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

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

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ 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/aio.h>
 54 #include <linux/mount.h>
 55 #include <linux/namei.h>
 56 #include <linux/slab.h>
 57 
 58 static 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 
 75         ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
 76                      ubifs_inode(inode)->creat_sqnum);
 77         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 
 98 dump:
 99         ubifs_err("bad data node (block %u, inode %lu)",
100                   block, inode->i_ino);
101         ubifs_dump_node(c, dn);
102         return -EINVAL;
103 }
104 
105 static 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;
150                         } 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);
156                         }
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 
175 out_free:
176         kfree(dn);
177 out:
178         SetPageUptodate(page);
179         ClearPageError(page);
180         flush_dcache_page(page);
181         kunmap(page);
182         return 0;
183 
184 error:
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  */
200 static 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  */
214 static void release_existing_page_budget(struct ubifs_info *c)
215 {
216         struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
217 
218         ubifs_release_budget(c, &req);
219 }
220 
221 static int write_begin_slow(struct address_space *mapping,
222                             loff_t pos, unsigned len, struct page **pagep,
223                             unsigned flags)
224 {
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 
251         page = grab_cache_page_write_begin(mapping, index, flags);
252         if (unlikely(!page)) {
253                 ubifs_release_budget(c, &req);
254                 return -ENOMEM;
255         }
256 
257         if (!PageUptodate(page)) {
258                 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
259                         SetPageChecked(page);
260                 else {
261                         err = do_readpage(page);
262                         if (err) {
263                                 unlock_page(page);
264                                 page_cache_release(page);
265                                 ubifs_release_budget(c, &req);
266                                 return err;
267                         }
268                 }
269 
270                 SetPageUptodate(page);
271                 ClearPageError(page);
272         }
273 
274         if (PagePrivate(page))
275                 /*
276                  * The page is dirty, which means it was budgeted twice:
277                  *   o first time the budget was allocated by the task which
278                  *     made the page dirty and set the PG_private flag;
279                  *   o and then we budgeted for it for the second time at the
280                  *     very beginning of this function.
281                  *
282                  * So what we have to do is to release the page budget we
283                  * allocated.
284                  */
285                 release_new_page_budget(c);
286         else if (!PageChecked(page))
287                 /*
288                  * We are changing a page which already exists on the media.
289                  * This means that changing the page does not make the amount
290                  * of indexing information larger, and this part of the budget
291                  * which we have already acquired may be released.
292                  */
293                 ubifs_convert_page_budget(c);
294 
295         if (appending) {
296                 struct ubifs_inode *ui = ubifs_inode(inode);
297 
298                 /*
299                  * 'ubifs_write_end()' is optimized from the fast-path part of
300                  * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
301                  * if data is appended.
302                  */
303                 mutex_lock(&ui->ui_mutex);
304                 if (ui->dirty)
305                         /*
306                          * The inode is dirty already, so we may free the
307                          * budget we allocated.
308                          */
309                         ubifs_release_dirty_inode_budget(c, ui);
310         }
311 
312         *pagep = page;
313         return 0;
314 }
315 
316 /**
317  * allocate_budget - allocate budget for 'ubifs_write_begin()'.
318  * @c: UBIFS file-system description object
319  * @page: page to allocate budget for
320  * @ui: UBIFS inode object the page belongs to
321  * @appending: non-zero if the page is appended
322  *
323  * This is a helper function for 'ubifs_write_begin()' which allocates budget
324  * for the operation. The budget is allocated differently depending on whether
325  * this is appending, whether the page is dirty or not, and so on. This
326  * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
327  * in case of success and %-ENOSPC in case of failure.
328  */
329 static int allocate_budget(struct ubifs_info *c, struct page *page,
330                            struct ubifs_inode *ui, int appending)
331 {
332         struct ubifs_budget_req req = { .fast = 1 };
333 
334         if (PagePrivate(page)) {
335                 if (!appending)
336                         /*
337                          * The page is dirty and we are not appending, which
338                          * means no budget is needed at all.
339                          */
340                         return 0;
341 
342                 mutex_lock(&ui->ui_mutex);
343                 if (ui->dirty)
344                         /*
345                          * The page is dirty and we are appending, so the inode
346                          * has to be marked as dirty. However, it is already
347                          * dirty, so we do not need any budget. We may return,
348                          * but @ui->ui_mutex hast to be left locked because we
349                          * should prevent write-back from flushing the inode
350                          * and freeing the budget. The lock will be released in
351                          * 'ubifs_write_end()'.
352                          */
353                         return 0;
354 
355                 /*
356                  * The page is dirty, we are appending, the inode is clean, so
357                  * we need to budget the inode change.
358                  */
359                 req.dirtied_ino = 1;
360         } else {
361                 if (PageChecked(page))
362                         /*
363                          * The page corresponds to a hole and does not
364                          * exist on the media. So changing it makes
365                          * make the amount of indexing information
366                          * larger, and we have to budget for a new
367                          * page.
368                          */
369                         req.new_page = 1;
370                 else
371                         /*
372                          * Not a hole, the change will not add any new
373                          * indexing information, budget for page
374                          * change.
375                          */
376                         req.dirtied_page = 1;
377 
378                 if (appending) {
379                         mutex_lock(&ui->ui_mutex);
380                         if (!ui->dirty)
381                                 /*
382                                  * The inode is clean but we will have to mark
383                                  * it as dirty because we are appending. This
384                                  * needs a budget.
385                                  */
386                                 req.dirtied_ino = 1;
387                 }
388         }
389 
390         return ubifs_budget_space(c, &req);
391 }
392 
393 /*
394  * This function is called when a page of data is going to be written. Since
395  * the page of data will not necessarily go to the flash straight away, UBIFS
396  * has to reserve space on the media for it, which is done by means of
397  * budgeting.
398  *
399  * This is the hot-path of the file-system and we are trying to optimize it as
400  * much as possible. For this reasons it is split on 2 parts - slow and fast.
401  *
402  * There many budgeting cases:
403  *     o a new page is appended - we have to budget for a new page and for
404  *       changing the inode; however, if the inode is already dirty, there is
405  *       no need to budget for it;
406  *     o an existing clean page is changed - we have budget for it; if the page
407  *       does not exist on the media (a hole), we have to budget for a new
408  *       page; otherwise, we may budget for changing an existing page; the
409  *       difference between these cases is that changing an existing page does
410  *       not introduce anything new to the FS indexing information, so it does
411  *       not grow, and smaller budget is acquired in this case;
412  *     o an existing dirty page is changed - no need to budget at all, because
413  *       the page budget has been acquired by earlier, when the page has been
414  *       marked dirty.
415  *
416  * UBIFS budgeting sub-system may force write-back if it thinks there is no
417  * space to reserve. This imposes some locking restrictions and makes it
418  * impossible to take into account the above cases, and makes it impossible to
419  * optimize budgeting.
420  *
421  * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
422  * there is a plenty of flash space and the budget will be acquired quickly,
423  * without forcing write-back. The slow path does not make this assumption.
424  */
425 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
426                              loff_t pos, unsigned len, unsigned flags,
427                              struct page **pagep, void **fsdata)
428 {
429         struct inode *inode = mapping->host;
430         struct ubifs_info *c = inode->i_sb->s_fs_info;
431         struct ubifs_inode *ui = ubifs_inode(inode);
432         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
433         int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
434         int skipped_read = 0;
435         struct page *page;
436 
437         ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
438         ubifs_assert(!c->ro_media && !c->ro_mount);
439 
440         if (unlikely(c->ro_error))
441                 return -EROFS;
442 
443         /* Try out the fast-path part first */
444         page = grab_cache_page_write_begin(mapping, index, flags);
445         if (unlikely(!page))
446                 return -ENOMEM;
447 
448         if (!PageUptodate(page)) {
449                 /* The page is not loaded from the flash */
450                 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
451                         /*
452                          * We change whole page so no need to load it. But we
453                          * do not know whether this page exists on the media or
454                          * not, so we assume the latter because it requires
455                          * larger budget. The assumption is that it is better
456                          * to budget a bit more than to read the page from the
457                          * media. Thus, we are setting the @PG_checked flag
458                          * here.
459                          */
460                         SetPageChecked(page);
461                         skipped_read = 1;
462                 } else {
463                         err = do_readpage(page);
464                         if (err) {
465                                 unlock_page(page);
466                                 page_cache_release(page);
467                                 return err;
468                         }
469                 }
470 
471                 SetPageUptodate(page);
472                 ClearPageError(page);
473         }
474 
475         err = allocate_budget(c, page, ui, appending);
476         if (unlikely(err)) {
477                 ubifs_assert(err == -ENOSPC);
478                 /*
479                  * If we skipped reading the page because we were going to
480                  * write all of it, then it is not up to date.
481                  */
482                 if (skipped_read) {
483                         ClearPageChecked(page);
484                         ClearPageUptodate(page);
485                 }
486                 /*
487                  * Budgeting failed which means it would have to force
488                  * write-back but didn't, because we set the @fast flag in the
489                  * request. Write-back cannot be done now, while we have the
490                  * page locked, because it would deadlock. Unlock and free
491                  * everything and fall-back to slow-path.
492                  */
493                 if (appending) {
494                         ubifs_assert(mutex_is_locked(&ui->ui_mutex));
495                         mutex_unlock(&ui->ui_mutex);
496                 }
497                 unlock_page(page);
498                 page_cache_release(page);
499 
500                 return write_begin_slow(mapping, pos, len, pagep, flags);
501         }
502 
503         /*
504          * Whee, we acquired budgeting quickly - without involving
505          * garbage-collection, committing or forcing write-back. We return
506          * with @ui->ui_mutex locked if we are appending pages, and unlocked
507          * otherwise. This is an optimization (slightly hacky though).
508          */
509         *pagep = page;
510         return 0;
511 
512 }
513 
514 /**
515  * cancel_budget - cancel budget.
516  * @c: UBIFS file-system description object
517  * @page: page to cancel budget for
518  * @ui: UBIFS inode object the page belongs to
519  * @appending: non-zero if the page is appended
520  *
521  * This is a helper function for a page write operation. It unlocks the
522  * @ui->ui_mutex in case of appending.
523  */
524 static void cancel_budget(struct ubifs_info *c, struct page *page,
525                           struct ubifs_inode *ui, int appending)
526 {
527         if (appending) {
528                 if (!ui->dirty)
529                         ubifs_release_dirty_inode_budget(c, ui);
530                 mutex_unlock(&ui->ui_mutex);
531         }
532         if (!PagePrivate(page)) {
533                 if (PageChecked(page))
534                         release_new_page_budget(c);
535                 else
536                         release_existing_page_budget(c);
537         }
538 }
539 
540 static int ubifs_write_end(struct file *file, struct address_space *mapping,
541                            loff_t pos, unsigned len, unsigned copied,
542                            struct page *page, void *fsdata)
543 {
544         struct inode *inode = mapping->host;
545         struct ubifs_inode *ui = ubifs_inode(inode);
546         struct ubifs_info *c = inode->i_sb->s_fs_info;
547         loff_t end_pos = pos + len;
548         int appending = !!(end_pos > inode->i_size);
549 
550         dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
551                 inode->i_ino, pos, page->index, len, copied, inode->i_size);
552 
553         if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
554                 /*
555                  * VFS copied less data to the page that it intended and
556                  * declared in its '->write_begin()' call via the @len
557                  * argument. If the page was not up-to-date, and @len was
558                  * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
559                  * not load it from the media (for optimization reasons). This
560                  * means that part of the page contains garbage. So read the
561                  * page now.
562                  */
563                 dbg_gen("copied %d instead of %d, read page and repeat",
564                         copied, len);
565                 cancel_budget(c, page, ui, appending);
566                 ClearPageChecked(page);
567 
568                 /*
569                  * Return 0 to force VFS to repeat the whole operation, or the
570                  * error code if 'do_readpage()' fails.
571                  */
572                 copied = do_readpage(page);
573                 goto out;
574         }
575 
576         if (!PagePrivate(page)) {
577                 SetPagePrivate(page);
578                 atomic_long_inc(&c->dirty_pg_cnt);
579                 __set_page_dirty_nobuffers(page);
580         }
581 
582         if (appending) {
583                 i_size_write(inode, end_pos);
584                 ui->ui_size = end_pos;
585                 /*
586                  * Note, we do not set @I_DIRTY_PAGES (which means that the
587                  * inode has dirty pages), this has been done in
588                  * '__set_page_dirty_nobuffers()'.
589                  */
590                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
591                 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
592                 mutex_unlock(&ui->ui_mutex);
593         }
594 
595 out:
596         unlock_page(page);
597         page_cache_release(page);
598         return copied;
599 }
600 
601 /**
602  * populate_page - copy data nodes into a page for bulk-read.
603  * @c: UBIFS file-system description object
604  * @page: page
605  * @bu: bulk-read information
606  * @n: next zbranch slot
607  *
608  * This function returns %0 on success and a negative error code on failure.
609  */
610 static int populate_page(struct ubifs_info *c, struct page *page,
611                          struct bu_info *bu, int *n)
612 {
613         int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
614         struct inode *inode = page->mapping->host;
615         loff_t i_size = i_size_read(inode);
616         unsigned int page_block;
617         void *addr, *zaddr;
618         pgoff_t end_index;
619 
620         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
621                 inode->i_ino, page->index, i_size, page->flags);
622 
623         addr = zaddr = kmap(page);
624 
625         end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
626         if (!i_size || page->index > end_index) {
627                 hole = 1;
628                 memset(addr, 0, PAGE_CACHE_SIZE);
629                 goto out_hole;
630         }
631 
632         page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
633         while (1) {
634                 int err, len, out_len, dlen;
635 
636                 if (nn >= bu->cnt) {
637                         hole = 1;
638                         memset(addr, 0, UBIFS_BLOCK_SIZE);
639                 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
640                         struct ubifs_data_node *dn;
641 
642                         dn = bu->buf + (bu->zbranch[nn].offs - offs);
643 
644                         ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
645                                      ubifs_inode(inode)->creat_sqnum);
646 
647                         len = le32_to_cpu(dn->size);
648                         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
649                                 goto out_err;
650 
651                         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
652                         out_len = UBIFS_BLOCK_SIZE;
653                         err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
654                                                le16_to_cpu(dn->compr_type));
655                         if (err || len != out_len)
656                                 goto out_err;
657 
658                         if (len < UBIFS_BLOCK_SIZE)
659                                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
660 
661                         nn += 1;
662                         read = (i << UBIFS_BLOCK_SHIFT) + len;
663                 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
664                         nn += 1;
665                         continue;
666                 } else {
667                         hole = 1;
668                         memset(addr, 0, UBIFS_BLOCK_SIZE);
669                 }
670                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
671                         break;
672                 addr += UBIFS_BLOCK_SIZE;
673                 page_block += 1;
674         }
675 
676         if (end_index == page->index) {
677                 int len = i_size & (PAGE_CACHE_SIZE - 1);
678 
679                 if (len && len < read)
680                         memset(zaddr + len, 0, read - len);
681         }
682 
683 out_hole:
684         if (hole) {
685                 SetPageChecked(page);
686                 dbg_gen("hole");
687         }
688 
689         SetPageUptodate(page);
690         ClearPageError(page);
691         flush_dcache_page(page);
692         kunmap(page);
693         *n = nn;
694         return 0;
695 
696 out_err:
697         ClearPageUptodate(page);
698         SetPageError(page);
699         flush_dcache_page(page);
700         kunmap(page);
701         ubifs_err("bad data node (block %u, inode %lu)",
702                   page_block, inode->i_ino);
703         return -EINVAL;
704 }
705 
706 /**
707  * ubifs_do_bulk_read - do bulk-read.
708  * @c: UBIFS file-system description object
709  * @bu: bulk-read information
710  * @page1: first page to read
711  *
712  * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
713  */
714 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
715                               struct page *page1)
716 {
717         pgoff_t offset = page1->index, end_index;
718         struct address_space *mapping = page1->mapping;
719         struct inode *inode = mapping->host;
720         struct ubifs_inode *ui = ubifs_inode(inode);
721         int err, page_idx, page_cnt, ret = 0, n = 0;
722         int allocate = bu->buf ? 0 : 1;
723         loff_t isize;
724 
725         err = ubifs_tnc_get_bu_keys(c, bu);
726         if (err)
727                 goto out_warn;
728 
729         if (bu->eof) {
730                 /* Turn off bulk-read at the end of the file */
731                 ui->read_in_a_row = 1;
732                 ui->bulk_read = 0;
733         }
734 
735         page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
736         if (!page_cnt) {
737                 /*
738                  * This happens when there are multiple blocks per page and the
739                  * blocks for the first page we are looking for, are not
740                  * together. If all the pages were like this, bulk-read would
741                  * reduce performance, so we turn it off for a while.
742                  */
743                 goto out_bu_off;
744         }
745 
746         if (bu->cnt) {
747                 if (allocate) {
748                         /*
749                          * Allocate bulk-read buffer depending on how many data
750                          * nodes we are going to read.
751                          */
752                         bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
753                                       bu->zbranch[bu->cnt - 1].len -
754                                       bu->zbranch[0].offs;
755                         ubifs_assert(bu->buf_len > 0);
756                         ubifs_assert(bu->buf_len <= c->leb_size);
757                         bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
758                         if (!bu->buf)
759                                 goto out_bu_off;
760                 }
761 
762                 err = ubifs_tnc_bulk_read(c, bu);
763                 if (err)
764                         goto out_warn;
765         }
766 
767         err = populate_page(c, page1, bu, &n);
768         if (err)
769                 goto out_warn;
770 
771         unlock_page(page1);
772         ret = 1;
773 
774         isize = i_size_read(inode);
775         if (isize == 0)
776                 goto out_free;
777         end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
778 
779         for (page_idx = 1; page_idx < page_cnt; page_idx++) {
780                 pgoff_t page_offset = offset + page_idx;
781                 struct page *page;
782 
783                 if (page_offset > end_index)
784                         break;
785                 page = find_or_create_page(mapping, page_offset,
786                                            GFP_NOFS | __GFP_COLD);
787                 if (!page)
788                         break;
789                 if (!PageUptodate(page))
790                         err = populate_page(c, page, bu, &n);
791                 unlock_page(page);
792                 page_cache_release(page);
793                 if (err)
794                         break;
795         }
796 
797         ui->last_page_read = offset + page_idx - 1;
798 
799 out_free:
800         if (allocate)
801                 kfree(bu->buf);
802         return ret;
803 
804 out_warn:
805         ubifs_warn("ignoring error %d and skipping bulk-read", err);
806         goto out_free;
807 
808 out_bu_off:
809         ui->read_in_a_row = ui->bulk_read = 0;
810         goto out_free;
811 }
812 
813 /**
814  * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
815  * @page: page from which to start bulk-read.
816  *
817  * Some flash media are capable of reading sequentially at faster rates. UBIFS
818  * bulk-read facility is designed to take advantage of that, by reading in one
819  * go consecutive data nodes that are also located consecutively in the same
820  * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
821  */
822 static int ubifs_bulk_read(struct page *page)
823 {
824         struct inode *inode = page->mapping->host;
825         struct ubifs_info *c = inode->i_sb->s_fs_info;
826         struct ubifs_inode *ui = ubifs_inode(inode);
827         pgoff_t index = page->index, last_page_read = ui->last_page_read;
828         struct bu_info *bu;
829         int err = 0, allocated = 0;
830 
831         ui->last_page_read = index;
832         if (!c->bulk_read)
833                 return 0;
834 
835         /*
836          * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
837          * so don't bother if we cannot lock the mutex.
838          */
839         if (!mutex_trylock(&ui->ui_mutex))
840                 return 0;
841 
842         if (index != last_page_read + 1) {
843                 /* Turn off bulk-read if we stop reading sequentially */
844                 ui->read_in_a_row = 1;
845                 if (ui->bulk_read)
846                         ui->bulk_read = 0;
847                 goto out_unlock;
848         }
849 
850         if (!ui->bulk_read) {
851                 ui->read_in_a_row += 1;
852                 if (ui->read_in_a_row < 3)
853                         goto out_unlock;
854                 /* Three reads in a row, so switch on bulk-read */
855                 ui->bulk_read = 1;
856         }
857 
858         /*
859          * If possible, try to use pre-allocated bulk-read information, which
860          * is protected by @c->bu_mutex.
861          */
862         if (mutex_trylock(&c->bu_mutex))
863                 bu = &c->bu;
864         else {
865                 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
866                 if (!bu)
867                         goto out_unlock;
868 
869                 bu->buf = NULL;
870                 allocated = 1;
871         }
872 
873         bu->buf_len = c->max_bu_buf_len;
874         data_key_init(c, &bu->key, inode->i_ino,
875                       page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
876         err = ubifs_do_bulk_read(c, bu, page);
877 
878         if (!allocated)
879                 mutex_unlock(&c->bu_mutex);
880         else
881                 kfree(bu);
882 
883 out_unlock:
884         mutex_unlock(&ui->ui_mutex);
885         return err;
886 }
887 
888 static int ubifs_readpage(struct file *file, struct page *page)
889 {
890         if (ubifs_bulk_read(page))
891                 return 0;
892         do_readpage(page);
893         unlock_page(page);
894         return 0;
895 }
896 
897 static int do_writepage(struct page *page, int len)
898 {
899         int err = 0, i, blen;
900         unsigned int block;
901         void *addr;
902         union ubifs_key key;
903         struct inode *inode = page->mapping->host;
904         struct ubifs_info *c = inode->i_sb->s_fs_info;
905 
906 #ifdef UBIFS_DEBUG
907         struct ubifs_inode *ui = ubifs_inode(inode);
908         spin_lock(&ui->ui_lock);
909         ubifs_assert(page->index <= ui->synced_i_size >> PAGE_CACHE_SHIFT);
910         spin_unlock(&ui->ui_lock);
911 #endif
912 
913         /* Update radix tree tags */
914         set_page_writeback(page);
915 
916         addr = kmap(page);
917         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
918         i = 0;
919         while (len) {
920                 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
921                 data_key_init(c, &key, inode->i_ino, block);
922                 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
923                 if (err)
924                         break;
925                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
926                         break;
927                 block += 1;
928                 addr += blen;
929                 len -= blen;
930         }
931         if (err) {
932                 SetPageError(page);
933                 ubifs_err("cannot write page %lu of inode %lu, error %d",
934                           page->index, inode->i_ino, err);
935                 ubifs_ro_mode(c, err);
936         }
937 
938         ubifs_assert(PagePrivate(page));
939         if (PageChecked(page))
940                 release_new_page_budget(c);
941         else
942                 release_existing_page_budget(c);
943 
944         atomic_long_dec(&c->dirty_pg_cnt);
945         ClearPagePrivate(page);
946         ClearPageChecked(page);
947 
948         kunmap(page);
949         unlock_page(page);
950         end_page_writeback(page);
951         return err;
952 }
953 
954 /*
955  * When writing-back dirty inodes, VFS first writes-back pages belonging to the
956  * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
957  * situation when a we have an inode with size 0, then a megabyte of data is
958  * appended to the inode, then write-back starts and flushes some amount of the
959  * dirty pages, the journal becomes full, commit happens and finishes, and then
960  * an unclean reboot happens. When the file system is mounted next time, the
961  * inode size would still be 0, but there would be many pages which are beyond
962  * the inode size, they would be indexed and consume flash space. Because the
963  * journal has been committed, the replay would not be able to detect this
964  * situation and correct the inode size. This means UBIFS would have to scan
965  * whole index and correct all inode sizes, which is long an unacceptable.
966  *
967  * To prevent situations like this, UBIFS writes pages back only if they are
968  * within the last synchronized inode size, i.e. the size which has been
969  * written to the flash media last time. Otherwise, UBIFS forces inode
970  * write-back, thus making sure the on-flash inode contains current inode size,
971  * and then keeps writing pages back.
972  *
973  * Some locking issues explanation. 'ubifs_writepage()' first is called with
974  * the page locked, and it locks @ui_mutex. However, write-back does take inode
975  * @i_mutex, which means other VFS operations may be run on this inode at the
976  * same time. And the problematic one is truncation to smaller size, from where
977  * we have to call 'truncate_setsize()', which first changes @inode->i_size,
978  * then drops the truncated pages. And while dropping the pages, it takes the
979  * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
980  * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
981  * This means that @inode->i_size is changed while @ui_mutex is unlocked.
982  *
983  * XXX(truncate): with the new truncate sequence this is not true anymore,
984  * and the calls to truncate_setsize can be move around freely.  They should
985  * be moved to the very end of the truncate sequence.
986  *
987  * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
988  * inode size. How do we do this if @inode->i_size may became smaller while we
989  * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
990  * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
991  * internally and updates it under @ui_mutex.
992  *
993  * Q: why we do not worry that if we race with truncation, we may end up with a
994  * situation when the inode is truncated while we are in the middle of
995  * 'do_writepage()', so we do write beyond inode size?
996  * A: If we are in the middle of 'do_writepage()', truncation would be locked
997  * on the page lock and it would not write the truncated inode node to the
998  * journal before we have finished.
999  */
1000 static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
1001 {
1002         struct inode *inode = page->mapping->host;
1003         struct ubifs_inode *ui = ubifs_inode(inode);
1004         loff_t i_size =  i_size_read(inode), synced_i_size;
1005         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1006         int err, len = i_size & (PAGE_CACHE_SIZE - 1);
1007         void *kaddr;
1008 
1009         dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1010                 inode->i_ino, page->index, page->flags);
1011         ubifs_assert(PagePrivate(page));
1012 
1013         /* Is the page fully outside @i_size? (truncate in progress) */
1014         if (page->index > end_index || (page->index == end_index && !len)) {
1015                 err = 0;
1016                 goto out_unlock;
1017         }
1018 
1019         spin_lock(&ui->ui_lock);
1020         synced_i_size = ui->synced_i_size;
1021         spin_unlock(&ui->ui_lock);
1022 
1023         /* Is the page fully inside @i_size? */
1024         if (page->index < end_index) {
1025                 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
1026                         err = inode->i_sb->s_op->write_inode(inode, NULL);
1027                         if (err)
1028                                 goto out_unlock;
1029                         /*
1030                          * The inode has been written, but the write-buffer has
1031                          * not been synchronized, so in case of an unclean
1032                          * reboot we may end up with some pages beyond inode
1033                          * size, but they would be in the journal (because
1034                          * commit flushes write buffers) and recovery would deal
1035                          * with this.
1036                          */
1037                 }
1038                 return do_writepage(page, PAGE_CACHE_SIZE);
1039         }
1040 
1041         /*
1042          * The page straddles @i_size. It must be zeroed out on each and every
1043          * writepage invocation because it may be mmapped. "A file is mapped
1044          * in multiples of the page size. For a file that is not a multiple of
1045          * the page size, the remaining memory is zeroed when mapped, and
1046          * writes to that region are not written out to the file."
1047          */
1048         kaddr = kmap_atomic(page);
1049         memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1050         flush_dcache_page(page);
1051         kunmap_atomic(kaddr);
1052 
1053         if (i_size > synced_i_size) {
1054                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1055                 if (err)
1056                         goto out_unlock;
1057         }
1058 
1059         return do_writepage(page, len);
1060 
1061 out_unlock:
1062         unlock_page(page);
1063         return err;
1064 }
1065 
1066 /**
1067  * do_attr_changes - change inode attributes.
1068  * @inode: inode to change attributes for
1069  * @attr: describes attributes to change
1070  */
1071 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1072 {
1073         if (attr->ia_valid & ATTR_UID)
1074                 inode->i_uid = attr->ia_uid;
1075         if (attr->ia_valid & ATTR_GID)
1076                 inode->i_gid = attr->ia_gid;
1077         if (attr->ia_valid & ATTR_ATIME)
1078                 inode->i_atime = timespec_trunc(attr->ia_atime,
1079                                                 inode->i_sb->s_time_gran);
1080         if (attr->ia_valid & ATTR_MTIME)
1081                 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1082                                                 inode->i_sb->s_time_gran);
1083         if (attr->ia_valid & ATTR_CTIME)
1084                 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1085                                                 inode->i_sb->s_time_gran);
1086         if (attr->ia_valid & ATTR_MODE) {
1087                 umode_t mode = attr->ia_mode;
1088 
1089                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1090                         mode &= ~S_ISGID;
1091                 inode->i_mode = mode;
1092         }
1093 }
1094 
1095 /**
1096  * do_truncation - truncate an inode.
1097  * @c: UBIFS file-system description object
1098  * @inode: inode to truncate
1099  * @attr: inode attribute changes description
1100  *
1101  * This function implements VFS '->setattr()' call when the inode is truncated
1102  * to a smaller size. Returns zero in case of success and a negative error code
1103  * in case of failure.
1104  */
1105 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1106                          const struct iattr *attr)
1107 {
1108         int err;
1109         struct ubifs_budget_req req;
1110         loff_t old_size = inode->i_size, new_size = attr->ia_size;
1111         int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1112         struct ubifs_inode *ui = ubifs_inode(inode);
1113 
1114         dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1115         memset(&req, 0, sizeof(struct ubifs_budget_req));
1116 
1117         /*
1118          * If this is truncation to a smaller size, and we do not truncate on a
1119          * block boundary, budget for changing one data block, because the last
1120          * block will be re-written.
1121          */
1122         if (new_size & (UBIFS_BLOCK_SIZE - 1))
1123                 req.dirtied_page = 1;
1124 
1125         req.dirtied_ino = 1;
1126         /* A funny way to budget for truncation node */
1127         req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1128         err = ubifs_budget_space(c, &req);
1129         if (err) {
1130                 /*
1131                  * Treat truncations to zero as deletion and always allow them,
1132                  * just like we do for '->unlink()'.
1133                  */
1134                 if (new_size || err != -ENOSPC)
1135                         return err;
1136                 budgeted = 0;
1137         }
1138 
1139         truncate_setsize(inode, new_size);
1140 
1141         if (offset) {
1142                 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1143                 struct page *page;
1144 
1145                 page = find_lock_page(inode->i_mapping, index);
1146                 if (page) {
1147                         if (PageDirty(page)) {
1148                                 /*
1149                                  * 'ubifs_jnl_truncate()' will try to truncate
1150                                  * the last data node, but it contains
1151                                  * out-of-date data because the page is dirty.
1152                                  * Write the page now, so that
1153                                  * 'ubifs_jnl_truncate()' will see an already
1154                                  * truncated (and up to date) data node.
1155                                  */
1156                                 ubifs_assert(PagePrivate(page));
1157 
1158                                 clear_page_dirty_for_io(page);
1159                                 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1160                                         offset = new_size &
1161                                                  (PAGE_CACHE_SIZE - 1);
1162                                 err = do_writepage(page, offset);
1163                                 page_cache_release(page);
1164                                 if (err)
1165                                         goto out_budg;
1166                                 /*
1167                                  * We could now tell 'ubifs_jnl_truncate()' not
1168                                  * to read the last block.
1169                                  */
1170                         } else {
1171                                 /*
1172                                  * We could 'kmap()' the page and pass the data
1173                                  * to 'ubifs_jnl_truncate()' to save it from
1174                                  * having to read it.
1175                                  */
1176                                 unlock_page(page);
1177                                 page_cache_release(page);
1178                         }
1179                 }
1180         }
1181 
1182         mutex_lock(&ui->ui_mutex);
1183         ui->ui_size = inode->i_size;
1184         /* Truncation changes inode [mc]time */
1185         inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1186         /* Other attributes may be changed at the same time as well */
1187         do_attr_changes(inode, attr);
1188         err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1189         mutex_unlock(&ui->ui_mutex);
1190 
1191 out_budg:
1192         if (budgeted)
1193                 ubifs_release_budget(c, &req);
1194         else {
1195                 c->bi.nospace = c->bi.nospace_rp = 0;
1196                 smp_wmb();
1197         }
1198         return err;
1199 }
1200 
1201 /**
1202  * do_setattr - change inode attributes.
1203  * @c: UBIFS file-system description object
1204  * @inode: inode to change attributes for
1205  * @attr: inode attribute changes description
1206  *
1207  * This function implements VFS '->setattr()' call for all cases except
1208  * truncations to smaller size. Returns zero in case of success and a negative
1209  * error code in case of failure.
1210  */
1211 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1212                       const struct iattr *attr)
1213 {
1214         int err, release;
1215         loff_t new_size = attr->ia_size;
1216         struct ubifs_inode *ui = ubifs_inode(inode);
1217         struct ubifs_budget_req req = { .dirtied_ino = 1,
1218                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1219 
1220         err = ubifs_budget_space(c, &req);
1221         if (err)
1222                 return err;
1223 
1224         if (attr->ia_valid & ATTR_SIZE) {
1225                 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1226                 truncate_setsize(inode, new_size);
1227         }
1228 
1229         mutex_lock(&ui->ui_mutex);
1230         if (attr->ia_valid & ATTR_SIZE) {
1231                 /* Truncation changes inode [mc]time */
1232                 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1233                 /* 'truncate_setsize()' changed @i_size, update @ui_size */
1234                 ui->ui_size = inode->i_size;
1235         }
1236 
1237         do_attr_changes(inode, attr);
1238 
1239         release = ui->dirty;
1240         if (attr->ia_valid & ATTR_SIZE)
1241                 /*
1242                  * Inode length changed, so we have to make sure
1243                  * @I_DIRTY_DATASYNC is set.
1244                  */
1245                  __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1246         else
1247                 mark_inode_dirty_sync(inode);
1248         mutex_unlock(&ui->ui_mutex);
1249 
1250         if (release)
1251                 ubifs_release_budget(c, &req);
1252         if (IS_SYNC(inode))
1253                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1254         return err;
1255 }
1256 
1257 int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1258 {
1259         int err;
1260         struct inode *inode = dentry->d_inode;
1261         struct ubifs_info *c = inode->i_sb->s_fs_info;
1262 
1263         dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1264                 inode->i_ino, inode->i_mode, attr->ia_valid);
1265         err = inode_change_ok(inode, attr);
1266         if (err)
1267                 return err;
1268 
1269         err = dbg_check_synced_i_size(c, inode);
1270         if (err)
1271                 return err;
1272 
1273         if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1274                 /* Truncation to a smaller size */
1275                 err = do_truncation(c, inode, attr);
1276         else
1277                 err = do_setattr(c, inode, attr);
1278 
1279         return err;
1280 }
1281 
1282 static void ubifs_invalidatepage(struct page *page, unsigned int offset,
1283                                  unsigned int length)
1284 {
1285         struct inode *inode = page->mapping->host;
1286         struct ubifs_info *c = inode->i_sb->s_fs_info;
1287 
1288         ubifs_assert(PagePrivate(page));
1289         if (offset || length < PAGE_CACHE_SIZE)
1290                 /* Partial page remains dirty */
1291                 return;
1292 
1293         if (PageChecked(page))
1294                 release_new_page_budget(c);
1295         else
1296                 release_existing_page_budget(c);
1297 
1298         atomic_long_dec(&c->dirty_pg_cnt);
1299         ClearPagePrivate(page);
1300         ClearPageChecked(page);
1301 }
1302 
1303 static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
1304 {
1305         struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
1306 
1307         nd_set_link(nd, ui->data);
1308         return NULL;
1309 }
1310 
1311 int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1312 {
1313         struct inode *inode = file->f_mapping->host;
1314         struct ubifs_info *c = inode->i_sb->s_fs_info;
1315         int err;
1316 
1317         dbg_gen("syncing inode %lu", inode->i_ino);
1318 
1319         if (c->ro_mount)
1320                 /*
1321                  * For some really strange reasons VFS does not filter out
1322                  * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1323                  */
1324                 return 0;
1325 
1326         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
1327         if (err)
1328                 return err;
1329         mutex_lock(&inode->i_mutex);
1330 
1331         /* Synchronize the inode unless this is a 'datasync()' call. */
1332         if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1333                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1334                 if (err)
1335                         goto out;
1336         }
1337 
1338         /*
1339          * Nodes related to this inode may still sit in a write-buffer. Flush
1340          * them.
1341          */
1342         err = ubifs_sync_wbufs_by_inode(c, inode);
1343 out:
1344         mutex_unlock(&inode->i_mutex);
1345         return err;
1346 }
1347 
1348 /**
1349  * mctime_update_needed - check if mtime or ctime update is needed.
1350  * @inode: the inode to do the check for
1351  * @now: current time
1352  *
1353  * This helper function checks if the inode mtime/ctime should be updated or
1354  * not. If current values of the time-stamps are within the UBIFS inode time
1355  * granularity, they are not updated. This is an optimization.
1356  */
1357 static inline int mctime_update_needed(const struct inode *inode,
1358                                        const struct timespec *now)
1359 {
1360         if (!timespec_equal(&inode->i_mtime, now) ||
1361             !timespec_equal(&inode->i_ctime, now))
1362                 return 1;
1363         return 0;
1364 }
1365 
1366 /**
1367  * update_ctime - update mtime and ctime of an inode.
1368  * @inode: inode to update
1369  *
1370  * This function updates mtime and ctime of the inode if it is not equivalent to
1371  * current time. Returns zero in case of success and a negative error code in
1372  * case of failure.
1373  */
1374 static int update_mctime(struct inode *inode)
1375 {
1376         struct timespec now = ubifs_current_time(inode);
1377         struct ubifs_inode *ui = ubifs_inode(inode);
1378         struct ubifs_info *c = inode->i_sb->s_fs_info;
1379 
1380         if (mctime_update_needed(inode, &now)) {
1381                 int err, release;
1382                 struct ubifs_budget_req req = { .dirtied_ino = 1,
1383                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1384 
1385                 err = ubifs_budget_space(c, &req);
1386                 if (err)
1387                         return err;
1388 
1389                 mutex_lock(&ui->ui_mutex);
1390                 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1391                 release = ui->dirty;
1392                 mark_inode_dirty_sync(inode);
1393                 mutex_unlock(&ui->ui_mutex);
1394                 if (release)
1395                         ubifs_release_budget(c, &req);
1396         }
1397 
1398         return 0;
1399 }
1400 
1401 static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1402 {
1403         int err = update_mctime(file_inode(iocb->ki_filp));
1404         if (err)
1405                 return err;
1406 
1407         return generic_file_write_iter(iocb, from);
1408 }
1409 
1410 static int ubifs_set_page_dirty(struct page *page)
1411 {
1412         int ret;
1413 
1414         ret = __set_page_dirty_nobuffers(page);
1415         /*
1416          * An attempt to dirty a page without budgeting for it - should not
1417          * happen.
1418          */
1419         ubifs_assert(ret == 0);
1420         return ret;
1421 }
1422 
1423 static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1424 {
1425         /*
1426          * An attempt to release a dirty page without budgeting for it - should
1427          * not happen.
1428          */
1429         if (PageWriteback(page))
1430                 return 0;
1431         ubifs_assert(PagePrivate(page));
1432         ubifs_assert(0);
1433         ClearPagePrivate(page);
1434         ClearPageChecked(page);
1435         return 1;
1436 }
1437 
1438 /*
1439  * mmap()d file has taken write protection fault and is being made writable.
1440  * UBIFS must ensure page is budgeted for.
1441  */
1442 static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma,
1443                                  struct vm_fault *vmf)
1444 {
1445         struct page *page = vmf->page;
1446         struct inode *inode = file_inode(vma->vm_file);
1447         struct ubifs_info *c = inode->i_sb->s_fs_info;
1448         struct timespec now = ubifs_current_time(inode);
1449         struct ubifs_budget_req req = { .new_page = 1 };
1450         int err, update_time;
1451 
1452         dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1453                 i_size_read(inode));
1454         ubifs_assert(!c->ro_media && !c->ro_mount);
1455 
1456         if (unlikely(c->ro_error))
1457                 return VM_FAULT_SIGBUS; /* -EROFS */
1458 
1459         /*
1460          * We have not locked @page so far so we may budget for changing the
1461          * page. Note, we cannot do this after we locked the page, because
1462          * budgeting may cause write-back which would cause deadlock.
1463          *
1464          * At the moment we do not know whether the page is dirty or not, so we
1465          * assume that it is not and budget for a new page. We could look at
1466          * the @PG_private flag and figure this out, but we may race with write
1467          * back and the page state may change by the time we lock it, so this
1468          * would need additional care. We do not bother with this at the
1469          * moment, although it might be good idea to do. Instead, we allocate
1470          * budget for a new page and amend it later on if the page was in fact
1471          * dirty.
1472          *
1473          * The budgeting-related logic of this function is similar to what we
1474          * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1475          * for more comments.
1476          */
1477         update_time = mctime_update_needed(inode, &now);
1478         if (update_time)
1479                 /*
1480                  * We have to change inode time stamp which requires extra
1481                  * budgeting.
1482                  */
1483                 req.dirtied_ino = 1;
1484 
1485         err = ubifs_budget_space(c, &req);
1486         if (unlikely(err)) {
1487                 if (err == -ENOSPC)
1488                         ubifs_warn("out of space for mmapped file (inode number %lu)",
1489                                    inode->i_ino);
1490                 return VM_FAULT_SIGBUS;
1491         }
1492 
1493         lock_page(page);
1494         if (unlikely(page->mapping != inode->i_mapping ||
1495                      page_offset(page) > i_size_read(inode))) {
1496                 /* Page got truncated out from underneath us */
1497                 err = -EINVAL;
1498                 goto out_unlock;
1499         }
1500 
1501         if (PagePrivate(page))
1502                 release_new_page_budget(c);
1503         else {
1504                 if (!PageChecked(page))
1505                         ubifs_convert_page_budget(c);
1506                 SetPagePrivate(page);
1507                 atomic_long_inc(&c->dirty_pg_cnt);
1508                 __set_page_dirty_nobuffers(page);
1509         }
1510 
1511         if (update_time) {
1512                 int release;
1513                 struct ubifs_inode *ui = ubifs_inode(inode);
1514 
1515                 mutex_lock(&ui->ui_mutex);
1516                 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1517                 release = ui->dirty;
1518                 mark_inode_dirty_sync(inode);
1519                 mutex_unlock(&ui->ui_mutex);
1520                 if (release)
1521                         ubifs_release_dirty_inode_budget(c, ui);
1522         }
1523 
1524         wait_for_stable_page(page);
1525         return VM_FAULT_LOCKED;
1526 
1527 out_unlock:
1528         unlock_page(page);
1529         ubifs_release_budget(c, &req);
1530         if (err)
1531                 err = VM_FAULT_SIGBUS;
1532         return err;
1533 }
1534 
1535 static const struct vm_operations_struct ubifs_file_vm_ops = {
1536         .fault        = filemap_fault,
1537         .map_pages = filemap_map_pages,
1538         .page_mkwrite = ubifs_vm_page_mkwrite,
1539 };
1540 
1541 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1542 {
1543         int err;
1544 
1545         err = generic_file_mmap(file, vma);
1546         if (err)
1547                 return err;
1548         vma->vm_ops = &ubifs_file_vm_ops;
1549         return 0;
1550 }
1551 
1552 const struct address_space_operations ubifs_file_address_operations = {
1553         .readpage       = ubifs_readpage,
1554         .writepage      = ubifs_writepage,
1555         .write_begin    = ubifs_write_begin,
1556         .write_end      = ubifs_write_end,
1557         .invalidatepage = ubifs_invalidatepage,
1558         .set_page_dirty = ubifs_set_page_dirty,
1559         .releasepage    = ubifs_releasepage,
1560 };
1561 
1562 const struct inode_operations ubifs_file_inode_operations = {
1563         .setattr     = ubifs_setattr,
1564         .getattr     = ubifs_getattr,
1565         .setxattr    = ubifs_setxattr,
1566         .getxattr    = ubifs_getxattr,
1567         .listxattr   = ubifs_listxattr,
1568         .removexattr = ubifs_removexattr,
1569 };
1570 
1571 const struct inode_operations ubifs_symlink_inode_operations = {
1572         .readlink    = generic_readlink,
1573         .follow_link = ubifs_follow_link,
1574         .setattr     = ubifs_setattr,
1575         .getattr     = ubifs_getattr,
1576         .setxattr    = ubifs_setxattr,
1577         .getxattr    = ubifs_getxattr,
1578         .listxattr   = ubifs_listxattr,
1579         .removexattr = ubifs_removexattr,
1580 };
1581 
1582 const struct file_operations ubifs_file_operations = {
1583         .llseek         = generic_file_llseek,
1584         .read           = new_sync_read,
1585         .write          = new_sync_write,
1586         .read_iter      = generic_file_read_iter,
1587         .write_iter     = ubifs_write_iter,
1588         .mmap           = ubifs_file_mmap,
1589         .fsync          = ubifs_fsync,
1590         .unlocked_ioctl = ubifs_ioctl,
1591         .splice_read    = generic_file_splice_read,
1592         .splice_write   = iter_file_splice_write,
1593 #ifdef CONFIG_COMPAT
1594         .compat_ioctl   = ubifs_compat_ioctl,
1595 #endif
1596 };
1597 

~ [ 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