~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ubifs/file.c

Version: ~ [ linux-5.5-rc7 ] ~ [ linux-5.4.13 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.97 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.166 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.210 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.210 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.81 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * This file is part of UBIFS.
  3  *
  4  * Copyright (C) 2006-2008 Nokia Corporation.
  5  *
  6  * This program is free software; you can redistribute it and/or modify it
  7  * under the terms of the GNU General Public License version 2 as published by
  8  * the Free Software Foundation.
  9  *
 10  * This program is distributed in the hope that it will be useful, but WITHOUT
 11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 13  * more details.
 14  *
 15  * You should have received a copy of the GNU General Public License along with
 16  * this program; if not, write to the Free Software Foundation, Inc., 51
 17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 18  *
 19  * Authors: Artem Bityutskiy (Битюцкий Артём)
 20  *          Adrian Hunter
 21  */
 22 
 23 /*
 24  * This file implements VFS file and inode operations for regular files, device
 25  * nodes and symlinks as well as address space operations.
 26  *
 27  * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
 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.
 37  *
 38  * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
 39  * implement. However, this is not true for 'ubifs_writepage()', which may be
 40  * called with @i_mutex unlocked. For example, when 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.
 45  *
 46  * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the
 47  * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
 48  * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
 49  * set as well. However, UBIFS disables readahead.
 50  */
 51 
 52 #include "ubifs.h"
 53 #include <linux/mount.h>
 54 #include <linux/slab.h>
 55 #include <linux/migrate.h>
 56 
 57 static int read_block(struct inode *inode, void *addr, unsigned int block,
 58                       struct ubifs_data_node *dn)
 59 {
 60         struct ubifs_info *c = inode->i_sb->s_fs_info;
 61         int err, len, out_len;
 62         union ubifs_key key;
 63         unsigned int dlen;
 64 
 65         data_key_init(c, &key, inode->i_ino, block);
 66         err = ubifs_tnc_lookup(c, &key, dn);
 67         if (err) {
 68                 if (err == -ENOENT)
 69                         /* Not found, so it must be a hole */
 70                         memset(addr, 0, UBIFS_BLOCK_SIZE);
 71                 return err;
 72         }
 73 
 74         ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
 75                      ubifs_inode(inode)->creat_sqnum);
 76         len = le32_to_cpu(dn->size);
 77         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
 78                 goto dump;
 79 
 80         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
 81 
 82         if (ubifs_crypt_is_encrypted(inode)) {
 83                 err = ubifs_decrypt(inode, dn, &dlen, block);
 84                 if (err)
 85                         goto dump;
 86         }
 87 
 88         out_len = UBIFS_BLOCK_SIZE;
 89         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
 90                                le16_to_cpu(dn->compr_type));
 91         if (err || len != out_len)
 92                 goto dump;
 93 
 94         /*
 95          * Data length can be less than a full block, even for blocks that are
 96          * not the last in the file (e.g., as a result of making a hole and
 97          * appending data). Ensure that the remainder is zeroed out.
 98          */
 99         if (len < UBIFS_BLOCK_SIZE)
100                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
101 
102         return 0;
103 
104 dump:
105         ubifs_err(c, "bad data node (block %u, inode %lu)",
106                   block, inode->i_ino);
107         ubifs_dump_node(c, dn);
108         return -EINVAL;
109 }
110 
111 static int do_readpage(struct page *page)
112 {
113         void *addr;
114         int err = 0, i;
115         unsigned int block, beyond;
116         struct ubifs_data_node *dn;
117         struct inode *inode = page->mapping->host;
118         struct ubifs_info *c = inode->i_sb->s_fs_info;
119         loff_t i_size = i_size_read(inode);
120 
121         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
122                 inode->i_ino, page->index, i_size, page->flags);
123         ubifs_assert(c, !PageChecked(page));
124         ubifs_assert(c, !PagePrivate(page));
125 
126         addr = kmap(page);
127 
128         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
129         beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
130         if (block >= beyond) {
131                 /* Reading beyond inode */
132                 SetPageChecked(page);
133                 memset(addr, 0, PAGE_SIZE);
134                 goto out;
135         }
136 
137         dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
138         if (!dn) {
139                 err = -ENOMEM;
140                 goto error;
141         }
142 
143         i = 0;
144         while (1) {
145                 int ret;
146 
147                 if (block >= beyond) {
148                         /* Reading beyond inode */
149                         err = -ENOENT;
150                         memset(addr, 0, UBIFS_BLOCK_SIZE);
151                 } else {
152                         ret = read_block(inode, addr, block, dn);
153                         if (ret) {
154                                 err = ret;
155                                 if (err != -ENOENT)
156                                         break;
157                         } else if (block + 1 == beyond) {
158                                 int dlen = le32_to_cpu(dn->size);
159                                 int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
160 
161                                 if (ilen && ilen < dlen)
162                                         memset(addr + ilen, 0, dlen - ilen);
163                         }
164                 }
165                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
166                         break;
167                 block += 1;
168                 addr += UBIFS_BLOCK_SIZE;
169         }
170         if (err) {
171                 struct ubifs_info *c = inode->i_sb->s_fs_info;
172                 if (err == -ENOENT) {
173                         /* Not found, so it must be a hole */
174                         SetPageChecked(page);
175                         dbg_gen("hole");
176                         goto out_free;
177                 }
178                 ubifs_err(c, "cannot read page %lu of inode %lu, error %d",
179                           page->index, inode->i_ino, err);
180                 goto error;
181         }
182 
183 out_free:
184         kfree(dn);
185 out:
186         SetPageUptodate(page);
187         ClearPageError(page);
188         flush_dcache_page(page);
189         kunmap(page);
190         return 0;
191 
192 error:
193         kfree(dn);
194         ClearPageUptodate(page);
195         SetPageError(page);
196         flush_dcache_page(page);
197         kunmap(page);
198         return err;
199 }
200 
201 /**
202  * release_new_page_budget - release budget of a new page.
203  * @c: UBIFS file-system description object
204  *
205  * This is a helper function which releases budget corresponding to the budget
206  * of one new page of data.
207  */
208 static void release_new_page_budget(struct ubifs_info *c)
209 {
210         struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
211 
212         ubifs_release_budget(c, &req);
213 }
214 
215 /**
216  * release_existing_page_budget - release budget of an existing page.
217  * @c: UBIFS file-system description object
218  *
219  * This is a helper function which releases budget corresponding to the budget
220  * of changing one one page of data which already exists on the flash media.
221  */
222 static void release_existing_page_budget(struct ubifs_info *c)
223 {
224         struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
225 
226         ubifs_release_budget(c, &req);
227 }
228 
229 static int write_begin_slow(struct address_space *mapping,
230                             loff_t pos, unsigned len, struct page **pagep,
231                             unsigned flags)
232 {
233         struct inode *inode = mapping->host;
234         struct ubifs_info *c = inode->i_sb->s_fs_info;
235         pgoff_t index = pos >> PAGE_SHIFT;
236         struct ubifs_budget_req req = { .new_page = 1 };
237         int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
238         struct page *page;
239 
240         dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
241                 inode->i_ino, pos, len, inode->i_size);
242 
243         /*
244          * At the slow path we have to budget before locking the page, because
245          * budgeting may force write-back, which would wait on locked pages and
246          * deadlock if we had the page locked. At this point we do not know
247          * anything about the page, so assume that this is a new page which is
248          * written to a hole. This corresponds to largest budget. Later the
249          * budget will be amended if this is not true.
250          */
251         if (appending)
252                 /* We are appending data, budget for inode change */
253                 req.dirtied_ino = 1;
254 
255         err = ubifs_budget_space(c, &req);
256         if (unlikely(err))
257                 return err;
258 
259         page = grab_cache_page_write_begin(mapping, index, flags);
260         if (unlikely(!page)) {
261                 ubifs_release_budget(c, &req);
262                 return -ENOMEM;
263         }
264 
265         if (!PageUptodate(page)) {
266                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE)
267                         SetPageChecked(page);
268                 else {
269                         err = do_readpage(page);
270                         if (err) {
271                                 unlock_page(page);
272                                 put_page(page);
273                                 ubifs_release_budget(c, &req);
274                                 return err;
275                         }
276                 }
277 
278                 SetPageUptodate(page);
279                 ClearPageError(page);
280         }
281 
282         if (PagePrivate(page))
283                 /*
284                  * The page is dirty, which means it was budgeted twice:
285                  *   o first time the budget was allocated by the task which
286                  *     made the page dirty and set the PG_private flag;
287                  *   o and then we budgeted for it for the second time at the
288                  *     very beginning of this function.
289                  *
290                  * So what we have to do is to release the page budget we
291                  * allocated.
292                  */
293                 release_new_page_budget(c);
294         else if (!PageChecked(page))
295                 /*
296                  * We are changing a page which already exists on the media.
297                  * This means that changing the page does not make the amount
298                  * of indexing information larger, and this part of the budget
299                  * which we have already acquired may be released.
300                  */
301                 ubifs_convert_page_budget(c);
302 
303         if (appending) {
304                 struct ubifs_inode *ui = ubifs_inode(inode);
305 
306                 /*
307                  * 'ubifs_write_end()' is optimized from the fast-path part of
308                  * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
309                  * if data is appended.
310                  */
311                 mutex_lock(&ui->ui_mutex);
312                 if (ui->dirty)
313                         /*
314                          * The inode is dirty already, so we may free the
315                          * budget we allocated.
316                          */
317                         ubifs_release_dirty_inode_budget(c, ui);
318         }
319 
320         *pagep = page;
321         return 0;
322 }
323 
324 /**
325  * allocate_budget - allocate budget for 'ubifs_write_begin()'.
326  * @c: UBIFS file-system description object
327  * @page: page to allocate budget for
328  * @ui: UBIFS inode object the page belongs to
329  * @appending: non-zero if the page is appended
330  *
331  * This is a helper function for 'ubifs_write_begin()' which allocates budget
332  * for the operation. The budget is allocated differently depending on whether
333  * this is appending, whether the page is dirty or not, and so on. This
334  * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
335  * in case of success and %-ENOSPC in case of failure.
336  */
337 static int allocate_budget(struct ubifs_info *c, struct page *page,
338                            struct ubifs_inode *ui, int appending)
339 {
340         struct ubifs_budget_req req = { .fast = 1 };
341 
342         if (PagePrivate(page)) {
343                 if (!appending)
344                         /*
345                          * The page is dirty and we are not appending, which
346                          * means no budget is needed at all.
347                          */
348                         return 0;
349 
350                 mutex_lock(&ui->ui_mutex);
351                 if (ui->dirty)
352                         /*
353                          * The page is dirty and we are appending, so the inode
354                          * has to be marked as dirty. However, it is already
355                          * dirty, so we do not need any budget. We may return,
356                          * but @ui->ui_mutex hast to be left locked because we
357                          * should prevent write-back from flushing the inode
358                          * and freeing the budget. The lock will be released in
359                          * 'ubifs_write_end()'.
360                          */
361                         return 0;
362 
363                 /*
364                  * The page is dirty, we are appending, the inode is clean, so
365                  * we need to budget the inode change.
366                  */
367                 req.dirtied_ino = 1;
368         } else {
369                 if (PageChecked(page))
370                         /*
371                          * The page corresponds to a hole and does not
372                          * exist on the media. So changing it makes
373                          * make the amount of indexing information
374                          * larger, and we have to budget for a new
375                          * page.
376                          */
377                         req.new_page = 1;
378                 else
379                         /*
380                          * Not a hole, the change will not add any new
381                          * indexing information, budget for page
382                          * change.
383                          */
384                         req.dirtied_page = 1;
385 
386                 if (appending) {
387                         mutex_lock(&ui->ui_mutex);
388                         if (!ui->dirty)
389                                 /*
390                                  * The inode is clean but we will have to mark
391                                  * it as dirty because we are appending. This
392                                  * needs a budget.
393                                  */
394                                 req.dirtied_ino = 1;
395                 }
396         }
397 
398         return ubifs_budget_space(c, &req);
399 }
400 
401 /*
402  * This function is called when a page of data is going to be written. Since
403  * the page of data will not necessarily go to the flash straight away, UBIFS
404  * has to reserve space on the media for it, which is done by means of
405  * budgeting.
406  *
407  * This is the hot-path of the file-system and we are trying to optimize it as
408  * much as possible. For this reasons it is split on 2 parts - slow and fast.
409  *
410  * There many budgeting cases:
411  *     o a new page is appended - we have to budget for a new page and for
412  *       changing the inode; however, if the inode is already dirty, there is
413  *       no need to budget for it;
414  *     o an existing clean page is changed - we have budget for it; if the page
415  *       does not exist on the media (a hole), we have to budget for a new
416  *       page; otherwise, we may budget for changing an existing page; the
417  *       difference between these cases is that changing an existing page does
418  *       not introduce anything new to the FS indexing information, so it does
419  *       not grow, and smaller budget is acquired in this case;
420  *     o an existing dirty page is changed - no need to budget at all, because
421  *       the page budget has been acquired by earlier, when the page has been
422  *       marked dirty.
423  *
424  * UBIFS budgeting sub-system may force write-back if it thinks there is no
425  * space to reserve. This imposes some locking restrictions and makes it
426  * impossible to take into account the above cases, and makes it impossible to
427  * optimize budgeting.
428  *
429  * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
430  * there is a plenty of flash space and the budget will be acquired quickly,
431  * without forcing write-back. The slow path does not make this assumption.
432  */
433 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
434                              loff_t pos, unsigned len, unsigned flags,
435                              struct page **pagep, void **fsdata)
436 {
437         struct inode *inode = mapping->host;
438         struct ubifs_info *c = inode->i_sb->s_fs_info;
439         struct ubifs_inode *ui = ubifs_inode(inode);
440         pgoff_t index = pos >> PAGE_SHIFT;
441         int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
442         int skipped_read = 0;
443         struct page *page;
444 
445         ubifs_assert(c, ubifs_inode(inode)->ui_size == inode->i_size);
446         ubifs_assert(c, !c->ro_media && !c->ro_mount);
447 
448         if (unlikely(c->ro_error))
449                 return -EROFS;
450 
451         /* Try out the fast-path part first */
452         page = grab_cache_page_write_begin(mapping, index, flags);
453         if (unlikely(!page))
454                 return -ENOMEM;
455 
456         if (!PageUptodate(page)) {
457                 /* The page is not loaded from the flash */
458                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE) {
459                         /*
460                          * We change whole page so no need to load it. But we
461                          * do not know whether this page exists on the media or
462                          * not, so we assume the latter because it requires
463                          * larger budget. The assumption is that it is better
464                          * to budget a bit more than to read the page from the
465                          * media. Thus, we are setting the @PG_checked flag
466                          * here.
467                          */
468                         SetPageChecked(page);
469                         skipped_read = 1;
470                 } else {
471                         err = do_readpage(page);
472                         if (err) {
473                                 unlock_page(page);
474                                 put_page(page);
475                                 return err;
476                         }
477                 }
478 
479                 SetPageUptodate(page);
480                 ClearPageError(page);
481         }
482 
483         err = allocate_budget(c, page, ui, appending);
484         if (unlikely(err)) {
485                 ubifs_assert(c, err == -ENOSPC);
486                 /*
487                  * If we skipped reading the page because we were going to
488                  * write all of it, then it is not up to date.
489                  */
490                 if (skipped_read) {
491                         ClearPageChecked(page);
492                         ClearPageUptodate(page);
493                 }
494                 /*
495                  * Budgeting failed which means it would have to force
496                  * write-back but didn't, because we set the @fast flag in the
497                  * request. Write-back cannot be done now, while we have the
498                  * page locked, because it would deadlock. Unlock and free
499                  * everything and fall-back to slow-path.
500                  */
501                 if (appending) {
502                         ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
503                         mutex_unlock(&ui->ui_mutex);
504                 }
505                 unlock_page(page);
506                 put_page(page);
507 
508                 return write_begin_slow(mapping, pos, len, pagep, flags);
509         }
510 
511         /*
512          * Whee, we acquired budgeting quickly - without involving
513          * garbage-collection, committing or forcing write-back. We return
514          * with @ui->ui_mutex locked if we are appending pages, and unlocked
515          * otherwise. This is an optimization (slightly hacky though).
516          */
517         *pagep = page;
518         return 0;
519 
520 }
521 
522 /**
523  * cancel_budget - cancel budget.
524  * @c: UBIFS file-system description object
525  * @page: page to cancel budget for
526  * @ui: UBIFS inode object the page belongs to
527  * @appending: non-zero if the page is appended
528  *
529  * This is a helper function for a page write operation. It unlocks the
530  * @ui->ui_mutex in case of appending.
531  */
532 static void cancel_budget(struct ubifs_info *c, struct page *page,
533                           struct ubifs_inode *ui, int appending)
534 {
535         if (appending) {
536                 if (!ui->dirty)
537                         ubifs_release_dirty_inode_budget(c, ui);
538                 mutex_unlock(&ui->ui_mutex);
539         }
540         if (!PagePrivate(page)) {
541                 if (PageChecked(page))
542                         release_new_page_budget(c);
543                 else
544                         release_existing_page_budget(c);
545         }
546 }
547 
548 static int ubifs_write_end(struct file *file, struct address_space *mapping,
549                            loff_t pos, unsigned len, unsigned copied,
550                            struct page *page, void *fsdata)
551 {
552         struct inode *inode = mapping->host;
553         struct ubifs_inode *ui = ubifs_inode(inode);
554         struct ubifs_info *c = inode->i_sb->s_fs_info;
555         loff_t end_pos = pos + len;
556         int appending = !!(end_pos > inode->i_size);
557 
558         dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
559                 inode->i_ino, pos, page->index, len, copied, inode->i_size);
560 
561         if (unlikely(copied < len && len == PAGE_SIZE)) {
562                 /*
563                  * VFS copied less data to the page that it intended and
564                  * declared in its '->write_begin()' call via the @len
565                  * argument. If the page was not up-to-date, and @len was
566                  * @PAGE_SIZE, the 'ubifs_write_begin()' function did
567                  * not load it from the media (for optimization reasons). This
568                  * means that part of the page contains garbage. So read the
569                  * page now.
570                  */
571                 dbg_gen("copied %d instead of %d, read page and repeat",
572                         copied, len);
573                 cancel_budget(c, page, ui, appending);
574                 ClearPageChecked(page);
575 
576                 /*
577                  * Return 0 to force VFS to repeat the whole operation, or the
578                  * error code if 'do_readpage()' fails.
579                  */
580                 copied = do_readpage(page);
581                 goto out;
582         }
583 
584         if (!PagePrivate(page)) {
585                 SetPagePrivate(page);
586                 atomic_long_inc(&c->dirty_pg_cnt);
587                 __set_page_dirty_nobuffers(page);
588         }
589 
590         if (appending) {
591                 i_size_write(inode, end_pos);
592                 ui->ui_size = end_pos;
593                 /*
594                  * Note, we do not set @I_DIRTY_PAGES (which means that the
595                  * inode has dirty pages), this has been done in
596                  * '__set_page_dirty_nobuffers()'.
597                  */
598                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
599                 ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
600                 mutex_unlock(&ui->ui_mutex);
601         }
602 
603 out:
604         unlock_page(page);
605         put_page(page);
606         return copied;
607 }
608 
609 /**
610  * populate_page - copy data nodes into a page for bulk-read.
611  * @c: UBIFS file-system description object
612  * @page: page
613  * @bu: bulk-read information
614  * @n: next zbranch slot
615  *
616  * This function returns %0 on success and a negative error code on failure.
617  */
618 static int populate_page(struct ubifs_info *c, struct page *page,
619                          struct bu_info *bu, int *n)
620 {
621         int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
622         struct inode *inode = page->mapping->host;
623         loff_t i_size = i_size_read(inode);
624         unsigned int page_block;
625         void *addr, *zaddr;
626         pgoff_t end_index;
627 
628         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
629                 inode->i_ino, page->index, i_size, page->flags);
630 
631         addr = zaddr = kmap(page);
632 
633         end_index = (i_size - 1) >> PAGE_SHIFT;
634         if (!i_size || page->index > end_index) {
635                 hole = 1;
636                 memset(addr, 0, PAGE_SIZE);
637                 goto out_hole;
638         }
639 
640         page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
641         while (1) {
642                 int err, len, out_len, dlen;
643 
644                 if (nn >= bu->cnt) {
645                         hole = 1;
646                         memset(addr, 0, UBIFS_BLOCK_SIZE);
647                 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
648                         struct ubifs_data_node *dn;
649 
650                         dn = bu->buf + (bu->zbranch[nn].offs - offs);
651 
652                         ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
653                                      ubifs_inode(inode)->creat_sqnum);
654 
655                         len = le32_to_cpu(dn->size);
656                         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
657                                 goto out_err;
658 
659                         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
660                         out_len = UBIFS_BLOCK_SIZE;
661 
662                         if (ubifs_crypt_is_encrypted(inode)) {
663                                 err = ubifs_decrypt(inode, dn, &dlen, page_block);
664                                 if (err)
665                                         goto out_err;
666                         }
667 
668                         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
669                                                le16_to_cpu(dn->compr_type));
670                         if (err || len != out_len)
671                                 goto out_err;
672 
673                         if (len < UBIFS_BLOCK_SIZE)
674                                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
675 
676                         nn += 1;
677                         read = (i << UBIFS_BLOCK_SHIFT) + len;
678                 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
679                         nn += 1;
680                         continue;
681                 } else {
682                         hole = 1;
683                         memset(addr, 0, UBIFS_BLOCK_SIZE);
684                 }
685                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
686                         break;
687                 addr += UBIFS_BLOCK_SIZE;
688                 page_block += 1;
689         }
690 
691         if (end_index == page->index) {
692                 int len = i_size & (PAGE_SIZE - 1);
693 
694                 if (len && len < read)
695                         memset(zaddr + len, 0, read - len);
696         }
697 
698 out_hole:
699         if (hole) {
700                 SetPageChecked(page);
701                 dbg_gen("hole");
702         }
703 
704         SetPageUptodate(page);
705         ClearPageError(page);
706         flush_dcache_page(page);
707         kunmap(page);
708         *n = nn;
709         return 0;
710 
711 out_err:
712         ClearPageUptodate(page);
713         SetPageError(page);
714         flush_dcache_page(page);
715         kunmap(page);
716         ubifs_err(c, "bad data node (block %u, inode %lu)",
717                   page_block, inode->i_ino);
718         return -EINVAL;
719 }
720 
721 /**
722  * ubifs_do_bulk_read - do bulk-read.
723  * @c: UBIFS file-system description object
724  * @bu: bulk-read information
725  * @page1: first page to read
726  *
727  * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
728  */
729 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
730                               struct page *page1)
731 {
732         pgoff_t offset = page1->index, end_index;
733         struct address_space *mapping = page1->mapping;
734         struct inode *inode = mapping->host;
735         struct ubifs_inode *ui = ubifs_inode(inode);
736         int err, page_idx, page_cnt, ret = 0, n = 0;
737         int allocate = bu->buf ? 0 : 1;
738         loff_t isize;
739         gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS;
740 
741         err = ubifs_tnc_get_bu_keys(c, bu);
742         if (err)
743                 goto out_warn;
744 
745         if (bu->eof) {
746                 /* Turn off bulk-read at the end of the file */
747                 ui->read_in_a_row = 1;
748                 ui->bulk_read = 0;
749         }
750 
751         page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
752         if (!page_cnt) {
753                 /*
754                  * This happens when there are multiple blocks per page and the
755                  * blocks for the first page we are looking for, are not
756                  * together. If all the pages were like this, bulk-read would
757                  * reduce performance, so we turn it off for a while.
758                  */
759                 goto out_bu_off;
760         }
761 
762         if (bu->cnt) {
763                 if (allocate) {
764                         /*
765                          * Allocate bulk-read buffer depending on how many data
766                          * nodes we are going to read.
767                          */
768                         bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
769                                       bu->zbranch[bu->cnt - 1].len -
770                                       bu->zbranch[0].offs;
771                         ubifs_assert(c, bu->buf_len > 0);
772                         ubifs_assert(c, bu->buf_len <= c->leb_size);
773                         bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
774                         if (!bu->buf)
775                                 goto out_bu_off;
776                 }
777 
778                 err = ubifs_tnc_bulk_read(c, bu);
779                 if (err)
780                         goto out_warn;
781         }
782 
783         err = populate_page(c, page1, bu, &n);
784         if (err)
785                 goto out_warn;
786 
787         unlock_page(page1);
788         ret = 1;
789 
790         isize = i_size_read(inode);
791         if (isize == 0)
792                 goto out_free;
793         end_index = ((isize - 1) >> PAGE_SHIFT);
794 
795         for (page_idx = 1; page_idx < page_cnt; page_idx++) {
796                 pgoff_t page_offset = offset + page_idx;
797                 struct page *page;
798 
799                 if (page_offset > end_index)
800                         break;
801                 page = find_or_create_page(mapping, page_offset, ra_gfp_mask);
802                 if (!page)
803                         break;
804                 if (!PageUptodate(page))
805                         err = populate_page(c, page, bu, &n);
806                 unlock_page(page);
807                 put_page(page);
808                 if (err)
809                         break;
810         }
811 
812         ui->last_page_read = offset + page_idx - 1;
813 
814 out_free:
815         if (allocate)
816                 kfree(bu->buf);
817         return ret;
818 
819 out_warn:
820         ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
821         goto out_free;
822 
823 out_bu_off:
824         ui->read_in_a_row = ui->bulk_read = 0;
825         goto out_free;
826 }
827 
828 /**
829  * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
830  * @page: page from which to start bulk-read.
831  *
832  * Some flash media are capable of reading sequentially at faster rates. UBIFS
833  * bulk-read facility is designed to take advantage of that, by reading in one
834  * go consecutive data nodes that are also located consecutively in the same
835  * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
836  */
837 static int ubifs_bulk_read(struct page *page)
838 {
839         struct inode *inode = page->mapping->host;
840         struct ubifs_info *c = inode->i_sb->s_fs_info;
841         struct ubifs_inode *ui = ubifs_inode(inode);
842         pgoff_t index = page->index, last_page_read = ui->last_page_read;
843         struct bu_info *bu;
844         int err = 0, allocated = 0;
845 
846         ui->last_page_read = index;
847         if (!c->bulk_read)
848                 return 0;
849 
850         /*
851          * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
852          * so don't bother if we cannot lock the mutex.
853          */
854         if (!mutex_trylock(&ui->ui_mutex))
855                 return 0;
856 
857         if (index != last_page_read + 1) {
858                 /* Turn off bulk-read if we stop reading sequentially */
859                 ui->read_in_a_row = 1;
860                 if (ui->bulk_read)
861                         ui->bulk_read = 0;
862                 goto out_unlock;
863         }
864 
865         if (!ui->bulk_read) {
866                 ui->read_in_a_row += 1;
867                 if (ui->read_in_a_row < 3)
868                         goto out_unlock;
869                 /* Three reads in a row, so switch on bulk-read */
870                 ui->bulk_read = 1;
871         }
872 
873         /*
874          * If possible, try to use pre-allocated bulk-read information, which
875          * is protected by @c->bu_mutex.
876          */
877         if (mutex_trylock(&c->bu_mutex))
878                 bu = &c->bu;
879         else {
880                 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
881                 if (!bu)
882                         goto out_unlock;
883 
884                 bu->buf = NULL;
885                 allocated = 1;
886         }
887 
888         bu->buf_len = c->max_bu_buf_len;
889         data_key_init(c, &bu->key, inode->i_ino,
890                       page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
891         err = ubifs_do_bulk_read(c, bu, page);
892 
893         if (!allocated)
894                 mutex_unlock(&c->bu_mutex);
895         else
896                 kfree(bu);
897 
898 out_unlock:
899         mutex_unlock(&ui->ui_mutex);
900         return err;
901 }
902 
903 static int ubifs_readpage(struct file *file, struct page *page)
904 {
905         if (ubifs_bulk_read(page))
906                 return 0;
907         do_readpage(page);
908         unlock_page(page);
909         return 0;
910 }
911 
912 static int do_writepage(struct page *page, int len)
913 {
914         int err = 0, i, blen;
915         unsigned int block;
916         void *addr;
917         union ubifs_key key;
918         struct inode *inode = page->mapping->host;
919         struct ubifs_info *c = inode->i_sb->s_fs_info;
920 
921 #ifdef UBIFS_DEBUG
922         struct ubifs_inode *ui = ubifs_inode(inode);
923         spin_lock(&ui->ui_lock);
924         ubifs_assert(c, page->index <= ui->synced_i_size >> PAGE_SHIFT);
925         spin_unlock(&ui->ui_lock);
926 #endif
927 
928         /* Update radix tree tags */
929         set_page_writeback(page);
930 
931         addr = kmap(page);
932         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
933         i = 0;
934         while (len) {
935                 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
936                 data_key_init(c, &key, inode->i_ino, block);
937                 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
938                 if (err)
939                         break;
940                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
941                         break;
942                 block += 1;
943                 addr += blen;
944                 len -= blen;
945         }
946         if (err) {
947                 SetPageError(page);
948                 ubifs_err(c, "cannot write page %lu of inode %lu, error %d",
949                           page->index, inode->i_ino, err);
950                 ubifs_ro_mode(c, err);
951         }
952 
953         ubifs_assert(c, PagePrivate(page));
954         if (PageChecked(page))
955                 release_new_page_budget(c);
956         else
957                 release_existing_page_budget(c);
958 
959         atomic_long_dec(&c->dirty_pg_cnt);
960         ClearPagePrivate(page);
961         ClearPageChecked(page);
962 
963         kunmap(page);
964         unlock_page(page);
965         end_page_writeback(page);
966         return err;
967 }
968 
969 /*
970  * When writing-back dirty inodes, VFS first writes-back pages belonging to the
971  * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
972  * situation when a we have an inode with size 0, then a megabyte of data is
973  * appended to the inode, then write-back starts and flushes some amount of the
974  * dirty pages, the journal becomes full, commit happens and finishes, and then
975  * an unclean reboot happens. When the file system is mounted next time, the
976  * inode size would still be 0, but there would be many pages which are beyond
977  * the inode size, they would be indexed and consume flash space. Because the
978  * journal has been committed, the replay would not be able to detect this
979  * situation and correct the inode size. This means UBIFS would have to scan
980  * whole index and correct all inode sizes, which is long an unacceptable.
981  *
982  * To prevent situations like this, UBIFS writes pages back only if they are
983  * within the last synchronized inode size, i.e. the size which has been
984  * written to the flash media last time. Otherwise, UBIFS forces inode
985  * write-back, thus making sure the on-flash inode contains current inode size,
986  * and then keeps writing pages back.
987  *
988  * Some locking issues explanation. 'ubifs_writepage()' first is called with
989  * the page locked, and it locks @ui_mutex. However, write-back does take inode
990  * @i_mutex, which means other VFS operations may be run on this inode at the
991  * same time. And the problematic one is truncation to smaller size, from where
992  * we have to call 'truncate_setsize()', which first changes @inode->i_size,
993  * then drops the truncated pages. And while dropping the pages, it takes the
994  * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
995  * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
996  * This means that @inode->i_size is changed while @ui_mutex is unlocked.
997  *
998  * XXX(truncate): with the new truncate sequence this is not true anymore,
999  * and the calls to truncate_setsize can be move around freely.  They should
1000  * be moved to the very end of the truncate sequence.
1001  *
1002  * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
1003  * inode size. How do we do this if @inode->i_size may became smaller while we
1004  * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
1005  * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
1006  * internally and updates it under @ui_mutex.
1007  *
1008  * Q: why we do not worry that if we race with truncation, we may end up with a
1009  * situation when the inode is truncated while we are in the middle of
1010  * 'do_writepage()', so we do write beyond inode size?
1011  * A: If we are in the middle of 'do_writepage()', truncation would be locked
1012  * on the page lock and it would not write the truncated inode node to the
1013  * journal before we have finished.
1014  */
1015 static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
1016 {
1017         struct inode *inode = page->mapping->host;
1018         struct ubifs_info *c = inode->i_sb->s_fs_info;
1019         struct ubifs_inode *ui = ubifs_inode(inode);
1020         loff_t i_size =  i_size_read(inode), synced_i_size;
1021         pgoff_t end_index = i_size >> PAGE_SHIFT;
1022         int err, len = i_size & (PAGE_SIZE - 1);
1023         void *kaddr;
1024 
1025         dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1026                 inode->i_ino, page->index, page->flags);
1027         ubifs_assert(c, PagePrivate(page));
1028 
1029         /* Is the page fully outside @i_size? (truncate in progress) */
1030         if (page->index > end_index || (page->index == end_index && !len)) {
1031                 err = 0;
1032                 goto out_unlock;
1033         }
1034 
1035         spin_lock(&ui->ui_lock);
1036         synced_i_size = ui->synced_i_size;
1037         spin_unlock(&ui->ui_lock);
1038 
1039         /* Is the page fully inside @i_size? */
1040         if (page->index < end_index) {
1041                 if (page->index >= synced_i_size >> PAGE_SHIFT) {
1042                         err = inode->i_sb->s_op->write_inode(inode, NULL);
1043                         if (err)
1044                                 goto out_unlock;
1045                         /*
1046                          * The inode has been written, but the write-buffer has
1047                          * not been synchronized, so in case of an unclean
1048                          * reboot we may end up with some pages beyond inode
1049                          * size, but they would be in the journal (because
1050                          * commit flushes write buffers) and recovery would deal
1051                          * with this.
1052                          */
1053                 }
1054                 return do_writepage(page, PAGE_SIZE);
1055         }
1056 
1057         /*
1058          * The page straddles @i_size. It must be zeroed out on each and every
1059          * writepage invocation because it may be mmapped. "A file is mapped
1060          * in multiples of the page size. For a file that is not a multiple of
1061          * the page size, the remaining memory is zeroed when mapped, and
1062          * writes to that region are not written out to the file."
1063          */
1064         kaddr = kmap_atomic(page);
1065         memset(kaddr + len, 0, PAGE_SIZE - len);
1066         flush_dcache_page(page);
1067         kunmap_atomic(kaddr);
1068 
1069         if (i_size > synced_i_size) {
1070                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1071                 if (err)
1072                         goto out_unlock;
1073         }
1074 
1075         return do_writepage(page, len);
1076 
1077 out_unlock:
1078         unlock_page(page);
1079         return err;
1080 }
1081 
1082 /**
1083  * do_attr_changes - change inode attributes.
1084  * @inode: inode to change attributes for
1085  * @attr: describes attributes to change
1086  */
1087 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1088 {
1089         if (attr->ia_valid & ATTR_UID)
1090                 inode->i_uid = attr->ia_uid;
1091         if (attr->ia_valid & ATTR_GID)
1092                 inode->i_gid = attr->ia_gid;
1093         if (attr->ia_valid & ATTR_ATIME)
1094                 inode->i_atime = timespec64_trunc(attr->ia_atime,
1095                                                   inode->i_sb->s_time_gran);
1096         if (attr->ia_valid & ATTR_MTIME)
1097                 inode->i_mtime = timespec64_trunc(attr->ia_mtime,
1098                                                   inode->i_sb->s_time_gran);
1099         if (attr->ia_valid & ATTR_CTIME)
1100                 inode->i_ctime = timespec64_trunc(attr->ia_ctime,
1101                                                   inode->i_sb->s_time_gran);
1102         if (attr->ia_valid & ATTR_MODE) {
1103                 umode_t mode = attr->ia_mode;
1104 
1105                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1106                         mode &= ~S_ISGID;
1107                 inode->i_mode = mode;
1108         }
1109 }
1110 
1111 /**
1112  * do_truncation - truncate an inode.
1113  * @c: UBIFS file-system description object
1114  * @inode: inode to truncate
1115  * @attr: inode attribute changes description
1116  *
1117  * This function implements VFS '->setattr()' call when the inode is truncated
1118  * to a smaller size. Returns zero in case of success and a negative error code
1119  * in case of failure.
1120  */
1121 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1122                          const struct iattr *attr)
1123 {
1124         int err;
1125         struct ubifs_budget_req req;
1126         loff_t old_size = inode->i_size, new_size = attr->ia_size;
1127         int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1128         struct ubifs_inode *ui = ubifs_inode(inode);
1129 
1130         dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1131         memset(&req, 0, sizeof(struct ubifs_budget_req));
1132 
1133         /*
1134          * If this is truncation to a smaller size, and we do not truncate on a
1135          * block boundary, budget for changing one data block, because the last
1136          * block will be re-written.
1137          */
1138         if (new_size & (UBIFS_BLOCK_SIZE - 1))
1139                 req.dirtied_page = 1;
1140 
1141         req.dirtied_ino = 1;
1142         /* A funny way to budget for truncation node */
1143         req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1144         err = ubifs_budget_space(c, &req);
1145         if (err) {
1146                 /*
1147                  * Treat truncations to zero as deletion and always allow them,
1148                  * just like we do for '->unlink()'.
1149                  */
1150                 if (new_size || err != -ENOSPC)
1151                         return err;
1152                 budgeted = 0;
1153         }
1154 
1155         truncate_setsize(inode, new_size);
1156 
1157         if (offset) {
1158                 pgoff_t index = new_size >> PAGE_SHIFT;
1159                 struct page *page;
1160 
1161                 page = find_lock_page(inode->i_mapping, index);
1162                 if (page) {
1163                         if (PageDirty(page)) {
1164                                 /*
1165                                  * 'ubifs_jnl_truncate()' will try to truncate
1166                                  * the last data node, but it contains
1167                                  * out-of-date data because the page is dirty.
1168                                  * Write the page now, so that
1169                                  * 'ubifs_jnl_truncate()' will see an already
1170                                  * truncated (and up to date) data node.
1171                                  */
1172                                 ubifs_assert(c, PagePrivate(page));
1173 
1174                                 clear_page_dirty_for_io(page);
1175                                 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1176                                         offset = new_size &
1177                                                  (PAGE_SIZE - 1);
1178                                 err = do_writepage(page, offset);
1179                                 put_page(page);
1180                                 if (err)
1181                                         goto out_budg;
1182                                 /*
1183                                  * We could now tell 'ubifs_jnl_truncate()' not
1184                                  * to read the last block.
1185                                  */
1186                         } else {
1187                                 /*
1188                                  * We could 'kmap()' the page and pass the data
1189                                  * to 'ubifs_jnl_truncate()' to save it from
1190                                  * having to read it.
1191                                  */
1192                                 unlock_page(page);
1193                                 put_page(page);
1194                         }
1195                 }
1196         }
1197 
1198         mutex_lock(&ui->ui_mutex);
1199         ui->ui_size = inode->i_size;
1200         /* Truncation changes inode [mc]time */
1201         inode->i_mtime = inode->i_ctime = current_time(inode);
1202         /* Other attributes may be changed at the same time as well */
1203         do_attr_changes(inode, attr);
1204         err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1205         mutex_unlock(&ui->ui_mutex);
1206 
1207 out_budg:
1208         if (budgeted)
1209                 ubifs_release_budget(c, &req);
1210         else {
1211                 c->bi.nospace = c->bi.nospace_rp = 0;
1212                 smp_wmb();
1213         }
1214         return err;
1215 }
1216 
1217 /**
1218  * do_setattr - change inode attributes.
1219  * @c: UBIFS file-system description object
1220  * @inode: inode to change attributes for
1221  * @attr: inode attribute changes description
1222  *
1223  * This function implements VFS '->setattr()' call for all cases except
1224  * truncations to smaller size. Returns zero in case of success and a negative
1225  * error code in case of failure.
1226  */
1227 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1228                       const struct iattr *attr)
1229 {
1230         int err, release;
1231         loff_t new_size = attr->ia_size;
1232         struct ubifs_inode *ui = ubifs_inode(inode);
1233         struct ubifs_budget_req req = { .dirtied_ino = 1,
1234                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1235 
1236         err = ubifs_budget_space(c, &req);
1237         if (err)
1238                 return err;
1239 
1240         if (attr->ia_valid & ATTR_SIZE) {
1241                 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1242                 truncate_setsize(inode, new_size);
1243         }
1244 
1245         mutex_lock(&ui->ui_mutex);
1246         if (attr->ia_valid & ATTR_SIZE) {
1247                 /* Truncation changes inode [mc]time */
1248                 inode->i_mtime = inode->i_ctime = current_time(inode);
1249                 /* 'truncate_setsize()' changed @i_size, update @ui_size */
1250                 ui->ui_size = inode->i_size;
1251         }
1252 
1253         do_attr_changes(inode, attr);
1254 
1255         release = ui->dirty;
1256         if (attr->ia_valid & ATTR_SIZE)
1257                 /*
1258                  * Inode length changed, so we have to make sure
1259                  * @I_DIRTY_DATASYNC is set.
1260                  */
1261                  __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1262         else
1263                 mark_inode_dirty_sync(inode);
1264         mutex_unlock(&ui->ui_mutex);
1265 
1266         if (release)
1267                 ubifs_release_budget(c, &req);
1268         if (IS_SYNC(inode))
1269                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1270         return err;
1271 }
1272 
1273 int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1274 {
1275         int err;
1276         struct inode *inode = d_inode(dentry);
1277         struct ubifs_info *c = inode->i_sb->s_fs_info;
1278 
1279         dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1280                 inode->i_ino, inode->i_mode, attr->ia_valid);
1281         err = setattr_prepare(dentry, attr);
1282         if (err)
1283                 return err;
1284 
1285         err = dbg_check_synced_i_size(c, inode);
1286         if (err)
1287                 return err;
1288 
1289         err = fscrypt_prepare_setattr(dentry, attr);
1290         if (err)
1291                 return err;
1292 
1293         if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1294                 /* Truncation to a smaller size */
1295                 err = do_truncation(c, inode, attr);
1296         else
1297                 err = do_setattr(c, inode, attr);
1298 
1299         return err;
1300 }
1301 
1302 static void ubifs_invalidatepage(struct page *page, unsigned int offset,
1303                                  unsigned int length)
1304 {
1305         struct inode *inode = page->mapping->host;
1306         struct ubifs_info *c = inode->i_sb->s_fs_info;
1307 
1308         ubifs_assert(c, PagePrivate(page));
1309         if (offset || length < PAGE_SIZE)
1310                 /* Partial page remains dirty */
1311                 return;
1312 
1313         if (PageChecked(page))
1314                 release_new_page_budget(c);
1315         else
1316                 release_existing_page_budget(c);
1317 
1318         atomic_long_dec(&c->dirty_pg_cnt);
1319         ClearPagePrivate(page);
1320         ClearPageChecked(page);
1321 }
1322 
1323 int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1324 {
1325         struct inode *inode = file->f_mapping->host;
1326         struct ubifs_info *c = inode->i_sb->s_fs_info;
1327         int err;
1328 
1329         dbg_gen("syncing inode %lu", inode->i_ino);
1330 
1331         if (c->ro_mount)
1332                 /*
1333                  * For some really strange reasons VFS does not filter out
1334                  * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1335                  */
1336                 return 0;
1337 
1338         err = file_write_and_wait_range(file, start, end);
1339         if (err)
1340                 return err;
1341         inode_lock(inode);
1342 
1343         /* Synchronize the inode unless this is a 'datasync()' call. */
1344         if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1345                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1346                 if (err)
1347                         goto out;
1348         }
1349 
1350         /*
1351          * Nodes related to this inode may still sit in a write-buffer. Flush
1352          * them.
1353          */
1354         err = ubifs_sync_wbufs_by_inode(c, inode);
1355 out:
1356         inode_unlock(inode);
1357         return err;
1358 }
1359 
1360 /**
1361  * mctime_update_needed - check if mtime or ctime update is needed.
1362  * @inode: the inode to do the check for
1363  * @now: current time
1364  *
1365  * This helper function checks if the inode mtime/ctime should be updated or
1366  * not. If current values of the time-stamps are within the UBIFS inode time
1367  * granularity, they are not updated. This is an optimization.
1368  */
1369 static inline int mctime_update_needed(const struct inode *inode,
1370                                        const struct timespec64 *now)
1371 {
1372         if (!timespec64_equal(&inode->i_mtime, now) ||
1373             !timespec64_equal(&inode->i_ctime, now))
1374                 return 1;
1375         return 0;
1376 }
1377 
1378 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1379 /**
1380  * ubifs_update_time - update time of inode.
1381  * @inode: inode to update
1382  *
1383  * This function updates time of the inode.
1384  */
1385 int ubifs_update_time(struct inode *inode, struct timespec64 *time,
1386                              int flags)
1387 {
1388         struct ubifs_inode *ui = ubifs_inode(inode);
1389         struct ubifs_info *c = inode->i_sb->s_fs_info;
1390         struct ubifs_budget_req req = { .dirtied_ino = 1,
1391                         .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1392         int iflags = I_DIRTY_TIME;
1393         int err, release;
1394 
1395         err = ubifs_budget_space(c, &req);
1396         if (err)
1397                 return err;
1398 
1399         mutex_lock(&ui->ui_mutex);
1400         if (flags & S_ATIME)
1401                 inode->i_atime = *time;
1402         if (flags & S_CTIME)
1403                 inode->i_ctime = *time;
1404         if (flags & S_MTIME)
1405                 inode->i_mtime = *time;
1406 
1407         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
1408                 iflags |= I_DIRTY_SYNC;
1409 
1410         release = ui->dirty;
1411         __mark_inode_dirty(inode, iflags);
1412         mutex_unlock(&ui->ui_mutex);
1413         if (release)
1414                 ubifs_release_budget(c, &req);
1415         return 0;
1416 }
1417 #endif
1418 
1419 /**
1420  * update_mctime - update mtime and ctime of an inode.
1421  * @inode: inode to update
1422  *
1423  * This function updates mtime and ctime of the inode if it is not equivalent to
1424  * current time. Returns zero in case of success and a negative error code in
1425  * case of failure.
1426  */
1427 static int update_mctime(struct inode *inode)
1428 {
1429         struct timespec64 now = current_time(inode);
1430         struct ubifs_inode *ui = ubifs_inode(inode);
1431         struct ubifs_info *c = inode->i_sb->s_fs_info;
1432 
1433         if (mctime_update_needed(inode, &now)) {
1434                 int err, release;
1435                 struct ubifs_budget_req req = { .dirtied_ino = 1,
1436                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1437 
1438                 err = ubifs_budget_space(c, &req);
1439                 if (err)
1440                         return err;
1441 
1442                 mutex_lock(&ui->ui_mutex);
1443                 inode->i_mtime = inode->i_ctime = current_time(inode);
1444                 release = ui->dirty;
1445                 mark_inode_dirty_sync(inode);
1446                 mutex_unlock(&ui->ui_mutex);
1447                 if (release)
1448                         ubifs_release_budget(c, &req);
1449         }
1450 
1451         return 0;
1452 }
1453 
1454 static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1455 {
1456         int err = update_mctime(file_inode(iocb->ki_filp));
1457         if (err)
1458                 return err;
1459 
1460         return generic_file_write_iter(iocb, from);
1461 }
1462 
1463 static int ubifs_set_page_dirty(struct page *page)
1464 {
1465         int ret;
1466         struct inode *inode = page->mapping->host;
1467         struct ubifs_info *c = inode->i_sb->s_fs_info;
1468 
1469         ret = __set_page_dirty_nobuffers(page);
1470         /*
1471          * An attempt to dirty a page without budgeting for it - should not
1472          * happen.
1473          */
1474         ubifs_assert(c, ret == 0);
1475         return ret;
1476 }
1477 
1478 #ifdef CONFIG_MIGRATION
1479 static int ubifs_migrate_page(struct address_space *mapping,
1480                 struct page *newpage, struct page *page, enum migrate_mode mode)
1481 {
1482         int rc;
1483 
1484         rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
1485         if (rc != MIGRATEPAGE_SUCCESS)
1486                 return rc;
1487 
1488         if (PagePrivate(page)) {
1489                 ClearPagePrivate(page);
1490                 SetPagePrivate(newpage);
1491         }
1492 
1493         if (mode != MIGRATE_SYNC_NO_COPY)
1494                 migrate_page_copy(newpage, page);
1495         else
1496                 migrate_page_states(newpage, page);
1497         return MIGRATEPAGE_SUCCESS;
1498 }
1499 #endif
1500 
1501 static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1502 {
1503         struct inode *inode = page->mapping->host;
1504         struct ubifs_info *c = inode->i_sb->s_fs_info;
1505 
1506         /*
1507          * An attempt to release a dirty page without budgeting for it - should
1508          * not happen.
1509          */
1510         if (PageWriteback(page))
1511                 return 0;
1512         ubifs_assert(c, PagePrivate(page));
1513         ubifs_assert(c, 0);
1514         ClearPagePrivate(page);
1515         ClearPageChecked(page);
1516         return 1;
1517 }
1518 
1519 /*
1520  * mmap()d file has taken write protection fault and is being made writable.
1521  * UBIFS must ensure page is budgeted for.
1522  */
1523 static vm_fault_t ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1524 {
1525         struct page *page = vmf->page;
1526         struct inode *inode = file_inode(vmf->vma->vm_file);
1527         struct ubifs_info *c = inode->i_sb->s_fs_info;
1528         struct timespec64 now = current_time(inode);
1529         struct ubifs_budget_req req = { .new_page = 1 };
1530         int err, update_time;
1531 
1532         dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1533                 i_size_read(inode));
1534         ubifs_assert(c, !c->ro_media && !c->ro_mount);
1535 
1536         if (unlikely(c->ro_error))
1537                 return VM_FAULT_SIGBUS; /* -EROFS */
1538 
1539         /*
1540          * We have not locked @page so far so we may budget for changing the
1541          * page. Note, we cannot do this after we locked the page, because
1542          * budgeting may cause write-back which would cause deadlock.
1543          *
1544          * At the moment we do not know whether the page is dirty or not, so we
1545          * assume that it is not and budget for a new page. We could look at
1546          * the @PG_private flag and figure this out, but we may race with write
1547          * back and the page state may change by the time we lock it, so this
1548          * would need additional care. We do not bother with this at the
1549          * moment, although it might be good idea to do. Instead, we allocate
1550          * budget for a new page and amend it later on if the page was in fact
1551          * dirty.
1552          *
1553          * The budgeting-related logic of this function is similar to what we
1554          * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1555          * for more comments.
1556          */
1557         update_time = mctime_update_needed(inode, &now);
1558         if (update_time)
1559                 /*
1560                  * We have to change inode time stamp which requires extra
1561                  * budgeting.
1562                  */
1563                 req.dirtied_ino = 1;
1564 
1565         err = ubifs_budget_space(c, &req);
1566         if (unlikely(err)) {
1567                 if (err == -ENOSPC)
1568                         ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1569                                    inode->i_ino);
1570                 return VM_FAULT_SIGBUS;
1571         }
1572 
1573         lock_page(page);
1574         if (unlikely(page->mapping != inode->i_mapping ||
1575                      page_offset(page) > i_size_read(inode))) {
1576                 /* Page got truncated out from underneath us */
1577                 goto sigbus;
1578         }
1579 
1580         if (PagePrivate(page))
1581                 release_new_page_budget(c);
1582         else {
1583                 if (!PageChecked(page))
1584                         ubifs_convert_page_budget(c);
1585                 SetPagePrivate(page);
1586                 atomic_long_inc(&c->dirty_pg_cnt);
1587                 __set_page_dirty_nobuffers(page);
1588         }
1589 
1590         if (update_time) {
1591                 int release;
1592                 struct ubifs_inode *ui = ubifs_inode(inode);
1593 
1594                 mutex_lock(&ui->ui_mutex);
1595                 inode->i_mtime = inode->i_ctime = current_time(inode);
1596                 release = ui->dirty;
1597                 mark_inode_dirty_sync(inode);
1598                 mutex_unlock(&ui->ui_mutex);
1599                 if (release)
1600                         ubifs_release_dirty_inode_budget(c, ui);
1601         }
1602 
1603         wait_for_stable_page(page);
1604         return VM_FAULT_LOCKED;
1605 
1606 sigbus:
1607         unlock_page(page);
1608         ubifs_release_budget(c, &req);
1609         return VM_FAULT_SIGBUS;
1610 }
1611 
1612 static const struct vm_operations_struct ubifs_file_vm_ops = {
1613         .fault        = filemap_fault,
1614         .map_pages = filemap_map_pages,
1615         .page_mkwrite = ubifs_vm_page_mkwrite,
1616 };
1617 
1618 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1619 {
1620         int err;
1621 
1622         err = generic_file_mmap(file, vma);
1623         if (err)
1624                 return err;
1625         vma->vm_ops = &ubifs_file_vm_ops;
1626 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1627         file_accessed(file);
1628 #endif
1629         return 0;
1630 }
1631 
1632 static const char *ubifs_get_link(struct dentry *dentry,
1633                                             struct inode *inode,
1634                                             struct delayed_call *done)
1635 {
1636         struct ubifs_inode *ui = ubifs_inode(inode);
1637 
1638         if (!IS_ENCRYPTED(inode))
1639                 return ui->data;
1640 
1641         if (!dentry)
1642                 return ERR_PTR(-ECHILD);
1643 
1644         return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1645 }
1646 
1647 const struct address_space_operations ubifs_file_address_operations = {
1648         .readpage       = ubifs_readpage,
1649         .writepage      = ubifs_writepage,
1650         .write_begin    = ubifs_write_begin,
1651         .write_end      = ubifs_write_end,
1652         .invalidatepage = ubifs_invalidatepage,
1653         .set_page_dirty = ubifs_set_page_dirty,
1654 #ifdef CONFIG_MIGRATION
1655         .migratepage    = ubifs_migrate_page,
1656 #endif
1657         .releasepage    = ubifs_releasepage,
1658 };
1659 
1660 const struct inode_operations ubifs_file_inode_operations = {
1661         .setattr     = ubifs_setattr,
1662         .getattr     = ubifs_getattr,
1663 #ifdef CONFIG_UBIFS_FS_XATTR
1664         .listxattr   = ubifs_listxattr,
1665 #endif
1666 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1667         .update_time = ubifs_update_time,
1668 #endif
1669 };
1670 
1671 const struct inode_operations ubifs_symlink_inode_operations = {
1672         .get_link    = ubifs_get_link,
1673         .setattr     = ubifs_setattr,
1674         .getattr     = ubifs_getattr,
1675 #ifdef CONFIG_UBIFS_FS_XATTR
1676         .listxattr   = ubifs_listxattr,
1677 #endif
1678 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1679         .update_time = ubifs_update_time,
1680 #endif
1681 };
1682 
1683 const struct file_operations ubifs_file_operations = {
1684         .llseek         = generic_file_llseek,
1685         .read_iter      = generic_file_read_iter,
1686         .write_iter     = ubifs_write_iter,
1687         .mmap           = ubifs_file_mmap,
1688         .fsync          = ubifs_fsync,
1689         .unlocked_ioctl = ubifs_ioctl,
1690         .splice_read    = generic_file_splice_read,
1691         .splice_write   = iter_file_splice_write,
1692         .open           = fscrypt_file_open,
1693 #ifdef CONFIG_COMPAT
1694         .compat_ioctl   = ubifs_compat_ioctl,
1695 #endif
1696 };
1697 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp