .aio_read = generic_file_aio_read,
1586 .aio_write = ubifs_aio_write,
1587 .mmap = ubifs_file_mmap,
1588 .fsync = ubifs_fsync,
1589 .unlocked_ioctl = ubifs_ioctl,
1590 .splice_read = generic_file_splice_read,
Zoltan Sogor22bc7fa2008-07-28 16:28:49 +02001591 .splice_write = generic_file_splice_write,
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001592#ifdef CONFIG_COMPAT
1593 .compat_ioctl = ubifs_compat_ioctl,
1594#endif
1595};