4ddab67867eb273553b6afbf0a42399f56c51735
[linux-2.6.git] / fs / gfs2 / ops_address.c
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/lm_interface.h>
23 #include <linux/backing-dev.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 "ops_address.h"
33 #include "quota.h"
34 #include "trans.h"
35 #include "rgrp.h"
36 #include "super.h"
37 #include "util.h"
38 #include "glops.h"
39
40
41 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
42                                    unsigned int from, unsigned int to)
43 {
44         struct buffer_head *head = page_buffers(page);
45         unsigned int bsize = head->b_size;
46         struct buffer_head *bh;
47         unsigned int start, end;
48
49         for (bh = head, start = 0; bh != head || !start;
50              bh = bh->b_this_page, start = end) {
51                 end = start + bsize;
52                 if (end <= from || start >= to)
53                         continue;
54                 if (gfs2_is_jdata(ip))
55                         set_buffer_uptodate(bh);
56                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
57         }
58 }
59
60 /**
61  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
62  * @inode: The inode
63  * @lblock: The block number to look up
64  * @bh_result: The buffer head to return the result in
65  * @create: Non-zero if we may add block to the file
66  *
67  * Returns: errno
68  */
69
70 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
71                                   struct buffer_head *bh_result, int create)
72 {
73         int error;
74
75         error = gfs2_block_map(inode, lblock, bh_result, 0);
76         if (error)
77                 return error;
78         if (!buffer_mapped(bh_result))
79                 return -EIO;
80         return 0;
81 }
82
83 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
84                                  struct buffer_head *bh_result, int create)
85 {
86         return gfs2_block_map(inode, lblock, bh_result, 0);
87 }
88
89 /**
90  * gfs2_writepage_common - Common bits of writepage
91  * @page: The page to be written
92  * @wbc: The writeback control
93  *
94  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
95  */
96
97 static int gfs2_writepage_common(struct page *page,
98                                  struct writeback_control *wbc)
99 {
100         struct inode *inode = page->mapping->host;
101         struct gfs2_inode *ip = GFS2_I(inode);
102         struct gfs2_sbd *sdp = GFS2_SB(inode);
103         loff_t i_size = i_size_read(inode);
104         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
105         unsigned offset;
106
107         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
108                 goto out;
109         if (current->journal_info)
110                 goto redirty;
111         /* Is the page fully outside i_size? (truncate in progress) */
112         offset = i_size & (PAGE_CACHE_SIZE-1);
113         if (page->index > end_index || (page->index == end_index && !offset)) {
114                 page->mapping->a_ops->invalidatepage(page, 0);
115                 goto out;
116         }
117         return 1;
118 redirty:
119         redirty_page_for_writepage(wbc, page);
120 out:
121         unlock_page(page);
122         return 0;
123 }
124
125 /**
126  * gfs2_writeback_writepage - Write page for writeback mappings
127  * @page: The page
128  * @wbc: The writeback control
129  *
130  */
131
132 static int gfs2_writeback_writepage(struct page *page,
133                                     struct writeback_control *wbc)
134 {
135         int ret;
136
137         ret = gfs2_writepage_common(page, wbc);
138         if (ret <= 0)
139                 return ret;
140
141         ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
142         if (ret == -EAGAIN)
143                 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
144         return ret;
145 }
146
147 /**
148  * gfs2_ordered_writepage - Write page for ordered data files
149  * @page: The page to write
150  * @wbc: The writeback control
151  *
152  */
153
154 static int gfs2_ordered_writepage(struct page *page,
155                                   struct writeback_control *wbc)
156 {
157         struct inode *inode = page->mapping->host;
158         struct gfs2_inode *ip = GFS2_I(inode);
159         int ret;
160
161         ret = gfs2_writepage_common(page, wbc);
162         if (ret <= 0)
163                 return ret;
164
165         if (!page_has_buffers(page)) {
166                 create_empty_buffers(page, inode->i_sb->s_blocksize,
167                                      (1 << BH_Dirty)|(1 << BH_Uptodate));
168         }
169         gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
170         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
171 }
172
173 /**
174  * __gfs2_jdata_writepage - The core of jdata writepage
175  * @page: The page to write
176  * @wbc: The writeback control
177  *
178  * This is shared between writepage and writepages and implements the
179  * core of the writepage operation. If a transaction is required then
180  * PageChecked will have been set and the transaction will have
181  * already been started before this is called.
182  */
183
184 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
185 {
186         struct inode *inode = page->mapping->host;
187         struct gfs2_inode *ip = GFS2_I(inode);
188         struct gfs2_sbd *sdp = GFS2_SB(inode);
189
190         if (PageChecked(page)) {
191                 ClearPageChecked(page);
192                 if (!page_has_buffers(page)) {
193                         create_empty_buffers(page, inode->i_sb->s_blocksize,
194                                              (1 << BH_Dirty)|(1 << BH_Uptodate));
195                 }
196                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
197         }
198         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
199 }
200
201 /**
202  * gfs2_jdata_writepage - Write complete page
203  * @page: Page to write
204  *
205  * Returns: errno
206  *
207  */
208
209 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
210 {
211         struct inode *inode = page->mapping->host;
212         struct gfs2_sbd *sdp = GFS2_SB(inode);
213         int ret;
214         int done_trans = 0;
215
216         if (PageChecked(page)) {
217                 if (wbc->sync_mode != WB_SYNC_ALL)
218                         goto out_ignore;
219                 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
220                 if (ret)
221                         goto out_ignore;
222                 done_trans = 1;
223         }
224         ret = gfs2_writepage_common(page, wbc);
225         if (ret > 0)
226                 ret = __gfs2_jdata_writepage(page, wbc);
227         if (done_trans)
228                 gfs2_trans_end(sdp);
229         return ret;
230
231 out_ignore:
232         redirty_page_for_writepage(wbc, page);
233         unlock_page(page);
234         return 0;
235 }
236
237 /**
238  * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
239  * @mapping: The mapping to write
240  * @wbc: Write-back control
241  *
242  * For the data=writeback case we can already ignore buffer heads
243  * and write whole extents at once. This is a big reduction in the
244  * number of I/O requests we send and the bmap calls we make in this case.
245  */
246 static int gfs2_writeback_writepages(struct address_space *mapping,
247                                      struct writeback_control *wbc)
248 {
249         return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
250 }
251
252 /**
253  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
254  * @mapping: The mapping
255  * @wbc: The writeback control
256  * @writepage: The writepage function to call for each page
257  * @pvec: The vector of pages
258  * @nr_pages: The number of pages to write
259  *
260  * Returns: non-zero if loop should terminate, zero otherwise
261  */
262
263 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
264                                     struct writeback_control *wbc,
265                                     struct pagevec *pvec,
266                                     int nr_pages, pgoff_t end)
267 {
268         struct inode *inode = mapping->host;
269         struct gfs2_sbd *sdp = GFS2_SB(inode);
270         loff_t i_size = i_size_read(inode);
271         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
272         unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
273         unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
274         struct backing_dev_info *bdi = mapping->backing_dev_info;
275         int i;
276         int ret;
277
278         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
279         if (ret < 0)
280                 return ret;
281
282         for(i = 0; i < nr_pages; i++) {
283                 struct page *page = pvec->pages[i];
284
285                 lock_page(page);
286
287                 if (unlikely(page->mapping != mapping)) {
288                         unlock_page(page);
289                         continue;
290                 }
291
292                 if (!wbc->range_cyclic && page->index > end) {
293                         ret = 1;
294                         unlock_page(page);
295                         continue;
296                 }
297
298                 if (wbc->sync_mode != WB_SYNC_NONE)
299                         wait_on_page_writeback(page);
300
301                 if (PageWriteback(page) ||
302                     !clear_page_dirty_for_io(page)) {
303                         unlock_page(page);
304                         continue;
305                 }
306
307                 /* Is the page fully outside i_size? (truncate in progress) */
308                 if (page->index > end_index || (page->index == end_index && !offset)) {
309                         page->mapping->a_ops->invalidatepage(page, 0);
310                         unlock_page(page);
311                         continue;
312                 }
313
314                 ret = __gfs2_jdata_writepage(page, wbc);
315
316                 if (ret || (--(wbc->nr_to_write) <= 0))
317                         ret = 1;
318                 if (wbc->nonblocking && bdi_write_congested(bdi)) {
319                         wbc->encountered_congestion = 1;
320                         ret = 1;
321                 }
322
323         }
324         gfs2_trans_end(sdp);
325         return ret;
326 }
327
328 /**
329  * gfs2_write_cache_jdata - Like write_cache_pages but different
330  * @mapping: The mapping to write
331  * @wbc: The writeback control
332  * @writepage: The writepage function to call
333  * @data: The data to pass to writepage
334  *
335  * The reason that we use our own function here is that we need to
336  * start transactions before we grab page locks. This allows us
337  * to get the ordering right.
338  */
339
340 static int gfs2_write_cache_jdata(struct address_space *mapping,
341                                   struct writeback_control *wbc)
342 {
343         struct backing_dev_info *bdi = mapping->backing_dev_info;
344         int ret = 0;
345         int done = 0;
346         struct pagevec pvec;
347         int nr_pages;
348         pgoff_t index;
349         pgoff_t end;
350         int scanned = 0;
351         int range_whole = 0;
352
353         if (wbc->nonblocking && bdi_write_congested(bdi)) {
354                 wbc->encountered_congestion = 1;
355                 return 0;
356         }
357
358         pagevec_init(&pvec, 0);
359         if (wbc->range_cyclic) {
360                 index = mapping->writeback_index; /* Start from prev offset */
361                 end = -1;
362         } else {
363                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
364                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
365                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
366                         range_whole = 1;
367                 scanned = 1;
368         }
369
370 retry:
371          while (!done && (index <= end) &&
372                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
373                                                PAGECACHE_TAG_DIRTY,
374                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
375                 scanned = 1;
376                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
377                 if (ret)
378                         done = 1;
379                 if (ret > 0)
380                         ret = 0;
381
382                 pagevec_release(&pvec);
383                 cond_resched();
384         }
385
386         if (!scanned && !done) {
387                 /*
388                  * We hit the last page and there is more work to be done: wrap
389                  * back to the start of the file
390                  */
391                 scanned = 1;
392                 index = 0;
393                 goto retry;
394         }
395
396         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
397                 mapping->writeback_index = index;
398         return ret;
399 }
400
401
402 /**
403  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
404  * @mapping: The mapping to write
405  * @wbc: The writeback control
406  * 
407  */
408
409 static int gfs2_jdata_writepages(struct address_space *mapping,
410                                  struct writeback_control *wbc)
411 {
412         struct gfs2_inode *ip = GFS2_I(mapping->host);
413         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
414         int ret;
415
416         ret = gfs2_write_cache_jdata(mapping, wbc);
417         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
418                 gfs2_log_flush(sdp, ip->i_gl);
419                 ret = gfs2_write_cache_jdata(mapping, wbc);
420         }
421         return ret;
422 }
423
424 /**
425  * stuffed_readpage - Fill in a Linux page with stuffed file data
426  * @ip: the inode
427  * @page: the page
428  *
429  * Returns: errno
430  */
431
432 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
433 {
434         struct buffer_head *dibh;
435         void *kaddr;
436         int error;
437
438         /*
439          * Due to the order of unstuffing files and ->fault(), we can be
440          * asked for a zero page in the case of a stuffed file being extended,
441          * so we need to supply one here. It doesn't happen often.
442          */
443         if (unlikely(page->index)) {
444                 zero_user(page, 0, PAGE_CACHE_SIZE);
445                 return 0;
446         }
447
448         error = gfs2_meta_inode_buffer(ip, &dibh);
449         if (error)
450                 return error;
451
452         kaddr = kmap_atomic(page, KM_USER0);
453         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
454                ip->i_disksize);
455         memset(kaddr + ip->i_disksize, 0, PAGE_CACHE_SIZE - ip->i_disksize);
456         kunmap_atomic(kaddr, KM_USER0);
457         flush_dcache_page(page);
458         brelse(dibh);
459         SetPageUptodate(page);
460
461         return 0;
462 }
463
464
465 /**
466  * __gfs2_readpage - readpage
467  * @file: The file to read a page for
468  * @page: The page to read
469  *
470  * This is the core of gfs2's readpage. Its used by the internal file
471  * reading code as in that case we already hold the glock. Also its
472  * called by gfs2_readpage() once the required lock has been granted.
473  *
474  */
475
476 static int __gfs2_readpage(void *file, struct page *page)
477 {
478         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
479         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
480         int error;
481
482         if (gfs2_is_stuffed(ip)) {
483                 error = stuffed_readpage(ip, page);
484                 unlock_page(page);
485         } else {
486                 error = mpage_readpage(page, gfs2_block_map);
487         }
488
489         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
490                 return -EIO;
491
492         return error;
493 }
494
495 /**
496  * gfs2_readpage - read a page of a file
497  * @file: The file to read
498  * @page: The page of the file
499  *
500  * This deals with the locking required. We have to unlock and
501  * relock the page in order to get the locking in the right
502  * order.
503  */
504
505 static int gfs2_readpage(struct file *file, struct page *page)
506 {
507         struct address_space *mapping = page->mapping;
508         struct gfs2_inode *ip = GFS2_I(mapping->host);
509         struct gfs2_holder gh;
510         int error;
511
512         unlock_page(page);
513         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
514         error = gfs2_glock_nq(&gh);
515         if (unlikely(error))
516                 goto out;
517         error = AOP_TRUNCATED_PAGE;
518         lock_page(page);
519         if (page->mapping == mapping && !PageUptodate(page))
520                 error = __gfs2_readpage(file, page);
521         else
522                 unlock_page(page);
523         gfs2_glock_dq(&gh);
524 out:
525         gfs2_holder_uninit(&gh);
526         if (error && error != AOP_TRUNCATED_PAGE)
527                 lock_page(page);
528         return error;
529 }
530
531 /**
532  * gfs2_internal_read - read an internal file
533  * @ip: The gfs2 inode
534  * @ra_state: The readahead state (or NULL for no readahead)
535  * @buf: The buffer to fill
536  * @pos: The file position
537  * @size: The amount to read
538  *
539  */
540
541 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
542                        char *buf, loff_t *pos, unsigned size)
543 {
544         struct address_space *mapping = ip->i_inode.i_mapping;
545         unsigned long index = *pos / PAGE_CACHE_SIZE;
546         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
547         unsigned copied = 0;
548         unsigned amt;
549         struct page *page;
550         void *p;
551
552         do {
553                 amt = size - copied;
554                 if (offset + size > PAGE_CACHE_SIZE)
555                         amt = PAGE_CACHE_SIZE - offset;
556                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
557                 if (IS_ERR(page))
558                         return PTR_ERR(page);
559                 p = kmap_atomic(page, KM_USER0);
560                 memcpy(buf + copied, p + offset, amt);
561                 kunmap_atomic(p, KM_USER0);
562                 mark_page_accessed(page);
563                 page_cache_release(page);
564                 copied += amt;
565                 index++;
566                 offset = 0;
567         } while(copied < size);
568         (*pos) += size;
569         return size;
570 }
571
572 /**
573  * gfs2_readpages - Read a bunch of pages at once
574  *
575  * Some notes:
576  * 1. This is only for readahead, so we can simply ignore any things
577  *    which are slightly inconvenient (such as locking conflicts between
578  *    the page lock and the glock) and return having done no I/O. Its
579  *    obviously not something we'd want to do on too regular a basis.
580  *    Any I/O we ignore at this time will be done via readpage later.
581  * 2. We don't handle stuffed files here we let readpage do the honours.
582  * 3. mpage_readpages() does most of the heavy lifting in the common case.
583  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
584  */
585
586 static int gfs2_readpages(struct file *file, struct address_space *mapping,
587                           struct list_head *pages, unsigned nr_pages)
588 {
589         struct inode *inode = mapping->host;
590         struct gfs2_inode *ip = GFS2_I(inode);
591         struct gfs2_sbd *sdp = GFS2_SB(inode);
592         struct gfs2_holder gh;
593         int ret;
594
595         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
596         ret = gfs2_glock_nq(&gh);
597         if (unlikely(ret))
598                 goto out_uninit;
599         if (!gfs2_is_stuffed(ip))
600                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
601         gfs2_glock_dq(&gh);
602 out_uninit:
603         gfs2_holder_uninit(&gh);
604         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
605                 ret = -EIO;
606         return ret;
607 }
608
609 /**
610  * gfs2_write_begin - Begin to write to a file
611  * @file: The file to write to
612  * @mapping: The mapping in which to write
613  * @pos: The file offset at which to start writing
614  * @len: Length of the write
615  * @flags: Various flags
616  * @pagep: Pointer to return the page
617  * @fsdata: Pointer to return fs data (unused by GFS2)
618  *
619  * Returns: errno
620  */
621
622 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
623                             loff_t pos, unsigned len, unsigned flags,
624                             struct page **pagep, void **fsdata)
625 {
626         struct gfs2_inode *ip = GFS2_I(mapping->host);
627         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
628         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
629         int alloc_required;
630         int error = 0;
631         struct gfs2_alloc *al;
632         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
633         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
634         unsigned to = from + len;
635         struct page *page;
636
637         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
638         error = gfs2_glock_nq(&ip->i_gh);
639         if (unlikely(error))
640                 goto out_uninit;
641
642         error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
643         if (error)
644                 goto out_unlock;
645
646         if (alloc_required || gfs2_is_jdata(ip))
647                 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
648
649         if (alloc_required) {
650                 al = gfs2_alloc_get(ip);
651                 if (!al) {
652                         error = -ENOMEM;
653                         goto out_unlock;
654                 }
655
656                 error = gfs2_quota_lock_check(ip);
657                 if (error)
658                         goto out_alloc_put;
659
660                 al->al_requested = data_blocks + ind_blocks;
661                 error = gfs2_inplace_reserve(ip);
662                 if (error)
663                         goto out_qunlock;
664         }
665
666         rblocks = RES_DINODE + ind_blocks;
667         if (gfs2_is_jdata(ip))
668                 rblocks += data_blocks ? data_blocks : 1;
669         if (ind_blocks || data_blocks)
670                 rblocks += RES_STATFS + RES_QUOTA;
671
672         error = gfs2_trans_begin(sdp, rblocks,
673                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
674         if (error)
675                 goto out_trans_fail;
676
677         error = -ENOMEM;
678         flags |= AOP_FLAG_NOFS;
679         page = grab_cache_page_write_begin(mapping, index, flags);
680         *pagep = page;
681         if (unlikely(!page))
682                 goto out_endtrans;
683
684         if (gfs2_is_stuffed(ip)) {
685                 error = 0;
686                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
687                         error = gfs2_unstuff_dinode(ip, page);
688                         if (error == 0)
689                                 goto prepare_write;
690                 } else if (!PageUptodate(page)) {
691                         error = stuffed_readpage(ip, page);
692                 }
693                 goto out;
694         }
695
696 prepare_write:
697         error = block_prepare_write(page, from, to, gfs2_block_map);
698 out:
699         if (error == 0)
700                 return 0;
701
702         page_cache_release(page);
703         if (pos + len > ip->i_inode.i_size)
704                 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
705 out_endtrans:
706         gfs2_trans_end(sdp);
707 out_trans_fail:
708         if (alloc_required) {
709                 gfs2_inplace_release(ip);
710 out_qunlock:
711                 gfs2_quota_unlock(ip);
712 out_alloc_put:
713                 gfs2_alloc_put(ip);
714         }
715 out_unlock:
716         gfs2_glock_dq(&ip->i_gh);
717 out_uninit:
718         gfs2_holder_uninit(&ip->i_gh);
719         return error;
720 }
721
722 /**
723  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
724  * @inode: the rindex inode
725  */
726 static void adjust_fs_space(struct inode *inode)
727 {
728         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
729         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
730         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
731         u64 fs_total, new_free;
732
733         /* Total up the file system space, according to the latest rindex. */
734         fs_total = gfs2_ri_total(sdp);
735
736         spin_lock(&sdp->sd_statfs_spin);
737         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
738                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
739         else
740                 new_free = 0;
741         spin_unlock(&sdp->sd_statfs_spin);
742         fs_warn(sdp, "File system extended by %llu blocks.\n",
743                 (unsigned long long)new_free);
744         gfs2_statfs_change(sdp, new_free, new_free, 0);
745 }
746
747 /**
748  * gfs2_stuffed_write_end - Write end for stuffed files
749  * @inode: The inode
750  * @dibh: The buffer_head containing the on-disk inode
751  * @pos: The file position
752  * @len: The length of the write
753  * @copied: How much was actually copied by the VFS
754  * @page: The page
755  *
756  * This copies the data from the page into the inode block after
757  * the inode data structure itself.
758  *
759  * Returns: errno
760  */
761 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
762                                   loff_t pos, unsigned len, unsigned copied,
763                                   struct page *page)
764 {
765         struct gfs2_inode *ip = GFS2_I(inode);
766         struct gfs2_sbd *sdp = GFS2_SB(inode);
767         u64 to = pos + copied;
768         void *kaddr;
769         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
770         struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
771
772         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
773         kaddr = kmap_atomic(page, KM_USER0);
774         memcpy(buf + pos, kaddr + pos, copied);
775         memset(kaddr + pos + copied, 0, len - copied);
776         flush_dcache_page(page);
777         kunmap_atomic(kaddr, KM_USER0);
778
779         if (!PageUptodate(page))
780                 SetPageUptodate(page);
781         unlock_page(page);
782         page_cache_release(page);
783
784         if (inode->i_size < to) {
785                 i_size_write(inode, to);
786                 ip->i_disksize = inode->i_size;
787                 di->di_size = cpu_to_be64(inode->i_size);
788                 mark_inode_dirty(inode);
789         }
790
791         if (inode == sdp->sd_rindex)
792                 adjust_fs_space(inode);
793
794         brelse(dibh);
795         gfs2_trans_end(sdp);
796         gfs2_glock_dq(&ip->i_gh);
797         gfs2_holder_uninit(&ip->i_gh);
798         return copied;
799 }
800
801 /**
802  * gfs2_write_end
803  * @file: The file to write to
804  * @mapping: The address space to write to
805  * @pos: The file position
806  * @len: The length of the data
807  * @copied:
808  * @page: The page that has been written
809  * @fsdata: The fsdata (unused in GFS2)
810  *
811  * The main write_end function for GFS2. We have a separate one for
812  * stuffed files as they are slightly different, otherwise we just
813  * put our locking around the VFS provided functions.
814  *
815  * Returns: errno
816  */
817
818 static int gfs2_write_end(struct file *file, struct address_space *mapping,
819                           loff_t pos, unsigned len, unsigned copied,
820                           struct page *page, void *fsdata)
821 {
822         struct inode *inode = page->mapping->host;
823         struct gfs2_inode *ip = GFS2_I(inode);
824         struct gfs2_sbd *sdp = GFS2_SB(inode);
825         struct buffer_head *dibh;
826         struct gfs2_alloc *al = ip->i_alloc;
827         struct gfs2_dinode *di;
828         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
829         unsigned int to = from + len;
830         int ret;
831
832         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
833
834         ret = gfs2_meta_inode_buffer(ip, &dibh);
835         if (unlikely(ret)) {
836                 unlock_page(page);
837                 page_cache_release(page);
838                 goto failed;
839         }
840
841         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
842
843         if (gfs2_is_stuffed(ip))
844                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
845
846         if (!gfs2_is_writeback(ip))
847                 gfs2_page_add_databufs(ip, page, from, to);
848
849         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
850
851         if (likely(ret >= 0) && (inode->i_size > ip->i_disksize)) {
852                 di = (struct gfs2_dinode *)dibh->b_data;
853                 ip->i_disksize = inode->i_size;
854                 di->di_size = cpu_to_be64(inode->i_size);
855                 mark_inode_dirty(inode);
856         }
857
858         if (inode == sdp->sd_rindex)
859                 adjust_fs_space(inode);
860
861         brelse(dibh);
862         gfs2_trans_end(sdp);
863 failed:
864         if (al) {
865                 gfs2_inplace_release(ip);
866                 gfs2_quota_unlock(ip);
867                 gfs2_alloc_put(ip);
868         }
869         gfs2_glock_dq(&ip->i_gh);
870         gfs2_holder_uninit(&ip->i_gh);
871         return ret;
872 }
873
874 /**
875  * gfs2_set_page_dirty - Page dirtying function
876  * @page: The page to dirty
877  *
878  * Returns: 1 if it dirtyed the page, or 0 otherwise
879  */
880  
881 static int gfs2_set_page_dirty(struct page *page)
882 {
883         SetPageChecked(page);
884         return __set_page_dirty_buffers(page);
885 }
886
887 /**
888  * gfs2_bmap - Block map function
889  * @mapping: Address space info
890  * @lblock: The block to map
891  *
892  * Returns: The disk address for the block or 0 on hole or error
893  */
894
895 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
896 {
897         struct gfs2_inode *ip = GFS2_I(mapping->host);
898         struct gfs2_holder i_gh;
899         sector_t dblock = 0;
900         int error;
901
902         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
903         if (error)
904                 return 0;
905
906         if (!gfs2_is_stuffed(ip))
907                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
908
909         gfs2_glock_dq_uninit(&i_gh);
910
911         return dblock;
912 }
913
914 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
915 {
916         struct gfs2_bufdata *bd;
917
918         lock_buffer(bh);
919         gfs2_log_lock(sdp);
920         clear_buffer_dirty(bh);
921         bd = bh->b_private;
922         if (bd) {
923                 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
924                         list_del_init(&bd->bd_le.le_list);
925                 else
926                         gfs2_remove_from_journal(bh, current->journal_info, 0);
927         }
928         bh->b_bdev = NULL;
929         clear_buffer_mapped(bh);
930         clear_buffer_req(bh);
931         clear_buffer_new(bh);
932         gfs2_log_unlock(sdp);
933         unlock_buffer(bh);
934 }
935
936 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
937 {
938         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
939         struct buffer_head *bh, *head;
940         unsigned long pos = 0;
941
942         BUG_ON(!PageLocked(page));
943         if (offset == 0)
944                 ClearPageChecked(page);
945         if (!page_has_buffers(page))
946                 goto out;
947
948         bh = head = page_buffers(page);
949         do {
950                 if (offset <= pos)
951                         gfs2_discard(sdp, bh);
952                 pos += bh->b_size;
953                 bh = bh->b_this_page;
954         } while (bh != head);
955 out:
956         if (offset == 0)
957                 try_to_release_page(page, 0);
958 }
959
960 /**
961  * gfs2_ok_for_dio - check that dio is valid on this file
962  * @ip: The inode
963  * @rw: READ or WRITE
964  * @offset: The offset at which we are reading or writing
965  *
966  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
967  *          1 (to accept the i/o request)
968  */
969 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
970 {
971         /*
972          * Should we return an error here? I can't see that O_DIRECT for
973          * a stuffed file makes any sense. For now we'll silently fall
974          * back to buffered I/O
975          */
976         if (gfs2_is_stuffed(ip))
977                 return 0;
978
979         if (offset >= i_size_read(&ip->i_inode))
980                 return 0;
981         return 1;
982 }
983
984
985
986 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
987                               const struct iovec *iov, loff_t offset,
988                               unsigned long nr_segs)
989 {
990         struct file *file = iocb->ki_filp;
991         struct inode *inode = file->f_mapping->host;
992         struct gfs2_inode *ip = GFS2_I(inode);
993         struct gfs2_holder gh;
994         int rv;
995
996         /*
997          * Deferred lock, even if its a write, since we do no allocation
998          * on this path. All we need change is atime, and this lock mode
999          * ensures that other nodes have flushed their buffered read caches
1000          * (i.e. their page cache entries for this inode). We do not,
1001          * unfortunately have the option of only flushing a range like
1002          * the VFS does.
1003          */
1004         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1005         rv = gfs2_glock_nq(&gh);
1006         if (rv)
1007                 return rv;
1008         rv = gfs2_ok_for_dio(ip, rw, offset);
1009         if (rv != 1)
1010                 goto out; /* dio not valid, fall back to buffered i/o */
1011
1012         rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1013                                            iov, offset, nr_segs,
1014                                            gfs2_get_block_direct, NULL);
1015 out:
1016         gfs2_glock_dq_m(1, &gh);
1017         gfs2_holder_uninit(&gh);
1018         return rv;
1019 }
1020
1021 /**
1022  * gfs2_releasepage - free the metadata associated with a page
1023  * @page: the page that's being released
1024  * @gfp_mask: passed from Linux VFS, ignored by us
1025  *
1026  * Call try_to_free_buffers() if the buffers in this page can be
1027  * released.
1028  *
1029  * Returns: 0
1030  */
1031
1032 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1033 {
1034         struct inode *aspace = page->mapping->host;
1035         struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1036         struct buffer_head *bh, *head;
1037         struct gfs2_bufdata *bd;
1038
1039         if (!page_has_buffers(page))
1040                 return 0;
1041
1042         gfs2_log_lock(sdp);
1043         head = bh = page_buffers(page);
1044         do {
1045                 if (atomic_read(&bh->b_count))
1046                         goto cannot_release;
1047                 bd = bh->b_private;
1048                 if (bd && bd->bd_ail)
1049                         goto cannot_release;
1050                 gfs2_assert_warn(sdp, !buffer_pinned(bh));
1051                 gfs2_assert_warn(sdp, !buffer_dirty(bh));
1052                 bh = bh->b_this_page;
1053         } while(bh != head);
1054         gfs2_log_unlock(sdp);
1055
1056         head = bh = page_buffers(page);
1057         do {
1058                 gfs2_log_lock(sdp);
1059                 bd = bh->b_private;
1060                 if (bd) {
1061                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1062                         gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
1063                         if (!list_empty(&bd->bd_le.le_list)) {
1064                                 if (!buffer_pinned(bh))
1065                                         list_del_init(&bd->bd_le.le_list);
1066                                 else
1067                                         bd = NULL;
1068                         }
1069                         if (bd)
1070                                 bd->bd_bh = NULL;
1071                         bh->b_private = NULL;
1072                 }
1073                 gfs2_log_unlock(sdp);
1074                 if (bd)
1075                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1076
1077                 bh = bh->b_this_page;
1078         } while (bh != head);
1079
1080         return try_to_free_buffers(page);
1081 cannot_release:
1082         gfs2_log_unlock(sdp);
1083         return 0;
1084 }
1085
1086 static const struct address_space_operations gfs2_writeback_aops = {
1087         .writepage = gfs2_writeback_writepage,
1088         .writepages = gfs2_writeback_writepages,
1089         .readpage = gfs2_readpage,
1090         .readpages = gfs2_readpages,
1091         .sync_page = block_sync_page,
1092         .write_begin = gfs2_write_begin,
1093         .write_end = gfs2_write_end,
1094         .bmap = gfs2_bmap,
1095         .invalidatepage = gfs2_invalidatepage,
1096         .releasepage = gfs2_releasepage,
1097         .direct_IO = gfs2_direct_IO,
1098         .migratepage = buffer_migrate_page,
1099 };
1100
1101 static const struct address_space_operations gfs2_ordered_aops = {
1102         .writepage = gfs2_ordered_writepage,
1103         .readpage = gfs2_readpage,
1104         .readpages = gfs2_readpages,
1105         .sync_page = block_sync_page,
1106         .write_begin = gfs2_write_begin,
1107         .write_end = gfs2_write_end,
1108         .set_page_dirty = gfs2_set_page_dirty,
1109         .bmap = gfs2_bmap,
1110         .invalidatepage = gfs2_invalidatepage,
1111         .releasepage = gfs2_releasepage,
1112         .direct_IO = gfs2_direct_IO,
1113         .migratepage = buffer_migrate_page,
1114 };
1115
1116 static const struct address_space_operations gfs2_jdata_aops = {
1117         .writepage = gfs2_jdata_writepage,
1118         .writepages = gfs2_jdata_writepages,
1119         .readpage = gfs2_readpage,
1120         .readpages = gfs2_readpages,
1121         .sync_page = block_sync_page,
1122         .write_begin = gfs2_write_begin,
1123         .write_end = gfs2_write_end,
1124         .set_page_dirty = gfs2_set_page_dirty,
1125         .bmap = gfs2_bmap,
1126         .invalidatepage = gfs2_invalidatepage,
1127         .releasepage = gfs2_releasepage,
1128 };
1129
1130 void gfs2_set_aops(struct inode *inode)
1131 {
1132         struct gfs2_inode *ip = GFS2_I(inode);
1133
1134         if (gfs2_is_writeback(ip))
1135                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1136         else if (gfs2_is_ordered(ip))
1137                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1138         else if (gfs2_is_jdata(ip))
1139                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1140         else
1141                 BUG();
1142 }
1143