Loading...
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#include <trace/events/writeback.h>
25
26#include "gfs2.h"
27#include "incore.h"
28#include "bmap.h"
29#include "glock.h"
30#include "inode.h"
31#include "log.h"
32#include "meta_io.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
41static 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_data(ip->i_gl, bh);
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
70static 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
83static 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
97static 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, PAGE_CACHE_SIZE);
115 goto out;
116 }
117 return 1;
118redirty:
119 redirty_page_for_writepage(wbc, page);
120out:
121 unlock_page(page);
122 return 0;
123}
124
125/**
126 * gfs2_writepage - Write page for writeback mappings
127 * @page: The page
128 * @wbc: The writeback control
129 *
130 */
131
132static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
133{
134 int ret;
135
136 ret = gfs2_writepage_common(page, wbc);
137 if (ret <= 0)
138 return ret;
139
140 return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
141}
142
143/**
144 * __gfs2_jdata_writepage - The core of jdata writepage
145 * @page: The page to write
146 * @wbc: The writeback control
147 *
148 * This is shared between writepage and writepages and implements the
149 * core of the writepage operation. If a transaction is required then
150 * PageChecked will have been set and the transaction will have
151 * already been started before this is called.
152 */
153
154static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
155{
156 struct inode *inode = page->mapping->host;
157 struct gfs2_inode *ip = GFS2_I(inode);
158 struct gfs2_sbd *sdp = GFS2_SB(inode);
159
160 if (PageChecked(page)) {
161 ClearPageChecked(page);
162 if (!page_has_buffers(page)) {
163 create_empty_buffers(page, inode->i_sb->s_blocksize,
164 (1 << BH_Dirty)|(1 << BH_Uptodate));
165 }
166 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
167 }
168 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
169}
170
171/**
172 * gfs2_jdata_writepage - Write complete page
173 * @page: Page to write
174 *
175 * Returns: errno
176 *
177 */
178
179static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
180{
181 struct inode *inode = page->mapping->host;
182 struct gfs2_sbd *sdp = GFS2_SB(inode);
183 int ret;
184 int done_trans = 0;
185
186 if (PageChecked(page)) {
187 if (wbc->sync_mode != WB_SYNC_ALL)
188 goto out_ignore;
189 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
190 if (ret)
191 goto out_ignore;
192 done_trans = 1;
193 }
194 ret = gfs2_writepage_common(page, wbc);
195 if (ret > 0)
196 ret = __gfs2_jdata_writepage(page, wbc);
197 if (done_trans)
198 gfs2_trans_end(sdp);
199 return ret;
200
201out_ignore:
202 redirty_page_for_writepage(wbc, page);
203 unlock_page(page);
204 return 0;
205}
206
207/**
208 * gfs2_writepages - Write a bunch of dirty pages back to disk
209 * @mapping: The mapping to write
210 * @wbc: Write-back control
211 *
212 * Used for both ordered and writeback modes.
213 */
214static int gfs2_writepages(struct address_space *mapping,
215 struct writeback_control *wbc)
216{
217 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
218}
219
220/**
221 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
222 * @mapping: The mapping
223 * @wbc: The writeback control
224 * @writepage: The writepage function to call for each page
225 * @pvec: The vector of pages
226 * @nr_pages: The number of pages to write
227 *
228 * Returns: non-zero if loop should terminate, zero otherwise
229 */
230
231static int gfs2_write_jdata_pagevec(struct address_space *mapping,
232 struct writeback_control *wbc,
233 struct pagevec *pvec,
234 int nr_pages, pgoff_t end,
235 pgoff_t *done_index)
236{
237 struct inode *inode = mapping->host;
238 struct gfs2_sbd *sdp = GFS2_SB(inode);
239 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
240 int i;
241 int ret;
242
243 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
244 if (ret < 0)
245 return ret;
246
247 for(i = 0; i < nr_pages; i++) {
248 struct page *page = pvec->pages[i];
249
250 /*
251 * At this point, the page may be truncated or
252 * invalidated (changing page->mapping to NULL), or
253 * even swizzled back from swapper_space to tmpfs file
254 * mapping. However, page->index will not change
255 * because we have a reference on the page.
256 */
257 if (page->index > end) {
258 /*
259 * can't be range_cyclic (1st pass) because
260 * end == -1 in that case.
261 */
262 ret = 1;
263 break;
264 }
265
266 *done_index = page->index;
267
268 lock_page(page);
269
270 if (unlikely(page->mapping != mapping)) {
271continue_unlock:
272 unlock_page(page);
273 continue;
274 }
275
276 if (!PageDirty(page)) {
277 /* someone wrote it for us */
278 goto continue_unlock;
279 }
280
281 if (PageWriteback(page)) {
282 if (wbc->sync_mode != WB_SYNC_NONE)
283 wait_on_page_writeback(page);
284 else
285 goto continue_unlock;
286 }
287
288 BUG_ON(PageWriteback(page));
289 if (!clear_page_dirty_for_io(page))
290 goto continue_unlock;
291
292 trace_wbc_writepage(wbc, mapping->backing_dev_info);
293
294 ret = __gfs2_jdata_writepage(page, wbc);
295 if (unlikely(ret)) {
296 if (ret == AOP_WRITEPAGE_ACTIVATE) {
297 unlock_page(page);
298 ret = 0;
299 } else {
300
301 /*
302 * done_index is set past this page,
303 * so media errors will not choke
304 * background writeout for the entire
305 * file. This has consequences for
306 * range_cyclic semantics (ie. it may
307 * not be suitable for data integrity
308 * writeout).
309 */
310 *done_index = page->index + 1;
311 ret = 1;
312 break;
313 }
314 }
315
316 /*
317 * We stop writing back only if we are not doing
318 * integrity sync. In case of integrity sync we have to
319 * keep going until we have written all the pages
320 * we tagged for writeback prior to entering this loop.
321 */
322 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
323 ret = 1;
324 break;
325 }
326
327 }
328 gfs2_trans_end(sdp);
329 return ret;
330}
331
332/**
333 * gfs2_write_cache_jdata - Like write_cache_pages but different
334 * @mapping: The mapping to write
335 * @wbc: The writeback control
336 * @writepage: The writepage function to call
337 * @data: The data to pass to writepage
338 *
339 * The reason that we use our own function here is that we need to
340 * start transactions before we grab page locks. This allows us
341 * to get the ordering right.
342 */
343
344static int gfs2_write_cache_jdata(struct address_space *mapping,
345 struct writeback_control *wbc)
346{
347 int ret = 0;
348 int done = 0;
349 struct pagevec pvec;
350 int nr_pages;
351 pgoff_t uninitialized_var(writeback_index);
352 pgoff_t index;
353 pgoff_t end;
354 pgoff_t done_index;
355 int cycled;
356 int range_whole = 0;
357 int tag;
358
359 pagevec_init(&pvec, 0);
360 if (wbc->range_cyclic) {
361 writeback_index = mapping->writeback_index; /* prev offset */
362 index = writeback_index;
363 if (index == 0)
364 cycled = 1;
365 else
366 cycled = 0;
367 end = -1;
368 } else {
369 index = wbc->range_start >> PAGE_CACHE_SHIFT;
370 end = wbc->range_end >> PAGE_CACHE_SHIFT;
371 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
372 range_whole = 1;
373 cycled = 1; /* ignore range_cyclic tests */
374 }
375 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
376 tag = PAGECACHE_TAG_TOWRITE;
377 else
378 tag = PAGECACHE_TAG_DIRTY;
379
380retry:
381 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
382 tag_pages_for_writeback(mapping, index, end);
383 done_index = index;
384 while (!done && (index <= end)) {
385 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
386 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
387 if (nr_pages == 0)
388 break;
389
390 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end, &done_index);
391 if (ret)
392 done = 1;
393 if (ret > 0)
394 ret = 0;
395 pagevec_release(&pvec);
396 cond_resched();
397 }
398
399 if (!cycled && !done) {
400 /*
401 * range_cyclic:
402 * We hit the last page and there is more work to be done: wrap
403 * back to the start of the file
404 */
405 cycled = 1;
406 index = 0;
407 end = writeback_index - 1;
408 goto retry;
409 }
410
411 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
412 mapping->writeback_index = done_index;
413
414 return ret;
415}
416
417
418/**
419 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
420 * @mapping: The mapping to write
421 * @wbc: The writeback control
422 *
423 */
424
425static int gfs2_jdata_writepages(struct address_space *mapping,
426 struct writeback_control *wbc)
427{
428 struct gfs2_inode *ip = GFS2_I(mapping->host);
429 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
430 int ret;
431
432 ret = gfs2_write_cache_jdata(mapping, wbc);
433 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
434 gfs2_log_flush(sdp, ip->i_gl);
435 ret = gfs2_write_cache_jdata(mapping, wbc);
436 }
437 return ret;
438}
439
440/**
441 * stuffed_readpage - Fill in a Linux page with stuffed file data
442 * @ip: the inode
443 * @page: the page
444 *
445 * Returns: errno
446 */
447
448static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
449{
450 struct buffer_head *dibh;
451 u64 dsize = i_size_read(&ip->i_inode);
452 void *kaddr;
453 int error;
454
455 /*
456 * Due to the order of unstuffing files and ->fault(), we can be
457 * asked for a zero page in the case of a stuffed file being extended,
458 * so we need to supply one here. It doesn't happen often.
459 */
460 if (unlikely(page->index)) {
461 zero_user(page, 0, PAGE_CACHE_SIZE);
462 SetPageUptodate(page);
463 return 0;
464 }
465
466 error = gfs2_meta_inode_buffer(ip, &dibh);
467 if (error)
468 return error;
469
470 kaddr = kmap_atomic(page);
471 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
472 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
473 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
474 memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
475 kunmap_atomic(kaddr);
476 flush_dcache_page(page);
477 brelse(dibh);
478 SetPageUptodate(page);
479
480 return 0;
481}
482
483
484/**
485 * __gfs2_readpage - readpage
486 * @file: The file to read a page for
487 * @page: The page to read
488 *
489 * This is the core of gfs2's readpage. Its used by the internal file
490 * reading code as in that case we already hold the glock. Also its
491 * called by gfs2_readpage() once the required lock has been granted.
492 *
493 */
494
495static int __gfs2_readpage(void *file, struct page *page)
496{
497 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
498 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
499 int error;
500
501 if (gfs2_is_stuffed(ip)) {
502 error = stuffed_readpage(ip, page);
503 unlock_page(page);
504 } else {
505 error = mpage_readpage(page, gfs2_block_map);
506 }
507
508 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
509 return -EIO;
510
511 return error;
512}
513
514/**
515 * gfs2_readpage - read a page of a file
516 * @file: The file to read
517 * @page: The page of the file
518 *
519 * This deals with the locking required. We have to unlock and
520 * relock the page in order to get the locking in the right
521 * order.
522 */
523
524static int gfs2_readpage(struct file *file, struct page *page)
525{
526 struct address_space *mapping = page->mapping;
527 struct gfs2_inode *ip = GFS2_I(mapping->host);
528 struct gfs2_holder gh;
529 int error;
530
531 unlock_page(page);
532 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
533 error = gfs2_glock_nq(&gh);
534 if (unlikely(error))
535 goto out;
536 error = AOP_TRUNCATED_PAGE;
537 lock_page(page);
538 if (page->mapping == mapping && !PageUptodate(page))
539 error = __gfs2_readpage(file, page);
540 else
541 unlock_page(page);
542 gfs2_glock_dq(&gh);
543out:
544 gfs2_holder_uninit(&gh);
545 if (error && error != AOP_TRUNCATED_PAGE)
546 lock_page(page);
547 return error;
548}
549
550/**
551 * gfs2_internal_read - read an internal file
552 * @ip: The gfs2 inode
553 * @buf: The buffer to fill
554 * @pos: The file position
555 * @size: The amount to read
556 *
557 */
558
559int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
560 unsigned size)
561{
562 struct address_space *mapping = ip->i_inode.i_mapping;
563 unsigned long index = *pos / PAGE_CACHE_SIZE;
564 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
565 unsigned copied = 0;
566 unsigned amt;
567 struct page *page;
568 void *p;
569
570 do {
571 amt = size - copied;
572 if (offset + size > PAGE_CACHE_SIZE)
573 amt = PAGE_CACHE_SIZE - offset;
574 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
575 if (IS_ERR(page))
576 return PTR_ERR(page);
577 p = kmap_atomic(page);
578 memcpy(buf + copied, p + offset, amt);
579 kunmap_atomic(p);
580 mark_page_accessed(page);
581 page_cache_release(page);
582 copied += amt;
583 index++;
584 offset = 0;
585 } while(copied < size);
586 (*pos) += size;
587 return size;
588}
589
590/**
591 * gfs2_readpages - Read a bunch of pages at once
592 *
593 * Some notes:
594 * 1. This is only for readahead, so we can simply ignore any things
595 * which are slightly inconvenient (such as locking conflicts between
596 * the page lock and the glock) and return having done no I/O. Its
597 * obviously not something we'd want to do on too regular a basis.
598 * Any I/O we ignore at this time will be done via readpage later.
599 * 2. We don't handle stuffed files here we let readpage do the honours.
600 * 3. mpage_readpages() does most of the heavy lifting in the common case.
601 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
602 */
603
604static int gfs2_readpages(struct file *file, struct address_space *mapping,
605 struct list_head *pages, unsigned nr_pages)
606{
607 struct inode *inode = mapping->host;
608 struct gfs2_inode *ip = GFS2_I(inode);
609 struct gfs2_sbd *sdp = GFS2_SB(inode);
610 struct gfs2_holder gh;
611 int ret;
612
613 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
614 ret = gfs2_glock_nq(&gh);
615 if (unlikely(ret))
616 goto out_uninit;
617 if (!gfs2_is_stuffed(ip))
618 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
619 gfs2_glock_dq(&gh);
620out_uninit:
621 gfs2_holder_uninit(&gh);
622 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
623 ret = -EIO;
624 return ret;
625}
626
627/**
628 * gfs2_write_begin - Begin to write to a file
629 * @file: The file to write to
630 * @mapping: The mapping in which to write
631 * @pos: The file offset at which to start writing
632 * @len: Length of the write
633 * @flags: Various flags
634 * @pagep: Pointer to return the page
635 * @fsdata: Pointer to return fs data (unused by GFS2)
636 *
637 * Returns: errno
638 */
639
640static int gfs2_write_begin(struct file *file, struct address_space *mapping,
641 loff_t pos, unsigned len, unsigned flags,
642 struct page **pagep, void **fsdata)
643{
644 struct gfs2_inode *ip = GFS2_I(mapping->host);
645 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
646 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
647 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
648 unsigned requested = 0;
649 int alloc_required;
650 int error = 0;
651 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
652 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
653 struct page *page;
654
655 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
656 error = gfs2_glock_nq(&ip->i_gh);
657 if (unlikely(error))
658 goto out_uninit;
659 if (&ip->i_inode == sdp->sd_rindex) {
660 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
661 GL_NOCACHE, &m_ip->i_gh);
662 if (unlikely(error)) {
663 gfs2_glock_dq(&ip->i_gh);
664 goto out_uninit;
665 }
666 }
667
668 alloc_required = gfs2_write_alloc_required(ip, pos, len);
669
670 if (alloc_required || gfs2_is_jdata(ip))
671 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
672
673 if (alloc_required) {
674 struct gfs2_alloc_parms ap = { .aflags = 0, };
675 error = gfs2_quota_lock_check(ip);
676 if (error)
677 goto out_unlock;
678
679 requested = data_blocks + ind_blocks;
680 ap.target = requested;
681 error = gfs2_inplace_reserve(ip, &ap);
682 if (error)
683 goto out_qunlock;
684 }
685
686 rblocks = RES_DINODE + ind_blocks;
687 if (gfs2_is_jdata(ip))
688 rblocks += data_blocks ? data_blocks : 1;
689 if (ind_blocks || data_blocks)
690 rblocks += RES_STATFS + RES_QUOTA;
691 if (&ip->i_inode == sdp->sd_rindex)
692 rblocks += 2 * RES_STATFS;
693 if (alloc_required)
694 rblocks += gfs2_rg_blocks(ip, requested);
695
696 error = gfs2_trans_begin(sdp, rblocks,
697 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
698 if (error)
699 goto out_trans_fail;
700
701 error = -ENOMEM;
702 flags |= AOP_FLAG_NOFS;
703 page = grab_cache_page_write_begin(mapping, index, flags);
704 *pagep = page;
705 if (unlikely(!page))
706 goto out_endtrans;
707
708 if (gfs2_is_stuffed(ip)) {
709 error = 0;
710 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
711 error = gfs2_unstuff_dinode(ip, page);
712 if (error == 0)
713 goto prepare_write;
714 } else if (!PageUptodate(page)) {
715 error = stuffed_readpage(ip, page);
716 }
717 goto out;
718 }
719
720prepare_write:
721 error = __block_write_begin(page, from, len, gfs2_block_map);
722out:
723 if (error == 0)
724 return 0;
725
726 unlock_page(page);
727 page_cache_release(page);
728
729 gfs2_trans_end(sdp);
730 if (pos + len > ip->i_inode.i_size)
731 gfs2_trim_blocks(&ip->i_inode);
732 goto out_trans_fail;
733
734out_endtrans:
735 gfs2_trans_end(sdp);
736out_trans_fail:
737 if (alloc_required) {
738 gfs2_inplace_release(ip);
739out_qunlock:
740 gfs2_quota_unlock(ip);
741 }
742out_unlock:
743 if (&ip->i_inode == sdp->sd_rindex) {
744 gfs2_glock_dq(&m_ip->i_gh);
745 gfs2_holder_uninit(&m_ip->i_gh);
746 }
747 gfs2_glock_dq(&ip->i_gh);
748out_uninit:
749 gfs2_holder_uninit(&ip->i_gh);
750 return error;
751}
752
753/**
754 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
755 * @inode: the rindex inode
756 */
757static void adjust_fs_space(struct inode *inode)
758{
759 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
760 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
761 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
762 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
763 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
764 struct buffer_head *m_bh, *l_bh;
765 u64 fs_total, new_free;
766
767 /* Total up the file system space, according to the latest rindex. */
768 fs_total = gfs2_ri_total(sdp);
769 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
770 return;
771
772 spin_lock(&sdp->sd_statfs_spin);
773 gfs2_statfs_change_in(m_sc, m_bh->b_data +
774 sizeof(struct gfs2_dinode));
775 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
776 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
777 else
778 new_free = 0;
779 spin_unlock(&sdp->sd_statfs_spin);
780 fs_warn(sdp, "File system extended by %llu blocks.\n",
781 (unsigned long long)new_free);
782 gfs2_statfs_change(sdp, new_free, new_free, 0);
783
784 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
785 goto out;
786 update_statfs(sdp, m_bh, l_bh);
787 brelse(l_bh);
788out:
789 brelse(m_bh);
790}
791
792/**
793 * gfs2_stuffed_write_end - Write end for stuffed files
794 * @inode: The inode
795 * @dibh: The buffer_head containing the on-disk inode
796 * @pos: The file position
797 * @len: The length of the write
798 * @copied: How much was actually copied by the VFS
799 * @page: The page
800 *
801 * This copies the data from the page into the inode block after
802 * the inode data structure itself.
803 *
804 * Returns: errno
805 */
806static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
807 loff_t pos, unsigned len, unsigned copied,
808 struct page *page)
809{
810 struct gfs2_inode *ip = GFS2_I(inode);
811 struct gfs2_sbd *sdp = GFS2_SB(inode);
812 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
813 u64 to = pos + copied;
814 void *kaddr;
815 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
816
817 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
818 kaddr = kmap_atomic(page);
819 memcpy(buf + pos, kaddr + pos, copied);
820 memset(kaddr + pos + copied, 0, len - copied);
821 flush_dcache_page(page);
822 kunmap_atomic(kaddr);
823
824 if (!PageUptodate(page))
825 SetPageUptodate(page);
826 unlock_page(page);
827 page_cache_release(page);
828
829 if (copied) {
830 if (inode->i_size < to)
831 i_size_write(inode, to);
832 mark_inode_dirty(inode);
833 }
834
835 if (inode == sdp->sd_rindex) {
836 adjust_fs_space(inode);
837 sdp->sd_rindex_uptodate = 0;
838 }
839
840 brelse(dibh);
841 gfs2_trans_end(sdp);
842 if (inode == sdp->sd_rindex) {
843 gfs2_glock_dq(&m_ip->i_gh);
844 gfs2_holder_uninit(&m_ip->i_gh);
845 }
846 gfs2_glock_dq(&ip->i_gh);
847 gfs2_holder_uninit(&ip->i_gh);
848 return copied;
849}
850
851/**
852 * gfs2_write_end
853 * @file: The file to write to
854 * @mapping: The address space to write to
855 * @pos: The file position
856 * @len: The length of the data
857 * @copied:
858 * @page: The page that has been written
859 * @fsdata: The fsdata (unused in GFS2)
860 *
861 * The main write_end function for GFS2. We have a separate one for
862 * stuffed files as they are slightly different, otherwise we just
863 * put our locking around the VFS provided functions.
864 *
865 * Returns: errno
866 */
867
868static int gfs2_write_end(struct file *file, struct address_space *mapping,
869 loff_t pos, unsigned len, unsigned copied,
870 struct page *page, void *fsdata)
871{
872 struct inode *inode = page->mapping->host;
873 struct gfs2_inode *ip = GFS2_I(inode);
874 struct gfs2_sbd *sdp = GFS2_SB(inode);
875 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
876 struct buffer_head *dibh;
877 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
878 unsigned int to = from + len;
879 int ret;
880 struct gfs2_trans *tr = current->journal_info;
881 BUG_ON(!tr);
882
883 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
884
885 ret = gfs2_meta_inode_buffer(ip, &dibh);
886 if (unlikely(ret)) {
887 unlock_page(page);
888 page_cache_release(page);
889 goto failed;
890 }
891
892 if (gfs2_is_stuffed(ip))
893 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
894
895 if (!gfs2_is_writeback(ip))
896 gfs2_page_add_databufs(ip, page, from, to);
897
898 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
899 if (tr->tr_num_buf_new)
900 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
901 else
902 gfs2_trans_add_meta(ip->i_gl, dibh);
903
904
905 if (inode == sdp->sd_rindex) {
906 adjust_fs_space(inode);
907 sdp->sd_rindex_uptodate = 0;
908 }
909
910 brelse(dibh);
911failed:
912 gfs2_trans_end(sdp);
913 gfs2_inplace_release(ip);
914 if (ip->i_res->rs_qa_qd_num)
915 gfs2_quota_unlock(ip);
916 if (inode == sdp->sd_rindex) {
917 gfs2_glock_dq(&m_ip->i_gh);
918 gfs2_holder_uninit(&m_ip->i_gh);
919 }
920 gfs2_glock_dq(&ip->i_gh);
921 gfs2_holder_uninit(&ip->i_gh);
922 return ret;
923}
924
925/**
926 * gfs2_set_page_dirty - Page dirtying function
927 * @page: The page to dirty
928 *
929 * Returns: 1 if it dirtyed the page, or 0 otherwise
930 */
931
932static int gfs2_set_page_dirty(struct page *page)
933{
934 SetPageChecked(page);
935 return __set_page_dirty_buffers(page);
936}
937
938/**
939 * gfs2_bmap - Block map function
940 * @mapping: Address space info
941 * @lblock: The block to map
942 *
943 * Returns: The disk address for the block or 0 on hole or error
944 */
945
946static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
947{
948 struct gfs2_inode *ip = GFS2_I(mapping->host);
949 struct gfs2_holder i_gh;
950 sector_t dblock = 0;
951 int error;
952
953 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
954 if (error)
955 return 0;
956
957 if (!gfs2_is_stuffed(ip))
958 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
959
960 gfs2_glock_dq_uninit(&i_gh);
961
962 return dblock;
963}
964
965static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
966{
967 struct gfs2_bufdata *bd;
968
969 lock_buffer(bh);
970 gfs2_log_lock(sdp);
971 clear_buffer_dirty(bh);
972 bd = bh->b_private;
973 if (bd) {
974 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
975 list_del_init(&bd->bd_list);
976 else
977 gfs2_remove_from_journal(bh, current->journal_info, 0);
978 }
979 bh->b_bdev = NULL;
980 clear_buffer_mapped(bh);
981 clear_buffer_req(bh);
982 clear_buffer_new(bh);
983 gfs2_log_unlock(sdp);
984 unlock_buffer(bh);
985}
986
987static void gfs2_invalidatepage(struct page *page, unsigned int offset,
988 unsigned int length)
989{
990 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
991 unsigned int stop = offset + length;
992 int partial_page = (offset || length < PAGE_CACHE_SIZE);
993 struct buffer_head *bh, *head;
994 unsigned long pos = 0;
995
996 BUG_ON(!PageLocked(page));
997 if (!partial_page)
998 ClearPageChecked(page);
999 if (!page_has_buffers(page))
1000 goto out;
1001
1002 bh = head = page_buffers(page);
1003 do {
1004 if (pos + bh->b_size > stop)
1005 return;
1006
1007 if (offset <= pos)
1008 gfs2_discard(sdp, bh);
1009 pos += bh->b_size;
1010 bh = bh->b_this_page;
1011 } while (bh != head);
1012out:
1013 if (!partial_page)
1014 try_to_release_page(page, 0);
1015}
1016
1017/**
1018 * gfs2_ok_for_dio - check that dio is valid on this file
1019 * @ip: The inode
1020 * @rw: READ or WRITE
1021 * @offset: The offset at which we are reading or writing
1022 *
1023 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1024 * 1 (to accept the i/o request)
1025 */
1026static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
1027{
1028 /*
1029 * Should we return an error here? I can't see that O_DIRECT for
1030 * a stuffed file makes any sense. For now we'll silently fall
1031 * back to buffered I/O
1032 */
1033 if (gfs2_is_stuffed(ip))
1034 return 0;
1035
1036 if (offset >= i_size_read(&ip->i_inode))
1037 return 0;
1038 return 1;
1039}
1040
1041
1042
1043static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1044 const struct iovec *iov, loff_t offset,
1045 unsigned long nr_segs)
1046{
1047 struct file *file = iocb->ki_filp;
1048 struct inode *inode = file->f_mapping->host;
1049 struct address_space *mapping = inode->i_mapping;
1050 struct gfs2_inode *ip = GFS2_I(inode);
1051 struct gfs2_holder gh;
1052 int rv;
1053
1054 /*
1055 * Deferred lock, even if its a write, since we do no allocation
1056 * on this path. All we need change is atime, and this lock mode
1057 * ensures that other nodes have flushed their buffered read caches
1058 * (i.e. their page cache entries for this inode). We do not,
1059 * unfortunately have the option of only flushing a range like
1060 * the VFS does.
1061 */
1062 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1063 rv = gfs2_glock_nq(&gh);
1064 if (rv)
1065 return rv;
1066 rv = gfs2_ok_for_dio(ip, rw, offset);
1067 if (rv != 1)
1068 goto out; /* dio not valid, fall back to buffered i/o */
1069
1070 /*
1071 * Now since we are holding a deferred (CW) lock at this point, you
1072 * might be wondering why this is ever needed. There is a case however
1073 * where we've granted a deferred local lock against a cached exclusive
1074 * glock. That is ok provided all granted local locks are deferred, but
1075 * it also means that it is possible to encounter pages which are
1076 * cached and possibly also mapped. So here we check for that and sort
1077 * them out ahead of the dio. The glock state machine will take care of
1078 * everything else.
1079 *
1080 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
1081 * the first place, mapping->nr_pages will always be zero.
1082 */
1083 if (mapping->nrpages) {
1084 loff_t lstart = offset & (PAGE_CACHE_SIZE - 1);
1085 loff_t len = iov_length(iov, nr_segs);
1086 loff_t end = PAGE_ALIGN(offset + len) - 1;
1087
1088 rv = 0;
1089 if (len == 0)
1090 goto out;
1091 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
1092 unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
1093 rv = filemap_write_and_wait_range(mapping, lstart, end);
1094 if (rv)
1095 goto out;
1096 if (rw == WRITE)
1097 truncate_inode_pages_range(mapping, lstart, end);
1098 }
1099
1100 rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1101 offset, nr_segs, gfs2_get_block_direct,
1102 NULL, NULL, 0);
1103out:
1104 gfs2_glock_dq(&gh);
1105 gfs2_holder_uninit(&gh);
1106 return rv;
1107}
1108
1109/**
1110 * gfs2_releasepage - free the metadata associated with a page
1111 * @page: the page that's being released
1112 * @gfp_mask: passed from Linux VFS, ignored by us
1113 *
1114 * Call try_to_free_buffers() if the buffers in this page can be
1115 * released.
1116 *
1117 * Returns: 0
1118 */
1119
1120int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1121{
1122 struct address_space *mapping = page->mapping;
1123 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1124 struct buffer_head *bh, *head;
1125 struct gfs2_bufdata *bd;
1126
1127 if (!page_has_buffers(page))
1128 return 0;
1129
1130 gfs2_log_lock(sdp);
1131 spin_lock(&sdp->sd_ail_lock);
1132 head = bh = page_buffers(page);
1133 do {
1134 if (atomic_read(&bh->b_count))
1135 goto cannot_release;
1136 bd = bh->b_private;
1137 if (bd && bd->bd_tr)
1138 goto cannot_release;
1139 if (buffer_pinned(bh) || buffer_dirty(bh))
1140 goto not_possible;
1141 bh = bh->b_this_page;
1142 } while(bh != head);
1143 spin_unlock(&sdp->sd_ail_lock);
1144
1145 head = bh = page_buffers(page);
1146 do {
1147 bd = bh->b_private;
1148 if (bd) {
1149 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1150 if (!list_empty(&bd->bd_list))
1151 list_del_init(&bd->bd_list);
1152 bd->bd_bh = NULL;
1153 bh->b_private = NULL;
1154 kmem_cache_free(gfs2_bufdata_cachep, bd);
1155 }
1156
1157 bh = bh->b_this_page;
1158 } while (bh != head);
1159 gfs2_log_unlock(sdp);
1160
1161 return try_to_free_buffers(page);
1162
1163not_possible: /* Should never happen */
1164 WARN_ON(buffer_dirty(bh));
1165 WARN_ON(buffer_pinned(bh));
1166cannot_release:
1167 spin_unlock(&sdp->sd_ail_lock);
1168 gfs2_log_unlock(sdp);
1169 return 0;
1170}
1171
1172static const struct address_space_operations gfs2_writeback_aops = {
1173 .writepage = gfs2_writepage,
1174 .writepages = gfs2_writepages,
1175 .readpage = gfs2_readpage,
1176 .readpages = gfs2_readpages,
1177 .write_begin = gfs2_write_begin,
1178 .write_end = gfs2_write_end,
1179 .bmap = gfs2_bmap,
1180 .invalidatepage = gfs2_invalidatepage,
1181 .releasepage = gfs2_releasepage,
1182 .direct_IO = gfs2_direct_IO,
1183 .migratepage = buffer_migrate_page,
1184 .is_partially_uptodate = block_is_partially_uptodate,
1185 .error_remove_page = generic_error_remove_page,
1186};
1187
1188static const struct address_space_operations gfs2_ordered_aops = {
1189 .writepage = gfs2_writepage,
1190 .writepages = gfs2_writepages,
1191 .readpage = gfs2_readpage,
1192 .readpages = gfs2_readpages,
1193 .write_begin = gfs2_write_begin,
1194 .write_end = gfs2_write_end,
1195 .set_page_dirty = gfs2_set_page_dirty,
1196 .bmap = gfs2_bmap,
1197 .invalidatepage = gfs2_invalidatepage,
1198 .releasepage = gfs2_releasepage,
1199 .direct_IO = gfs2_direct_IO,
1200 .migratepage = buffer_migrate_page,
1201 .is_partially_uptodate = block_is_partially_uptodate,
1202 .error_remove_page = generic_error_remove_page,
1203};
1204
1205static const struct address_space_operations gfs2_jdata_aops = {
1206 .writepage = gfs2_jdata_writepage,
1207 .writepages = gfs2_jdata_writepages,
1208 .readpage = gfs2_readpage,
1209 .readpages = gfs2_readpages,
1210 .write_begin = gfs2_write_begin,
1211 .write_end = gfs2_write_end,
1212 .set_page_dirty = gfs2_set_page_dirty,
1213 .bmap = gfs2_bmap,
1214 .invalidatepage = gfs2_invalidatepage,
1215 .releasepage = gfs2_releasepage,
1216 .is_partially_uptodate = block_is_partially_uptodate,
1217 .error_remove_page = generic_error_remove_page,
1218};
1219
1220void gfs2_set_aops(struct inode *inode)
1221{
1222 struct gfs2_inode *ip = GFS2_I(inode);
1223
1224 if (gfs2_is_writeback(ip))
1225 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1226 else if (gfs2_is_ordered(ip))
1227 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1228 else if (gfs2_is_jdata(ip))
1229 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1230 else
1231 BUG();
1232}
1233
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
39static 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
68static 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
81static 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
95static 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;
116redirty:
117 redirty_page_for_writepage(wbc, page);
118out:
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
130static 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
149static 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
179static 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
204static 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
226out_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 */
241static 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
258static 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
329static 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
353retry:
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
392static 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
415static 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
462static 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
491static 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);
510out:
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
526int 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
571static 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);
587out_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
607static 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 int alloc_required;
616 int error = 0;
617 struct gfs2_qadata *qa = NULL;
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 qa = gfs2_qadata_get(ip);
642 if (!qa) {
643 error = -ENOMEM;
644 goto out_unlock;
645 }
646
647 error = gfs2_quota_lock_check(ip);
648 if (error)
649 goto out_alloc_put;
650
651 error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
652 if (error)
653 goto out_qunlock;
654 }
655
656 rblocks = RES_DINODE + ind_blocks;
657 if (gfs2_is_jdata(ip))
658 rblocks += data_blocks ? data_blocks : 1;
659 if (ind_blocks || data_blocks)
660 rblocks += RES_STATFS + RES_QUOTA;
661 if (&ip->i_inode == sdp->sd_rindex)
662 rblocks += 2 * RES_STATFS;
663 if (alloc_required)
664 rblocks += gfs2_rg_blocks(ip);
665
666 error = gfs2_trans_begin(sdp, rblocks,
667 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
668 if (error)
669 goto out_trans_fail;
670
671 error = -ENOMEM;
672 flags |= AOP_FLAG_NOFS;
673 page = grab_cache_page_write_begin(mapping, index, flags);
674 *pagep = page;
675 if (unlikely(!page))
676 goto out_endtrans;
677
678 if (gfs2_is_stuffed(ip)) {
679 error = 0;
680 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
681 error = gfs2_unstuff_dinode(ip, page);
682 if (error == 0)
683 goto prepare_write;
684 } else if (!PageUptodate(page)) {
685 error = stuffed_readpage(ip, page);
686 }
687 goto out;
688 }
689
690prepare_write:
691 error = __block_write_begin(page, from, len, gfs2_block_map);
692out:
693 if (error == 0)
694 return 0;
695
696 unlock_page(page);
697 page_cache_release(page);
698
699 gfs2_trans_end(sdp);
700 if (pos + len > ip->i_inode.i_size)
701 gfs2_trim_blocks(&ip->i_inode);
702 goto out_trans_fail;
703
704out_endtrans:
705 gfs2_trans_end(sdp);
706out_trans_fail:
707 if (alloc_required) {
708 gfs2_inplace_release(ip);
709out_qunlock:
710 gfs2_quota_unlock(ip);
711out_alloc_put:
712 gfs2_qadata_put(ip);
713 }
714out_unlock:
715 if (&ip->i_inode == sdp->sd_rindex) {
716 gfs2_glock_dq(&m_ip->i_gh);
717 gfs2_holder_uninit(&m_ip->i_gh);
718 }
719 gfs2_glock_dq(&ip->i_gh);
720out_uninit:
721 gfs2_holder_uninit(&ip->i_gh);
722 return error;
723}
724
725/**
726 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
727 * @inode: the rindex inode
728 */
729static void adjust_fs_space(struct inode *inode)
730{
731 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
732 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
733 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
734 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
735 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
736 struct buffer_head *m_bh, *l_bh;
737 u64 fs_total, new_free;
738
739 /* Total up the file system space, according to the latest rindex. */
740 fs_total = gfs2_ri_total(sdp);
741 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
742 return;
743
744 spin_lock(&sdp->sd_statfs_spin);
745 gfs2_statfs_change_in(m_sc, m_bh->b_data +
746 sizeof(struct gfs2_dinode));
747 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
748 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
749 else
750 new_free = 0;
751 spin_unlock(&sdp->sd_statfs_spin);
752 fs_warn(sdp, "File system extended by %llu blocks.\n",
753 (unsigned long long)new_free);
754 gfs2_statfs_change(sdp, new_free, new_free, 0);
755
756 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
757 goto out;
758 update_statfs(sdp, m_bh, l_bh);
759 brelse(l_bh);
760out:
761 brelse(m_bh);
762}
763
764/**
765 * gfs2_stuffed_write_end - Write end for stuffed files
766 * @inode: The inode
767 * @dibh: The buffer_head containing the on-disk inode
768 * @pos: The file position
769 * @len: The length of the write
770 * @copied: How much was actually copied by the VFS
771 * @page: The page
772 *
773 * This copies the data from the page into the inode block after
774 * the inode data structure itself.
775 *
776 * Returns: errno
777 */
778static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
779 loff_t pos, unsigned len, unsigned copied,
780 struct page *page)
781{
782 struct gfs2_inode *ip = GFS2_I(inode);
783 struct gfs2_sbd *sdp = GFS2_SB(inode);
784 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
785 u64 to = pos + copied;
786 void *kaddr;
787 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
788
789 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
790 kaddr = kmap_atomic(page);
791 memcpy(buf + pos, kaddr + pos, copied);
792 memset(kaddr + pos + copied, 0, len - copied);
793 flush_dcache_page(page);
794 kunmap_atomic(kaddr);
795
796 if (!PageUptodate(page))
797 SetPageUptodate(page);
798 unlock_page(page);
799 page_cache_release(page);
800
801 if (copied) {
802 if (inode->i_size < to)
803 i_size_write(inode, to);
804 mark_inode_dirty(inode);
805 }
806
807 if (inode == sdp->sd_rindex) {
808 adjust_fs_space(inode);
809 sdp->sd_rindex_uptodate = 0;
810 }
811
812 brelse(dibh);
813 gfs2_trans_end(sdp);
814 if (inode == sdp->sd_rindex) {
815 gfs2_glock_dq(&m_ip->i_gh);
816 gfs2_holder_uninit(&m_ip->i_gh);
817 }
818 gfs2_glock_dq(&ip->i_gh);
819 gfs2_holder_uninit(&ip->i_gh);
820 return copied;
821}
822
823/**
824 * gfs2_write_end
825 * @file: The file to write to
826 * @mapping: The address space to write to
827 * @pos: The file position
828 * @len: The length of the data
829 * @copied:
830 * @page: The page that has been written
831 * @fsdata: The fsdata (unused in GFS2)
832 *
833 * The main write_end function for GFS2. We have a separate one for
834 * stuffed files as they are slightly different, otherwise we just
835 * put our locking around the VFS provided functions.
836 *
837 * Returns: errno
838 */
839
840static int gfs2_write_end(struct file *file, struct address_space *mapping,
841 loff_t pos, unsigned len, unsigned copied,
842 struct page *page, void *fsdata)
843{
844 struct inode *inode = page->mapping->host;
845 struct gfs2_inode *ip = GFS2_I(inode);
846 struct gfs2_sbd *sdp = GFS2_SB(inode);
847 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
848 struct buffer_head *dibh;
849 struct gfs2_qadata *qa = ip->i_qadata;
850 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
851 unsigned int to = from + len;
852 int ret;
853
854 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
855
856 ret = gfs2_meta_inode_buffer(ip, &dibh);
857 if (unlikely(ret)) {
858 unlock_page(page);
859 page_cache_release(page);
860 goto failed;
861 }
862
863 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
864
865 if (gfs2_is_stuffed(ip))
866 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
867
868 if (!gfs2_is_writeback(ip))
869 gfs2_page_add_databufs(ip, page, from, to);
870
871 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
872
873 if (inode == sdp->sd_rindex) {
874 adjust_fs_space(inode);
875 sdp->sd_rindex_uptodate = 0;
876 }
877
878 brelse(dibh);
879failed:
880 gfs2_trans_end(sdp);
881 if (ip->i_res)
882 gfs2_inplace_release(ip);
883 if (qa) {
884 gfs2_quota_unlock(ip);
885 gfs2_qadata_put(ip);
886 }
887 if (inode == sdp->sd_rindex) {
888 gfs2_glock_dq(&m_ip->i_gh);
889 gfs2_holder_uninit(&m_ip->i_gh);
890 }
891 gfs2_glock_dq(&ip->i_gh);
892 gfs2_holder_uninit(&ip->i_gh);
893 return ret;
894}
895
896/**
897 * gfs2_set_page_dirty - Page dirtying function
898 * @page: The page to dirty
899 *
900 * Returns: 1 if it dirtyed the page, or 0 otherwise
901 */
902
903static int gfs2_set_page_dirty(struct page *page)
904{
905 SetPageChecked(page);
906 return __set_page_dirty_buffers(page);
907}
908
909/**
910 * gfs2_bmap - Block map function
911 * @mapping: Address space info
912 * @lblock: The block to map
913 *
914 * Returns: The disk address for the block or 0 on hole or error
915 */
916
917static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
918{
919 struct gfs2_inode *ip = GFS2_I(mapping->host);
920 struct gfs2_holder i_gh;
921 sector_t dblock = 0;
922 int error;
923
924 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
925 if (error)
926 return 0;
927
928 if (!gfs2_is_stuffed(ip))
929 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
930
931 gfs2_glock_dq_uninit(&i_gh);
932
933 return dblock;
934}
935
936static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
937{
938 struct gfs2_bufdata *bd;
939
940 lock_buffer(bh);
941 gfs2_log_lock(sdp);
942 clear_buffer_dirty(bh);
943 bd = bh->b_private;
944 if (bd) {
945 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
946 list_del_init(&bd->bd_list);
947 else
948 gfs2_remove_from_journal(bh, current->journal_info, 0);
949 }
950 bh->b_bdev = NULL;
951 clear_buffer_mapped(bh);
952 clear_buffer_req(bh);
953 clear_buffer_new(bh);
954 gfs2_log_unlock(sdp);
955 unlock_buffer(bh);
956}
957
958static void gfs2_invalidatepage(struct page *page, unsigned long offset)
959{
960 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
961 struct buffer_head *bh, *head;
962 unsigned long pos = 0;
963
964 BUG_ON(!PageLocked(page));
965 if (offset == 0)
966 ClearPageChecked(page);
967 if (!page_has_buffers(page))
968 goto out;
969
970 bh = head = page_buffers(page);
971 do {
972 if (offset <= pos)
973 gfs2_discard(sdp, bh);
974 pos += bh->b_size;
975 bh = bh->b_this_page;
976 } while (bh != head);
977out:
978 if (offset == 0)
979 try_to_release_page(page, 0);
980}
981
982/**
983 * gfs2_ok_for_dio - check that dio is valid on this file
984 * @ip: The inode
985 * @rw: READ or WRITE
986 * @offset: The offset at which we are reading or writing
987 *
988 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
989 * 1 (to accept the i/o request)
990 */
991static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
992{
993 /*
994 * Should we return an error here? I can't see that O_DIRECT for
995 * a stuffed file makes any sense. For now we'll silently fall
996 * back to buffered I/O
997 */
998 if (gfs2_is_stuffed(ip))
999 return 0;
1000
1001 if (offset >= i_size_read(&ip->i_inode))
1002 return 0;
1003 return 1;
1004}
1005
1006
1007
1008static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1009 const struct iovec *iov, loff_t offset,
1010 unsigned long nr_segs)
1011{
1012 struct file *file = iocb->ki_filp;
1013 struct inode *inode = file->f_mapping->host;
1014 struct gfs2_inode *ip = GFS2_I(inode);
1015 struct gfs2_holder gh;
1016 int rv;
1017
1018 /*
1019 * Deferred lock, even if its a write, since we do no allocation
1020 * on this path. All we need change is atime, and this lock mode
1021 * ensures that other nodes have flushed their buffered read caches
1022 * (i.e. their page cache entries for this inode). We do not,
1023 * unfortunately have the option of only flushing a range like
1024 * the VFS does.
1025 */
1026 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1027 rv = gfs2_glock_nq(&gh);
1028 if (rv)
1029 return rv;
1030 rv = gfs2_ok_for_dio(ip, rw, offset);
1031 if (rv != 1)
1032 goto out; /* dio not valid, fall back to buffered i/o */
1033
1034 rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1035 offset, nr_segs, gfs2_get_block_direct,
1036 NULL, NULL, 0);
1037out:
1038 gfs2_glock_dq_m(1, &gh);
1039 gfs2_holder_uninit(&gh);
1040 return rv;
1041}
1042
1043/**
1044 * gfs2_releasepage - free the metadata associated with a page
1045 * @page: the page that's being released
1046 * @gfp_mask: passed from Linux VFS, ignored by us
1047 *
1048 * Call try_to_free_buffers() if the buffers in this page can be
1049 * released.
1050 *
1051 * Returns: 0
1052 */
1053
1054int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1055{
1056 struct address_space *mapping = page->mapping;
1057 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1058 struct buffer_head *bh, *head;
1059 struct gfs2_bufdata *bd;
1060
1061 if (!page_has_buffers(page))
1062 return 0;
1063
1064 gfs2_log_lock(sdp);
1065 spin_lock(&sdp->sd_ail_lock);
1066 head = bh = page_buffers(page);
1067 do {
1068 if (atomic_read(&bh->b_count))
1069 goto cannot_release;
1070 bd = bh->b_private;
1071 if (bd && bd->bd_ail)
1072 goto cannot_release;
1073 if (buffer_pinned(bh) || buffer_dirty(bh))
1074 goto not_possible;
1075 bh = bh->b_this_page;
1076 } while(bh != head);
1077 spin_unlock(&sdp->sd_ail_lock);
1078 gfs2_log_unlock(sdp);
1079
1080 head = bh = page_buffers(page);
1081 do {
1082 gfs2_log_lock(sdp);
1083 bd = bh->b_private;
1084 if (bd) {
1085 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1086 if (!list_empty(&bd->bd_list)) {
1087 if (!buffer_pinned(bh))
1088 list_del_init(&bd->bd_list);
1089 else
1090 bd = NULL;
1091 }
1092 if (bd)
1093 bd->bd_bh = NULL;
1094 bh->b_private = NULL;
1095 }
1096 gfs2_log_unlock(sdp);
1097 if (bd)
1098 kmem_cache_free(gfs2_bufdata_cachep, bd);
1099
1100 bh = bh->b_this_page;
1101 } while (bh != head);
1102
1103 return try_to_free_buffers(page);
1104
1105not_possible: /* Should never happen */
1106 WARN_ON(buffer_dirty(bh));
1107 WARN_ON(buffer_pinned(bh));
1108cannot_release:
1109 spin_unlock(&sdp->sd_ail_lock);
1110 gfs2_log_unlock(sdp);
1111 return 0;
1112}
1113
1114static const struct address_space_operations gfs2_writeback_aops = {
1115 .writepage = gfs2_writeback_writepage,
1116 .writepages = gfs2_writeback_writepages,
1117 .readpage = gfs2_readpage,
1118 .readpages = gfs2_readpages,
1119 .write_begin = gfs2_write_begin,
1120 .write_end = gfs2_write_end,
1121 .bmap = gfs2_bmap,
1122 .invalidatepage = gfs2_invalidatepage,
1123 .releasepage = gfs2_releasepage,
1124 .direct_IO = gfs2_direct_IO,
1125 .migratepage = buffer_migrate_page,
1126 .is_partially_uptodate = block_is_partially_uptodate,
1127 .error_remove_page = generic_error_remove_page,
1128};
1129
1130static const struct address_space_operations gfs2_ordered_aops = {
1131 .writepage = gfs2_ordered_writepage,
1132 .readpage = gfs2_readpage,
1133 .readpages = gfs2_readpages,
1134 .write_begin = gfs2_write_begin,
1135 .write_end = gfs2_write_end,
1136 .set_page_dirty = gfs2_set_page_dirty,
1137 .bmap = gfs2_bmap,
1138 .invalidatepage = gfs2_invalidatepage,
1139 .releasepage = gfs2_releasepage,
1140 .direct_IO = gfs2_direct_IO,
1141 .migratepage = buffer_migrate_page,
1142 .is_partially_uptodate = block_is_partially_uptodate,
1143 .error_remove_page = generic_error_remove_page,
1144};
1145
1146static const struct address_space_operations gfs2_jdata_aops = {
1147 .writepage = gfs2_jdata_writepage,
1148 .writepages = gfs2_jdata_writepages,
1149 .readpage = gfs2_readpage,
1150 .readpages = gfs2_readpages,
1151 .write_begin = gfs2_write_begin,
1152 .write_end = gfs2_write_end,
1153 .set_page_dirty = gfs2_set_page_dirty,
1154 .bmap = gfs2_bmap,
1155 .invalidatepage = gfs2_invalidatepage,
1156 .releasepage = gfs2_releasepage,
1157 .is_partially_uptodate = block_is_partially_uptodate,
1158 .error_remove_page = generic_error_remove_page,
1159};
1160
1161void gfs2_set_aops(struct inode *inode)
1162{
1163 struct gfs2_inode *ip = GFS2_I(inode);
1164
1165 if (gfs2_is_writeback(ip))
1166 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1167 else if (gfs2_is_ordered(ip))
1168 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1169 else if (gfs2_is_jdata(ip))
1170 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1171 else
1172 BUG();
1173}
1174