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
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
39void 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, KM_USER0);
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, KM_USER0);
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 * @ra_state: The readahead state (or NULL for no readahead)
521 * @buf: The buffer to fill
522 * @pos: The file position
523 * @size: The amount to read
524 *
525 */
526
527int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
528 char *buf, loff_t *pos, unsigned size)
529{
530 struct address_space *mapping = ip->i_inode.i_mapping;
531 unsigned long index = *pos / PAGE_CACHE_SIZE;
532 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
533 unsigned copied = 0;
534 unsigned amt;
535 struct page *page;
536 void *p;
537
538 do {
539 amt = size - copied;
540 if (offset + size > PAGE_CACHE_SIZE)
541 amt = PAGE_CACHE_SIZE - offset;
542 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
543 if (IS_ERR(page))
544 return PTR_ERR(page);
545 p = kmap_atomic(page, KM_USER0);
546 memcpy(buf + copied, p + offset, amt);
547 kunmap_atomic(p, KM_USER0);
548 mark_page_accessed(page);
549 page_cache_release(page);
550 copied += amt;
551 index++;
552 offset = 0;
553 } while(copied < size);
554 (*pos) += size;
555 return size;
556}
557
558/**
559 * gfs2_readpages - Read a bunch of pages at once
560 *
561 * Some notes:
562 * 1. This is only for readahead, so we can simply ignore any things
563 * which are slightly inconvenient (such as locking conflicts between
564 * the page lock and the glock) and return having done no I/O. Its
565 * obviously not something we'd want to do on too regular a basis.
566 * Any I/O we ignore at this time will be done via readpage later.
567 * 2. We don't handle stuffed files here we let readpage do the honours.
568 * 3. mpage_readpages() does most of the heavy lifting in the common case.
569 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
570 */
571
572static int gfs2_readpages(struct file *file, struct address_space *mapping,
573 struct list_head *pages, unsigned nr_pages)
574{
575 struct inode *inode = mapping->host;
576 struct gfs2_inode *ip = GFS2_I(inode);
577 struct gfs2_sbd *sdp = GFS2_SB(inode);
578 struct gfs2_holder gh;
579 int ret;
580
581 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
582 ret = gfs2_glock_nq(&gh);
583 if (unlikely(ret))
584 goto out_uninit;
585 if (!gfs2_is_stuffed(ip))
586 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
587 gfs2_glock_dq(&gh);
588out_uninit:
589 gfs2_holder_uninit(&gh);
590 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
591 ret = -EIO;
592 return ret;
593}
594
595/**
596 * gfs2_write_begin - Begin to write to a file
597 * @file: The file to write to
598 * @mapping: The mapping in which to write
599 * @pos: The file offset at which to start writing
600 * @len: Length of the write
601 * @flags: Various flags
602 * @pagep: Pointer to return the page
603 * @fsdata: Pointer to return fs data (unused by GFS2)
604 *
605 * Returns: errno
606 */
607
608static int gfs2_write_begin(struct file *file, struct address_space *mapping,
609 loff_t pos, unsigned len, unsigned flags,
610 struct page **pagep, void **fsdata)
611{
612 struct gfs2_inode *ip = GFS2_I(mapping->host);
613 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
614 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
615 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
616 int alloc_required;
617 int error = 0;
618 struct gfs2_alloc *al = NULL;
619 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
620 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
621 struct page *page;
622
623 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
624 error = gfs2_glock_nq(&ip->i_gh);
625 if (unlikely(error))
626 goto out_uninit;
627 if (&ip->i_inode == sdp->sd_rindex) {
628 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
629 GL_NOCACHE, &m_ip->i_gh);
630 if (unlikely(error)) {
631 gfs2_glock_dq(&ip->i_gh);
632 goto out_uninit;
633 }
634 }
635
636 alloc_required = gfs2_write_alloc_required(ip, pos, len);
637
638 if (alloc_required || gfs2_is_jdata(ip))
639 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
640
641 if (alloc_required) {
642 al = gfs2_alloc_get(ip);
643 if (!al) {
644 error = -ENOMEM;
645 goto out_unlock;
646 }
647
648 error = gfs2_quota_lock_check(ip);
649 if (error)
650 goto out_alloc_put;
651
652 al->al_requested = data_blocks + ind_blocks;
653 error = gfs2_inplace_reserve(ip);
654 if (error)
655 goto out_qunlock;
656 }
657
658 rblocks = RES_DINODE + ind_blocks;
659 if (gfs2_is_jdata(ip))
660 rblocks += data_blocks ? data_blocks : 1;
661 if (ind_blocks || data_blocks)
662 rblocks += RES_STATFS + RES_QUOTA;
663 if (&ip->i_inode == sdp->sd_rindex)
664 rblocks += 2 * RES_STATFS;
665 if (alloc_required)
666 rblocks += gfs2_rg_blocks(al);
667
668 error = gfs2_trans_begin(sdp, rblocks,
669 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
670 if (error)
671 goto out_trans_fail;
672
673 error = -ENOMEM;
674 flags |= AOP_FLAG_NOFS;
675 page = grab_cache_page_write_begin(mapping, index, flags);
676 *pagep = page;
677 if (unlikely(!page))
678 goto out_endtrans;
679
680 if (gfs2_is_stuffed(ip)) {
681 error = 0;
682 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
683 error = gfs2_unstuff_dinode(ip, page);
684 if (error == 0)
685 goto prepare_write;
686 } else if (!PageUptodate(page)) {
687 error = stuffed_readpage(ip, page);
688 }
689 goto out;
690 }
691
692prepare_write:
693 error = __block_write_begin(page, from, len, gfs2_block_map);
694out:
695 if (error == 0)
696 return 0;
697
698 unlock_page(page);
699 page_cache_release(page);
700
701 gfs2_trans_end(sdp);
702 if (pos + len > ip->i_inode.i_size)
703 gfs2_trim_blocks(&ip->i_inode);
704 goto out_trans_fail;
705
706out_endtrans:
707 gfs2_trans_end(sdp);
708out_trans_fail:
709 if (alloc_required) {
710 gfs2_inplace_release(ip);
711out_qunlock:
712 gfs2_quota_unlock(ip);
713out_alloc_put:
714 gfs2_alloc_put(ip);
715 }
716out_unlock:
717 if (&ip->i_inode == sdp->sd_rindex) {
718 gfs2_glock_dq(&m_ip->i_gh);
719 gfs2_holder_uninit(&m_ip->i_gh);
720 }
721 gfs2_glock_dq(&ip->i_gh);
722out_uninit:
723 gfs2_holder_uninit(&ip->i_gh);
724 return error;
725}
726
727/**
728 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
729 * @inode: the rindex inode
730 */
731static void adjust_fs_space(struct inode *inode)
732{
733 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
734 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
735 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
736 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
737 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
738 struct buffer_head *m_bh, *l_bh;
739 u64 fs_total, new_free;
740
741 /* Total up the file system space, according to the latest rindex. */
742 fs_total = gfs2_ri_total(sdp);
743 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
744 return;
745
746 spin_lock(&sdp->sd_statfs_spin);
747 gfs2_statfs_change_in(m_sc, m_bh->b_data +
748 sizeof(struct gfs2_dinode));
749 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
750 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
751 else
752 new_free = 0;
753 spin_unlock(&sdp->sd_statfs_spin);
754 fs_warn(sdp, "File system extended by %llu blocks.\n",
755 (unsigned long long)new_free);
756 gfs2_statfs_change(sdp, new_free, new_free, 0);
757
758 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
759 goto out;
760 update_statfs(sdp, m_bh, l_bh);
761 brelse(l_bh);
762out:
763 brelse(m_bh);
764}
765
766/**
767 * gfs2_stuffed_write_end - Write end for stuffed files
768 * @inode: The inode
769 * @dibh: The buffer_head containing the on-disk inode
770 * @pos: The file position
771 * @len: The length of the write
772 * @copied: How much was actually copied by the VFS
773 * @page: The page
774 *
775 * This copies the data from the page into the inode block after
776 * the inode data structure itself.
777 *
778 * Returns: errno
779 */
780static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
781 loff_t pos, unsigned len, unsigned copied,
782 struct page *page)
783{
784 struct gfs2_inode *ip = GFS2_I(inode);
785 struct gfs2_sbd *sdp = GFS2_SB(inode);
786 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
787 u64 to = pos + copied;
788 void *kaddr;
789 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
790 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
791
792 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
793 kaddr = kmap_atomic(page, KM_USER0);
794 memcpy(buf + pos, kaddr + pos, copied);
795 memset(kaddr + pos + copied, 0, len - copied);
796 flush_dcache_page(page);
797 kunmap_atomic(kaddr, KM_USER0);
798
799 if (!PageUptodate(page))
800 SetPageUptodate(page);
801 unlock_page(page);
802 page_cache_release(page);
803
804 if (copied) {
805 if (inode->i_size < to)
806 i_size_write(inode, to);
807 gfs2_dinode_out(ip, di);
808 mark_inode_dirty(inode);
809 }
810
811 if (inode == sdp->sd_rindex) {
812 adjust_fs_space(inode);
813 ip->i_gh.gh_flags |= GL_NOCACHE;
814 }
815
816 brelse(dibh);
817 gfs2_trans_end(sdp);
818 if (inode == sdp->sd_rindex) {
819 gfs2_glock_dq(&m_ip->i_gh);
820 gfs2_holder_uninit(&m_ip->i_gh);
821 }
822 gfs2_glock_dq(&ip->i_gh);
823 gfs2_holder_uninit(&ip->i_gh);
824 return copied;
825}
826
827/**
828 * gfs2_write_end
829 * @file: The file to write to
830 * @mapping: The address space to write to
831 * @pos: The file position
832 * @len: The length of the data
833 * @copied:
834 * @page: The page that has been written
835 * @fsdata: The fsdata (unused in GFS2)
836 *
837 * The main write_end function for GFS2. We have a separate one for
838 * stuffed files as they are slightly different, otherwise we just
839 * put our locking around the VFS provided functions.
840 *
841 * Returns: errno
842 */
843
844static int gfs2_write_end(struct file *file, struct address_space *mapping,
845 loff_t pos, unsigned len, unsigned copied,
846 struct page *page, void *fsdata)
847{
848 struct inode *inode = page->mapping->host;
849 struct gfs2_inode *ip = GFS2_I(inode);
850 struct gfs2_sbd *sdp = GFS2_SB(inode);
851 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
852 struct buffer_head *dibh;
853 struct gfs2_alloc *al = ip->i_alloc;
854 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
855 unsigned int to = from + len;
856 int ret;
857
858 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
859
860 ret = gfs2_meta_inode_buffer(ip, &dibh);
861 if (unlikely(ret)) {
862 unlock_page(page);
863 page_cache_release(page);
864 goto failed;
865 }
866
867 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
868
869 if (gfs2_is_stuffed(ip))
870 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
871
872 if (!gfs2_is_writeback(ip))
873 gfs2_page_add_databufs(ip, page, from, to);
874
875 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
876 if (ret > 0) {
877 gfs2_dinode_out(ip, dibh->b_data);
878 mark_inode_dirty(inode);
879 }
880
881 if (inode == sdp->sd_rindex) {
882 adjust_fs_space(inode);
883 ip->i_gh.gh_flags |= GL_NOCACHE;
884 }
885
886 brelse(dibh);
887failed:
888 gfs2_trans_end(sdp);
889 if (al) {
890 gfs2_inplace_release(ip);
891 gfs2_quota_unlock(ip);
892 gfs2_alloc_put(ip);
893 }
894 if (inode == sdp->sd_rindex) {
895 gfs2_glock_dq(&m_ip->i_gh);
896 gfs2_holder_uninit(&m_ip->i_gh);
897 }
898 gfs2_glock_dq(&ip->i_gh);
899 gfs2_holder_uninit(&ip->i_gh);
900 return ret;
901}
902
903/**
904 * gfs2_set_page_dirty - Page dirtying function
905 * @page: The page to dirty
906 *
907 * Returns: 1 if it dirtyed the page, or 0 otherwise
908 */
909
910static int gfs2_set_page_dirty(struct page *page)
911{
912 SetPageChecked(page);
913 return __set_page_dirty_buffers(page);
914}
915
916/**
917 * gfs2_bmap - Block map function
918 * @mapping: Address space info
919 * @lblock: The block to map
920 *
921 * Returns: The disk address for the block or 0 on hole or error
922 */
923
924static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
925{
926 struct gfs2_inode *ip = GFS2_I(mapping->host);
927 struct gfs2_holder i_gh;
928 sector_t dblock = 0;
929 int error;
930
931 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
932 if (error)
933 return 0;
934
935 if (!gfs2_is_stuffed(ip))
936 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
937
938 gfs2_glock_dq_uninit(&i_gh);
939
940 return dblock;
941}
942
943static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
944{
945 struct gfs2_bufdata *bd;
946
947 lock_buffer(bh);
948 gfs2_log_lock(sdp);
949 clear_buffer_dirty(bh);
950 bd = bh->b_private;
951 if (bd) {
952 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
953 list_del_init(&bd->bd_le.le_list);
954 else
955 gfs2_remove_from_journal(bh, current->journal_info, 0);
956 }
957 bh->b_bdev = NULL;
958 clear_buffer_mapped(bh);
959 clear_buffer_req(bh);
960 clear_buffer_new(bh);
961 gfs2_log_unlock(sdp);
962 unlock_buffer(bh);
963}
964
965static void gfs2_invalidatepage(struct page *page, unsigned long offset)
966{
967 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
968 struct buffer_head *bh, *head;
969 unsigned long pos = 0;
970
971 BUG_ON(!PageLocked(page));
972 if (offset == 0)
973 ClearPageChecked(page);
974 if (!page_has_buffers(page))
975 goto out;
976
977 bh = head = page_buffers(page);
978 do {
979 if (offset <= pos)
980 gfs2_discard(sdp, bh);
981 pos += bh->b_size;
982 bh = bh->b_this_page;
983 } while (bh != head);
984out:
985 if (offset == 0)
986 try_to_release_page(page, 0);
987}
988
989/**
990 * gfs2_ok_for_dio - check that dio is valid on this file
991 * @ip: The inode
992 * @rw: READ or WRITE
993 * @offset: The offset at which we are reading or writing
994 *
995 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
996 * 1 (to accept the i/o request)
997 */
998static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
999{
1000 /*
1001 * Should we return an error here? I can't see that O_DIRECT for
1002 * a stuffed file makes any sense. For now we'll silently fall
1003 * back to buffered I/O
1004 */
1005 if (gfs2_is_stuffed(ip))
1006 return 0;
1007
1008 if (offset >= i_size_read(&ip->i_inode))
1009 return 0;
1010 return 1;
1011}
1012
1013
1014
1015static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1016 const struct iovec *iov, loff_t offset,
1017 unsigned long nr_segs)
1018{
1019 struct file *file = iocb->ki_filp;
1020 struct inode *inode = file->f_mapping->host;
1021 struct gfs2_inode *ip = GFS2_I(inode);
1022 struct gfs2_holder gh;
1023 int rv;
1024
1025 /*
1026 * Deferred lock, even if its a write, since we do no allocation
1027 * on this path. All we need change is atime, and this lock mode
1028 * ensures that other nodes have flushed their buffered read caches
1029 * (i.e. their page cache entries for this inode). We do not,
1030 * unfortunately have the option of only flushing a range like
1031 * the VFS does.
1032 */
1033 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1034 rv = gfs2_glock_nq(&gh);
1035 if (rv)
1036 return rv;
1037 rv = gfs2_ok_for_dio(ip, rw, offset);
1038 if (rv != 1)
1039 goto out; /* dio not valid, fall back to buffered i/o */
1040
1041 rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1042 offset, nr_segs, gfs2_get_block_direct,
1043 NULL, NULL, 0);
1044out:
1045 gfs2_glock_dq_m(1, &gh);
1046 gfs2_holder_uninit(&gh);
1047 return rv;
1048}
1049
1050/**
1051 * gfs2_releasepage - free the metadata associated with a page
1052 * @page: the page that's being released
1053 * @gfp_mask: passed from Linux VFS, ignored by us
1054 *
1055 * Call try_to_free_buffers() if the buffers in this page can be
1056 * released.
1057 *
1058 * Returns: 0
1059 */
1060
1061int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1062{
1063 struct address_space *mapping = page->mapping;
1064 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1065 struct buffer_head *bh, *head;
1066 struct gfs2_bufdata *bd;
1067
1068 if (!page_has_buffers(page))
1069 return 0;
1070
1071 gfs2_log_lock(sdp);
1072 spin_lock(&sdp->sd_ail_lock);
1073 head = bh = page_buffers(page);
1074 do {
1075 if (atomic_read(&bh->b_count))
1076 goto cannot_release;
1077 bd = bh->b_private;
1078 if (bd && bd->bd_ail)
1079 goto cannot_release;
1080 if (buffer_pinned(bh) || buffer_dirty(bh))
1081 goto not_possible;
1082 bh = bh->b_this_page;
1083 } while(bh != head);
1084 spin_unlock(&sdp->sd_ail_lock);
1085 gfs2_log_unlock(sdp);
1086
1087 head = bh = page_buffers(page);
1088 do {
1089 gfs2_log_lock(sdp);
1090 bd = bh->b_private;
1091 if (bd) {
1092 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1093 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
1094 if (!list_empty(&bd->bd_le.le_list)) {
1095 if (!buffer_pinned(bh))
1096 list_del_init(&bd->bd_le.le_list);
1097 else
1098 bd = NULL;
1099 }
1100 if (bd)
1101 bd->bd_bh = NULL;
1102 bh->b_private = NULL;
1103 }
1104 gfs2_log_unlock(sdp);
1105 if (bd)
1106 kmem_cache_free(gfs2_bufdata_cachep, bd);
1107
1108 bh = bh->b_this_page;
1109 } while (bh != head);
1110
1111 return try_to_free_buffers(page);
1112
1113not_possible: /* Should never happen */
1114 WARN_ON(buffer_dirty(bh));
1115 WARN_ON(buffer_pinned(bh));
1116cannot_release:
1117 spin_unlock(&sdp->sd_ail_lock);
1118 gfs2_log_unlock(sdp);
1119 return 0;
1120}
1121
1122static const struct address_space_operations gfs2_writeback_aops = {
1123 .writepage = gfs2_writeback_writepage,
1124 .writepages = gfs2_writeback_writepages,
1125 .readpage = gfs2_readpage,
1126 .readpages = gfs2_readpages,
1127 .write_begin = gfs2_write_begin,
1128 .write_end = gfs2_write_end,
1129 .bmap = gfs2_bmap,
1130 .invalidatepage = gfs2_invalidatepage,
1131 .releasepage = gfs2_releasepage,
1132 .direct_IO = gfs2_direct_IO,
1133 .migratepage = buffer_migrate_page,
1134 .is_partially_uptodate = block_is_partially_uptodate,
1135 .error_remove_page = generic_error_remove_page,
1136};
1137
1138static const struct address_space_operations gfs2_ordered_aops = {
1139 .writepage = gfs2_ordered_writepage,
1140 .readpage = gfs2_readpage,
1141 .readpages = gfs2_readpages,
1142 .write_begin = gfs2_write_begin,
1143 .write_end = gfs2_write_end,
1144 .set_page_dirty = gfs2_set_page_dirty,
1145 .bmap = gfs2_bmap,
1146 .invalidatepage = gfs2_invalidatepage,
1147 .releasepage = gfs2_releasepage,
1148 .direct_IO = gfs2_direct_IO,
1149 .migratepage = buffer_migrate_page,
1150 .is_partially_uptodate = block_is_partially_uptodate,
1151 .error_remove_page = generic_error_remove_page,
1152};
1153
1154static const struct address_space_operations gfs2_jdata_aops = {
1155 .writepage = gfs2_jdata_writepage,
1156 .writepages = gfs2_jdata_writepages,
1157 .readpage = gfs2_readpage,
1158 .readpages = gfs2_readpages,
1159 .write_begin = gfs2_write_begin,
1160 .write_end = gfs2_write_end,
1161 .set_page_dirty = gfs2_set_page_dirty,
1162 .bmap = gfs2_bmap,
1163 .invalidatepage = gfs2_invalidatepage,
1164 .releasepage = gfs2_releasepage,
1165 .is_partially_uptodate = block_is_partially_uptodate,
1166 .error_remove_page = generic_error_remove_page,
1167};
1168
1169void gfs2_set_aops(struct inode *inode)
1170{
1171 struct gfs2_inode *ip = GFS2_I(inode);
1172
1173 if (gfs2_is_writeback(ip))
1174 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1175 else if (gfs2_is_ordered(ip))
1176 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1177 else if (gfs2_is_jdata(ip))
1178 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1179 else
1180 BUG();
1181}
1182
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/uio.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_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_SIZE-1);
113 if (page->index > end_index || (page->index == end_index && !offset)) {
114 page->mapping->a_ops->invalidatepage(page, 0, PAGE_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/* This is the same as calling block_write_full_page, but it also
144 * writes pages outside of i_size
145 */
146int gfs2_write_full_page(struct page *page, get_block_t *get_block,
147 struct writeback_control *wbc)
148{
149 struct inode * const inode = page->mapping->host;
150 loff_t i_size = i_size_read(inode);
151 const pgoff_t end_index = i_size >> PAGE_SHIFT;
152 unsigned offset;
153
154 /*
155 * The page straddles i_size. It must be zeroed out on each and every
156 * writepage invocation because it may be mmapped. "A file is mapped
157 * in multiples of the page size. For a file that is not a multiple of
158 * the page size, the remaining memory is zeroed when mapped, and
159 * writes to that region are not written out to the file."
160 */
161 offset = i_size & (PAGE_SIZE-1);
162 if (page->index == end_index && offset)
163 zero_user_segment(page, offset, PAGE_SIZE);
164
165 return __block_write_full_page(inode, page, get_block, wbc,
166 end_buffer_async_write);
167}
168
169/**
170 * __gfs2_jdata_writepage - The core of jdata writepage
171 * @page: The page to write
172 * @wbc: The writeback control
173 *
174 * This is shared between writepage and writepages and implements the
175 * core of the writepage operation. If a transaction is required then
176 * PageChecked will have been set and the transaction will have
177 * already been started before this is called.
178 */
179
180static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
181{
182 struct inode *inode = page->mapping->host;
183 struct gfs2_inode *ip = GFS2_I(inode);
184 struct gfs2_sbd *sdp = GFS2_SB(inode);
185
186 if (PageChecked(page)) {
187 ClearPageChecked(page);
188 if (!page_has_buffers(page)) {
189 create_empty_buffers(page, inode->i_sb->s_blocksize,
190 BIT(BH_Dirty)|BIT(BH_Uptodate));
191 }
192 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
193 }
194 return gfs2_write_full_page(page, gfs2_get_block_noalloc, wbc);
195}
196
197/**
198 * gfs2_jdata_writepage - Write complete page
199 * @page: Page to write
200 * @wbc: The writeback control
201 *
202 * Returns: errno
203 *
204 */
205
206static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
207{
208 struct inode *inode = page->mapping->host;
209 struct gfs2_inode *ip = GFS2_I(inode);
210 struct gfs2_sbd *sdp = GFS2_SB(inode);
211 int ret;
212
213 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
214 goto out;
215 if (PageChecked(page) || current->journal_info)
216 goto out_ignore;
217 ret = __gfs2_jdata_writepage(page, wbc);
218 return ret;
219
220out_ignore:
221 redirty_page_for_writepage(wbc, page);
222out:
223 unlock_page(page);
224 return 0;
225}
226
227/**
228 * gfs2_writepages - Write a bunch of dirty pages back to disk
229 * @mapping: The mapping to write
230 * @wbc: Write-back control
231 *
232 * Used for both ordered and writeback modes.
233 */
234static int gfs2_writepages(struct address_space *mapping,
235 struct writeback_control *wbc)
236{
237 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
238}
239
240/**
241 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
242 * @mapping: The mapping
243 * @wbc: The writeback control
244 * @pvec: The vector of pages
245 * @nr_pages: The number of pages to write
246 * @end: End position
247 * @done_index: Page index
248 *
249 * Returns: non-zero if loop should terminate, zero otherwise
250 */
251
252static int gfs2_write_jdata_pagevec(struct address_space *mapping,
253 struct writeback_control *wbc,
254 struct pagevec *pvec,
255 int nr_pages, pgoff_t end,
256 pgoff_t *done_index)
257{
258 struct inode *inode = mapping->host;
259 struct gfs2_sbd *sdp = GFS2_SB(inode);
260 unsigned nrblocks = nr_pages * (PAGE_SIZE/inode->i_sb->s_blocksize);
261 int i;
262 int ret;
263
264 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
265 if (ret < 0)
266 return ret;
267
268 for(i = 0; i < nr_pages; i++) {
269 struct page *page = pvec->pages[i];
270
271 /*
272 * At this point, the page may be truncated or
273 * invalidated (changing page->mapping to NULL), or
274 * even swizzled back from swapper_space to tmpfs file
275 * mapping. However, page->index will not change
276 * because we have a reference on the page.
277 */
278 if (page->index > end) {
279 /*
280 * can't be range_cyclic (1st pass) because
281 * end == -1 in that case.
282 */
283 ret = 1;
284 break;
285 }
286
287 *done_index = page->index;
288
289 lock_page(page);
290
291 if (unlikely(page->mapping != mapping)) {
292continue_unlock:
293 unlock_page(page);
294 continue;
295 }
296
297 if (!PageDirty(page)) {
298 /* someone wrote it for us */
299 goto continue_unlock;
300 }
301
302 if (PageWriteback(page)) {
303 if (wbc->sync_mode != WB_SYNC_NONE)
304 wait_on_page_writeback(page);
305 else
306 goto continue_unlock;
307 }
308
309 BUG_ON(PageWriteback(page));
310 if (!clear_page_dirty_for_io(page))
311 goto continue_unlock;
312
313 trace_wbc_writepage(wbc, inode_to_bdi(inode));
314
315 ret = __gfs2_jdata_writepage(page, wbc);
316 if (unlikely(ret)) {
317 if (ret == AOP_WRITEPAGE_ACTIVATE) {
318 unlock_page(page);
319 ret = 0;
320 } else {
321
322 /*
323 * done_index is set past this page,
324 * so media errors will not choke
325 * background writeout for the entire
326 * file. This has consequences for
327 * range_cyclic semantics (ie. it may
328 * not be suitable for data integrity
329 * writeout).
330 */
331 *done_index = page->index + 1;
332 ret = 1;
333 break;
334 }
335 }
336
337 /*
338 * We stop writing back only if we are not doing
339 * integrity sync. In case of integrity sync we have to
340 * keep going until we have written all the pages
341 * we tagged for writeback prior to entering this loop.
342 */
343 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
344 ret = 1;
345 break;
346 }
347
348 }
349 gfs2_trans_end(sdp);
350 return ret;
351}
352
353/**
354 * gfs2_write_cache_jdata - Like write_cache_pages but different
355 * @mapping: The mapping to write
356 * @wbc: The writeback control
357 *
358 * The reason that we use our own function here is that we need to
359 * start transactions before we grab page locks. This allows us
360 * to get the ordering right.
361 */
362
363static int gfs2_write_cache_jdata(struct address_space *mapping,
364 struct writeback_control *wbc)
365{
366 int ret = 0;
367 int done = 0;
368 struct pagevec pvec;
369 int nr_pages;
370 pgoff_t uninitialized_var(writeback_index);
371 pgoff_t index;
372 pgoff_t end;
373 pgoff_t done_index;
374 int cycled;
375 int range_whole = 0;
376 int tag;
377
378 pagevec_init(&pvec, 0);
379 if (wbc->range_cyclic) {
380 writeback_index = mapping->writeback_index; /* prev offset */
381 index = writeback_index;
382 if (index == 0)
383 cycled = 1;
384 else
385 cycled = 0;
386 end = -1;
387 } else {
388 index = wbc->range_start >> PAGE_SHIFT;
389 end = wbc->range_end >> PAGE_SHIFT;
390 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
391 range_whole = 1;
392 cycled = 1; /* ignore range_cyclic tests */
393 }
394 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
395 tag = PAGECACHE_TAG_TOWRITE;
396 else
397 tag = PAGECACHE_TAG_DIRTY;
398
399retry:
400 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
401 tag_pages_for_writeback(mapping, index, end);
402 done_index = index;
403 while (!done && (index <= end)) {
404 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
405 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
406 if (nr_pages == 0)
407 break;
408
409 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end, &done_index);
410 if (ret)
411 done = 1;
412 if (ret > 0)
413 ret = 0;
414 pagevec_release(&pvec);
415 cond_resched();
416 }
417
418 if (!cycled && !done) {
419 /*
420 * range_cyclic:
421 * We hit the last page and there is more work to be done: wrap
422 * back to the start of the file
423 */
424 cycled = 1;
425 index = 0;
426 end = writeback_index - 1;
427 goto retry;
428 }
429
430 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
431 mapping->writeback_index = done_index;
432
433 return ret;
434}
435
436
437/**
438 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
439 * @mapping: The mapping to write
440 * @wbc: The writeback control
441 *
442 */
443
444static int gfs2_jdata_writepages(struct address_space *mapping,
445 struct writeback_control *wbc)
446{
447 struct gfs2_inode *ip = GFS2_I(mapping->host);
448 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
449 int ret;
450
451 ret = gfs2_write_cache_jdata(mapping, wbc);
452 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
453 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
454 ret = gfs2_write_cache_jdata(mapping, wbc);
455 }
456 return ret;
457}
458
459/**
460 * stuffed_readpage - Fill in a Linux page with stuffed file data
461 * @ip: the inode
462 * @page: the page
463 *
464 * Returns: errno
465 */
466
467static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
468{
469 struct buffer_head *dibh;
470 u64 dsize = i_size_read(&ip->i_inode);
471 void *kaddr;
472 int error;
473
474 /*
475 * Due to the order of unstuffing files and ->fault(), we can be
476 * asked for a zero page in the case of a stuffed file being extended,
477 * so we need to supply one here. It doesn't happen often.
478 */
479 if (unlikely(page->index)) {
480 zero_user(page, 0, PAGE_SIZE);
481 SetPageUptodate(page);
482 return 0;
483 }
484
485 error = gfs2_meta_inode_buffer(ip, &dibh);
486 if (error)
487 return error;
488
489 kaddr = kmap_atomic(page);
490 if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
491 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
492 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
493 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
494 kunmap_atomic(kaddr);
495 flush_dcache_page(page);
496 brelse(dibh);
497 SetPageUptodate(page);
498
499 return 0;
500}
501
502
503/**
504 * __gfs2_readpage - readpage
505 * @file: The file to read a page for
506 * @page: The page to read
507 *
508 * This is the core of gfs2's readpage. Its used by the internal file
509 * reading code as in that case we already hold the glock. Also its
510 * called by gfs2_readpage() once the required lock has been granted.
511 *
512 */
513
514static int __gfs2_readpage(void *file, struct page *page)
515{
516 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
517 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
518 int error;
519
520 if (gfs2_is_stuffed(ip)) {
521 error = stuffed_readpage(ip, page);
522 unlock_page(page);
523 } else {
524 error = mpage_readpage(page, gfs2_block_map);
525 }
526
527 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
528 return -EIO;
529
530 return error;
531}
532
533/**
534 * gfs2_readpage - read a page of a file
535 * @file: The file to read
536 * @page: The page of the file
537 *
538 * This deals with the locking required. We have to unlock and
539 * relock the page in order to get the locking in the right
540 * order.
541 */
542
543static int gfs2_readpage(struct file *file, struct page *page)
544{
545 struct address_space *mapping = page->mapping;
546 struct gfs2_inode *ip = GFS2_I(mapping->host);
547 struct gfs2_holder gh;
548 int error;
549
550 unlock_page(page);
551 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
552 error = gfs2_glock_nq(&gh);
553 if (unlikely(error))
554 goto out;
555 error = AOP_TRUNCATED_PAGE;
556 lock_page(page);
557 if (page->mapping == mapping && !PageUptodate(page))
558 error = __gfs2_readpage(file, page);
559 else
560 unlock_page(page);
561 gfs2_glock_dq(&gh);
562out:
563 gfs2_holder_uninit(&gh);
564 if (error && error != AOP_TRUNCATED_PAGE)
565 lock_page(page);
566 return error;
567}
568
569/**
570 * gfs2_internal_read - read an internal file
571 * @ip: The gfs2 inode
572 * @buf: The buffer to fill
573 * @pos: The file position
574 * @size: The amount to read
575 *
576 */
577
578int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
579 unsigned size)
580{
581 struct address_space *mapping = ip->i_inode.i_mapping;
582 unsigned long index = *pos / PAGE_SIZE;
583 unsigned offset = *pos & (PAGE_SIZE - 1);
584 unsigned copied = 0;
585 unsigned amt;
586 struct page *page;
587 void *p;
588
589 do {
590 amt = size - copied;
591 if (offset + size > PAGE_SIZE)
592 amt = PAGE_SIZE - offset;
593 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
594 if (IS_ERR(page))
595 return PTR_ERR(page);
596 p = kmap_atomic(page);
597 memcpy(buf + copied, p + offset, amt);
598 kunmap_atomic(p);
599 put_page(page);
600 copied += amt;
601 index++;
602 offset = 0;
603 } while(copied < size);
604 (*pos) += size;
605 return size;
606}
607
608/**
609 * gfs2_readpages - Read a bunch of pages at once
610 * @file: The file to read from
611 * @mapping: Address space info
612 * @pages: List of pages to read
613 * @nr_pages: Number of pages to read
614 *
615 * Some notes:
616 * 1. This is only for readahead, so we can simply ignore any things
617 * which are slightly inconvenient (such as locking conflicts between
618 * the page lock and the glock) and return having done no I/O. Its
619 * obviously not something we'd want to do on too regular a basis.
620 * Any I/O we ignore at this time will be done via readpage later.
621 * 2. We don't handle stuffed files here we let readpage do the honours.
622 * 3. mpage_readpages() does most of the heavy lifting in the common case.
623 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
624 */
625
626static int gfs2_readpages(struct file *file, struct address_space *mapping,
627 struct list_head *pages, unsigned nr_pages)
628{
629 struct inode *inode = mapping->host;
630 struct gfs2_inode *ip = GFS2_I(inode);
631 struct gfs2_sbd *sdp = GFS2_SB(inode);
632 struct gfs2_holder gh;
633 int ret;
634
635 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
636 ret = gfs2_glock_nq(&gh);
637 if (unlikely(ret))
638 goto out_uninit;
639 if (!gfs2_is_stuffed(ip))
640 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
641 gfs2_glock_dq(&gh);
642out_uninit:
643 gfs2_holder_uninit(&gh);
644 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
645 ret = -EIO;
646 return ret;
647}
648
649/**
650 * gfs2_write_begin - Begin to write to a file
651 * @file: The file to write to
652 * @mapping: The mapping in which to write
653 * @pos: The file offset at which to start writing
654 * @len: Length of the write
655 * @flags: Various flags
656 * @pagep: Pointer to return the page
657 * @fsdata: Pointer to return fs data (unused by GFS2)
658 *
659 * Returns: errno
660 */
661
662static int gfs2_write_begin(struct file *file, struct address_space *mapping,
663 loff_t pos, unsigned len, unsigned flags,
664 struct page **pagep, void **fsdata)
665{
666 struct gfs2_inode *ip = GFS2_I(mapping->host);
667 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
668 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
669 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
670 unsigned requested = 0;
671 int alloc_required;
672 int error = 0;
673 pgoff_t index = pos >> PAGE_SHIFT;
674 unsigned from = pos & (PAGE_SIZE - 1);
675 struct page *page;
676
677 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
678 error = gfs2_glock_nq(&ip->i_gh);
679 if (unlikely(error))
680 goto out_uninit;
681 if (&ip->i_inode == sdp->sd_rindex) {
682 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
683 GL_NOCACHE, &m_ip->i_gh);
684 if (unlikely(error)) {
685 gfs2_glock_dq(&ip->i_gh);
686 goto out_uninit;
687 }
688 }
689
690 alloc_required = gfs2_write_alloc_required(ip, pos, len);
691
692 if (alloc_required || gfs2_is_jdata(ip))
693 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
694
695 if (alloc_required) {
696 struct gfs2_alloc_parms ap = { .aflags = 0, };
697 requested = data_blocks + ind_blocks;
698 ap.target = requested;
699 error = gfs2_quota_lock_check(ip, &ap);
700 if (error)
701 goto out_unlock;
702
703 error = gfs2_inplace_reserve(ip, &ap);
704 if (error)
705 goto out_qunlock;
706 }
707
708 rblocks = RES_DINODE + ind_blocks;
709 if (gfs2_is_jdata(ip))
710 rblocks += data_blocks ? data_blocks : 1;
711 if (ind_blocks || data_blocks)
712 rblocks += RES_STATFS + RES_QUOTA;
713 if (&ip->i_inode == sdp->sd_rindex)
714 rblocks += 2 * RES_STATFS;
715 if (alloc_required)
716 rblocks += gfs2_rg_blocks(ip, requested);
717
718 error = gfs2_trans_begin(sdp, rblocks,
719 PAGE_SIZE/sdp->sd_sb.sb_bsize);
720 if (error)
721 goto out_trans_fail;
722
723 error = -ENOMEM;
724 flags |= AOP_FLAG_NOFS;
725 page = grab_cache_page_write_begin(mapping, index, flags);
726 *pagep = page;
727 if (unlikely(!page))
728 goto out_endtrans;
729
730 if (gfs2_is_stuffed(ip)) {
731 error = 0;
732 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
733 error = gfs2_unstuff_dinode(ip, page);
734 if (error == 0)
735 goto prepare_write;
736 } else if (!PageUptodate(page)) {
737 error = stuffed_readpage(ip, page);
738 }
739 goto out;
740 }
741
742prepare_write:
743 error = __block_write_begin(page, from, len, gfs2_block_map);
744out:
745 if (error == 0)
746 return 0;
747
748 unlock_page(page);
749 put_page(page);
750
751 gfs2_trans_end(sdp);
752 if (pos + len > ip->i_inode.i_size)
753 gfs2_trim_blocks(&ip->i_inode);
754 goto out_trans_fail;
755
756out_endtrans:
757 gfs2_trans_end(sdp);
758out_trans_fail:
759 if (alloc_required) {
760 gfs2_inplace_release(ip);
761out_qunlock:
762 gfs2_quota_unlock(ip);
763 }
764out_unlock:
765 if (&ip->i_inode == sdp->sd_rindex) {
766 gfs2_glock_dq(&m_ip->i_gh);
767 gfs2_holder_uninit(&m_ip->i_gh);
768 }
769 gfs2_glock_dq(&ip->i_gh);
770out_uninit:
771 gfs2_holder_uninit(&ip->i_gh);
772 return error;
773}
774
775/**
776 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
777 * @inode: the rindex inode
778 */
779static void adjust_fs_space(struct inode *inode)
780{
781 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
782 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
783 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
784 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
785 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
786 struct buffer_head *m_bh, *l_bh;
787 u64 fs_total, new_free;
788
789 /* Total up the file system space, according to the latest rindex. */
790 fs_total = gfs2_ri_total(sdp);
791 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
792 return;
793
794 spin_lock(&sdp->sd_statfs_spin);
795 gfs2_statfs_change_in(m_sc, m_bh->b_data +
796 sizeof(struct gfs2_dinode));
797 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
798 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
799 else
800 new_free = 0;
801 spin_unlock(&sdp->sd_statfs_spin);
802 fs_warn(sdp, "File system extended by %llu blocks.\n",
803 (unsigned long long)new_free);
804 gfs2_statfs_change(sdp, new_free, new_free, 0);
805
806 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
807 goto out;
808 update_statfs(sdp, m_bh, l_bh);
809 brelse(l_bh);
810out:
811 brelse(m_bh);
812}
813
814/**
815 * gfs2_stuffed_write_end - Write end for stuffed files
816 * @inode: The inode
817 * @dibh: The buffer_head containing the on-disk inode
818 * @pos: The file position
819 * @len: The length of the write
820 * @copied: How much was actually copied by the VFS
821 * @page: The page
822 *
823 * This copies the data from the page into the inode block after
824 * the inode data structure itself.
825 *
826 * Returns: errno
827 */
828static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
829 loff_t pos, unsigned len, unsigned copied,
830 struct page *page)
831{
832 struct gfs2_inode *ip = GFS2_I(inode);
833 struct gfs2_sbd *sdp = GFS2_SB(inode);
834 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
835 u64 to = pos + copied;
836 void *kaddr;
837 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
838
839 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
840 kaddr = kmap_atomic(page);
841 memcpy(buf + pos, kaddr + pos, copied);
842 flush_dcache_page(page);
843 kunmap_atomic(kaddr);
844
845 WARN_ON(!PageUptodate(page));
846 unlock_page(page);
847 put_page(page);
848
849 if (copied) {
850 if (inode->i_size < to)
851 i_size_write(inode, to);
852 mark_inode_dirty(inode);
853 }
854
855 if (inode == sdp->sd_rindex) {
856 adjust_fs_space(inode);
857 sdp->sd_rindex_uptodate = 0;
858 }
859
860 brelse(dibh);
861 gfs2_trans_end(sdp);
862 if (inode == sdp->sd_rindex) {
863 gfs2_glock_dq(&m_ip->i_gh);
864 gfs2_holder_uninit(&m_ip->i_gh);
865 }
866 gfs2_glock_dq(&ip->i_gh);
867 gfs2_holder_uninit(&ip->i_gh);
868 return copied;
869}
870
871/**
872 * gfs2_write_end
873 * @file: The file to write to
874 * @mapping: The address space to write to
875 * @pos: The file position
876 * @len: The length of the data
877 * @copied: How much was actually copied by the VFS
878 * @page: The page that has been written
879 * @fsdata: The fsdata (unused in GFS2)
880 *
881 * The main write_end function for GFS2. We have a separate one for
882 * stuffed files as they are slightly different, otherwise we just
883 * put our locking around the VFS provided functions.
884 *
885 * Returns: errno
886 */
887
888static int gfs2_write_end(struct file *file, struct address_space *mapping,
889 loff_t pos, unsigned len, unsigned copied,
890 struct page *page, void *fsdata)
891{
892 struct inode *inode = page->mapping->host;
893 struct gfs2_inode *ip = GFS2_I(inode);
894 struct gfs2_sbd *sdp = GFS2_SB(inode);
895 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
896 struct buffer_head *dibh;
897 unsigned int from = pos & (PAGE_SIZE - 1);
898 unsigned int to = from + len;
899 int ret;
900 struct gfs2_trans *tr = current->journal_info;
901 BUG_ON(!tr);
902
903 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
904
905 ret = gfs2_meta_inode_buffer(ip, &dibh);
906 if (unlikely(ret)) {
907 unlock_page(page);
908 put_page(page);
909 goto failed;
910 }
911
912 if (gfs2_is_stuffed(ip))
913 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
914
915 if (!gfs2_is_writeback(ip))
916 gfs2_page_add_databufs(ip, page, from, to);
917
918 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
919 if (tr->tr_num_buf_new)
920 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
921 else
922 gfs2_trans_add_meta(ip->i_gl, dibh);
923
924
925 if (inode == sdp->sd_rindex) {
926 adjust_fs_space(inode);
927 sdp->sd_rindex_uptodate = 0;
928 }
929
930 brelse(dibh);
931failed:
932 gfs2_trans_end(sdp);
933 gfs2_inplace_release(ip);
934 if (ip->i_qadata && ip->i_qadata->qa_qd_num)
935 gfs2_quota_unlock(ip);
936 if (inode == sdp->sd_rindex) {
937 gfs2_glock_dq(&m_ip->i_gh);
938 gfs2_holder_uninit(&m_ip->i_gh);
939 }
940 gfs2_glock_dq(&ip->i_gh);
941 gfs2_holder_uninit(&ip->i_gh);
942 return ret;
943}
944
945/**
946 * gfs2_set_page_dirty - Page dirtying function
947 * @page: The page to dirty
948 *
949 * Returns: 1 if it dirtyed the page, or 0 otherwise
950 */
951
952static int gfs2_set_page_dirty(struct page *page)
953{
954 SetPageChecked(page);
955 return __set_page_dirty_buffers(page);
956}
957
958/**
959 * gfs2_bmap - Block map function
960 * @mapping: Address space info
961 * @lblock: The block to map
962 *
963 * Returns: The disk address for the block or 0 on hole or error
964 */
965
966static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
967{
968 struct gfs2_inode *ip = GFS2_I(mapping->host);
969 struct gfs2_holder i_gh;
970 sector_t dblock = 0;
971 int error;
972
973 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
974 if (error)
975 return 0;
976
977 if (!gfs2_is_stuffed(ip))
978 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
979
980 gfs2_glock_dq_uninit(&i_gh);
981
982 return dblock;
983}
984
985static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
986{
987 struct gfs2_bufdata *bd;
988
989 lock_buffer(bh);
990 gfs2_log_lock(sdp);
991 clear_buffer_dirty(bh);
992 bd = bh->b_private;
993 if (bd) {
994 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
995 list_del_init(&bd->bd_list);
996 else
997 gfs2_remove_from_journal(bh, REMOVE_JDATA);
998 }
999 bh->b_bdev = NULL;
1000 clear_buffer_mapped(bh);
1001 clear_buffer_req(bh);
1002 clear_buffer_new(bh);
1003 gfs2_log_unlock(sdp);
1004 unlock_buffer(bh);
1005}
1006
1007static void gfs2_invalidatepage(struct page *page, unsigned int offset,
1008 unsigned int length)
1009{
1010 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
1011 unsigned int stop = offset + length;
1012 int partial_page = (offset || length < PAGE_SIZE);
1013 struct buffer_head *bh, *head;
1014 unsigned long pos = 0;
1015
1016 BUG_ON(!PageLocked(page));
1017 if (!partial_page)
1018 ClearPageChecked(page);
1019 if (!page_has_buffers(page))
1020 goto out;
1021
1022 bh = head = page_buffers(page);
1023 do {
1024 if (pos + bh->b_size > stop)
1025 return;
1026
1027 if (offset <= pos)
1028 gfs2_discard(sdp, bh);
1029 pos += bh->b_size;
1030 bh = bh->b_this_page;
1031 } while (bh != head);
1032out:
1033 if (!partial_page)
1034 try_to_release_page(page, 0);
1035}
1036
1037/**
1038 * gfs2_ok_for_dio - check that dio is valid on this file
1039 * @ip: The inode
1040 * @offset: The offset at which we are reading or writing
1041 *
1042 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1043 * 1 (to accept the i/o request)
1044 */
1045static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset)
1046{
1047 /*
1048 * Should we return an error here? I can't see that O_DIRECT for
1049 * a stuffed file makes any sense. For now we'll silently fall
1050 * back to buffered I/O
1051 */
1052 if (gfs2_is_stuffed(ip))
1053 return 0;
1054
1055 if (offset >= i_size_read(&ip->i_inode))
1056 return 0;
1057 return 1;
1058}
1059
1060
1061
1062static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1063{
1064 struct file *file = iocb->ki_filp;
1065 struct inode *inode = file->f_mapping->host;
1066 struct address_space *mapping = inode->i_mapping;
1067 struct gfs2_inode *ip = GFS2_I(inode);
1068 loff_t offset = iocb->ki_pos;
1069 struct gfs2_holder gh;
1070 int rv;
1071
1072 /*
1073 * Deferred lock, even if its a write, since we do no allocation
1074 * on this path. All we need change is atime, and this lock mode
1075 * ensures that other nodes have flushed their buffered read caches
1076 * (i.e. their page cache entries for this inode). We do not,
1077 * unfortunately have the option of only flushing a range like
1078 * the VFS does.
1079 */
1080 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1081 rv = gfs2_glock_nq(&gh);
1082 if (rv)
1083 goto out_uninit;
1084 rv = gfs2_ok_for_dio(ip, offset);
1085 if (rv != 1)
1086 goto out; /* dio not valid, fall back to buffered i/o */
1087
1088 /*
1089 * Now since we are holding a deferred (CW) lock at this point, you
1090 * might be wondering why this is ever needed. There is a case however
1091 * where we've granted a deferred local lock against a cached exclusive
1092 * glock. That is ok provided all granted local locks are deferred, but
1093 * it also means that it is possible to encounter pages which are
1094 * cached and possibly also mapped. So here we check for that and sort
1095 * them out ahead of the dio. The glock state machine will take care of
1096 * everything else.
1097 *
1098 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
1099 * the first place, mapping->nr_pages will always be zero.
1100 */
1101 if (mapping->nrpages) {
1102 loff_t lstart = offset & ~(PAGE_SIZE - 1);
1103 loff_t len = iov_iter_count(iter);
1104 loff_t end = PAGE_ALIGN(offset + len) - 1;
1105
1106 rv = 0;
1107 if (len == 0)
1108 goto out;
1109 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
1110 unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
1111 rv = filemap_write_and_wait_range(mapping, lstart, end);
1112 if (rv)
1113 goto out;
1114 if (iov_iter_rw(iter) == WRITE)
1115 truncate_inode_pages_range(mapping, lstart, end);
1116 }
1117
1118 rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
1119 gfs2_get_block_direct, NULL, NULL, 0);
1120out:
1121 gfs2_glock_dq(&gh);
1122out_uninit:
1123 gfs2_holder_uninit(&gh);
1124 return rv;
1125}
1126
1127/**
1128 * gfs2_releasepage - free the metadata associated with a page
1129 * @page: the page that's being released
1130 * @gfp_mask: passed from Linux VFS, ignored by us
1131 *
1132 * Call try_to_free_buffers() if the buffers in this page can be
1133 * released.
1134 *
1135 * Returns: 0
1136 */
1137
1138int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1139{
1140 struct address_space *mapping = page->mapping;
1141 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1142 struct buffer_head *bh, *head;
1143 struct gfs2_bufdata *bd;
1144
1145 if (!page_has_buffers(page))
1146 return 0;
1147
1148 /*
1149 * From xfs_vm_releasepage: mm accommodates an old ext3 case where
1150 * clean pages might not have had the dirty bit cleared. Thus, it can
1151 * send actual dirty pages to ->releasepage() via shrink_active_list().
1152 *
1153 * As a workaround, we skip pages that contain dirty buffers below.
1154 * Once ->releasepage isn't called on dirty pages anymore, we can warn
1155 * on dirty buffers like we used to here again.
1156 */
1157
1158 gfs2_log_lock(sdp);
1159 spin_lock(&sdp->sd_ail_lock);
1160 head = bh = page_buffers(page);
1161 do {
1162 if (atomic_read(&bh->b_count))
1163 goto cannot_release;
1164 bd = bh->b_private;
1165 if (bd && bd->bd_tr)
1166 goto cannot_release;
1167 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
1168 goto cannot_release;
1169 bh = bh->b_this_page;
1170 } while(bh != head);
1171 spin_unlock(&sdp->sd_ail_lock);
1172
1173 head = bh = page_buffers(page);
1174 do {
1175 bd = bh->b_private;
1176 if (bd) {
1177 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1178 if (!list_empty(&bd->bd_list))
1179 list_del_init(&bd->bd_list);
1180 bd->bd_bh = NULL;
1181 bh->b_private = NULL;
1182 kmem_cache_free(gfs2_bufdata_cachep, bd);
1183 }
1184
1185 bh = bh->b_this_page;
1186 } while (bh != head);
1187 gfs2_log_unlock(sdp);
1188
1189 return try_to_free_buffers(page);
1190
1191cannot_release:
1192 spin_unlock(&sdp->sd_ail_lock);
1193 gfs2_log_unlock(sdp);
1194 return 0;
1195}
1196
1197static const struct address_space_operations gfs2_writeback_aops = {
1198 .writepage = gfs2_writepage,
1199 .writepages = gfs2_writepages,
1200 .readpage = gfs2_readpage,
1201 .readpages = gfs2_readpages,
1202 .write_begin = gfs2_write_begin,
1203 .write_end = gfs2_write_end,
1204 .bmap = gfs2_bmap,
1205 .invalidatepage = gfs2_invalidatepage,
1206 .releasepage = gfs2_releasepage,
1207 .direct_IO = gfs2_direct_IO,
1208 .migratepage = buffer_migrate_page,
1209 .is_partially_uptodate = block_is_partially_uptodate,
1210 .error_remove_page = generic_error_remove_page,
1211};
1212
1213static const struct address_space_operations gfs2_ordered_aops = {
1214 .writepage = gfs2_writepage,
1215 .writepages = gfs2_writepages,
1216 .readpage = gfs2_readpage,
1217 .readpages = gfs2_readpages,
1218 .write_begin = gfs2_write_begin,
1219 .write_end = gfs2_write_end,
1220 .set_page_dirty = gfs2_set_page_dirty,
1221 .bmap = gfs2_bmap,
1222 .invalidatepage = gfs2_invalidatepage,
1223 .releasepage = gfs2_releasepage,
1224 .direct_IO = gfs2_direct_IO,
1225 .migratepage = buffer_migrate_page,
1226 .is_partially_uptodate = block_is_partially_uptodate,
1227 .error_remove_page = generic_error_remove_page,
1228};
1229
1230static const struct address_space_operations gfs2_jdata_aops = {
1231 .writepage = gfs2_jdata_writepage,
1232 .writepages = gfs2_jdata_writepages,
1233 .readpage = gfs2_readpage,
1234 .readpages = gfs2_readpages,
1235 .write_begin = gfs2_write_begin,
1236 .write_end = gfs2_write_end,
1237 .set_page_dirty = gfs2_set_page_dirty,
1238 .bmap = gfs2_bmap,
1239 .invalidatepage = gfs2_invalidatepage,
1240 .releasepage = gfs2_releasepage,
1241 .is_partially_uptodate = block_is_partially_uptodate,
1242 .error_remove_page = generic_error_remove_page,
1243};
1244
1245void gfs2_set_aops(struct inode *inode)
1246{
1247 struct gfs2_inode *ip = GFS2_I(inode);
1248
1249 if (gfs2_is_writeback(ip))
1250 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1251 else if (gfs2_is_ordered(ip))
1252 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1253 else if (gfs2_is_jdata(ip))
1254 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1255 else
1256 BUG();
1257}
1258