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

TOMOYO Linux Cross Reference
Linux/fs/gfs2/aops.c

Version: ~ [ linux-5.12-rc7 ] ~ [ linux-5.11.13 ] ~ [ linux-5.10.29 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.111 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.186 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.230 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.266 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.266 ] ~ [ 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  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
  3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
  4  *
  5  * This copyrighted material is made available to anyone wishing to use,
  6  * modify, copy, or redistribute it subject to the terms and conditions
  7  * of the GNU General Public License version 2.
  8  */
  9 
 10 #include <linux/sched.h>
 11 #include <linux/slab.h>
 12 #include <linux/spinlock.h>
 13 #include <linux/completion.h>
 14 #include <linux/buffer_head.h>
 15 #include <linux/pagemap.h>
 16 #include <linux/pagevec.h>
 17 #include <linux/mpage.h>
 18 #include <linux/fs.h>
 19 #include <linux/writeback.h>
 20 #include <linux/swap.h>
 21 #include <linux/gfs2_ondisk.h>
 22 #include <linux/backing-dev.h>
 23 #include <linux/aio.h>
 24 
 25 #include "gfs2.h"
 26 #include "incore.h"
 27 #include "bmap.h"
 28 #include "glock.h"
 29 #include "inode.h"
 30 #include "log.h"
 31 #include "meta_io.h"
 32 #include "quota.h"
 33 #include "trans.h"
 34 #include "rgrp.h"
 35 #include "super.h"
 36 #include "util.h"
 37 #include "glops.h"
 38 
 39 
 40 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
 41                                    unsigned int from, unsigned int to)
 42 {
 43         struct buffer_head *head = page_buffers(page);
 44         unsigned int bsize = head->b_size;
 45         struct buffer_head *bh;
 46         unsigned int start, end;
 47 
 48         for (bh = head, start = 0; bh != head || !start;
 49              bh = bh->b_this_page, start = end) {
 50                 end = start + bsize;
 51                 if (end <= from || start >= to)
 52                         continue;
 53                 if (gfs2_is_jdata(ip))
 54                         set_buffer_uptodate(bh);
 55                 gfs2_trans_add_data(ip->i_gl, bh);
 56         }
 57 }
 58 
 59 /**
 60  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
 61  * @inode: The inode
 62  * @lblock: The block number to look up
 63  * @bh_result: The buffer head to return the result in
 64  * @create: Non-zero if we may add block to the file
 65  *
 66  * Returns: errno
 67  */
 68 
 69 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
 70                                   struct buffer_head *bh_result, int create)
 71 {
 72         int error;
 73 
 74         error = gfs2_block_map(inode, lblock, bh_result, 0);
 75         if (error)
 76                 return error;
 77         if (!buffer_mapped(bh_result))
 78                 return -EIO;
 79         return 0;
 80 }
 81 
 82 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
 83                                  struct buffer_head *bh_result, int create)
 84 {
 85         return gfs2_block_map(inode, lblock, bh_result, 0);
 86 }
 87 
 88 /**
 89  * gfs2_writepage_common - Common bits of writepage
 90  * @page: The page to be written
 91  * @wbc: The writeback control
 92  *
 93  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
 94  */
 95 
 96 static int gfs2_writepage_common(struct page *page,
 97                                  struct writeback_control *wbc)
 98 {
 99         struct inode *inode = page->mapping->host;
100         struct gfs2_inode *ip = GFS2_I(inode);
101         struct gfs2_sbd *sdp = GFS2_SB(inode);
102         loff_t i_size = i_size_read(inode);
103         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
104         unsigned offset;
105 
106         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
107                 goto out;
108         if (current->journal_info)
109                 goto redirty;
110         /* Is the page fully outside i_size? (truncate in progress) */
111         offset = i_size & (PAGE_CACHE_SIZE-1);
112         if (page->index > end_index || (page->index == end_index && !offset)) {
113                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
114                 goto out;
115         }
116         return 1;
117 redirty:
118         redirty_page_for_writepage(wbc, page);
119 out:
120         unlock_page(page);
121         return 0;
122 }
123 
124 /**
125  * gfs2_writepage - Write page for writeback mappings
126  * @page: The page
127  * @wbc: The writeback control
128  *
129  */
130 
131 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
132 {
133         int ret;
134 
135         ret = gfs2_writepage_common(page, wbc);
136         if (ret <= 0)
137                 return ret;
138 
139         return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
140 }
141 
142 /**
143  * __gfs2_jdata_writepage - The core of jdata writepage
144  * @page: The page to write
145  * @wbc: The writeback control
146  *
147  * This is shared between writepage and writepages and implements the
148  * core of the writepage operation. If a transaction is required then
149  * PageChecked will have been set and the transaction will have
150  * already been started before this is called.
151  */
152 
153 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
154 {
155         struct inode *inode = page->mapping->host;
156         struct gfs2_inode *ip = GFS2_I(inode);
157         struct gfs2_sbd *sdp = GFS2_SB(inode);
158 
159         if (PageChecked(page)) {
160                 ClearPageChecked(page);
161                 if (!page_has_buffers(page)) {
162                         create_empty_buffers(page, inode->i_sb->s_blocksize,
163                                              (1 << BH_Dirty)|(1 << BH_Uptodate));
164                 }
165                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
166         }
167         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
168 }
169 
170 /**
171  * gfs2_jdata_writepage - Write complete page
172  * @page: Page to write
173  *
174  * Returns: errno
175  *
176  */
177 
178 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
179 {
180         struct inode *inode = page->mapping->host;
181         struct gfs2_sbd *sdp = GFS2_SB(inode);
182         int ret;
183         int done_trans = 0;
184 
185         if (PageChecked(page)) {
186                 if (wbc->sync_mode != WB_SYNC_ALL)
187                         goto out_ignore;
188                 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
189                 if (ret)
190                         goto out_ignore;
191                 done_trans = 1;
192         }
193         ret = gfs2_writepage_common(page, wbc);
194         if (ret > 0)
195                 ret = __gfs2_jdata_writepage(page, wbc);
196         if (done_trans)
197                 gfs2_trans_end(sdp);
198         return ret;
199 
200 out_ignore:
201         redirty_page_for_writepage(wbc, page);
202         unlock_page(page);
203         return 0;
204 }
205 
206 /**
207  * gfs2_writepages - Write a bunch of dirty pages back to disk
208  * @mapping: The mapping to write
209  * @wbc: Write-back control
210  *
211  * Used for both ordered and writeback modes.
212  */
213 static int gfs2_writepages(struct address_space *mapping,
214                            struct writeback_control *wbc)
215 {
216         return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
217 }
218 
219 /**
220  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
221  * @mapping: The mapping
222  * @wbc: The writeback control
223  * @writepage: The writepage function to call for each page
224  * @pvec: The vector of pages
225  * @nr_pages: The number of pages to write
226  *
227  * Returns: non-zero if loop should terminate, zero otherwise
228  */
229 
230 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
231                                     struct writeback_control *wbc,
232                                     struct pagevec *pvec,
233                                     int nr_pages, pgoff_t end)
234 {
235         struct inode *inode = mapping->host;
236         struct gfs2_sbd *sdp = GFS2_SB(inode);
237         loff_t i_size = i_size_read(inode);
238         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
239         unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
240         unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
241         int i;
242         int ret;
243 
244         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
245         if (ret < 0)
246                 return ret;
247 
248         for(i = 0; i < nr_pages; i++) {
249                 struct page *page = pvec->pages[i];
250 
251                 lock_page(page);
252 
253                 if (unlikely(page->mapping != mapping)) {
254                         unlock_page(page);
255                         continue;
256                 }
257 
258                 if (!wbc->range_cyclic && page->index > end) {
259                         ret = 1;
260                         unlock_page(page);
261                         continue;
262                 }
263 
264                 if (wbc->sync_mode != WB_SYNC_NONE)
265                         wait_on_page_writeback(page);
266 
267                 if (PageWriteback(page) ||
268                     !clear_page_dirty_for_io(page)) {
269                         unlock_page(page);
270                         continue;
271                 }
272 
273                 /* Is the page fully outside i_size? (truncate in progress) */
274                 if (page->index > end_index || (page->index == end_index && !offset)) {
275                         page->mapping->a_ops->invalidatepage(page, 0,
276                                                              PAGE_CACHE_SIZE);
277                         unlock_page(page);
278                         continue;
279                 }
280 
281                 ret = __gfs2_jdata_writepage(page, wbc);
282 
283                 if (ret || (--(wbc->nr_to_write) <= 0))
284                         ret = 1;
285         }
286         gfs2_trans_end(sdp);
287         return ret;
288 }
289 
290 /**
291  * gfs2_write_cache_jdata - Like write_cache_pages but different
292  * @mapping: The mapping to write
293  * @wbc: The writeback control
294  * @writepage: The writepage function to call
295  * @data: The data to pass to writepage
296  *
297  * The reason that we use our own function here is that we need to
298  * start transactions before we grab page locks. This allows us
299  * to get the ordering right.
300  */
301 
302 static int gfs2_write_cache_jdata(struct address_space *mapping,
303                                   struct writeback_control *wbc)
304 {
305         int ret = 0;
306         int done = 0;
307         struct pagevec pvec;
308         int nr_pages;
309         pgoff_t index;
310         pgoff_t end;
311         int scanned = 0;
312         int range_whole = 0;
313 
314         pagevec_init(&pvec, 0);
315         if (wbc->range_cyclic) {
316                 index = mapping->writeback_index; /* Start from prev offset */
317                 end = -1;
318         } else {
319                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
320                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
321                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
322                         range_whole = 1;
323                 scanned = 1;
324         }
325 
326 retry:
327          while (!done && (index <= end) &&
328                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
329                                                PAGECACHE_TAG_DIRTY,
330                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
331                 scanned = 1;
332                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
333                 if (ret)
334                         done = 1;
335                 if (ret > 0)
336                         ret = 0;
337 
338                 pagevec_release(&pvec);
339                 cond_resched();
340         }
341 
342         if (!scanned && !done) {
343                 /*
344                  * We hit the last page and there is more work to be done: wrap
345                  * back to the start of the file
346                  */
347                 scanned = 1;
348                 index = 0;
349                 goto retry;
350         }
351 
352         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
353                 mapping->writeback_index = index;
354         return ret;
355 }
356 
357 
358 /**
359  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
360  * @mapping: The mapping to write
361  * @wbc: The writeback control
362  * 
363  */
364 
365 static int gfs2_jdata_writepages(struct address_space *mapping,
366                                  struct writeback_control *wbc)
367 {
368         struct gfs2_inode *ip = GFS2_I(mapping->host);
369         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
370         int ret;
371 
372         ret = gfs2_write_cache_jdata(mapping, wbc);
373         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
374                 gfs2_log_flush(sdp, ip->i_gl);
375                 ret = gfs2_write_cache_jdata(mapping, wbc);
376         }
377         return ret;
378 }
379 
380 /**
381  * stuffed_readpage - Fill in a Linux page with stuffed file data
382  * @ip: the inode
383  * @page: the page
384  *
385  * Returns: errno
386  */
387 
388 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
389 {
390         struct buffer_head *dibh;
391         u64 dsize = i_size_read(&ip->i_inode);
392         void *kaddr;
393         int error;
394 
395         /*
396          * Due to the order of unstuffing files and ->fault(), we can be
397          * asked for a zero page in the case of a stuffed file being extended,
398          * so we need to supply one here. It doesn't happen often.
399          */
400         if (unlikely(page->index)) {
401                 zero_user(page, 0, PAGE_CACHE_SIZE);
402                 SetPageUptodate(page);
403                 return 0;
404         }
405 
406         error = gfs2_meta_inode_buffer(ip, &dibh);
407         if (error)
408                 return error;
409 
410         kaddr = kmap_atomic(page);
411         if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
412                 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
413         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
414         memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
415         kunmap_atomic(kaddr);
416         flush_dcache_page(page);
417         brelse(dibh);
418         SetPageUptodate(page);
419 
420         return 0;
421 }
422 
423 
424 /**
425  * __gfs2_readpage - readpage
426  * @file: The file to read a page for
427  * @page: The page to read
428  *
429  * This is the core of gfs2's readpage. Its used by the internal file
430  * reading code as in that case we already hold the glock. Also its
431  * called by gfs2_readpage() once the required lock has been granted.
432  *
433  */
434 
435 static int __gfs2_readpage(void *file, struct page *page)
436 {
437         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
438         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
439         int error;
440 
441         if (gfs2_is_stuffed(ip)) {
442                 error = stuffed_readpage(ip, page);
443                 unlock_page(page);
444         } else {
445                 error = mpage_readpage(page, gfs2_block_map);
446         }
447 
448         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
449                 return -EIO;
450 
451         return error;
452 }
453 
454 /**
455  * gfs2_readpage - read a page of a file
456  * @file: The file to read
457  * @page: The page of the file
458  *
459  * This deals with the locking required. We have to unlock and
460  * relock the page in order to get the locking in the right
461  * order.
462  */
463 
464 static int gfs2_readpage(struct file *file, struct page *page)
465 {
466         struct address_space *mapping = page->mapping;
467         struct gfs2_inode *ip = GFS2_I(mapping->host);
468         struct gfs2_holder gh;
469         int error;
470 
471         unlock_page(page);
472         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
473         error = gfs2_glock_nq(&gh);
474         if (unlikely(error))
475                 goto out;
476         error = AOP_TRUNCATED_PAGE;
477         lock_page(page);
478         if (page->mapping == mapping && !PageUptodate(page))
479                 error = __gfs2_readpage(file, page);
480         else
481                 unlock_page(page);
482         gfs2_glock_dq(&gh);
483 out:
484         gfs2_holder_uninit(&gh);
485         if (error && error != AOP_TRUNCATED_PAGE)
486                 lock_page(page);
487         return error;
488 }
489 
490 /**
491  * gfs2_internal_read - read an internal file
492  * @ip: The gfs2 inode
493  * @buf: The buffer to fill
494  * @pos: The file position
495  * @size: The amount to read
496  *
497  */
498 
499 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
500                        unsigned size)
501 {
502         struct address_space *mapping = ip->i_inode.i_mapping;
503         unsigned long index = *pos / PAGE_CACHE_SIZE;
504         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
505         unsigned copied = 0;
506         unsigned amt;
507         struct page *page;
508         void *p;
509 
510         do {
511                 amt = size - copied;
512                 if (offset + size > PAGE_CACHE_SIZE)
513                         amt = PAGE_CACHE_SIZE - offset;
514                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
515                 if (IS_ERR(page))
516                         return PTR_ERR(page);
517                 p = kmap_atomic(page);
518                 memcpy(buf + copied, p + offset, amt);
519                 kunmap_atomic(p);
520                 page_cache_release(page);
521                 copied += amt;
522                 index++;
523                 offset = 0;
524         } while(copied < size);
525         (*pos) += size;
526         return size;
527 }
528 
529 /**
530  * gfs2_readpages - Read a bunch of pages at once
531  *
532  * Some notes:
533  * 1. This is only for readahead, so we can simply ignore any things
534  *    which are slightly inconvenient (such as locking conflicts between
535  *    the page lock and the glock) and return having done no I/O. Its
536  *    obviously not something we'd want to do on too regular a basis.
537  *    Any I/O we ignore at this time will be done via readpage later.
538  * 2. We don't handle stuffed files here we let readpage do the honours.
539  * 3. mpage_readpages() does most of the heavy lifting in the common case.
540  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
541  */
542 
543 static int gfs2_readpages(struct file *file, struct address_space *mapping,
544                           struct list_head *pages, unsigned nr_pages)
545 {
546         struct inode *inode = mapping->host;
547         struct gfs2_inode *ip = GFS2_I(inode);
548         struct gfs2_sbd *sdp = GFS2_SB(inode);
549         struct gfs2_holder gh;
550         int ret;
551 
552         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
553         ret = gfs2_glock_nq(&gh);
554         if (unlikely(ret))
555                 goto out_uninit;
556         if (!gfs2_is_stuffed(ip))
557                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
558         gfs2_glock_dq(&gh);
559 out_uninit:
560         gfs2_holder_uninit(&gh);
561         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
562                 ret = -EIO;
563         return ret;
564 }
565 
566 /**
567  * gfs2_write_begin - Begin to write to a file
568  * @file: The file to write to
569  * @mapping: The mapping in which to write
570  * @pos: The file offset at which to start writing
571  * @len: Length of the write
572  * @flags: Various flags
573  * @pagep: Pointer to return the page
574  * @fsdata: Pointer to return fs data (unused by GFS2)
575  *
576  * Returns: errno
577  */
578 
579 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
580                             loff_t pos, unsigned len, unsigned flags,
581                             struct page **pagep, void **fsdata)
582 {
583         struct gfs2_inode *ip = GFS2_I(mapping->host);
584         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
585         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
586         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
587         unsigned requested = 0;
588         int alloc_required;
589         int error = 0;
590         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
591         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
592         struct page *page;
593 
594         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
595         error = gfs2_glock_nq(&ip->i_gh);
596         if (unlikely(error))
597                 goto out_uninit;
598         if (&ip->i_inode == sdp->sd_rindex) {
599                 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
600                                            GL_NOCACHE, &m_ip->i_gh);
601                 if (unlikely(error)) {
602                         gfs2_glock_dq(&ip->i_gh);
603                         goto out_uninit;
604                 }
605         }
606 
607         alloc_required = gfs2_write_alloc_required(ip, pos, len);
608 
609         if (alloc_required || gfs2_is_jdata(ip))
610                 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
611 
612         if (alloc_required) {
613                 struct gfs2_alloc_parms ap = { .aflags = 0, };
614                 error = gfs2_quota_lock_check(ip);
615                 if (error)
616                         goto out_unlock;
617 
618                 requested = data_blocks + ind_blocks;
619                 ap.target = requested;
620                 error = gfs2_inplace_reserve(ip, &ap);
621                 if (error)
622                         goto out_qunlock;
623         }
624 
625         rblocks = RES_DINODE + ind_blocks;
626         if (gfs2_is_jdata(ip))
627                 rblocks += data_blocks ? data_blocks : 1;
628         if (ind_blocks || data_blocks)
629                 rblocks += RES_STATFS + RES_QUOTA;
630         if (&ip->i_inode == sdp->sd_rindex)
631                 rblocks += 2 * RES_STATFS;
632         if (alloc_required)
633                 rblocks += gfs2_rg_blocks(ip, requested);
634 
635         error = gfs2_trans_begin(sdp, rblocks,
636                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
637         if (error)
638                 goto out_trans_fail;
639 
640         error = -ENOMEM;
641         flags |= AOP_FLAG_NOFS;
642         page = grab_cache_page_write_begin(mapping, index, flags);
643         *pagep = page;
644         if (unlikely(!page))
645                 goto out_endtrans;
646 
647         if (gfs2_is_stuffed(ip)) {
648                 error = 0;
649                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
650                         error = gfs2_unstuff_dinode(ip, page);
651                         if (error == 0)
652                                 goto prepare_write;
653                 } else if (!PageUptodate(page)) {
654                         error = stuffed_readpage(ip, page);
655                 }
656                 goto out;
657         }
658 
659 prepare_write:
660         error = __block_write_begin(page, from, len, gfs2_block_map);
661 out:
662         if (error == 0)
663                 return 0;
664 
665         unlock_page(page);
666         page_cache_release(page);
667 
668         gfs2_trans_end(sdp);
669         if (pos + len > ip->i_inode.i_size)
670                 gfs2_trim_blocks(&ip->i_inode);
671         goto out_trans_fail;
672 
673 out_endtrans:
674         gfs2_trans_end(sdp);
675 out_trans_fail:
676         if (alloc_required) {
677                 gfs2_inplace_release(ip);
678 out_qunlock:
679                 gfs2_quota_unlock(ip);
680         }
681 out_unlock:
682         if (&ip->i_inode == sdp->sd_rindex) {
683                 gfs2_glock_dq(&m_ip->i_gh);
684                 gfs2_holder_uninit(&m_ip->i_gh);
685         }
686         gfs2_glock_dq(&ip->i_gh);
687 out_uninit:
688         gfs2_holder_uninit(&ip->i_gh);
689         return error;
690 }
691 
692 /**
693  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
694  * @inode: the rindex inode
695  */
696 static void adjust_fs_space(struct inode *inode)
697 {
698         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
699         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
700         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
701         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
702         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
703         struct buffer_head *m_bh, *l_bh;
704         u64 fs_total, new_free;
705 
706         /* Total up the file system space, according to the latest rindex. */
707         fs_total = gfs2_ri_total(sdp);
708         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
709                 return;
710 
711         spin_lock(&sdp->sd_statfs_spin);
712         gfs2_statfs_change_in(m_sc, m_bh->b_data +
713                               sizeof(struct gfs2_dinode));
714         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
715                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
716         else
717                 new_free = 0;
718         spin_unlock(&sdp->sd_statfs_spin);
719         fs_warn(sdp, "File system extended by %llu blocks.\n",
720                 (unsigned long long)new_free);
721         gfs2_statfs_change(sdp, new_free, new_free, 0);
722 
723         if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
724                 goto out;
725         update_statfs(sdp, m_bh, l_bh);
726         brelse(l_bh);
727 out:
728         brelse(m_bh);
729 }
730 
731 /**
732  * gfs2_stuffed_write_end - Write end for stuffed files
733  * @inode: The inode
734  * @dibh: The buffer_head containing the on-disk inode
735  * @pos: The file position
736  * @len: The length of the write
737  * @copied: How much was actually copied by the VFS
738  * @page: The page
739  *
740  * This copies the data from the page into the inode block after
741  * the inode data structure itself.
742  *
743  * Returns: errno
744  */
745 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
746                                   loff_t pos, unsigned len, unsigned copied,
747                                   struct page *page)
748 {
749         struct gfs2_inode *ip = GFS2_I(inode);
750         struct gfs2_sbd *sdp = GFS2_SB(inode);
751         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
752         u64 to = pos + copied;
753         void *kaddr;
754         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
755 
756         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
757         kaddr = kmap_atomic(page);
758         memcpy(buf + pos, kaddr + pos, copied);
759         memset(kaddr + pos + copied, 0, len - copied);
760         flush_dcache_page(page);
761         kunmap_atomic(kaddr);
762 
763         if (!PageUptodate(page))
764                 SetPageUptodate(page);
765         unlock_page(page);
766         page_cache_release(page);
767 
768         if (copied) {
769                 if (inode->i_size < to)
770                         i_size_write(inode, to);
771                 mark_inode_dirty(inode);
772         }
773 
774         if (inode == sdp->sd_rindex) {
775                 adjust_fs_space(inode);
776                 sdp->sd_rindex_uptodate = 0;
777         }
778 
779         brelse(dibh);
780         gfs2_trans_end(sdp);
781         if (inode == sdp->sd_rindex) {
782                 gfs2_glock_dq(&m_ip->i_gh);
783                 gfs2_holder_uninit(&m_ip->i_gh);
784         }
785         gfs2_glock_dq(&ip->i_gh);
786         gfs2_holder_uninit(&ip->i_gh);
787         return copied;
788 }
789 
790 /**
791  * gfs2_write_end
792  * @file: The file to write to
793  * @mapping: The address space to write to
794  * @pos: The file position
795  * @len: The length of the data
796  * @copied:
797  * @page: The page that has been written
798  * @fsdata: The fsdata (unused in GFS2)
799  *
800  * The main write_end function for GFS2. We have a separate one for
801  * stuffed files as they are slightly different, otherwise we just
802  * put our locking around the VFS provided functions.
803  *
804  * Returns: errno
805  */
806 
807 static int gfs2_write_end(struct file *file, struct address_space *mapping,
808                           loff_t pos, unsigned len, unsigned copied,
809                           struct page *page, void *fsdata)
810 {
811         struct inode *inode = page->mapping->host;
812         struct gfs2_inode *ip = GFS2_I(inode);
813         struct gfs2_sbd *sdp = GFS2_SB(inode);
814         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
815         struct buffer_head *dibh;
816         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
817         unsigned int to = from + len;
818         int ret;
819         struct gfs2_trans *tr = current->journal_info;
820         BUG_ON(!tr);
821 
822         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
823 
824         ret = gfs2_meta_inode_buffer(ip, &dibh);
825         if (unlikely(ret)) {
826                 unlock_page(page);
827                 page_cache_release(page);
828                 goto failed;
829         }
830 
831         if (gfs2_is_stuffed(ip))
832                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
833 
834         if (!gfs2_is_writeback(ip))
835                 gfs2_page_add_databufs(ip, page, from, to);
836 
837         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
838         if (tr->tr_num_buf_new)
839                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
840         else
841                 gfs2_trans_add_meta(ip->i_gl, dibh);
842 
843 
844         if (inode == sdp->sd_rindex) {
845                 adjust_fs_space(inode);
846                 sdp->sd_rindex_uptodate = 0;
847         }
848 
849         brelse(dibh);
850 failed:
851         gfs2_trans_end(sdp);
852         gfs2_inplace_release(ip);
853         if (ip->i_res->rs_qa_qd_num)
854                 gfs2_quota_unlock(ip);
855         if (inode == sdp->sd_rindex) {
856                 gfs2_glock_dq(&m_ip->i_gh);
857                 gfs2_holder_uninit(&m_ip->i_gh);
858         }
859         gfs2_glock_dq(&ip->i_gh);
860         gfs2_holder_uninit(&ip->i_gh);
861         return ret;
862 }
863 
864 /**
865  * gfs2_set_page_dirty - Page dirtying function
866  * @page: The page to dirty
867  *
868  * Returns: 1 if it dirtyed the page, or 0 otherwise
869  */
870  
871 static int gfs2_set_page_dirty(struct page *page)
872 {
873         SetPageChecked(page);
874         return __set_page_dirty_buffers(page);
875 }
876 
877 /**
878  * gfs2_bmap - Block map function
879  * @mapping: Address space info
880  * @lblock: The block to map
881  *
882  * Returns: The disk address for the block or 0 on hole or error
883  */
884 
885 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
886 {
887         struct gfs2_inode *ip = GFS2_I(mapping->host);
888         struct gfs2_holder i_gh;
889         sector_t dblock = 0;
890         int error;
891 
892         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
893         if (error)
894                 return 0;
895 
896         if (!gfs2_is_stuffed(ip))
897                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
898 
899         gfs2_glock_dq_uninit(&i_gh);
900 
901         return dblock;
902 }
903 
904 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
905 {
906         struct gfs2_bufdata *bd;
907 
908         lock_buffer(bh);
909         gfs2_log_lock(sdp);
910         clear_buffer_dirty(bh);
911         bd = bh->b_private;
912         if (bd) {
913                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
914                         list_del_init(&bd->bd_list);
915                 else
916                         gfs2_remove_from_journal(bh, current->journal_info, 0);
917         }
918         bh->b_bdev = NULL;
919         clear_buffer_mapped(bh);
920         clear_buffer_req(bh);
921         clear_buffer_new(bh);
922         gfs2_log_unlock(sdp);
923         unlock_buffer(bh);
924 }
925 
926 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
927                                 unsigned int length)
928 {
929         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
930         unsigned int stop = offset + length;
931         int partial_page = (offset || length < PAGE_CACHE_SIZE);
932         struct buffer_head *bh, *head;
933         unsigned long pos = 0;
934 
935         BUG_ON(!PageLocked(page));
936         if (!partial_page)
937                 ClearPageChecked(page);
938         if (!page_has_buffers(page))
939                 goto out;
940 
941         bh = head = page_buffers(page);
942         do {
943                 if (pos + bh->b_size > stop)
944                         return;
945 
946                 if (offset <= pos)
947                         gfs2_discard(sdp, bh);
948                 pos += bh->b_size;
949                 bh = bh->b_this_page;
950         } while (bh != head);
951 out:
952         if (!partial_page)
953                 try_to_release_page(page, 0);
954 }
955 
956 /**
957  * gfs2_ok_for_dio - check that dio is valid on this file
958  * @ip: The inode
959  * @rw: READ or WRITE
960  * @offset: The offset at which we are reading or writing
961  *
962  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
963  *          1 (to accept the i/o request)
964  */
965 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
966 {
967         /*
968          * Should we return an error here? I can't see that O_DIRECT for
969          * a stuffed file makes any sense. For now we'll silently fall
970          * back to buffered I/O
971          */
972         if (gfs2_is_stuffed(ip))
973                 return 0;
974 
975         if (offset >= i_size_read(&ip->i_inode))
976                 return 0;
977         return 1;
978 }
979 
980 
981 
982 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
983                               const struct iovec *iov, loff_t offset,
984                               unsigned long nr_segs)
985 {
986         struct file *file = iocb->ki_filp;
987         struct inode *inode = file->f_mapping->host;
988         struct address_space *mapping = inode->i_mapping;
989         struct gfs2_inode *ip = GFS2_I(inode);
990         struct gfs2_holder gh;
991         int rv;
992 
993         /*
994          * Deferred lock, even if its a write, since we do no allocation
995          * on this path. All we need change is atime, and this lock mode
996          * ensures that other nodes have flushed their buffered read caches
997          * (i.e. their page cache entries for this inode). We do not,
998          * unfortunately have the option of only flushing a range like
999          * the VFS does.
1000          */
1001         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1002         rv = gfs2_glock_nq(&gh);
1003         if (rv)
1004                 return rv;
1005         rv = gfs2_ok_for_dio(ip, rw, offset);
1006         if (rv != 1)
1007                 goto out; /* dio not valid, fall back to buffered i/o */
1008 
1009         /*
1010          * Now since we are holding a deferred (CW) lock at this point, you
1011          * might be wondering why this is ever needed. There is a case however
1012          * where we've granted a deferred local lock against a cached exclusive
1013          * glock. That is ok provided all granted local locks are deferred, but
1014          * it also means that it is possible to encounter pages which are
1015          * cached and possibly also mapped. So here we check for that and sort
1016          * them out ahead of the dio. The glock state machine will take care of
1017          * everything else.
1018          *
1019          * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
1020          * the first place, mapping->nr_pages will always be zero.
1021          */
1022         if (mapping->nrpages) {
1023                 loff_t lstart = offset & (PAGE_CACHE_SIZE - 1);
1024                 loff_t len = iov_length(iov, nr_segs);
1025                 loff_t end = PAGE_ALIGN(offset + len) - 1;
1026 
1027                 rv = 0;
1028                 if (len == 0)
1029                         goto out;
1030                 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
1031                         unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
1032                 rv = filemap_write_and_wait_range(mapping, lstart, end);
1033                 if (rv)
1034                         goto out;
1035                 if (rw == WRITE)
1036                         truncate_inode_pages_range(mapping, lstart, end);
1037         }
1038 
1039         rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1040                                   offset, nr_segs, gfs2_get_block_direct,
1041                                   NULL, NULL, 0);
1042 out:
1043         gfs2_glock_dq(&gh);
1044         gfs2_holder_uninit(&gh);
1045         return rv;
1046 }
1047 
1048 /**
1049  * gfs2_releasepage - free the metadata associated with a page
1050  * @page: the page that's being released
1051  * @gfp_mask: passed from Linux VFS, ignored by us
1052  *
1053  * Call try_to_free_buffers() if the buffers in this page can be
1054  * released.
1055  *
1056  * Returns: 0
1057  */
1058 
1059 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1060 {
1061         struct address_space *mapping = page->mapping;
1062         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1063         struct buffer_head *bh, *head;
1064         struct gfs2_bufdata *bd;
1065 
1066         if (!page_has_buffers(page))
1067                 return 0;
1068 
1069         gfs2_log_lock(sdp);
1070         spin_lock(&sdp->sd_ail_lock);
1071         head = bh = page_buffers(page);
1072         do {
1073                 if (atomic_read(&bh->b_count))
1074                         goto cannot_release;
1075                 bd = bh->b_private;
1076                 if (bd && bd->bd_tr)
1077                         goto cannot_release;
1078                 if (buffer_pinned(bh) || buffer_dirty(bh))
1079                         goto not_possible;
1080                 bh = bh->b_this_page;
1081         } while(bh != head);
1082         spin_unlock(&sdp->sd_ail_lock);
1083 
1084         head = bh = page_buffers(page);
1085         do {
1086                 bd = bh->b_private;
1087                 if (bd) {
1088                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1089                         if (!list_empty(&bd->bd_list))
1090                                 list_del_init(&bd->bd_list);
1091                         bd->bd_bh = NULL;
1092                         bh->b_private = NULL;
1093                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1094                 }
1095 
1096                 bh = bh->b_this_page;
1097         } while (bh != head);
1098         gfs2_log_unlock(sdp);
1099 
1100         return try_to_free_buffers(page);
1101 
1102 not_possible: /* Should never happen */
1103         WARN_ON(buffer_dirty(bh));
1104         WARN_ON(buffer_pinned(bh));
1105 cannot_release:
1106         spin_unlock(&sdp->sd_ail_lock);
1107         gfs2_log_unlock(sdp);
1108         return 0;
1109 }
1110 
1111 static const struct address_space_operations gfs2_writeback_aops = {
1112         .writepage = gfs2_writepage,
1113         .writepages = gfs2_writepages,
1114         .readpage = gfs2_readpage,
1115         .readpages = gfs2_readpages,
1116         .write_begin = gfs2_write_begin,
1117         .write_end = gfs2_write_end,
1118         .bmap = gfs2_bmap,
1119         .invalidatepage = gfs2_invalidatepage,
1120         .releasepage = gfs2_releasepage,
1121         .direct_IO = gfs2_direct_IO,
1122         .migratepage = buffer_migrate_page,
1123         .is_partially_uptodate = block_is_partially_uptodate,
1124         .error_remove_page = generic_error_remove_page,
1125 };
1126 
1127 static const struct address_space_operations gfs2_ordered_aops = {
1128         .writepage = gfs2_writepage,
1129         .writepages = gfs2_writepages,
1130         .readpage = gfs2_readpage,
1131         .readpages = gfs2_readpages,
1132         .write_begin = gfs2_write_begin,
1133         .write_end = gfs2_write_end,
1134         .set_page_dirty = gfs2_set_page_dirty,
1135         .bmap = gfs2_bmap,
1136         .invalidatepage = gfs2_invalidatepage,
1137         .releasepage = gfs2_releasepage,
1138         .direct_IO = gfs2_direct_IO,
1139         .migratepage = buffer_migrate_page,
1140         .is_partially_uptodate = block_is_partially_uptodate,
1141         .error_remove_page = generic_error_remove_page,
1142 };
1143 
1144 static const struct address_space_operations gfs2_jdata_aops = {
1145         .writepage = gfs2_jdata_writepage,
1146         .writepages = gfs2_jdata_writepages,
1147         .readpage = gfs2_readpage,
1148         .readpages = gfs2_readpages,
1149         .write_begin = gfs2_write_begin,
1150         .write_end = gfs2_write_end,
1151         .set_page_dirty = gfs2_set_page_dirty,
1152         .bmap = gfs2_bmap,
1153         .invalidatepage = gfs2_invalidatepage,
1154         .releasepage = gfs2_releasepage,
1155         .is_partially_uptodate = block_is_partially_uptodate,
1156         .error_remove_page = generic_error_remove_page,
1157 };
1158 
1159 void gfs2_set_aops(struct inode *inode)
1160 {
1161         struct gfs2_inode *ip = GFS2_I(inode);
1162 
1163         if (gfs2_is_writeback(ip))
1164                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1165         else if (gfs2_is_ordered(ip))
1166                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1167         else if (gfs2_is_jdata(ip))
1168                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1169         else
1170                 BUG();
1171 }
1172 
1173 

~ [ 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