Merge tag 'for-linus-v3.9-rc1' of git://oss.sgi.com/xfs/xfs
[~shefty/rdma-dev.git] / fs / gfs2 / aops.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/backing-dev.h>
23
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "bmap.h"
27 #include "glock.h"
28 #include "inode.h"
29 #include "log.h"
30 #include "meta_io.h"
31 #include "quota.h"
32 #include "trans.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "util.h"
36 #include "glops.h"
37
38
39 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
40                                    unsigned int from, unsigned int to)
41 {
42         struct buffer_head *head = page_buffers(page);
43         unsigned int bsize = head->b_size;
44         struct buffer_head *bh;
45         unsigned int start, end;
46
47         for (bh = head, start = 0; bh != head || !start;
48              bh = bh->b_this_page, start = end) {
49                 end = start + bsize;
50                 if (end <= from || start >= to)
51                         continue;
52                 if (gfs2_is_jdata(ip))
53                         set_buffer_uptodate(bh);
54                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
55         }
56 }
57
58 /**
59  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
60  * @inode: The inode
61  * @lblock: The block number to look up
62  * @bh_result: The buffer head to return the result in
63  * @create: Non-zero if we may add block to the file
64  *
65  * Returns: errno
66  */
67
68 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
69                                   struct buffer_head *bh_result, int create)
70 {
71         int error;
72
73         error = gfs2_block_map(inode, lblock, bh_result, 0);
74         if (error)
75                 return error;
76         if (!buffer_mapped(bh_result))
77                 return -EIO;
78         return 0;
79 }
80
81 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
82                                  struct buffer_head *bh_result, int create)
83 {
84         return gfs2_block_map(inode, lblock, bh_result, 0);
85 }
86
87 /**
88  * gfs2_writepage_common - Common bits of writepage
89  * @page: The page to be written
90  * @wbc: The writeback control
91  *
92  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
93  */
94
95 static int gfs2_writepage_common(struct page *page,
96                                  struct writeback_control *wbc)
97 {
98         struct inode *inode = page->mapping->host;
99         struct gfs2_inode *ip = GFS2_I(inode);
100         struct gfs2_sbd *sdp = GFS2_SB(inode);
101         loff_t i_size = i_size_read(inode);
102         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
103         unsigned offset;
104
105         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
106                 goto out;
107         if (current->journal_info)
108                 goto redirty;
109         /* Is the page fully outside i_size? (truncate in progress) */
110         offset = i_size & (PAGE_CACHE_SIZE-1);
111         if (page->index > end_index || (page->index == end_index && !offset)) {
112                 page->mapping->a_ops->invalidatepage(page, 0);
113                 goto out;
114         }
115         return 1;
116 redirty:
117         redirty_page_for_writepage(wbc, page);
118 out:
119         unlock_page(page);
120         return 0;
121 }
122
123 /**
124  * gfs2_writeback_writepage - Write page for writeback mappings
125  * @page: The page
126  * @wbc: The writeback control
127  *
128  */
129
130 static int gfs2_writeback_writepage(struct page *page,
131                                     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_ordered_writepage - Write page for ordered data files
144  * @page: The page to write
145  * @wbc: The writeback control
146  *
147  */
148
149 static int gfs2_ordered_writepage(struct page *page,
150                                   struct writeback_control *wbc)
151 {
152         struct inode *inode = page->mapping->host;
153         struct gfs2_inode *ip = GFS2_I(inode);
154         int ret;
155
156         ret = gfs2_writepage_common(page, wbc);
157         if (ret <= 0)
158                 return ret;
159
160         if (!page_has_buffers(page)) {
161                 create_empty_buffers(page, inode->i_sb->s_blocksize,
162                                      (1 << BH_Dirty)|(1 << BH_Uptodate));
163         }
164         gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
165         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
166 }
167
168 /**
169  * __gfs2_jdata_writepage - The core of jdata writepage
170  * @page: The page to write
171  * @wbc: The writeback control
172  *
173  * This is shared between writepage and writepages and implements the
174  * core of the writepage operation. If a transaction is required then
175  * PageChecked will have been set and the transaction will have
176  * already been started before this is called.
177  */
178
179 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
180 {
181         struct inode *inode = page->mapping->host;
182         struct gfs2_inode *ip = GFS2_I(inode);
183         struct gfs2_sbd *sdp = GFS2_SB(inode);
184
185         if (PageChecked(page)) {
186                 ClearPageChecked(page);
187                 if (!page_has_buffers(page)) {
188                         create_empty_buffers(page, inode->i_sb->s_blocksize,
189                                              (1 << BH_Dirty)|(1 << BH_Uptodate));
190                 }
191                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
192         }
193         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
194 }
195
196 /**
197  * gfs2_jdata_writepage - Write complete page
198  * @page: Page to write
199  *
200  * Returns: errno
201  *
202  */
203
204 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
205 {
206         struct inode *inode = page->mapping->host;
207         struct gfs2_sbd *sdp = GFS2_SB(inode);
208         int ret;
209         int done_trans = 0;
210
211         if (PageChecked(page)) {
212                 if (wbc->sync_mode != WB_SYNC_ALL)
213                         goto out_ignore;
214                 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
215                 if (ret)
216                         goto out_ignore;
217                 done_trans = 1;
218         }
219         ret = gfs2_writepage_common(page, wbc);
220         if (ret > 0)
221                 ret = __gfs2_jdata_writepage(page, wbc);
222         if (done_trans)
223                 gfs2_trans_end(sdp);
224         return ret;
225
226 out_ignore:
227         redirty_page_for_writepage(wbc, page);
228         unlock_page(page);
229         return 0;
230 }
231
232 /**
233  * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
234  * @mapping: The mapping to write
235  * @wbc: Write-back control
236  *
237  * For the data=writeback case we can already ignore buffer heads
238  * and write whole extents at once. This is a big reduction in the
239  * number of I/O requests we send and the bmap calls we make in this case.
240  */
241 static int gfs2_writeback_writepages(struct address_space *mapping,
242                                      struct writeback_control *wbc)
243 {
244         return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
245 }
246
247 /**
248  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
249  * @mapping: The mapping
250  * @wbc: The writeback control
251  * @writepage: The writepage function to call for each page
252  * @pvec: The vector of pages
253  * @nr_pages: The number of pages to write
254  *
255  * Returns: non-zero if loop should terminate, zero otherwise
256  */
257
258 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
259                                     struct writeback_control *wbc,
260                                     struct pagevec *pvec,
261                                     int nr_pages, pgoff_t end)
262 {
263         struct inode *inode = mapping->host;
264         struct gfs2_sbd *sdp = GFS2_SB(inode);
265         loff_t i_size = i_size_read(inode);
266         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
267         unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
268         unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
269         int i;
270         int ret;
271
272         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
273         if (ret < 0)
274                 return ret;
275
276         for(i = 0; i < nr_pages; i++) {
277                 struct page *page = pvec->pages[i];
278
279                 lock_page(page);
280
281                 if (unlikely(page->mapping != mapping)) {
282                         unlock_page(page);
283                         continue;
284                 }
285
286                 if (!wbc->range_cyclic && page->index > end) {
287                         ret = 1;
288                         unlock_page(page);
289                         continue;
290                 }
291
292                 if (wbc->sync_mode != WB_SYNC_NONE)
293                         wait_on_page_writeback(page);
294
295                 if (PageWriteback(page) ||
296                     !clear_page_dirty_for_io(page)) {
297                         unlock_page(page);
298                         continue;
299                 }
300
301                 /* Is the page fully outside i_size? (truncate in progress) */
302                 if (page->index > end_index || (page->index == end_index && !offset)) {
303                         page->mapping->a_ops->invalidatepage(page, 0);
304                         unlock_page(page);
305                         continue;
306                 }
307
308                 ret = __gfs2_jdata_writepage(page, wbc);
309
310                 if (ret || (--(wbc->nr_to_write) <= 0))
311                         ret = 1;
312         }
313         gfs2_trans_end(sdp);
314         return ret;
315 }
316
317 /**
318  * gfs2_write_cache_jdata - Like write_cache_pages but different
319  * @mapping: The mapping to write
320  * @wbc: The writeback control
321  * @writepage: The writepage function to call
322  * @data: The data to pass to writepage
323  *
324  * The reason that we use our own function here is that we need to
325  * start transactions before we grab page locks. This allows us
326  * to get the ordering right.
327  */
328
329 static int gfs2_write_cache_jdata(struct address_space *mapping,
330                                   struct writeback_control *wbc)
331 {
332         int ret = 0;
333         int done = 0;
334         struct pagevec pvec;
335         int nr_pages;
336         pgoff_t index;
337         pgoff_t end;
338         int scanned = 0;
339         int range_whole = 0;
340
341         pagevec_init(&pvec, 0);
342         if (wbc->range_cyclic) {
343                 index = mapping->writeback_index; /* Start from prev offset */
344                 end = -1;
345         } else {
346                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
347                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
348                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
349                         range_whole = 1;
350                 scanned = 1;
351         }
352
353 retry:
354          while (!done && (index <= end) &&
355                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
356                                                PAGECACHE_TAG_DIRTY,
357                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
358                 scanned = 1;
359                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
360                 if (ret)
361                         done = 1;
362                 if (ret > 0)
363                         ret = 0;
364
365                 pagevec_release(&pvec);
366                 cond_resched();
367         }
368
369         if (!scanned && !done) {
370                 /*
371                  * We hit the last page and there is more work to be done: wrap
372                  * back to the start of the file
373                  */
374                 scanned = 1;
375                 index = 0;
376                 goto retry;
377         }
378
379         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
380                 mapping->writeback_index = index;
381         return ret;
382 }
383
384
385 /**
386  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
387  * @mapping: The mapping to write
388  * @wbc: The writeback control
389  * 
390  */
391
392 static int gfs2_jdata_writepages(struct address_space *mapping,
393                                  struct writeback_control *wbc)
394 {
395         struct gfs2_inode *ip = GFS2_I(mapping->host);
396         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
397         int ret;
398
399         ret = gfs2_write_cache_jdata(mapping, wbc);
400         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
401                 gfs2_log_flush(sdp, ip->i_gl);
402                 ret = gfs2_write_cache_jdata(mapping, wbc);
403         }
404         return ret;
405 }
406
407 /**
408  * stuffed_readpage - Fill in a Linux page with stuffed file data
409  * @ip: the inode
410  * @page: the page
411  *
412  * Returns: errno
413  */
414
415 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
416 {
417         struct buffer_head *dibh;
418         u64 dsize = i_size_read(&ip->i_inode);
419         void *kaddr;
420         int error;
421
422         /*
423          * Due to the order of unstuffing files and ->fault(), we can be
424          * asked for a zero page in the case of a stuffed file being extended,
425          * so we need to supply one here. It doesn't happen often.
426          */
427         if (unlikely(page->index)) {
428                 zero_user(page, 0, PAGE_CACHE_SIZE);
429                 SetPageUptodate(page);
430                 return 0;
431         }
432
433         error = gfs2_meta_inode_buffer(ip, &dibh);
434         if (error)
435                 return error;
436
437         kaddr = kmap_atomic(page);
438         if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
439                 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
440         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
441         memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
442         kunmap_atomic(kaddr);
443         flush_dcache_page(page);
444         brelse(dibh);
445         SetPageUptodate(page);
446
447         return 0;
448 }
449
450
451 /**
452  * __gfs2_readpage - readpage
453  * @file: The file to read a page for
454  * @page: The page to read
455  *
456  * This is the core of gfs2's readpage. Its used by the internal file
457  * reading code as in that case we already hold the glock. Also its
458  * called by gfs2_readpage() once the required lock has been granted.
459  *
460  */
461
462 static int __gfs2_readpage(void *file, struct page *page)
463 {
464         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
465         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
466         int error;
467
468         if (gfs2_is_stuffed(ip)) {
469                 error = stuffed_readpage(ip, page);
470                 unlock_page(page);
471         } else {
472                 error = mpage_readpage(page, gfs2_block_map);
473         }
474
475         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
476                 return -EIO;
477
478         return error;
479 }
480
481 /**
482  * gfs2_readpage - read a page of a file
483  * @file: The file to read
484  * @page: The page of the file
485  *
486  * This deals with the locking required. We have to unlock and
487  * relock the page in order to get the locking in the right
488  * order.
489  */
490
491 static int gfs2_readpage(struct file *file, struct page *page)
492 {
493         struct address_space *mapping = page->mapping;
494         struct gfs2_inode *ip = GFS2_I(mapping->host);
495         struct gfs2_holder gh;
496         int error;
497
498         unlock_page(page);
499         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
500         error = gfs2_glock_nq(&gh);
501         if (unlikely(error))
502                 goto out;
503         error = AOP_TRUNCATED_PAGE;
504         lock_page(page);
505         if (page->mapping == mapping && !PageUptodate(page))
506                 error = __gfs2_readpage(file, page);
507         else
508                 unlock_page(page);
509         gfs2_glock_dq(&gh);
510 out:
511         gfs2_holder_uninit(&gh);
512         if (error && error != AOP_TRUNCATED_PAGE)
513                 lock_page(page);
514         return error;
515 }
516
517 /**
518  * gfs2_internal_read - read an internal file
519  * @ip: The gfs2 inode
520  * @buf: The buffer to fill
521  * @pos: The file position
522  * @size: The amount to read
523  *
524  */
525
526 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
527                        unsigned size)
528 {
529         struct address_space *mapping = ip->i_inode.i_mapping;
530         unsigned long index = *pos / PAGE_CACHE_SIZE;
531         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
532         unsigned copied = 0;
533         unsigned amt;
534         struct page *page;
535         void *p;
536
537         do {
538                 amt = size - copied;
539                 if (offset + size > PAGE_CACHE_SIZE)
540                         amt = PAGE_CACHE_SIZE - offset;
541                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
542                 if (IS_ERR(page))
543                         return PTR_ERR(page);
544                 p = kmap_atomic(page);
545                 memcpy(buf + copied, p + offset, amt);
546                 kunmap_atomic(p);
547                 mark_page_accessed(page);
548                 page_cache_release(page);
549                 copied += amt;
550                 index++;
551                 offset = 0;
552         } while(copied < size);
553         (*pos) += size;
554         return size;
555 }
556
557 /**
558  * gfs2_readpages - Read a bunch of pages at once
559  *
560  * Some notes:
561  * 1. This is only for readahead, so we can simply ignore any things
562  *    which are slightly inconvenient (such as locking conflicts between
563  *    the page lock and the glock) and return having done no I/O. Its
564  *    obviously not something we'd want to do on too regular a basis.
565  *    Any I/O we ignore at this time will be done via readpage later.
566  * 2. We don't handle stuffed files here we let readpage do the honours.
567  * 3. mpage_readpages() does most of the heavy lifting in the common case.
568  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
569  */
570
571 static int gfs2_readpages(struct file *file, struct address_space *mapping,
572                           struct list_head *pages, unsigned nr_pages)
573 {
574         struct inode *inode = mapping->host;
575         struct gfs2_inode *ip = GFS2_I(inode);
576         struct gfs2_sbd *sdp = GFS2_SB(inode);
577         struct gfs2_holder gh;
578         int ret;
579
580         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
581         ret = gfs2_glock_nq(&gh);
582         if (unlikely(ret))
583                 goto out_uninit;
584         if (!gfs2_is_stuffed(ip))
585                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
586         gfs2_glock_dq(&gh);
587 out_uninit:
588         gfs2_holder_uninit(&gh);
589         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
590                 ret = -EIO;
591         return ret;
592 }
593
594 /**
595  * gfs2_write_begin - Begin to write to a file
596  * @file: The file to write to
597  * @mapping: The mapping in which to write
598  * @pos: The file offset at which to start writing
599  * @len: Length of the write
600  * @flags: Various flags
601  * @pagep: Pointer to return the page
602  * @fsdata: Pointer to return fs data (unused by GFS2)
603  *
604  * Returns: errno
605  */
606
607 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
608                             loff_t pos, unsigned len, unsigned flags,
609                             struct page **pagep, void **fsdata)
610 {
611         struct gfs2_inode *ip = GFS2_I(mapping->host);
612         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
613         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
614         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
615         unsigned requested = 0;
616         int alloc_required;
617         int error = 0;
618         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
619         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
620         struct page *page;
621
622         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
623         error = gfs2_glock_nq(&ip->i_gh);
624         if (unlikely(error))
625                 goto out_uninit;
626         if (&ip->i_inode == sdp->sd_rindex) {
627                 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
628                                            GL_NOCACHE, &m_ip->i_gh);
629                 if (unlikely(error)) {
630                         gfs2_glock_dq(&ip->i_gh);
631                         goto out_uninit;
632                 }
633         }
634
635         alloc_required = gfs2_write_alloc_required(ip, pos, len);
636
637         if (alloc_required || gfs2_is_jdata(ip))
638                 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
639
640         if (alloc_required) {
641                 error = gfs2_quota_lock_check(ip);
642                 if (error)
643                         goto out_unlock;
644
645                 requested = data_blocks + ind_blocks;
646                 error = gfs2_inplace_reserve(ip, requested, 0);
647                 if (error)
648                         goto out_qunlock;
649         }
650
651         rblocks = RES_DINODE + ind_blocks;
652         if (gfs2_is_jdata(ip))
653                 rblocks += data_blocks ? data_blocks : 1;
654         if (ind_blocks || data_blocks)
655                 rblocks += RES_STATFS + RES_QUOTA;
656         if (&ip->i_inode == sdp->sd_rindex)
657                 rblocks += 2 * RES_STATFS;
658         if (alloc_required)
659                 rblocks += gfs2_rg_blocks(ip, requested);
660
661         error = gfs2_trans_begin(sdp, rblocks,
662                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
663         if (error)
664                 goto out_trans_fail;
665
666         error = -ENOMEM;
667         flags |= AOP_FLAG_NOFS;
668         page = grab_cache_page_write_begin(mapping, index, flags);
669         *pagep = page;
670         if (unlikely(!page))
671                 goto out_endtrans;
672
673         if (gfs2_is_stuffed(ip)) {
674                 error = 0;
675                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
676                         error = gfs2_unstuff_dinode(ip, page);
677                         if (error == 0)
678                                 goto prepare_write;
679                 } else if (!PageUptodate(page)) {
680                         error = stuffed_readpage(ip, page);
681                 }
682                 goto out;
683         }
684
685 prepare_write:
686         error = __block_write_begin(page, from, len, gfs2_block_map);
687 out:
688         if (error == 0)
689                 return 0;
690
691         unlock_page(page);
692         page_cache_release(page);
693
694         gfs2_trans_end(sdp);
695         if (pos + len > ip->i_inode.i_size)
696                 gfs2_trim_blocks(&ip->i_inode);
697         goto out_trans_fail;
698
699 out_endtrans:
700         gfs2_trans_end(sdp);
701 out_trans_fail:
702         if (alloc_required) {
703                 gfs2_inplace_release(ip);
704 out_qunlock:
705                 gfs2_quota_unlock(ip);
706         }
707 out_unlock:
708         if (&ip->i_inode == sdp->sd_rindex) {
709                 gfs2_glock_dq(&m_ip->i_gh);
710                 gfs2_holder_uninit(&m_ip->i_gh);
711         }
712         gfs2_glock_dq(&ip->i_gh);
713 out_uninit:
714         gfs2_holder_uninit(&ip->i_gh);
715         return error;
716 }
717
718 /**
719  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
720  * @inode: the rindex inode
721  */
722 static void adjust_fs_space(struct inode *inode)
723 {
724         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
725         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
726         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
727         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
728         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
729         struct buffer_head *m_bh, *l_bh;
730         u64 fs_total, new_free;
731
732         /* Total up the file system space, according to the latest rindex. */
733         fs_total = gfs2_ri_total(sdp);
734         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
735                 return;
736
737         spin_lock(&sdp->sd_statfs_spin);
738         gfs2_statfs_change_in(m_sc, m_bh->b_data +
739                               sizeof(struct gfs2_dinode));
740         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
741                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
742         else
743                 new_free = 0;
744         spin_unlock(&sdp->sd_statfs_spin);
745         fs_warn(sdp, "File system extended by %llu blocks.\n",
746                 (unsigned long long)new_free);
747         gfs2_statfs_change(sdp, new_free, new_free, 0);
748
749         if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
750                 goto out;
751         update_statfs(sdp, m_bh, l_bh);
752         brelse(l_bh);
753 out:
754         brelse(m_bh);
755 }
756
757 /**
758  * gfs2_stuffed_write_end - Write end for stuffed files
759  * @inode: The inode
760  * @dibh: The buffer_head containing the on-disk inode
761  * @pos: The file position
762  * @len: The length of the write
763  * @copied: How much was actually copied by the VFS
764  * @page: The page
765  *
766  * This copies the data from the page into the inode block after
767  * the inode data structure itself.
768  *
769  * Returns: errno
770  */
771 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
772                                   loff_t pos, unsigned len, unsigned copied,
773                                   struct page *page)
774 {
775         struct gfs2_inode *ip = GFS2_I(inode);
776         struct gfs2_sbd *sdp = GFS2_SB(inode);
777         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
778         u64 to = pos + copied;
779         void *kaddr;
780         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
781
782         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
783         kaddr = kmap_atomic(page);
784         memcpy(buf + pos, kaddr + pos, copied);
785         memset(kaddr + pos + copied, 0, len - copied);
786         flush_dcache_page(page);
787         kunmap_atomic(kaddr);
788
789         if (!PageUptodate(page))
790                 SetPageUptodate(page);
791         unlock_page(page);
792         page_cache_release(page);
793
794         if (copied) {
795                 if (inode->i_size < to)
796                         i_size_write(inode, to);
797                 mark_inode_dirty(inode);
798         }
799
800         if (inode == sdp->sd_rindex) {
801                 adjust_fs_space(inode);
802                 sdp->sd_rindex_uptodate = 0;
803         }
804
805         brelse(dibh);
806         gfs2_trans_end(sdp);
807         if (inode == sdp->sd_rindex) {
808                 gfs2_glock_dq(&m_ip->i_gh);
809                 gfs2_holder_uninit(&m_ip->i_gh);
810         }
811         gfs2_glock_dq(&ip->i_gh);
812         gfs2_holder_uninit(&ip->i_gh);
813         return copied;
814 }
815
816 /**
817  * gfs2_write_end
818  * @file: The file to write to
819  * @mapping: The address space to write to
820  * @pos: The file position
821  * @len: The length of the data
822  * @copied:
823  * @page: The page that has been written
824  * @fsdata: The fsdata (unused in GFS2)
825  *
826  * The main write_end function for GFS2. We have a separate one for
827  * stuffed files as they are slightly different, otherwise we just
828  * put our locking around the VFS provided functions.
829  *
830  * Returns: errno
831  */
832
833 static int gfs2_write_end(struct file *file, struct address_space *mapping,
834                           loff_t pos, unsigned len, unsigned copied,
835                           struct page *page, void *fsdata)
836 {
837         struct inode *inode = page->mapping->host;
838         struct gfs2_inode *ip = GFS2_I(inode);
839         struct gfs2_sbd *sdp = GFS2_SB(inode);
840         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
841         struct buffer_head *dibh;
842         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
843         unsigned int to = from + len;
844         int ret;
845
846         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
847
848         ret = gfs2_meta_inode_buffer(ip, &dibh);
849         if (unlikely(ret)) {
850                 unlock_page(page);
851                 page_cache_release(page);
852                 goto failed;
853         }
854
855         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
856
857         if (gfs2_is_stuffed(ip))
858                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
859
860         if (!gfs2_is_writeback(ip))
861                 gfs2_page_add_databufs(ip, page, from, to);
862
863         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
864
865         if (inode == sdp->sd_rindex) {
866                 adjust_fs_space(inode);
867                 sdp->sd_rindex_uptodate = 0;
868         }
869
870         brelse(dibh);
871 failed:
872         gfs2_trans_end(sdp);
873         gfs2_inplace_release(ip);
874         if (ip->i_res->rs_qa_qd_num)
875                 gfs2_quota_unlock(ip);
876         if (inode == sdp->sd_rindex) {
877                 gfs2_glock_dq(&m_ip->i_gh);
878                 gfs2_holder_uninit(&m_ip->i_gh);
879         }
880         gfs2_glock_dq(&ip->i_gh);
881         gfs2_holder_uninit(&ip->i_gh);
882         return ret;
883 }
884
885 /**
886  * gfs2_set_page_dirty - Page dirtying function
887  * @page: The page to dirty
888  *
889  * Returns: 1 if it dirtyed the page, or 0 otherwise
890  */
891  
892 static int gfs2_set_page_dirty(struct page *page)
893 {
894         SetPageChecked(page);
895         return __set_page_dirty_buffers(page);
896 }
897
898 /**
899  * gfs2_bmap - Block map function
900  * @mapping: Address space info
901  * @lblock: The block to map
902  *
903  * Returns: The disk address for the block or 0 on hole or error
904  */
905
906 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
907 {
908         struct gfs2_inode *ip = GFS2_I(mapping->host);
909         struct gfs2_holder i_gh;
910         sector_t dblock = 0;
911         int error;
912
913         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
914         if (error)
915                 return 0;
916
917         if (!gfs2_is_stuffed(ip))
918                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
919
920         gfs2_glock_dq_uninit(&i_gh);
921
922         return dblock;
923 }
924
925 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
926 {
927         struct gfs2_bufdata *bd;
928
929         lock_buffer(bh);
930         gfs2_log_lock(sdp);
931         clear_buffer_dirty(bh);
932         bd = bh->b_private;
933         if (bd) {
934                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
935                         list_del_init(&bd->bd_list);
936                 else
937                         gfs2_remove_from_journal(bh, current->journal_info, 0);
938         }
939         bh->b_bdev = NULL;
940         clear_buffer_mapped(bh);
941         clear_buffer_req(bh);
942         clear_buffer_new(bh);
943         gfs2_log_unlock(sdp);
944         unlock_buffer(bh);
945 }
946
947 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
948 {
949         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
950         struct buffer_head *bh, *head;
951         unsigned long pos = 0;
952
953         BUG_ON(!PageLocked(page));
954         if (offset == 0)
955                 ClearPageChecked(page);
956         if (!page_has_buffers(page))
957                 goto out;
958
959         bh = head = page_buffers(page);
960         do {
961                 if (offset <= pos)
962                         gfs2_discard(sdp, bh);
963                 pos += bh->b_size;
964                 bh = bh->b_this_page;
965         } while (bh != head);
966 out:
967         if (offset == 0)
968                 try_to_release_page(page, 0);
969 }
970
971 /**
972  * gfs2_ok_for_dio - check that dio is valid on this file
973  * @ip: The inode
974  * @rw: READ or WRITE
975  * @offset: The offset at which we are reading or writing
976  *
977  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
978  *          1 (to accept the i/o request)
979  */
980 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
981 {
982         /*
983          * Should we return an error here? I can't see that O_DIRECT for
984          * a stuffed file makes any sense. For now we'll silently fall
985          * back to buffered I/O
986          */
987         if (gfs2_is_stuffed(ip))
988                 return 0;
989
990         if (offset >= i_size_read(&ip->i_inode))
991                 return 0;
992         return 1;
993 }
994
995
996
997 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
998                               const struct iovec *iov, loff_t offset,
999                               unsigned long nr_segs)
1000 {
1001         struct file *file = iocb->ki_filp;
1002         struct inode *inode = file->f_mapping->host;
1003         struct gfs2_inode *ip = GFS2_I(inode);
1004         struct gfs2_holder gh;
1005         int rv;
1006
1007         /*
1008          * Deferred lock, even if its a write, since we do no allocation
1009          * on this path. All we need change is atime, and this lock mode
1010          * ensures that other nodes have flushed their buffered read caches
1011          * (i.e. their page cache entries for this inode). We do not,
1012          * unfortunately have the option of only flushing a range like
1013          * the VFS does.
1014          */
1015         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1016         rv = gfs2_glock_nq(&gh);
1017         if (rv)
1018                 return rv;
1019         rv = gfs2_ok_for_dio(ip, rw, offset);
1020         if (rv != 1)
1021                 goto out; /* dio not valid, fall back to buffered i/o */
1022
1023         rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1024                                   offset, nr_segs, gfs2_get_block_direct,
1025                                   NULL, NULL, 0);
1026 out:
1027         gfs2_glock_dq(&gh);
1028         gfs2_holder_uninit(&gh);
1029         return rv;
1030 }
1031
1032 /**
1033  * gfs2_releasepage - free the metadata associated with a page
1034  * @page: the page that's being released
1035  * @gfp_mask: passed from Linux VFS, ignored by us
1036  *
1037  * Call try_to_free_buffers() if the buffers in this page can be
1038  * released.
1039  *
1040  * Returns: 0
1041  */
1042
1043 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1044 {
1045         struct address_space *mapping = page->mapping;
1046         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1047         struct buffer_head *bh, *head;
1048         struct gfs2_bufdata *bd;
1049
1050         if (!page_has_buffers(page))
1051                 return 0;
1052
1053         gfs2_log_lock(sdp);
1054         spin_lock(&sdp->sd_ail_lock);
1055         head = bh = page_buffers(page);
1056         do {
1057                 if (atomic_read(&bh->b_count))
1058                         goto cannot_release;
1059                 bd = bh->b_private;
1060                 if (bd && bd->bd_ail)
1061                         goto cannot_release;
1062                 if (buffer_pinned(bh) || buffer_dirty(bh))
1063                         goto not_possible;
1064                 bh = bh->b_this_page;
1065         } while(bh != head);
1066         spin_unlock(&sdp->sd_ail_lock);
1067         gfs2_log_unlock(sdp);
1068
1069         head = bh = page_buffers(page);
1070         do {
1071                 gfs2_log_lock(sdp);
1072                 bd = bh->b_private;
1073                 if (bd) {
1074                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1075                         if (!list_empty(&bd->bd_list)) {
1076                                 if (!buffer_pinned(bh))
1077                                         list_del_init(&bd->bd_list);
1078                                 else
1079                                         bd = NULL;
1080                         }
1081                         if (bd)
1082                                 bd->bd_bh = NULL;
1083                         bh->b_private = NULL;
1084                 }
1085                 gfs2_log_unlock(sdp);
1086                 if (bd)
1087                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1088
1089                 bh = bh->b_this_page;
1090         } while (bh != head);
1091
1092         return try_to_free_buffers(page);
1093
1094 not_possible: /* Should never happen */
1095         WARN_ON(buffer_dirty(bh));
1096         WARN_ON(buffer_pinned(bh));
1097 cannot_release:
1098         spin_unlock(&sdp->sd_ail_lock);
1099         gfs2_log_unlock(sdp);
1100         return 0;
1101 }
1102
1103 static const struct address_space_operations gfs2_writeback_aops = {
1104         .writepage = gfs2_writeback_writepage,
1105         .writepages = gfs2_writeback_writepages,
1106         .readpage = gfs2_readpage,
1107         .readpages = gfs2_readpages,
1108         .write_begin = gfs2_write_begin,
1109         .write_end = gfs2_write_end,
1110         .bmap = gfs2_bmap,
1111         .invalidatepage = gfs2_invalidatepage,
1112         .releasepage = gfs2_releasepage,
1113         .direct_IO = gfs2_direct_IO,
1114         .migratepage = buffer_migrate_page,
1115         .is_partially_uptodate = block_is_partially_uptodate,
1116         .error_remove_page = generic_error_remove_page,
1117 };
1118
1119 static const struct address_space_operations gfs2_ordered_aops = {
1120         .writepage = gfs2_ordered_writepage,
1121         .readpage = gfs2_readpage,
1122         .readpages = gfs2_readpages,
1123         .write_begin = gfs2_write_begin,
1124         .write_end = gfs2_write_end,
1125         .set_page_dirty = gfs2_set_page_dirty,
1126         .bmap = gfs2_bmap,
1127         .invalidatepage = gfs2_invalidatepage,
1128         .releasepage = gfs2_releasepage,
1129         .direct_IO = gfs2_direct_IO,
1130         .migratepage = buffer_migrate_page,
1131         .is_partially_uptodate = block_is_partially_uptodate,
1132         .error_remove_page = generic_error_remove_page,
1133 };
1134
1135 static const struct address_space_operations gfs2_jdata_aops = {
1136         .writepage = gfs2_jdata_writepage,
1137         .writepages = gfs2_jdata_writepages,
1138         .readpage = gfs2_readpage,
1139         .readpages = gfs2_readpages,
1140         .write_begin = gfs2_write_begin,
1141         .write_end = gfs2_write_end,
1142         .set_page_dirty = gfs2_set_page_dirty,
1143         .bmap = gfs2_bmap,
1144         .invalidatepage = gfs2_invalidatepage,
1145         .releasepage = gfs2_releasepage,
1146         .is_partially_uptodate = block_is_partially_uptodate,
1147         .error_remove_page = generic_error_remove_page,
1148 };
1149
1150 void gfs2_set_aops(struct inode *inode)
1151 {
1152         struct gfs2_inode *ip = GFS2_I(inode);
1153
1154         if (gfs2_is_writeback(ip))
1155                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1156         else if (gfs2_is_ordered(ip))
1157                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1158         else if (gfs2_is_jdata(ip))
1159                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1160         else
1161                 BUG();
1162 }
1163