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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7#include <linux/sched.h>
8#include <linux/slab.h>
9#include <linux/spinlock.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/pagemap.h>
13#include <linux/pagevec.h>
14#include <linux/mpage.h>
15#include <linux/fs.h>
16#include <linux/writeback.h>
17#include <linux/swap.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/backing-dev.h>
20#include <linux/uio.h>
21#include <trace/events/writeback.h>
22#include <linux/sched/signal.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#include "aops.h"
38
39
40void gfs2_trans_add_databufs(struct gfs2_inode *ip, struct folio *folio,
41 size_t from, size_t len)
42{
43 struct buffer_head *head = folio_buffers(folio);
44 unsigned int bsize = head->b_size;
45 struct buffer_head *bh;
46 size_t to = from + len;
47 size_t 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)
53 continue;
54 if (start >= to)
55 break;
56 set_buffer_uptodate(bh);
57 gfs2_trans_add_data(ip->i_gl, bh);
58 }
59}
60
61/**
62 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
63 * @inode: The inode
64 * @lblock: The block number to look up
65 * @bh_result: The buffer head to return the result in
66 * @create: Non-zero if we may add block to the file
67 *
68 * Returns: errno
69 */
70
71static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
72 struct buffer_head *bh_result, int create)
73{
74 int error;
75
76 error = gfs2_block_map(inode, lblock, bh_result, 0);
77 if (error)
78 return error;
79 if (!buffer_mapped(bh_result))
80 return -ENODATA;
81 return 0;
82}
83
84/**
85 * gfs2_write_jdata_folio - gfs2 jdata-specific version of block_write_full_folio
86 * @folio: The folio to write
87 * @wbc: The writeback control
88 *
89 * This is the same as calling block_write_full_folio, but it also
90 * writes pages outside of i_size
91 */
92static int gfs2_write_jdata_folio(struct folio *folio,
93 struct writeback_control *wbc)
94{
95 struct inode * const inode = folio->mapping->host;
96 loff_t i_size = i_size_read(inode);
97
98 /*
99 * The folio straddles i_size. It must be zeroed out on each and every
100 * writepage invocation because it may be mmapped. "A file is mapped
101 * in multiples of the page size. For a file that is not a multiple of
102 * the page size, the remaining memory is zeroed when mapped, and
103 * writes to that region are not written out to the file."
104 */
105 if (folio_pos(folio) < i_size &&
106 i_size < folio_pos(folio) + folio_size(folio))
107 folio_zero_segment(folio, offset_in_folio(folio, i_size),
108 folio_size(folio));
109
110 return __block_write_full_folio(inode, folio, gfs2_get_block_noalloc,
111 wbc);
112}
113
114/**
115 * __gfs2_jdata_write_folio - The core of jdata writepage
116 * @folio: The folio to write
117 * @wbc: The writeback control
118 *
119 * Implements the core of write back. If a transaction is required then
120 * the checked flag will have been set and the transaction will have
121 * already been started before this is called.
122 */
123static int __gfs2_jdata_write_folio(struct folio *folio,
124 struct writeback_control *wbc)
125{
126 struct inode *inode = folio->mapping->host;
127 struct gfs2_inode *ip = GFS2_I(inode);
128
129 if (folio_test_checked(folio)) {
130 folio_clear_checked(folio);
131 if (!folio_buffers(folio)) {
132 create_empty_buffers(folio,
133 inode->i_sb->s_blocksize,
134 BIT(BH_Dirty)|BIT(BH_Uptodate));
135 }
136 gfs2_trans_add_databufs(ip, folio, 0, folio_size(folio));
137 }
138 return gfs2_write_jdata_folio(folio, wbc);
139}
140
141/**
142 * gfs2_writepages - Write a bunch of dirty pages back to disk
143 * @mapping: The mapping to write
144 * @wbc: Write-back control
145 *
146 * Used for both ordered and writeback modes.
147 */
148static int gfs2_writepages(struct address_space *mapping,
149 struct writeback_control *wbc)
150{
151 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
152 struct iomap_writepage_ctx wpc = { };
153 int ret;
154
155 /*
156 * Even if we didn't write enough pages here, we might still be holding
157 * dirty pages in the ail. We forcibly flush the ail because we don't
158 * want balance_dirty_pages() to loop indefinitely trying to write out
159 * pages held in the ail that it can't find.
160 */
161 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
162 if (ret == 0 && wbc->nr_to_write > 0)
163 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
164 return ret;
165}
166
167/**
168 * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
169 * @mapping: The mapping
170 * @wbc: The writeback control
171 * @fbatch: The batch of folios
172 * @done_index: Page index
173 *
174 * Returns: non-zero if loop should terminate, zero otherwise
175 */
176
177static int gfs2_write_jdata_batch(struct address_space *mapping,
178 struct writeback_control *wbc,
179 struct folio_batch *fbatch,
180 pgoff_t *done_index)
181{
182 struct inode *inode = mapping->host;
183 struct gfs2_sbd *sdp = GFS2_SB(inode);
184 unsigned nrblocks;
185 int i;
186 int ret;
187 size_t size = 0;
188 int nr_folios = folio_batch_count(fbatch);
189
190 for (i = 0; i < nr_folios; i++)
191 size += folio_size(fbatch->folios[i]);
192 nrblocks = size >> inode->i_blkbits;
193
194 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
195 if (ret < 0)
196 return ret;
197
198 for (i = 0; i < nr_folios; i++) {
199 struct folio *folio = fbatch->folios[i];
200
201 *done_index = folio->index;
202
203 folio_lock(folio);
204
205 if (unlikely(folio->mapping != mapping)) {
206continue_unlock:
207 folio_unlock(folio);
208 continue;
209 }
210
211 if (!folio_test_dirty(folio)) {
212 /* someone wrote it for us */
213 goto continue_unlock;
214 }
215
216 if (folio_test_writeback(folio)) {
217 if (wbc->sync_mode != WB_SYNC_NONE)
218 folio_wait_writeback(folio);
219 else
220 goto continue_unlock;
221 }
222
223 BUG_ON(folio_test_writeback(folio));
224 if (!folio_clear_dirty_for_io(folio))
225 goto continue_unlock;
226
227 trace_wbc_writepage(wbc, inode_to_bdi(inode));
228
229 ret = __gfs2_jdata_write_folio(folio, wbc);
230 if (unlikely(ret)) {
231 if (ret == AOP_WRITEPAGE_ACTIVATE) {
232 folio_unlock(folio);
233 ret = 0;
234 } else {
235
236 /*
237 * done_index is set past this page,
238 * so media errors will not choke
239 * background writeout for the entire
240 * file. This has consequences for
241 * range_cyclic semantics (ie. it may
242 * not be suitable for data integrity
243 * writeout).
244 */
245 *done_index = folio_next_index(folio);
246 ret = 1;
247 break;
248 }
249 }
250
251 /*
252 * We stop writing back only if we are not doing
253 * integrity sync. In case of integrity sync we have to
254 * keep going until we have written all the pages
255 * we tagged for writeback prior to entering this loop.
256 */
257 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
258 ret = 1;
259 break;
260 }
261
262 }
263 gfs2_trans_end(sdp);
264 return ret;
265}
266
267/**
268 * gfs2_write_cache_jdata - Like write_cache_pages but different
269 * @mapping: The mapping to write
270 * @wbc: The writeback control
271 *
272 * The reason that we use our own function here is that we need to
273 * start transactions before we grab page locks. This allows us
274 * to get the ordering right.
275 */
276
277static int gfs2_write_cache_jdata(struct address_space *mapping,
278 struct writeback_control *wbc)
279{
280 int ret = 0;
281 int done = 0;
282 struct folio_batch fbatch;
283 int nr_folios;
284 pgoff_t writeback_index;
285 pgoff_t index;
286 pgoff_t end;
287 pgoff_t done_index;
288 int cycled;
289 int range_whole = 0;
290 xa_mark_t tag;
291
292 folio_batch_init(&fbatch);
293 if (wbc->range_cyclic) {
294 writeback_index = mapping->writeback_index; /* prev offset */
295 index = writeback_index;
296 if (index == 0)
297 cycled = 1;
298 else
299 cycled = 0;
300 end = -1;
301 } else {
302 index = wbc->range_start >> PAGE_SHIFT;
303 end = wbc->range_end >> PAGE_SHIFT;
304 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
305 range_whole = 1;
306 cycled = 1; /* ignore range_cyclic tests */
307 }
308 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
309 tag = PAGECACHE_TAG_TOWRITE;
310 else
311 tag = PAGECACHE_TAG_DIRTY;
312
313retry:
314 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
315 tag_pages_for_writeback(mapping, index, end);
316 done_index = index;
317 while (!done && (index <= end)) {
318 nr_folios = filemap_get_folios_tag(mapping, &index, end,
319 tag, &fbatch);
320 if (nr_folios == 0)
321 break;
322
323 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
324 &done_index);
325 if (ret)
326 done = 1;
327 if (ret > 0)
328 ret = 0;
329 folio_batch_release(&fbatch);
330 cond_resched();
331 }
332
333 if (!cycled && !done) {
334 /*
335 * range_cyclic:
336 * We hit the last page and there is more work to be done: wrap
337 * back to the start of the file
338 */
339 cycled = 1;
340 index = 0;
341 end = writeback_index - 1;
342 goto retry;
343 }
344
345 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
346 mapping->writeback_index = done_index;
347
348 return ret;
349}
350
351
352/**
353 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
354 * @mapping: The mapping to write
355 * @wbc: The writeback control
356 *
357 */
358
359static int gfs2_jdata_writepages(struct address_space *mapping,
360 struct writeback_control *wbc)
361{
362 struct gfs2_inode *ip = GFS2_I(mapping->host);
363 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
364 int ret;
365
366 ret = gfs2_write_cache_jdata(mapping, wbc);
367 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
368 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
369 GFS2_LFC_JDATA_WPAGES);
370 ret = gfs2_write_cache_jdata(mapping, wbc);
371 }
372 return ret;
373}
374
375/**
376 * stuffed_read_folio - Fill in a Linux folio with stuffed file data
377 * @ip: the inode
378 * @folio: the folio
379 *
380 * Returns: errno
381 */
382static int stuffed_read_folio(struct gfs2_inode *ip, struct folio *folio)
383{
384 struct buffer_head *dibh = NULL;
385 size_t dsize = i_size_read(&ip->i_inode);
386 void *from = NULL;
387 int error = 0;
388
389 /*
390 * Due to the order of unstuffing files and ->fault(), we can be
391 * asked for a zero folio in the case of a stuffed file being extended,
392 * so we need to supply one here. It doesn't happen often.
393 */
394 if (unlikely(folio->index)) {
395 dsize = 0;
396 } else {
397 error = gfs2_meta_inode_buffer(ip, &dibh);
398 if (error)
399 goto out;
400 from = dibh->b_data + sizeof(struct gfs2_dinode);
401 }
402
403 folio_fill_tail(folio, 0, from, dsize);
404 brelse(dibh);
405out:
406 folio_end_read(folio, error == 0);
407
408 return error;
409}
410
411/**
412 * gfs2_read_folio - read a folio from a file
413 * @file: The file to read
414 * @folio: The folio in the file
415 */
416static int gfs2_read_folio(struct file *file, struct folio *folio)
417{
418 struct inode *inode = folio->mapping->host;
419 struct gfs2_inode *ip = GFS2_I(inode);
420 struct gfs2_sbd *sdp = GFS2_SB(inode);
421 int error;
422
423 if (!gfs2_is_jdata(ip) ||
424 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
425 error = iomap_read_folio(folio, &gfs2_iomap_ops);
426 } else if (gfs2_is_stuffed(ip)) {
427 error = stuffed_read_folio(ip, folio);
428 } else {
429 error = mpage_read_folio(folio, gfs2_block_map);
430 }
431
432 if (gfs2_withdrawing_or_withdrawn(sdp))
433 return -EIO;
434
435 return error;
436}
437
438/**
439 * gfs2_internal_read - read an internal file
440 * @ip: The gfs2 inode
441 * @buf: The buffer to fill
442 * @pos: The file position
443 * @size: The amount to read
444 *
445 */
446
447ssize_t gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
448 size_t size)
449{
450 struct address_space *mapping = ip->i_inode.i_mapping;
451 unsigned long index = *pos >> PAGE_SHIFT;
452 size_t copied = 0;
453
454 do {
455 size_t offset, chunk;
456 struct folio *folio;
457
458 folio = read_cache_folio(mapping, index, gfs2_read_folio, NULL);
459 if (IS_ERR(folio)) {
460 if (PTR_ERR(folio) == -EINTR)
461 continue;
462 return PTR_ERR(folio);
463 }
464 offset = *pos + copied - folio_pos(folio);
465 chunk = min(size - copied, folio_size(folio) - offset);
466 memcpy_from_folio(buf + copied, folio, offset, chunk);
467 index = folio_next_index(folio);
468 folio_put(folio);
469 copied += chunk;
470 } while(copied < size);
471 (*pos) += size;
472 return size;
473}
474
475/**
476 * gfs2_readahead - Read a bunch of pages at once
477 * @rac: Read-ahead control structure
478 *
479 * Some notes:
480 * 1. This is only for readahead, so we can simply ignore any things
481 * which are slightly inconvenient (such as locking conflicts between
482 * the page lock and the glock) and return having done no I/O. Its
483 * obviously not something we'd want to do on too regular a basis.
484 * Any I/O we ignore at this time will be done via readpage later.
485 * 2. We don't handle stuffed files here we let readpage do the honours.
486 * 3. mpage_readahead() does most of the heavy lifting in the common case.
487 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
488 */
489
490static void gfs2_readahead(struct readahead_control *rac)
491{
492 struct inode *inode = rac->mapping->host;
493 struct gfs2_inode *ip = GFS2_I(inode);
494
495 if (gfs2_is_stuffed(ip))
496 ;
497 else if (gfs2_is_jdata(ip))
498 mpage_readahead(rac, gfs2_block_map);
499 else
500 iomap_readahead(rac, &gfs2_iomap_ops);
501}
502
503/**
504 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
505 * @inode: the rindex inode
506 */
507void adjust_fs_space(struct inode *inode)
508{
509 struct gfs2_sbd *sdp = GFS2_SB(inode);
510 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
511 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
512 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
513 struct buffer_head *m_bh;
514 u64 fs_total, new_free;
515
516 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
517 return;
518
519 /* Total up the file system space, according to the latest rindex. */
520 fs_total = gfs2_ri_total(sdp);
521 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
522 goto out;
523
524 spin_lock(&sdp->sd_statfs_spin);
525 gfs2_statfs_change_in(m_sc, m_bh->b_data +
526 sizeof(struct gfs2_dinode));
527 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
528 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
529 else
530 new_free = 0;
531 spin_unlock(&sdp->sd_statfs_spin);
532 fs_warn(sdp, "File system extended by %llu blocks.\n",
533 (unsigned long long)new_free);
534 gfs2_statfs_change(sdp, new_free, new_free, 0);
535
536 update_statfs(sdp, m_bh);
537 brelse(m_bh);
538out:
539 sdp->sd_rindex_uptodate = 0;
540 gfs2_trans_end(sdp);
541}
542
543static bool jdata_dirty_folio(struct address_space *mapping,
544 struct folio *folio)
545{
546 if (current->journal_info)
547 folio_set_checked(folio);
548 return block_dirty_folio(mapping, folio);
549}
550
551/**
552 * gfs2_bmap - Block map function
553 * @mapping: Address space info
554 * @lblock: The block to map
555 *
556 * Returns: The disk address for the block or 0 on hole or error
557 */
558
559static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
560{
561 struct gfs2_inode *ip = GFS2_I(mapping->host);
562 struct gfs2_holder i_gh;
563 sector_t dblock = 0;
564 int error;
565
566 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
567 if (error)
568 return 0;
569
570 if (!gfs2_is_stuffed(ip))
571 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
572
573 gfs2_glock_dq_uninit(&i_gh);
574
575 return dblock;
576}
577
578static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
579{
580 struct gfs2_bufdata *bd;
581
582 lock_buffer(bh);
583 gfs2_log_lock(sdp);
584 clear_buffer_dirty(bh);
585 bd = bh->b_private;
586 if (bd) {
587 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
588 list_del_init(&bd->bd_list);
589 else {
590 spin_lock(&sdp->sd_ail_lock);
591 gfs2_remove_from_journal(bh, REMOVE_JDATA);
592 spin_unlock(&sdp->sd_ail_lock);
593 }
594 }
595 bh->b_bdev = NULL;
596 clear_buffer_mapped(bh);
597 clear_buffer_req(bh);
598 clear_buffer_new(bh);
599 gfs2_log_unlock(sdp);
600 unlock_buffer(bh);
601}
602
603static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
604 size_t length)
605{
606 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
607 size_t stop = offset + length;
608 int partial_page = (offset || length < folio_size(folio));
609 struct buffer_head *bh, *head;
610 unsigned long pos = 0;
611
612 BUG_ON(!folio_test_locked(folio));
613 if (!partial_page)
614 folio_clear_checked(folio);
615 head = folio_buffers(folio);
616 if (!head)
617 goto out;
618
619 bh = head;
620 do {
621 if (pos + bh->b_size > stop)
622 return;
623
624 if (offset <= pos)
625 gfs2_discard(sdp, bh);
626 pos += bh->b_size;
627 bh = bh->b_this_page;
628 } while (bh != head);
629out:
630 if (!partial_page)
631 filemap_release_folio(folio, 0);
632}
633
634/**
635 * gfs2_release_folio - free the metadata associated with a folio
636 * @folio: the folio that's being released
637 * @gfp_mask: passed from Linux VFS, ignored by us
638 *
639 * Calls try_to_free_buffers() to free the buffers and put the folio if the
640 * buffers can be released.
641 *
642 * Returns: true if the folio was put or else false
643 */
644
645bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
646{
647 struct address_space *mapping = folio->mapping;
648 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
649 struct buffer_head *bh, *head;
650 struct gfs2_bufdata *bd;
651
652 head = folio_buffers(folio);
653 if (!head)
654 return false;
655
656 /*
657 * mm accommodates an old ext3 case where clean folios might
658 * not have had the dirty bit cleared. Thus, it can send actual
659 * dirty folios to ->release_folio() via shrink_active_list().
660 *
661 * As a workaround, we skip folios that contain dirty buffers
662 * below. Once ->release_folio isn't called on dirty folios
663 * anymore, we can warn on dirty buffers like we used to here
664 * again.
665 */
666
667 gfs2_log_lock(sdp);
668 bh = head;
669 do {
670 if (atomic_read(&bh->b_count))
671 goto cannot_release;
672 bd = bh->b_private;
673 if (bd && bd->bd_tr)
674 goto cannot_release;
675 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
676 goto cannot_release;
677 bh = bh->b_this_page;
678 } while (bh != head);
679
680 bh = head;
681 do {
682 bd = bh->b_private;
683 if (bd) {
684 gfs2_assert_warn(sdp, bd->bd_bh == bh);
685 bd->bd_bh = NULL;
686 bh->b_private = NULL;
687 /*
688 * The bd may still be queued as a revoke, in which
689 * case we must not dequeue nor free it.
690 */
691 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
692 list_del_init(&bd->bd_list);
693 if (list_empty(&bd->bd_list))
694 kmem_cache_free(gfs2_bufdata_cachep, bd);
695 }
696
697 bh = bh->b_this_page;
698 } while (bh != head);
699 gfs2_log_unlock(sdp);
700
701 return try_to_free_buffers(folio);
702
703cannot_release:
704 gfs2_log_unlock(sdp);
705 return false;
706}
707
708static const struct address_space_operations gfs2_aops = {
709 .writepages = gfs2_writepages,
710 .read_folio = gfs2_read_folio,
711 .readahead = gfs2_readahead,
712 .dirty_folio = iomap_dirty_folio,
713 .release_folio = iomap_release_folio,
714 .invalidate_folio = iomap_invalidate_folio,
715 .bmap = gfs2_bmap,
716 .migrate_folio = filemap_migrate_folio,
717 .is_partially_uptodate = iomap_is_partially_uptodate,
718 .error_remove_folio = generic_error_remove_folio,
719};
720
721static const struct address_space_operations gfs2_jdata_aops = {
722 .writepages = gfs2_jdata_writepages,
723 .read_folio = gfs2_read_folio,
724 .readahead = gfs2_readahead,
725 .dirty_folio = jdata_dirty_folio,
726 .bmap = gfs2_bmap,
727 .migrate_folio = buffer_migrate_folio,
728 .invalidate_folio = gfs2_invalidate_folio,
729 .release_folio = gfs2_release_folio,
730 .is_partially_uptodate = block_is_partially_uptodate,
731 .error_remove_folio = generic_error_remove_folio,
732};
733
734void gfs2_set_aops(struct inode *inode)
735{
736 if (gfs2_is_jdata(GFS2_I(inode)))
737 inode->i_mapping->a_ops = &gfs2_jdata_aops;
738 else
739 inode->i_mapping->a_ops = &gfs2_aops;
740}