Loading...
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_bit.h"
20#include "xfs_log.h"
21#include "xfs_inum.h"
22#include "xfs_sb.h"
23#include "xfs_ag.h"
24#include "xfs_trans.h"
25#include "xfs_mount.h"
26#include "xfs_bmap_btree.h"
27#include "xfs_dinode.h"
28#include "xfs_inode.h"
29#include "xfs_alloc.h"
30#include "xfs_error.h"
31#include "xfs_rw.h"
32#include "xfs_iomap.h"
33#include "xfs_vnodeops.h"
34#include "xfs_trace.h"
35#include "xfs_bmap.h"
36#include <linux/gfp.h>
37#include <linux/mpage.h>
38#include <linux/pagevec.h>
39#include <linux/writeback.h>
40
41
42/*
43 * Prime number of hash buckets since address is used as the key.
44 */
45#define NVSYNC 37
46#define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
47static wait_queue_head_t xfs_ioend_wq[NVSYNC];
48
49void __init
50xfs_ioend_init(void)
51{
52 int i;
53
54 for (i = 0; i < NVSYNC; i++)
55 init_waitqueue_head(&xfs_ioend_wq[i]);
56}
57
58void
59xfs_ioend_wait(
60 xfs_inode_t *ip)
61{
62 wait_queue_head_t *wq = to_ioend_wq(ip);
63
64 wait_event(*wq, (atomic_read(&ip->i_iocount) == 0));
65}
66
67STATIC void
68xfs_ioend_wake(
69 xfs_inode_t *ip)
70{
71 if (atomic_dec_and_test(&ip->i_iocount))
72 wake_up(to_ioend_wq(ip));
73}
74
75void
76xfs_count_page_state(
77 struct page *page,
78 int *delalloc,
79 int *unwritten)
80{
81 struct buffer_head *bh, *head;
82
83 *delalloc = *unwritten = 0;
84
85 bh = head = page_buffers(page);
86 do {
87 if (buffer_unwritten(bh))
88 (*unwritten) = 1;
89 else if (buffer_delay(bh))
90 (*delalloc) = 1;
91 } while ((bh = bh->b_this_page) != head);
92}
93
94STATIC struct block_device *
95xfs_find_bdev_for_inode(
96 struct inode *inode)
97{
98 struct xfs_inode *ip = XFS_I(inode);
99 struct xfs_mount *mp = ip->i_mount;
100
101 if (XFS_IS_REALTIME_INODE(ip))
102 return mp->m_rtdev_targp->bt_bdev;
103 else
104 return mp->m_ddev_targp->bt_bdev;
105}
106
107/*
108 * We're now finished for good with this ioend structure.
109 * Update the page state via the associated buffer_heads,
110 * release holds on the inode and bio, and finally free
111 * up memory. Do not use the ioend after this.
112 */
113STATIC void
114xfs_destroy_ioend(
115 xfs_ioend_t *ioend)
116{
117 struct buffer_head *bh, *next;
118 struct xfs_inode *ip = XFS_I(ioend->io_inode);
119
120 for (bh = ioend->io_buffer_head; bh; bh = next) {
121 next = bh->b_private;
122 bh->b_end_io(bh, !ioend->io_error);
123 }
124
125 /*
126 * Volume managers supporting multiple paths can send back ENODEV
127 * when the final path disappears. In this case continuing to fill
128 * the page cache with dirty data which cannot be written out is
129 * evil, so prevent that.
130 */
131 if (unlikely(ioend->io_error == -ENODEV)) {
132 xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ,
133 __FILE__, __LINE__);
134 }
135
136 xfs_ioend_wake(ip);
137 mempool_free(ioend, xfs_ioend_pool);
138}
139
140/*
141 * If the end of the current ioend is beyond the current EOF,
142 * return the new EOF value, otherwise zero.
143 */
144STATIC xfs_fsize_t
145xfs_ioend_new_eof(
146 xfs_ioend_t *ioend)
147{
148 xfs_inode_t *ip = XFS_I(ioend->io_inode);
149 xfs_fsize_t isize;
150 xfs_fsize_t bsize;
151
152 bsize = ioend->io_offset + ioend->io_size;
153 isize = MAX(ip->i_size, ip->i_new_size);
154 isize = MIN(isize, bsize);
155 return isize > ip->i_d.di_size ? isize : 0;
156}
157
158/*
159 * Update on-disk file size now that data has been written to disk. The
160 * current in-memory file size is i_size. If a write is beyond eof i_new_size
161 * will be the intended file size until i_size is updated. If this write does
162 * not extend all the way to the valid file size then restrict this update to
163 * the end of the write.
164 *
165 * This function does not block as blocking on the inode lock in IO completion
166 * can lead to IO completion order dependency deadlocks.. If it can't get the
167 * inode ilock it will return EAGAIN. Callers must handle this.
168 */
169STATIC int
170xfs_setfilesize(
171 xfs_ioend_t *ioend)
172{
173 xfs_inode_t *ip = XFS_I(ioend->io_inode);
174 xfs_fsize_t isize;
175
176 if (unlikely(ioend->io_error))
177 return 0;
178
179 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
180 return EAGAIN;
181
182 isize = xfs_ioend_new_eof(ioend);
183 if (isize) {
184 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
185 ip->i_d.di_size = isize;
186 xfs_mark_inode_dirty(ip);
187 }
188
189 xfs_iunlock(ip, XFS_ILOCK_EXCL);
190 return 0;
191}
192
193/*
194 * Schedule IO completion handling on the final put of an ioend.
195 */
196STATIC void
197xfs_finish_ioend(
198 struct xfs_ioend *ioend)
199{
200 if (atomic_dec_and_test(&ioend->io_remaining)) {
201 if (ioend->io_type == IO_UNWRITTEN)
202 queue_work(xfsconvertd_workqueue, &ioend->io_work);
203 else
204 queue_work(xfsdatad_workqueue, &ioend->io_work);
205 }
206}
207
208/*
209 * IO write completion.
210 */
211STATIC void
212xfs_end_io(
213 struct work_struct *work)
214{
215 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
216 struct xfs_inode *ip = XFS_I(ioend->io_inode);
217 int error = 0;
218
219 /*
220 * For unwritten extents we need to issue transactions to convert a
221 * range to normal written extens after the data I/O has finished.
222 */
223 if (ioend->io_type == IO_UNWRITTEN &&
224 likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) {
225
226 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
227 ioend->io_size);
228 if (error)
229 ioend->io_error = error;
230 }
231
232 /*
233 * We might have to update the on-disk file size after extending
234 * writes.
235 */
236 error = xfs_setfilesize(ioend);
237 ASSERT(!error || error == EAGAIN);
238
239 /*
240 * If we didn't complete processing of the ioend, requeue it to the
241 * tail of the workqueue for another attempt later. Otherwise destroy
242 * it.
243 */
244 if (error == EAGAIN) {
245 atomic_inc(&ioend->io_remaining);
246 xfs_finish_ioend(ioend);
247 /* ensure we don't spin on blocked ioends */
248 delay(1);
249 } else {
250 if (ioend->io_iocb)
251 aio_complete(ioend->io_iocb, ioend->io_result, 0);
252 xfs_destroy_ioend(ioend);
253 }
254}
255
256/*
257 * Call IO completion handling in caller context on the final put of an ioend.
258 */
259STATIC void
260xfs_finish_ioend_sync(
261 struct xfs_ioend *ioend)
262{
263 if (atomic_dec_and_test(&ioend->io_remaining))
264 xfs_end_io(&ioend->io_work);
265}
266
267/*
268 * Allocate and initialise an IO completion structure.
269 * We need to track unwritten extent write completion here initially.
270 * We'll need to extend this for updating the ondisk inode size later
271 * (vs. incore size).
272 */
273STATIC xfs_ioend_t *
274xfs_alloc_ioend(
275 struct inode *inode,
276 unsigned int type)
277{
278 xfs_ioend_t *ioend;
279
280 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
281
282 /*
283 * Set the count to 1 initially, which will prevent an I/O
284 * completion callback from happening before we have started
285 * all the I/O from calling the completion routine too early.
286 */
287 atomic_set(&ioend->io_remaining, 1);
288 ioend->io_error = 0;
289 ioend->io_list = NULL;
290 ioend->io_type = type;
291 ioend->io_inode = inode;
292 ioend->io_buffer_head = NULL;
293 ioend->io_buffer_tail = NULL;
294 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
295 ioend->io_offset = 0;
296 ioend->io_size = 0;
297 ioend->io_iocb = NULL;
298 ioend->io_result = 0;
299
300 INIT_WORK(&ioend->io_work, xfs_end_io);
301 return ioend;
302}
303
304STATIC int
305xfs_map_blocks(
306 struct inode *inode,
307 loff_t offset,
308 struct xfs_bmbt_irec *imap,
309 int type,
310 int nonblocking)
311{
312 struct xfs_inode *ip = XFS_I(inode);
313 struct xfs_mount *mp = ip->i_mount;
314 ssize_t count = 1 << inode->i_blkbits;
315 xfs_fileoff_t offset_fsb, end_fsb;
316 int error = 0;
317 int bmapi_flags = XFS_BMAPI_ENTIRE;
318 int nimaps = 1;
319
320 if (XFS_FORCED_SHUTDOWN(mp))
321 return -XFS_ERROR(EIO);
322
323 if (type == IO_UNWRITTEN)
324 bmapi_flags |= XFS_BMAPI_IGSTATE;
325
326 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
327 if (nonblocking)
328 return -XFS_ERROR(EAGAIN);
329 xfs_ilock(ip, XFS_ILOCK_SHARED);
330 }
331
332 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
333 (ip->i_df.if_flags & XFS_IFEXTENTS));
334 ASSERT(offset <= mp->m_maxioffset);
335
336 if (offset + count > mp->m_maxioffset)
337 count = mp->m_maxioffset - offset;
338 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
339 offset_fsb = XFS_B_TO_FSBT(mp, offset);
340 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
341 bmapi_flags, NULL, 0, imap, &nimaps, NULL);
342 xfs_iunlock(ip, XFS_ILOCK_SHARED);
343
344 if (error)
345 return -XFS_ERROR(error);
346
347 if (type == IO_DELALLOC &&
348 (!nimaps || isnullstartblock(imap->br_startblock))) {
349 error = xfs_iomap_write_allocate(ip, offset, count, imap);
350 if (!error)
351 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
352 return -XFS_ERROR(error);
353 }
354
355#ifdef DEBUG
356 if (type == IO_UNWRITTEN) {
357 ASSERT(nimaps);
358 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
359 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
360 }
361#endif
362 if (nimaps)
363 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
364 return 0;
365}
366
367STATIC int
368xfs_imap_valid(
369 struct inode *inode,
370 struct xfs_bmbt_irec *imap,
371 xfs_off_t offset)
372{
373 offset >>= inode->i_blkbits;
374
375 return offset >= imap->br_startoff &&
376 offset < imap->br_startoff + imap->br_blockcount;
377}
378
379/*
380 * BIO completion handler for buffered IO.
381 */
382STATIC void
383xfs_end_bio(
384 struct bio *bio,
385 int error)
386{
387 xfs_ioend_t *ioend = bio->bi_private;
388
389 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
390 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
391
392 /* Toss bio and pass work off to an xfsdatad thread */
393 bio->bi_private = NULL;
394 bio->bi_end_io = NULL;
395 bio_put(bio);
396
397 xfs_finish_ioend(ioend);
398}
399
400STATIC void
401xfs_submit_ioend_bio(
402 struct writeback_control *wbc,
403 xfs_ioend_t *ioend,
404 struct bio *bio)
405{
406 atomic_inc(&ioend->io_remaining);
407 bio->bi_private = ioend;
408 bio->bi_end_io = xfs_end_bio;
409
410 /*
411 * If the I/O is beyond EOF we mark the inode dirty immediately
412 * but don't update the inode size until I/O completion.
413 */
414 if (xfs_ioend_new_eof(ioend))
415 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
416
417 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
418}
419
420STATIC struct bio *
421xfs_alloc_ioend_bio(
422 struct buffer_head *bh)
423{
424 int nvecs = bio_get_nr_vecs(bh->b_bdev);
425 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
426
427 ASSERT(bio->bi_private == NULL);
428 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
429 bio->bi_bdev = bh->b_bdev;
430 return bio;
431}
432
433STATIC void
434xfs_start_buffer_writeback(
435 struct buffer_head *bh)
436{
437 ASSERT(buffer_mapped(bh));
438 ASSERT(buffer_locked(bh));
439 ASSERT(!buffer_delay(bh));
440 ASSERT(!buffer_unwritten(bh));
441
442 mark_buffer_async_write(bh);
443 set_buffer_uptodate(bh);
444 clear_buffer_dirty(bh);
445}
446
447STATIC void
448xfs_start_page_writeback(
449 struct page *page,
450 int clear_dirty,
451 int buffers)
452{
453 ASSERT(PageLocked(page));
454 ASSERT(!PageWriteback(page));
455 if (clear_dirty)
456 clear_page_dirty_for_io(page);
457 set_page_writeback(page);
458 unlock_page(page);
459 /* If no buffers on the page are to be written, finish it here */
460 if (!buffers)
461 end_page_writeback(page);
462}
463
464static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
465{
466 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
467}
468
469/*
470 * Submit all of the bios for all of the ioends we have saved up, covering the
471 * initial writepage page and also any probed pages.
472 *
473 * Because we may have multiple ioends spanning a page, we need to start
474 * writeback on all the buffers before we submit them for I/O. If we mark the
475 * buffers as we got, then we can end up with a page that only has buffers
476 * marked async write and I/O complete on can occur before we mark the other
477 * buffers async write.
478 *
479 * The end result of this is that we trip a bug in end_page_writeback() because
480 * we call it twice for the one page as the code in end_buffer_async_write()
481 * assumes that all buffers on the page are started at the same time.
482 *
483 * The fix is two passes across the ioend list - one to start writeback on the
484 * buffer_heads, and then submit them for I/O on the second pass.
485 */
486STATIC void
487xfs_submit_ioend(
488 struct writeback_control *wbc,
489 xfs_ioend_t *ioend)
490{
491 xfs_ioend_t *head = ioend;
492 xfs_ioend_t *next;
493 struct buffer_head *bh;
494 struct bio *bio;
495 sector_t lastblock = 0;
496
497 /* Pass 1 - start writeback */
498 do {
499 next = ioend->io_list;
500 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
501 xfs_start_buffer_writeback(bh);
502 } while ((ioend = next) != NULL);
503
504 /* Pass 2 - submit I/O */
505 ioend = head;
506 do {
507 next = ioend->io_list;
508 bio = NULL;
509
510 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
511
512 if (!bio) {
513 retry:
514 bio = xfs_alloc_ioend_bio(bh);
515 } else if (bh->b_blocknr != lastblock + 1) {
516 xfs_submit_ioend_bio(wbc, ioend, bio);
517 goto retry;
518 }
519
520 if (bio_add_buffer(bio, bh) != bh->b_size) {
521 xfs_submit_ioend_bio(wbc, ioend, bio);
522 goto retry;
523 }
524
525 lastblock = bh->b_blocknr;
526 }
527 if (bio)
528 xfs_submit_ioend_bio(wbc, ioend, bio);
529 xfs_finish_ioend(ioend);
530 } while ((ioend = next) != NULL);
531}
532
533/*
534 * Cancel submission of all buffer_heads so far in this endio.
535 * Toss the endio too. Only ever called for the initial page
536 * in a writepage request, so only ever one page.
537 */
538STATIC void
539xfs_cancel_ioend(
540 xfs_ioend_t *ioend)
541{
542 xfs_ioend_t *next;
543 struct buffer_head *bh, *next_bh;
544
545 do {
546 next = ioend->io_list;
547 bh = ioend->io_buffer_head;
548 do {
549 next_bh = bh->b_private;
550 clear_buffer_async_write(bh);
551 unlock_buffer(bh);
552 } while ((bh = next_bh) != NULL);
553
554 xfs_ioend_wake(XFS_I(ioend->io_inode));
555 mempool_free(ioend, xfs_ioend_pool);
556 } while ((ioend = next) != NULL);
557}
558
559/*
560 * Test to see if we've been building up a completion structure for
561 * earlier buffers -- if so, we try to append to this ioend if we
562 * can, otherwise we finish off any current ioend and start another.
563 * Return true if we've finished the given ioend.
564 */
565STATIC void
566xfs_add_to_ioend(
567 struct inode *inode,
568 struct buffer_head *bh,
569 xfs_off_t offset,
570 unsigned int type,
571 xfs_ioend_t **result,
572 int need_ioend)
573{
574 xfs_ioend_t *ioend = *result;
575
576 if (!ioend || need_ioend || type != ioend->io_type) {
577 xfs_ioend_t *previous = *result;
578
579 ioend = xfs_alloc_ioend(inode, type);
580 ioend->io_offset = offset;
581 ioend->io_buffer_head = bh;
582 ioend->io_buffer_tail = bh;
583 if (previous)
584 previous->io_list = ioend;
585 *result = ioend;
586 } else {
587 ioend->io_buffer_tail->b_private = bh;
588 ioend->io_buffer_tail = bh;
589 }
590
591 bh->b_private = NULL;
592 ioend->io_size += bh->b_size;
593}
594
595STATIC void
596xfs_map_buffer(
597 struct inode *inode,
598 struct buffer_head *bh,
599 struct xfs_bmbt_irec *imap,
600 xfs_off_t offset)
601{
602 sector_t bn;
603 struct xfs_mount *m = XFS_I(inode)->i_mount;
604 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
605 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
606
607 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
608 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
609
610 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
611 ((offset - iomap_offset) >> inode->i_blkbits);
612
613 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
614
615 bh->b_blocknr = bn;
616 set_buffer_mapped(bh);
617}
618
619STATIC void
620xfs_map_at_offset(
621 struct inode *inode,
622 struct buffer_head *bh,
623 struct xfs_bmbt_irec *imap,
624 xfs_off_t offset)
625{
626 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
627 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
628
629 xfs_map_buffer(inode, bh, imap, offset);
630 set_buffer_mapped(bh);
631 clear_buffer_delay(bh);
632 clear_buffer_unwritten(bh);
633}
634
635/*
636 * Test if a given page is suitable for writing as part of an unwritten
637 * or delayed allocate extent.
638 */
639STATIC int
640xfs_is_delayed_page(
641 struct page *page,
642 unsigned int type)
643{
644 if (PageWriteback(page))
645 return 0;
646
647 if (page->mapping && page_has_buffers(page)) {
648 struct buffer_head *bh, *head;
649 int acceptable = 0;
650
651 bh = head = page_buffers(page);
652 do {
653 if (buffer_unwritten(bh))
654 acceptable = (type == IO_UNWRITTEN);
655 else if (buffer_delay(bh))
656 acceptable = (type == IO_DELALLOC);
657 else if (buffer_dirty(bh) && buffer_mapped(bh))
658 acceptable = (type == IO_OVERWRITE);
659 else
660 break;
661 } while ((bh = bh->b_this_page) != head);
662
663 if (acceptable)
664 return 1;
665 }
666
667 return 0;
668}
669
670/*
671 * Allocate & map buffers for page given the extent map. Write it out.
672 * except for the original page of a writepage, this is called on
673 * delalloc/unwritten pages only, for the original page it is possible
674 * that the page has no mapping at all.
675 */
676STATIC int
677xfs_convert_page(
678 struct inode *inode,
679 struct page *page,
680 loff_t tindex,
681 struct xfs_bmbt_irec *imap,
682 xfs_ioend_t **ioendp,
683 struct writeback_control *wbc)
684{
685 struct buffer_head *bh, *head;
686 xfs_off_t end_offset;
687 unsigned long p_offset;
688 unsigned int type;
689 int len, page_dirty;
690 int count = 0, done = 0, uptodate = 1;
691 xfs_off_t offset = page_offset(page);
692
693 if (page->index != tindex)
694 goto fail;
695 if (!trylock_page(page))
696 goto fail;
697 if (PageWriteback(page))
698 goto fail_unlock_page;
699 if (page->mapping != inode->i_mapping)
700 goto fail_unlock_page;
701 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
702 goto fail_unlock_page;
703
704 /*
705 * page_dirty is initially a count of buffers on the page before
706 * EOF and is decremented as we move each into a cleanable state.
707 *
708 * Derivation:
709 *
710 * End offset is the highest offset that this page should represent.
711 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
712 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
713 * hence give us the correct page_dirty count. On any other page,
714 * it will be zero and in that case we need page_dirty to be the
715 * count of buffers on the page.
716 */
717 end_offset = min_t(unsigned long long,
718 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
719 i_size_read(inode));
720
721 len = 1 << inode->i_blkbits;
722 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
723 PAGE_CACHE_SIZE);
724 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
725 page_dirty = p_offset / len;
726
727 bh = head = page_buffers(page);
728 do {
729 if (offset >= end_offset)
730 break;
731 if (!buffer_uptodate(bh))
732 uptodate = 0;
733 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
734 done = 1;
735 continue;
736 }
737
738 if (buffer_unwritten(bh) || buffer_delay(bh) ||
739 buffer_mapped(bh)) {
740 if (buffer_unwritten(bh))
741 type = IO_UNWRITTEN;
742 else if (buffer_delay(bh))
743 type = IO_DELALLOC;
744 else
745 type = IO_OVERWRITE;
746
747 if (!xfs_imap_valid(inode, imap, offset)) {
748 done = 1;
749 continue;
750 }
751
752 lock_buffer(bh);
753 if (type != IO_OVERWRITE)
754 xfs_map_at_offset(inode, bh, imap, offset);
755 xfs_add_to_ioend(inode, bh, offset, type,
756 ioendp, done);
757
758 page_dirty--;
759 count++;
760 } else {
761 done = 1;
762 }
763 } while (offset += len, (bh = bh->b_this_page) != head);
764
765 if (uptodate && bh == head)
766 SetPageUptodate(page);
767
768 if (count) {
769 if (--wbc->nr_to_write <= 0 &&
770 wbc->sync_mode == WB_SYNC_NONE)
771 done = 1;
772 }
773 xfs_start_page_writeback(page, !page_dirty, count);
774
775 return done;
776 fail_unlock_page:
777 unlock_page(page);
778 fail:
779 return 1;
780}
781
782/*
783 * Convert & write out a cluster of pages in the same extent as defined
784 * by mp and following the start page.
785 */
786STATIC void
787xfs_cluster_write(
788 struct inode *inode,
789 pgoff_t tindex,
790 struct xfs_bmbt_irec *imap,
791 xfs_ioend_t **ioendp,
792 struct writeback_control *wbc,
793 pgoff_t tlast)
794{
795 struct pagevec pvec;
796 int done = 0, i;
797
798 pagevec_init(&pvec, 0);
799 while (!done && tindex <= tlast) {
800 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
801
802 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
803 break;
804
805 for (i = 0; i < pagevec_count(&pvec); i++) {
806 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
807 imap, ioendp, wbc);
808 if (done)
809 break;
810 }
811
812 pagevec_release(&pvec);
813 cond_resched();
814 }
815}
816
817STATIC void
818xfs_vm_invalidatepage(
819 struct page *page,
820 unsigned long offset)
821{
822 trace_xfs_invalidatepage(page->mapping->host, page, offset);
823 block_invalidatepage(page, offset);
824}
825
826/*
827 * If the page has delalloc buffers on it, we need to punch them out before we
828 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
829 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
830 * is done on that same region - the delalloc extent is returned when none is
831 * supposed to be there.
832 *
833 * We prevent this by truncating away the delalloc regions on the page before
834 * invalidating it. Because they are delalloc, we can do this without needing a
835 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
836 * truncation without a transaction as there is no space left for block
837 * reservation (typically why we see a ENOSPC in writeback).
838 *
839 * This is not a performance critical path, so for now just do the punching a
840 * buffer head at a time.
841 */
842STATIC void
843xfs_aops_discard_page(
844 struct page *page)
845{
846 struct inode *inode = page->mapping->host;
847 struct xfs_inode *ip = XFS_I(inode);
848 struct buffer_head *bh, *head;
849 loff_t offset = page_offset(page);
850
851 if (!xfs_is_delayed_page(page, IO_DELALLOC))
852 goto out_invalidate;
853
854 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
855 goto out_invalidate;
856
857 xfs_alert(ip->i_mount,
858 "page discard on page %p, inode 0x%llx, offset %llu.",
859 page, ip->i_ino, offset);
860
861 xfs_ilock(ip, XFS_ILOCK_EXCL);
862 bh = head = page_buffers(page);
863 do {
864 int error;
865 xfs_fileoff_t start_fsb;
866
867 if (!buffer_delay(bh))
868 goto next_buffer;
869
870 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
871 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
872 if (error) {
873 /* something screwed, just bail */
874 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
875 xfs_alert(ip->i_mount,
876 "page discard unable to remove delalloc mapping.");
877 }
878 break;
879 }
880next_buffer:
881 offset += 1 << inode->i_blkbits;
882
883 } while ((bh = bh->b_this_page) != head);
884
885 xfs_iunlock(ip, XFS_ILOCK_EXCL);
886out_invalidate:
887 xfs_vm_invalidatepage(page, 0);
888 return;
889}
890
891/*
892 * Write out a dirty page.
893 *
894 * For delalloc space on the page we need to allocate space and flush it.
895 * For unwritten space on the page we need to start the conversion to
896 * regular allocated space.
897 * For any other dirty buffer heads on the page we should flush them.
898 */
899STATIC int
900xfs_vm_writepage(
901 struct page *page,
902 struct writeback_control *wbc)
903{
904 struct inode *inode = page->mapping->host;
905 struct buffer_head *bh, *head;
906 struct xfs_bmbt_irec imap;
907 xfs_ioend_t *ioend = NULL, *iohead = NULL;
908 loff_t offset;
909 unsigned int type;
910 __uint64_t end_offset;
911 pgoff_t end_index, last_index;
912 ssize_t len;
913 int err, imap_valid = 0, uptodate = 1;
914 int count = 0;
915 int nonblocking = 0;
916
917 trace_xfs_writepage(inode, page, 0);
918
919 ASSERT(page_has_buffers(page));
920
921 /*
922 * Refuse to write the page out if we are called from reclaim context.
923 *
924 * This avoids stack overflows when called from deeply used stacks in
925 * random callers for direct reclaim or memcg reclaim. We explicitly
926 * allow reclaim from kswapd as the stack usage there is relatively low.
927 *
928 * This should really be done by the core VM, but until that happens
929 * filesystems like XFS, btrfs and ext4 have to take care of this
930 * by themselves.
931 */
932 if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC)
933 goto redirty;
934
935 /*
936 * Given that we do not allow direct reclaim to call us, we should
937 * never be called while in a filesystem transaction.
938 */
939 if (WARN_ON(current->flags & PF_FSTRANS))
940 goto redirty;
941
942 /* Is this page beyond the end of the file? */
943 offset = i_size_read(inode);
944 end_index = offset >> PAGE_CACHE_SHIFT;
945 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
946 if (page->index >= end_index) {
947 if ((page->index >= end_index + 1) ||
948 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
949 unlock_page(page);
950 return 0;
951 }
952 }
953
954 end_offset = min_t(unsigned long long,
955 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
956 offset);
957 len = 1 << inode->i_blkbits;
958
959 bh = head = page_buffers(page);
960 offset = page_offset(page);
961 type = IO_OVERWRITE;
962
963 if (wbc->sync_mode == WB_SYNC_NONE)
964 nonblocking = 1;
965
966 do {
967 int new_ioend = 0;
968
969 if (offset >= end_offset)
970 break;
971 if (!buffer_uptodate(bh))
972 uptodate = 0;
973
974 /*
975 * set_page_dirty dirties all buffers in a page, independent
976 * of their state. The dirty state however is entirely
977 * meaningless for holes (!mapped && uptodate), so skip
978 * buffers covering holes here.
979 */
980 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
981 imap_valid = 0;
982 continue;
983 }
984
985 if (buffer_unwritten(bh)) {
986 if (type != IO_UNWRITTEN) {
987 type = IO_UNWRITTEN;
988 imap_valid = 0;
989 }
990 } else if (buffer_delay(bh)) {
991 if (type != IO_DELALLOC) {
992 type = IO_DELALLOC;
993 imap_valid = 0;
994 }
995 } else if (buffer_uptodate(bh)) {
996 if (type != IO_OVERWRITE) {
997 type = IO_OVERWRITE;
998 imap_valid = 0;
999 }
1000 } else {
1001 if (PageUptodate(page)) {
1002 ASSERT(buffer_mapped(bh));
1003 imap_valid = 0;
1004 }
1005 continue;
1006 }
1007
1008 if (imap_valid)
1009 imap_valid = xfs_imap_valid(inode, &imap, offset);
1010 if (!imap_valid) {
1011 /*
1012 * If we didn't have a valid mapping then we need to
1013 * put the new mapping into a separate ioend structure.
1014 * This ensures non-contiguous extents always have
1015 * separate ioends, which is particularly important
1016 * for unwritten extent conversion at I/O completion
1017 * time.
1018 */
1019 new_ioend = 1;
1020 err = xfs_map_blocks(inode, offset, &imap, type,
1021 nonblocking);
1022 if (err)
1023 goto error;
1024 imap_valid = xfs_imap_valid(inode, &imap, offset);
1025 }
1026 if (imap_valid) {
1027 lock_buffer(bh);
1028 if (type != IO_OVERWRITE)
1029 xfs_map_at_offset(inode, bh, &imap, offset);
1030 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1031 new_ioend);
1032 count++;
1033 }
1034
1035 if (!iohead)
1036 iohead = ioend;
1037
1038 } while (offset += len, ((bh = bh->b_this_page) != head));
1039
1040 if (uptodate && bh == head)
1041 SetPageUptodate(page);
1042
1043 xfs_start_page_writeback(page, 1, count);
1044
1045 if (ioend && imap_valid) {
1046 xfs_off_t end_index;
1047
1048 end_index = imap.br_startoff + imap.br_blockcount;
1049
1050 /* to bytes */
1051 end_index <<= inode->i_blkbits;
1052
1053 /* to pages */
1054 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1055
1056 /* check against file size */
1057 if (end_index > last_index)
1058 end_index = last_index;
1059
1060 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1061 wbc, end_index);
1062 }
1063
1064 if (iohead)
1065 xfs_submit_ioend(wbc, iohead);
1066
1067 return 0;
1068
1069error:
1070 if (iohead)
1071 xfs_cancel_ioend(iohead);
1072
1073 if (err == -EAGAIN)
1074 goto redirty;
1075
1076 xfs_aops_discard_page(page);
1077 ClearPageUptodate(page);
1078 unlock_page(page);
1079 return err;
1080
1081redirty:
1082 redirty_page_for_writepage(wbc, page);
1083 unlock_page(page);
1084 return 0;
1085}
1086
1087STATIC int
1088xfs_vm_writepages(
1089 struct address_space *mapping,
1090 struct writeback_control *wbc)
1091{
1092 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1093 return generic_writepages(mapping, wbc);
1094}
1095
1096/*
1097 * Called to move a page into cleanable state - and from there
1098 * to be released. The page should already be clean. We always
1099 * have buffer heads in this call.
1100 *
1101 * Returns 1 if the page is ok to release, 0 otherwise.
1102 */
1103STATIC int
1104xfs_vm_releasepage(
1105 struct page *page,
1106 gfp_t gfp_mask)
1107{
1108 int delalloc, unwritten;
1109
1110 trace_xfs_releasepage(page->mapping->host, page, 0);
1111
1112 xfs_count_page_state(page, &delalloc, &unwritten);
1113
1114 if (WARN_ON(delalloc))
1115 return 0;
1116 if (WARN_ON(unwritten))
1117 return 0;
1118
1119 return try_to_free_buffers(page);
1120}
1121
1122STATIC int
1123__xfs_get_blocks(
1124 struct inode *inode,
1125 sector_t iblock,
1126 struct buffer_head *bh_result,
1127 int create,
1128 int direct)
1129{
1130 struct xfs_inode *ip = XFS_I(inode);
1131 struct xfs_mount *mp = ip->i_mount;
1132 xfs_fileoff_t offset_fsb, end_fsb;
1133 int error = 0;
1134 int lockmode = 0;
1135 struct xfs_bmbt_irec imap;
1136 int nimaps = 1;
1137 xfs_off_t offset;
1138 ssize_t size;
1139 int new = 0;
1140
1141 if (XFS_FORCED_SHUTDOWN(mp))
1142 return -XFS_ERROR(EIO);
1143
1144 offset = (xfs_off_t)iblock << inode->i_blkbits;
1145 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1146 size = bh_result->b_size;
1147
1148 if (!create && direct && offset >= i_size_read(inode))
1149 return 0;
1150
1151 if (create) {
1152 lockmode = XFS_ILOCK_EXCL;
1153 xfs_ilock(ip, lockmode);
1154 } else {
1155 lockmode = xfs_ilock_map_shared(ip);
1156 }
1157
1158 ASSERT(offset <= mp->m_maxioffset);
1159 if (offset + size > mp->m_maxioffset)
1160 size = mp->m_maxioffset - offset;
1161 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1162 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1163
1164 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
1165 XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL);
1166 if (error)
1167 goto out_unlock;
1168
1169 if (create &&
1170 (!nimaps ||
1171 (imap.br_startblock == HOLESTARTBLOCK ||
1172 imap.br_startblock == DELAYSTARTBLOCK))) {
1173 if (direct) {
1174 error = xfs_iomap_write_direct(ip, offset, size,
1175 &imap, nimaps);
1176 } else {
1177 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1178 }
1179 if (error)
1180 goto out_unlock;
1181
1182 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1183 } else if (nimaps) {
1184 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1185 } else {
1186 trace_xfs_get_blocks_notfound(ip, offset, size);
1187 goto out_unlock;
1188 }
1189 xfs_iunlock(ip, lockmode);
1190
1191 if (imap.br_startblock != HOLESTARTBLOCK &&
1192 imap.br_startblock != DELAYSTARTBLOCK) {
1193 /*
1194 * For unwritten extents do not report a disk address on
1195 * the read case (treat as if we're reading into a hole).
1196 */
1197 if (create || !ISUNWRITTEN(&imap))
1198 xfs_map_buffer(inode, bh_result, &imap, offset);
1199 if (create && ISUNWRITTEN(&imap)) {
1200 if (direct)
1201 bh_result->b_private = inode;
1202 set_buffer_unwritten(bh_result);
1203 }
1204 }
1205
1206 /*
1207 * If this is a realtime file, data may be on a different device.
1208 * to that pointed to from the buffer_head b_bdev currently.
1209 */
1210 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1211
1212 /*
1213 * If we previously allocated a block out beyond eof and we are now
1214 * coming back to use it then we will need to flag it as new even if it
1215 * has a disk address.
1216 *
1217 * With sub-block writes into unwritten extents we also need to mark
1218 * the buffer as new so that the unwritten parts of the buffer gets
1219 * correctly zeroed.
1220 */
1221 if (create &&
1222 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1223 (offset >= i_size_read(inode)) ||
1224 (new || ISUNWRITTEN(&imap))))
1225 set_buffer_new(bh_result);
1226
1227 if (imap.br_startblock == DELAYSTARTBLOCK) {
1228 BUG_ON(direct);
1229 if (create) {
1230 set_buffer_uptodate(bh_result);
1231 set_buffer_mapped(bh_result);
1232 set_buffer_delay(bh_result);
1233 }
1234 }
1235
1236 /*
1237 * If this is O_DIRECT or the mpage code calling tell them how large
1238 * the mapping is, so that we can avoid repeated get_blocks calls.
1239 */
1240 if (direct || size > (1 << inode->i_blkbits)) {
1241 xfs_off_t mapping_size;
1242
1243 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1244 mapping_size <<= inode->i_blkbits;
1245
1246 ASSERT(mapping_size > 0);
1247 if (mapping_size > size)
1248 mapping_size = size;
1249 if (mapping_size > LONG_MAX)
1250 mapping_size = LONG_MAX;
1251
1252 bh_result->b_size = mapping_size;
1253 }
1254
1255 return 0;
1256
1257out_unlock:
1258 xfs_iunlock(ip, lockmode);
1259 return -error;
1260}
1261
1262int
1263xfs_get_blocks(
1264 struct inode *inode,
1265 sector_t iblock,
1266 struct buffer_head *bh_result,
1267 int create)
1268{
1269 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1270}
1271
1272STATIC int
1273xfs_get_blocks_direct(
1274 struct inode *inode,
1275 sector_t iblock,
1276 struct buffer_head *bh_result,
1277 int create)
1278{
1279 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1280}
1281
1282/*
1283 * Complete a direct I/O write request.
1284 *
1285 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1286 * need to issue a transaction to convert the range from unwritten to written
1287 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1288 * to do this and we are done. But in case this was a successful AIO
1289 * request this handler is called from interrupt context, from which we
1290 * can't start transactions. In that case offload the I/O completion to
1291 * the workqueues we also use for buffered I/O completion.
1292 */
1293STATIC void
1294xfs_end_io_direct_write(
1295 struct kiocb *iocb,
1296 loff_t offset,
1297 ssize_t size,
1298 void *private,
1299 int ret,
1300 bool is_async)
1301{
1302 struct xfs_ioend *ioend = iocb->private;
1303 struct inode *inode = ioend->io_inode;
1304
1305 /*
1306 * blockdev_direct_IO can return an error even after the I/O
1307 * completion handler was called. Thus we need to protect
1308 * against double-freeing.
1309 */
1310 iocb->private = NULL;
1311
1312 ioend->io_offset = offset;
1313 ioend->io_size = size;
1314 if (private && size > 0)
1315 ioend->io_type = IO_UNWRITTEN;
1316
1317 if (is_async) {
1318 /*
1319 * If we are converting an unwritten extent we need to delay
1320 * the AIO completion until after the unwrittent extent
1321 * conversion has completed, otherwise do it ASAP.
1322 */
1323 if (ioend->io_type == IO_UNWRITTEN) {
1324 ioend->io_iocb = iocb;
1325 ioend->io_result = ret;
1326 } else {
1327 aio_complete(iocb, ret, 0);
1328 }
1329 xfs_finish_ioend(ioend);
1330 } else {
1331 xfs_finish_ioend_sync(ioend);
1332 }
1333
1334 /* XXX: probably should move into the real I/O completion handler */
1335 inode_dio_done(inode);
1336}
1337
1338STATIC ssize_t
1339xfs_vm_direct_IO(
1340 int rw,
1341 struct kiocb *iocb,
1342 const struct iovec *iov,
1343 loff_t offset,
1344 unsigned long nr_segs)
1345{
1346 struct inode *inode = iocb->ki_filp->f_mapping->host;
1347 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1348 ssize_t ret;
1349
1350 if (rw & WRITE) {
1351 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1352
1353 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1354 offset, nr_segs,
1355 xfs_get_blocks_direct,
1356 xfs_end_io_direct_write, NULL, 0);
1357 if (ret != -EIOCBQUEUED && iocb->private)
1358 xfs_destroy_ioend(iocb->private);
1359 } else {
1360 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1361 offset, nr_segs,
1362 xfs_get_blocks_direct,
1363 NULL, NULL, 0);
1364 }
1365
1366 return ret;
1367}
1368
1369STATIC void
1370xfs_vm_write_failed(
1371 struct address_space *mapping,
1372 loff_t to)
1373{
1374 struct inode *inode = mapping->host;
1375
1376 if (to > inode->i_size) {
1377 /*
1378 * punch out the delalloc blocks we have already allocated. We
1379 * don't call xfs_setattr() to do this as we may be in the
1380 * middle of a multi-iovec write and so the vfs inode->i_size
1381 * will not match the xfs ip->i_size and so it will zero too
1382 * much. Hence we jus truncate the page cache to zero what is
1383 * necessary and punch the delalloc blocks directly.
1384 */
1385 struct xfs_inode *ip = XFS_I(inode);
1386 xfs_fileoff_t start_fsb;
1387 xfs_fileoff_t end_fsb;
1388 int error;
1389
1390 truncate_pagecache(inode, to, inode->i_size);
1391
1392 /*
1393 * Check if there are any blocks that are outside of i_size
1394 * that need to be trimmed back.
1395 */
1396 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1397 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1398 if (end_fsb <= start_fsb)
1399 return;
1400
1401 xfs_ilock(ip, XFS_ILOCK_EXCL);
1402 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1403 end_fsb - start_fsb);
1404 if (error) {
1405 /* something screwed, just bail */
1406 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1407 xfs_alert(ip->i_mount,
1408 "xfs_vm_write_failed: unable to clean up ino %lld",
1409 ip->i_ino);
1410 }
1411 }
1412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1413 }
1414}
1415
1416STATIC int
1417xfs_vm_write_begin(
1418 struct file *file,
1419 struct address_space *mapping,
1420 loff_t pos,
1421 unsigned len,
1422 unsigned flags,
1423 struct page **pagep,
1424 void **fsdata)
1425{
1426 int ret;
1427
1428 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1429 pagep, xfs_get_blocks);
1430 if (unlikely(ret))
1431 xfs_vm_write_failed(mapping, pos + len);
1432 return ret;
1433}
1434
1435STATIC int
1436xfs_vm_write_end(
1437 struct file *file,
1438 struct address_space *mapping,
1439 loff_t pos,
1440 unsigned len,
1441 unsigned copied,
1442 struct page *page,
1443 void *fsdata)
1444{
1445 int ret;
1446
1447 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1448 if (unlikely(ret < len))
1449 xfs_vm_write_failed(mapping, pos + len);
1450 return ret;
1451}
1452
1453STATIC sector_t
1454xfs_vm_bmap(
1455 struct address_space *mapping,
1456 sector_t block)
1457{
1458 struct inode *inode = (struct inode *)mapping->host;
1459 struct xfs_inode *ip = XFS_I(inode);
1460
1461 trace_xfs_vm_bmap(XFS_I(inode));
1462 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1463 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1464 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1465 return generic_block_bmap(mapping, block, xfs_get_blocks);
1466}
1467
1468STATIC int
1469xfs_vm_readpage(
1470 struct file *unused,
1471 struct page *page)
1472{
1473 return mpage_readpage(page, xfs_get_blocks);
1474}
1475
1476STATIC int
1477xfs_vm_readpages(
1478 struct file *unused,
1479 struct address_space *mapping,
1480 struct list_head *pages,
1481 unsigned nr_pages)
1482{
1483 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1484}
1485
1486const struct address_space_operations xfs_address_space_operations = {
1487 .readpage = xfs_vm_readpage,
1488 .readpages = xfs_vm_readpages,
1489 .writepage = xfs_vm_writepage,
1490 .writepages = xfs_vm_writepages,
1491 .releasepage = xfs_vm_releasepage,
1492 .invalidatepage = xfs_vm_invalidatepage,
1493 .write_begin = xfs_vm_write_begin,
1494 .write_end = xfs_vm_write_end,
1495 .bmap = xfs_vm_bmap,
1496 .direct_IO = xfs_vm_direct_IO,
1497 .migratepage = buffer_migrate_page,
1498 .is_partially_uptodate = block_is_partially_uptodate,
1499 .error_remove_page = generic_error_remove_page,
1500};
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_shared.h"
20#include "xfs_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_inode.h"
25#include "xfs_trans.h"
26#include "xfs_inode_item.h"
27#include "xfs_alloc.h"
28#include "xfs_error.h"
29#include "xfs_iomap.h"
30#include "xfs_trace.h"
31#include "xfs_bmap.h"
32#include "xfs_bmap_util.h"
33#include "xfs_bmap_btree.h"
34#include "xfs_reflink.h"
35#include <linux/gfp.h>
36#include <linux/mpage.h>
37#include <linux/pagevec.h>
38#include <linux/writeback.h>
39
40/*
41 * structure owned by writepages passed to individual writepage calls
42 */
43struct xfs_writepage_ctx {
44 struct xfs_bmbt_irec imap;
45 bool imap_valid;
46 unsigned int io_type;
47 struct xfs_ioend *ioend;
48 sector_t last_block;
49};
50
51void
52xfs_count_page_state(
53 struct page *page,
54 int *delalloc,
55 int *unwritten)
56{
57 struct buffer_head *bh, *head;
58
59 *delalloc = *unwritten = 0;
60
61 bh = head = page_buffers(page);
62 do {
63 if (buffer_unwritten(bh))
64 (*unwritten) = 1;
65 else if (buffer_delay(bh))
66 (*delalloc) = 1;
67 } while ((bh = bh->b_this_page) != head);
68}
69
70struct block_device *
71xfs_find_bdev_for_inode(
72 struct inode *inode)
73{
74 struct xfs_inode *ip = XFS_I(inode);
75 struct xfs_mount *mp = ip->i_mount;
76
77 if (XFS_IS_REALTIME_INODE(ip))
78 return mp->m_rtdev_targp->bt_bdev;
79 else
80 return mp->m_ddev_targp->bt_bdev;
81}
82
83/*
84 * We're now finished for good with this page. Update the page state via the
85 * associated buffer_heads, paying attention to the start and end offsets that
86 * we need to process on the page.
87 *
88 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
89 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
90 * the page at all, as we may be racing with memory reclaim and it can free both
91 * the bufferhead chain and the page as it will see the page as clean and
92 * unused.
93 */
94static void
95xfs_finish_page_writeback(
96 struct inode *inode,
97 struct bio_vec *bvec,
98 int error)
99{
100 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
101 struct buffer_head *head, *bh, *next;
102 unsigned int off = 0;
103 unsigned int bsize;
104
105 ASSERT(bvec->bv_offset < PAGE_SIZE);
106 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
107 ASSERT(end < PAGE_SIZE);
108 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
109
110 bh = head = page_buffers(bvec->bv_page);
111
112 bsize = bh->b_size;
113 do {
114 next = bh->b_this_page;
115 if (off < bvec->bv_offset)
116 goto next_bh;
117 if (off > end)
118 break;
119 bh->b_end_io(bh, !error);
120next_bh:
121 off += bsize;
122 } while ((bh = next) != head);
123}
124
125/*
126 * We're now finished for good with this ioend structure. Update the page
127 * state, release holds on bios, and finally free up memory. Do not use the
128 * ioend after this.
129 */
130STATIC void
131xfs_destroy_ioend(
132 struct xfs_ioend *ioend,
133 int error)
134{
135 struct inode *inode = ioend->io_inode;
136 struct bio *last = ioend->io_bio;
137 struct bio *bio, *next;
138
139 for (bio = &ioend->io_inline_bio; bio; bio = next) {
140 struct bio_vec *bvec;
141 int i;
142
143 /*
144 * For the last bio, bi_private points to the ioend, so we
145 * need to explicitly end the iteration here.
146 */
147 if (bio == last)
148 next = NULL;
149 else
150 next = bio->bi_private;
151
152 /* walk each page on bio, ending page IO on them */
153 bio_for_each_segment_all(bvec, bio, i)
154 xfs_finish_page_writeback(inode, bvec, error);
155
156 bio_put(bio);
157 }
158}
159
160/*
161 * Fast and loose check if this write could update the on-disk inode size.
162 */
163static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
164{
165 return ioend->io_offset + ioend->io_size >
166 XFS_I(ioend->io_inode)->i_d.di_size;
167}
168
169STATIC int
170xfs_setfilesize_trans_alloc(
171 struct xfs_ioend *ioend)
172{
173 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
174 struct xfs_trans *tp;
175 int error;
176
177 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
178 if (error)
179 return error;
180
181 ioend->io_append_trans = tp;
182
183 /*
184 * We may pass freeze protection with a transaction. So tell lockdep
185 * we released it.
186 */
187 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
188 /*
189 * We hand off the transaction to the completion thread now, so
190 * clear the flag here.
191 */
192 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
193 return 0;
194}
195
196/*
197 * Update on-disk file size now that data has been written to disk.
198 */
199STATIC int
200__xfs_setfilesize(
201 struct xfs_inode *ip,
202 struct xfs_trans *tp,
203 xfs_off_t offset,
204 size_t size)
205{
206 xfs_fsize_t isize;
207
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209 isize = xfs_new_eof(ip, offset + size);
210 if (!isize) {
211 xfs_iunlock(ip, XFS_ILOCK_EXCL);
212 xfs_trans_cancel(tp);
213 return 0;
214 }
215
216 trace_xfs_setfilesize(ip, offset, size);
217
218 ip->i_d.di_size = isize;
219 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
220 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
221
222 return xfs_trans_commit(tp);
223}
224
225int
226xfs_setfilesize(
227 struct xfs_inode *ip,
228 xfs_off_t offset,
229 size_t size)
230{
231 struct xfs_mount *mp = ip->i_mount;
232 struct xfs_trans *tp;
233 int error;
234
235 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
236 if (error)
237 return error;
238
239 return __xfs_setfilesize(ip, tp, offset, size);
240}
241
242STATIC int
243xfs_setfilesize_ioend(
244 struct xfs_ioend *ioend,
245 int error)
246{
247 struct xfs_inode *ip = XFS_I(ioend->io_inode);
248 struct xfs_trans *tp = ioend->io_append_trans;
249
250 /*
251 * The transaction may have been allocated in the I/O submission thread,
252 * thus we need to mark ourselves as being in a transaction manually.
253 * Similarly for freeze protection.
254 */
255 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
256 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
257
258 /* we abort the update if there was an IO error */
259 if (error) {
260 xfs_trans_cancel(tp);
261 return error;
262 }
263
264 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
265}
266
267/*
268 * IO write completion.
269 */
270STATIC void
271xfs_end_io(
272 struct work_struct *work)
273{
274 struct xfs_ioend *ioend =
275 container_of(work, struct xfs_ioend, io_work);
276 struct xfs_inode *ip = XFS_I(ioend->io_inode);
277 xfs_off_t offset = ioend->io_offset;
278 size_t size = ioend->io_size;
279 int error = ioend->io_bio->bi_error;
280
281 /*
282 * Just clean up the in-memory strutures if the fs has been shut down.
283 */
284 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
285 error = -EIO;
286 goto done;
287 }
288
289 /*
290 * Clean up any COW blocks on an I/O error.
291 */
292 if (unlikely(error)) {
293 switch (ioend->io_type) {
294 case XFS_IO_COW:
295 xfs_reflink_cancel_cow_range(ip, offset, size, true);
296 break;
297 }
298
299 goto done;
300 }
301
302 /*
303 * Success: commit the COW or unwritten blocks if needed.
304 */
305 switch (ioend->io_type) {
306 case XFS_IO_COW:
307 error = xfs_reflink_end_cow(ip, offset, size);
308 break;
309 case XFS_IO_UNWRITTEN:
310 error = xfs_iomap_write_unwritten(ip, offset, size);
311 break;
312 default:
313 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
314 break;
315 }
316
317done:
318 if (ioend->io_append_trans)
319 error = xfs_setfilesize_ioend(ioend, error);
320 xfs_destroy_ioend(ioend, error);
321}
322
323STATIC void
324xfs_end_bio(
325 struct bio *bio)
326{
327 struct xfs_ioend *ioend = bio->bi_private;
328 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
329
330 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
331 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
332 else if (ioend->io_append_trans)
333 queue_work(mp->m_data_workqueue, &ioend->io_work);
334 else
335 xfs_destroy_ioend(ioend, bio->bi_error);
336}
337
338STATIC int
339xfs_map_blocks(
340 struct inode *inode,
341 loff_t offset,
342 struct xfs_bmbt_irec *imap,
343 int type)
344{
345 struct xfs_inode *ip = XFS_I(inode);
346 struct xfs_mount *mp = ip->i_mount;
347 ssize_t count = 1 << inode->i_blkbits;
348 xfs_fileoff_t offset_fsb, end_fsb;
349 int error = 0;
350 int bmapi_flags = XFS_BMAPI_ENTIRE;
351 int nimaps = 1;
352
353 if (XFS_FORCED_SHUTDOWN(mp))
354 return -EIO;
355
356 ASSERT(type != XFS_IO_COW);
357 if (type == XFS_IO_UNWRITTEN)
358 bmapi_flags |= XFS_BMAPI_IGSTATE;
359
360 xfs_ilock(ip, XFS_ILOCK_SHARED);
361 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
362 (ip->i_df.if_flags & XFS_IFEXTENTS));
363 ASSERT(offset <= mp->m_super->s_maxbytes);
364
365 if (offset + count > mp->m_super->s_maxbytes)
366 count = mp->m_super->s_maxbytes - offset;
367 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
368 offset_fsb = XFS_B_TO_FSBT(mp, offset);
369 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
370 imap, &nimaps, bmapi_flags);
371 /*
372 * Truncate an overwrite extent if there's a pending CoW
373 * reservation before the end of this extent. This forces us
374 * to come back to writepage to take care of the CoW.
375 */
376 if (nimaps && type == XFS_IO_OVERWRITE)
377 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
378 xfs_iunlock(ip, XFS_ILOCK_SHARED);
379
380 if (error)
381 return error;
382
383 if (type == XFS_IO_DELALLOC &&
384 (!nimaps || isnullstartblock(imap->br_startblock))) {
385 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
386 imap);
387 if (!error)
388 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
389 return error;
390 }
391
392#ifdef DEBUG
393 if (type == XFS_IO_UNWRITTEN) {
394 ASSERT(nimaps);
395 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
396 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
397 }
398#endif
399 if (nimaps)
400 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
401 return 0;
402}
403
404STATIC bool
405xfs_imap_valid(
406 struct inode *inode,
407 struct xfs_bmbt_irec *imap,
408 xfs_off_t offset)
409{
410 offset >>= inode->i_blkbits;
411
412 return offset >= imap->br_startoff &&
413 offset < imap->br_startoff + imap->br_blockcount;
414}
415
416STATIC void
417xfs_start_buffer_writeback(
418 struct buffer_head *bh)
419{
420 ASSERT(buffer_mapped(bh));
421 ASSERT(buffer_locked(bh));
422 ASSERT(!buffer_delay(bh));
423 ASSERT(!buffer_unwritten(bh));
424
425 mark_buffer_async_write(bh);
426 set_buffer_uptodate(bh);
427 clear_buffer_dirty(bh);
428}
429
430STATIC void
431xfs_start_page_writeback(
432 struct page *page,
433 int clear_dirty)
434{
435 ASSERT(PageLocked(page));
436 ASSERT(!PageWriteback(page));
437
438 /*
439 * if the page was not fully cleaned, we need to ensure that the higher
440 * layers come back to it correctly. That means we need to keep the page
441 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
442 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
443 * write this page in this writeback sweep will be made.
444 */
445 if (clear_dirty) {
446 clear_page_dirty_for_io(page);
447 set_page_writeback(page);
448 } else
449 set_page_writeback_keepwrite(page);
450
451 unlock_page(page);
452}
453
454static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
455{
456 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
457}
458
459/*
460 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
461 * it, and we submit that bio. The ioend may be used for multiple bio
462 * submissions, so we only want to allocate an append transaction for the ioend
463 * once. In the case of multiple bio submission, each bio will take an IO
464 * reference to the ioend to ensure that the ioend completion is only done once
465 * all bios have been submitted and the ioend is really done.
466 *
467 * If @fail is non-zero, it means that we have a situation where some part of
468 * the submission process has failed after we have marked paged for writeback
469 * and unlocked them. In this situation, we need to fail the bio and ioend
470 * rather than submit it to IO. This typically only happens on a filesystem
471 * shutdown.
472 */
473STATIC int
474xfs_submit_ioend(
475 struct writeback_control *wbc,
476 struct xfs_ioend *ioend,
477 int status)
478{
479 /* Convert CoW extents to regular */
480 if (!status && ioend->io_type == XFS_IO_COW) {
481 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
482 ioend->io_offset, ioend->io_size);
483 }
484
485 /* Reserve log space if we might write beyond the on-disk inode size. */
486 if (!status &&
487 ioend->io_type != XFS_IO_UNWRITTEN &&
488 xfs_ioend_is_append(ioend) &&
489 !ioend->io_append_trans)
490 status = xfs_setfilesize_trans_alloc(ioend);
491
492 ioend->io_bio->bi_private = ioend;
493 ioend->io_bio->bi_end_io = xfs_end_bio;
494 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
495
496 /*
497 * If we are failing the IO now, just mark the ioend with an
498 * error and finish it. This will run IO completion immediately
499 * as there is only one reference to the ioend at this point in
500 * time.
501 */
502 if (status) {
503 ioend->io_bio->bi_error = status;
504 bio_endio(ioend->io_bio);
505 return status;
506 }
507
508 submit_bio(ioend->io_bio);
509 return 0;
510}
511
512static void
513xfs_init_bio_from_bh(
514 struct bio *bio,
515 struct buffer_head *bh)
516{
517 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
518 bio->bi_bdev = bh->b_bdev;
519}
520
521static struct xfs_ioend *
522xfs_alloc_ioend(
523 struct inode *inode,
524 unsigned int type,
525 xfs_off_t offset,
526 struct buffer_head *bh)
527{
528 struct xfs_ioend *ioend;
529 struct bio *bio;
530
531 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
532 xfs_init_bio_from_bh(bio, bh);
533
534 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
535 INIT_LIST_HEAD(&ioend->io_list);
536 ioend->io_type = type;
537 ioend->io_inode = inode;
538 ioend->io_size = 0;
539 ioend->io_offset = offset;
540 INIT_WORK(&ioend->io_work, xfs_end_io);
541 ioend->io_append_trans = NULL;
542 ioend->io_bio = bio;
543 return ioend;
544}
545
546/*
547 * Allocate a new bio, and chain the old bio to the new one.
548 *
549 * Note that we have to do perform the chaining in this unintuitive order
550 * so that the bi_private linkage is set up in the right direction for the
551 * traversal in xfs_destroy_ioend().
552 */
553static void
554xfs_chain_bio(
555 struct xfs_ioend *ioend,
556 struct writeback_control *wbc,
557 struct buffer_head *bh)
558{
559 struct bio *new;
560
561 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
562 xfs_init_bio_from_bh(new, bh);
563
564 bio_chain(ioend->io_bio, new);
565 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
566 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
567 submit_bio(ioend->io_bio);
568 ioend->io_bio = new;
569}
570
571/*
572 * Test to see if we've been building up a completion structure for
573 * earlier buffers -- if so, we try to append to this ioend if we
574 * can, otherwise we finish off any current ioend and start another.
575 * Return the ioend we finished off so that the caller can submit it
576 * once it has finished processing the dirty page.
577 */
578STATIC void
579xfs_add_to_ioend(
580 struct inode *inode,
581 struct buffer_head *bh,
582 xfs_off_t offset,
583 struct xfs_writepage_ctx *wpc,
584 struct writeback_control *wbc,
585 struct list_head *iolist)
586{
587 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
588 bh->b_blocknr != wpc->last_block + 1 ||
589 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
590 if (wpc->ioend)
591 list_add(&wpc->ioend->io_list, iolist);
592 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
593 }
594
595 /*
596 * If the buffer doesn't fit into the bio we need to allocate a new
597 * one. This shouldn't happen more than once for a given buffer.
598 */
599 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
600 xfs_chain_bio(wpc->ioend, wbc, bh);
601
602 wpc->ioend->io_size += bh->b_size;
603 wpc->last_block = bh->b_blocknr;
604 xfs_start_buffer_writeback(bh);
605}
606
607STATIC void
608xfs_map_buffer(
609 struct inode *inode,
610 struct buffer_head *bh,
611 struct xfs_bmbt_irec *imap,
612 xfs_off_t offset)
613{
614 sector_t bn;
615 struct xfs_mount *m = XFS_I(inode)->i_mount;
616 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
617 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
618
619 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
620 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
621
622 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
623 ((offset - iomap_offset) >> inode->i_blkbits);
624
625 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
626
627 bh->b_blocknr = bn;
628 set_buffer_mapped(bh);
629}
630
631STATIC void
632xfs_map_at_offset(
633 struct inode *inode,
634 struct buffer_head *bh,
635 struct xfs_bmbt_irec *imap,
636 xfs_off_t offset)
637{
638 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
639 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
640
641 xfs_map_buffer(inode, bh, imap, offset);
642 set_buffer_mapped(bh);
643 clear_buffer_delay(bh);
644 clear_buffer_unwritten(bh);
645}
646
647/*
648 * Test if a given page contains at least one buffer of a given @type.
649 * If @check_all_buffers is true, then we walk all the buffers in the page to
650 * try to find one of the type passed in. If it is not set, then the caller only
651 * needs to check the first buffer on the page for a match.
652 */
653STATIC bool
654xfs_check_page_type(
655 struct page *page,
656 unsigned int type,
657 bool check_all_buffers)
658{
659 struct buffer_head *bh;
660 struct buffer_head *head;
661
662 if (PageWriteback(page))
663 return false;
664 if (!page->mapping)
665 return false;
666 if (!page_has_buffers(page))
667 return false;
668
669 bh = head = page_buffers(page);
670 do {
671 if (buffer_unwritten(bh)) {
672 if (type == XFS_IO_UNWRITTEN)
673 return true;
674 } else if (buffer_delay(bh)) {
675 if (type == XFS_IO_DELALLOC)
676 return true;
677 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
678 if (type == XFS_IO_OVERWRITE)
679 return true;
680 }
681
682 /* If we are only checking the first buffer, we are done now. */
683 if (!check_all_buffers)
684 break;
685 } while ((bh = bh->b_this_page) != head);
686
687 return false;
688}
689
690STATIC void
691xfs_vm_invalidatepage(
692 struct page *page,
693 unsigned int offset,
694 unsigned int length)
695{
696 trace_xfs_invalidatepage(page->mapping->host, page, offset,
697 length);
698 block_invalidatepage(page, offset, length);
699}
700
701/*
702 * If the page has delalloc buffers on it, we need to punch them out before we
703 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
704 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
705 * is done on that same region - the delalloc extent is returned when none is
706 * supposed to be there.
707 *
708 * We prevent this by truncating away the delalloc regions on the page before
709 * invalidating it. Because they are delalloc, we can do this without needing a
710 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
711 * truncation without a transaction as there is no space left for block
712 * reservation (typically why we see a ENOSPC in writeback).
713 *
714 * This is not a performance critical path, so for now just do the punching a
715 * buffer head at a time.
716 */
717STATIC void
718xfs_aops_discard_page(
719 struct page *page)
720{
721 struct inode *inode = page->mapping->host;
722 struct xfs_inode *ip = XFS_I(inode);
723 struct buffer_head *bh, *head;
724 loff_t offset = page_offset(page);
725
726 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
727 goto out_invalidate;
728
729 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
730 goto out_invalidate;
731
732 xfs_alert(ip->i_mount,
733 "page discard on page %p, inode 0x%llx, offset %llu.",
734 page, ip->i_ino, offset);
735
736 xfs_ilock(ip, XFS_ILOCK_EXCL);
737 bh = head = page_buffers(page);
738 do {
739 int error;
740 xfs_fileoff_t start_fsb;
741
742 if (!buffer_delay(bh))
743 goto next_buffer;
744
745 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
746 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
747 if (error) {
748 /* something screwed, just bail */
749 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
750 xfs_alert(ip->i_mount,
751 "page discard unable to remove delalloc mapping.");
752 }
753 break;
754 }
755next_buffer:
756 offset += 1 << inode->i_blkbits;
757
758 } while ((bh = bh->b_this_page) != head);
759
760 xfs_iunlock(ip, XFS_ILOCK_EXCL);
761out_invalidate:
762 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
763 return;
764}
765
766static int
767xfs_map_cow(
768 struct xfs_writepage_ctx *wpc,
769 struct inode *inode,
770 loff_t offset,
771 unsigned int *new_type)
772{
773 struct xfs_inode *ip = XFS_I(inode);
774 struct xfs_bmbt_irec imap;
775 bool is_cow = false;
776 int error;
777
778 /*
779 * If we already have a valid COW mapping keep using it.
780 */
781 if (wpc->io_type == XFS_IO_COW) {
782 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
783 if (wpc->imap_valid) {
784 *new_type = XFS_IO_COW;
785 return 0;
786 }
787 }
788
789 /*
790 * Else we need to check if there is a COW mapping at this offset.
791 */
792 xfs_ilock(ip, XFS_ILOCK_SHARED);
793 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
794 xfs_iunlock(ip, XFS_ILOCK_SHARED);
795
796 if (!is_cow)
797 return 0;
798
799 /*
800 * And if the COW mapping has a delayed extent here we need to
801 * allocate real space for it now.
802 */
803 if (isnullstartblock(imap.br_startblock)) {
804 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
805 &imap);
806 if (error)
807 return error;
808 }
809
810 wpc->io_type = *new_type = XFS_IO_COW;
811 wpc->imap_valid = true;
812 wpc->imap = imap;
813 return 0;
814}
815
816/*
817 * We implement an immediate ioend submission policy here to avoid needing to
818 * chain multiple ioends and hence nest mempool allocations which can violate
819 * forward progress guarantees we need to provide. The current ioend we are
820 * adding buffers to is cached on the writepage context, and if the new buffer
821 * does not append to the cached ioend it will create a new ioend and cache that
822 * instead.
823 *
824 * If a new ioend is created and cached, the old ioend is returned and queued
825 * locally for submission once the entire page is processed or an error has been
826 * detected. While ioends are submitted immediately after they are completed,
827 * batching optimisations are provided by higher level block plugging.
828 *
829 * At the end of a writeback pass, there will be a cached ioend remaining on the
830 * writepage context that the caller will need to submit.
831 */
832static int
833xfs_writepage_map(
834 struct xfs_writepage_ctx *wpc,
835 struct writeback_control *wbc,
836 struct inode *inode,
837 struct page *page,
838 loff_t offset,
839 __uint64_t end_offset)
840{
841 LIST_HEAD(submit_list);
842 struct xfs_ioend *ioend, *next;
843 struct buffer_head *bh, *head;
844 ssize_t len = 1 << inode->i_blkbits;
845 int error = 0;
846 int count = 0;
847 int uptodate = 1;
848 unsigned int new_type;
849
850 bh = head = page_buffers(page);
851 offset = page_offset(page);
852 do {
853 if (offset >= end_offset)
854 break;
855 if (!buffer_uptodate(bh))
856 uptodate = 0;
857
858 /*
859 * set_page_dirty dirties all buffers in a page, independent
860 * of their state. The dirty state however is entirely
861 * meaningless for holes (!mapped && uptodate), so skip
862 * buffers covering holes here.
863 */
864 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
865 wpc->imap_valid = false;
866 continue;
867 }
868
869 if (buffer_unwritten(bh))
870 new_type = XFS_IO_UNWRITTEN;
871 else if (buffer_delay(bh))
872 new_type = XFS_IO_DELALLOC;
873 else if (buffer_uptodate(bh))
874 new_type = XFS_IO_OVERWRITE;
875 else {
876 if (PageUptodate(page))
877 ASSERT(buffer_mapped(bh));
878 /*
879 * This buffer is not uptodate and will not be
880 * written to disk. Ensure that we will put any
881 * subsequent writeable buffers into a new
882 * ioend.
883 */
884 wpc->imap_valid = false;
885 continue;
886 }
887
888 if (xfs_is_reflink_inode(XFS_I(inode))) {
889 error = xfs_map_cow(wpc, inode, offset, &new_type);
890 if (error)
891 goto out;
892 }
893
894 if (wpc->io_type != new_type) {
895 wpc->io_type = new_type;
896 wpc->imap_valid = false;
897 }
898
899 if (wpc->imap_valid)
900 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
901 offset);
902 if (!wpc->imap_valid) {
903 error = xfs_map_blocks(inode, offset, &wpc->imap,
904 wpc->io_type);
905 if (error)
906 goto out;
907 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
908 offset);
909 }
910 if (wpc->imap_valid) {
911 lock_buffer(bh);
912 if (wpc->io_type != XFS_IO_OVERWRITE)
913 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
914 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
915 count++;
916 }
917
918 } while (offset += len, ((bh = bh->b_this_page) != head));
919
920 if (uptodate && bh == head)
921 SetPageUptodate(page);
922
923 ASSERT(wpc->ioend || list_empty(&submit_list));
924
925out:
926 /*
927 * On error, we have to fail the ioend here because we have locked
928 * buffers in the ioend. If we don't do this, we'll deadlock
929 * invalidating the page as that tries to lock the buffers on the page.
930 * Also, because we may have set pages under writeback, we have to make
931 * sure we run IO completion to mark the error state of the IO
932 * appropriately, so we can't cancel the ioend directly here. That means
933 * we have to mark this page as under writeback if we included any
934 * buffers from it in the ioend chain so that completion treats it
935 * correctly.
936 *
937 * If we didn't include the page in the ioend, the on error we can
938 * simply discard and unlock it as there are no other users of the page
939 * or it's buffers right now. The caller will still need to trigger
940 * submission of outstanding ioends on the writepage context so they are
941 * treated correctly on error.
942 */
943 if (count) {
944 xfs_start_page_writeback(page, !error);
945
946 /*
947 * Preserve the original error if there was one, otherwise catch
948 * submission errors here and propagate into subsequent ioend
949 * submissions.
950 */
951 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
952 int error2;
953
954 list_del_init(&ioend->io_list);
955 error2 = xfs_submit_ioend(wbc, ioend, error);
956 if (error2 && !error)
957 error = error2;
958 }
959 } else if (error) {
960 xfs_aops_discard_page(page);
961 ClearPageUptodate(page);
962 unlock_page(page);
963 } else {
964 /*
965 * We can end up here with no error and nothing to write if we
966 * race with a partial page truncate on a sub-page block sized
967 * filesystem. In that case we need to mark the page clean.
968 */
969 xfs_start_page_writeback(page, 1);
970 end_page_writeback(page);
971 }
972
973 mapping_set_error(page->mapping, error);
974 return error;
975}
976
977/*
978 * Write out a dirty page.
979 *
980 * For delalloc space on the page we need to allocate space and flush it.
981 * For unwritten space on the page we need to start the conversion to
982 * regular allocated space.
983 * For any other dirty buffer heads on the page we should flush them.
984 */
985STATIC int
986xfs_do_writepage(
987 struct page *page,
988 struct writeback_control *wbc,
989 void *data)
990{
991 struct xfs_writepage_ctx *wpc = data;
992 struct inode *inode = page->mapping->host;
993 loff_t offset;
994 __uint64_t end_offset;
995 pgoff_t end_index;
996
997 trace_xfs_writepage(inode, page, 0, 0);
998
999 ASSERT(page_has_buffers(page));
1000
1001 /*
1002 * Refuse to write the page out if we are called from reclaim context.
1003 *
1004 * This avoids stack overflows when called from deeply used stacks in
1005 * random callers for direct reclaim or memcg reclaim. We explicitly
1006 * allow reclaim from kswapd as the stack usage there is relatively low.
1007 *
1008 * This should never happen except in the case of a VM regression so
1009 * warn about it.
1010 */
1011 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1012 PF_MEMALLOC))
1013 goto redirty;
1014
1015 /*
1016 * Given that we do not allow direct reclaim to call us, we should
1017 * never be called while in a filesystem transaction.
1018 */
1019 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
1020 goto redirty;
1021
1022 /*
1023 * Is this page beyond the end of the file?
1024 *
1025 * The page index is less than the end_index, adjust the end_offset
1026 * to the highest offset that this page should represent.
1027 * -----------------------------------------------------
1028 * | file mapping | <EOF> |
1029 * -----------------------------------------------------
1030 * | Page ... | Page N-2 | Page N-1 | Page N | |
1031 * ^--------------------------------^----------|--------
1032 * | desired writeback range | see else |
1033 * ---------------------------------^------------------|
1034 */
1035 offset = i_size_read(inode);
1036 end_index = offset >> PAGE_SHIFT;
1037 if (page->index < end_index)
1038 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1039 else {
1040 /*
1041 * Check whether the page to write out is beyond or straddles
1042 * i_size or not.
1043 * -------------------------------------------------------
1044 * | file mapping | <EOF> |
1045 * -------------------------------------------------------
1046 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1047 * ^--------------------------------^-----------|---------
1048 * | | Straddles |
1049 * ---------------------------------^-----------|--------|
1050 */
1051 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1052
1053 /*
1054 * Skip the page if it is fully outside i_size, e.g. due to a
1055 * truncate operation that is in progress. We must redirty the
1056 * page so that reclaim stops reclaiming it. Otherwise
1057 * xfs_vm_releasepage() is called on it and gets confused.
1058 *
1059 * Note that the end_index is unsigned long, it would overflow
1060 * if the given offset is greater than 16TB on 32-bit system
1061 * and if we do check the page is fully outside i_size or not
1062 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1063 * will be evaluated to 0. Hence this page will be redirtied
1064 * and be written out repeatedly which would result in an
1065 * infinite loop, the user program that perform this operation
1066 * will hang. Instead, we can verify this situation by checking
1067 * if the page to write is totally beyond the i_size or if it's
1068 * offset is just equal to the EOF.
1069 */
1070 if (page->index > end_index ||
1071 (page->index == end_index && offset_into_page == 0))
1072 goto redirty;
1073
1074 /*
1075 * The page straddles i_size. It must be zeroed out on each
1076 * and every writepage invocation because it may be mmapped.
1077 * "A file is mapped in multiples of the page size. For a file
1078 * that is not a multiple of the page size, the remaining
1079 * memory is zeroed when mapped, and writes to that region are
1080 * not written out to the file."
1081 */
1082 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1083
1084 /* Adjust the end_offset to the end of file */
1085 end_offset = offset;
1086 }
1087
1088 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1089
1090redirty:
1091 redirty_page_for_writepage(wbc, page);
1092 unlock_page(page);
1093 return 0;
1094}
1095
1096STATIC int
1097xfs_vm_writepage(
1098 struct page *page,
1099 struct writeback_control *wbc)
1100{
1101 struct xfs_writepage_ctx wpc = {
1102 .io_type = XFS_IO_INVALID,
1103 };
1104 int ret;
1105
1106 ret = xfs_do_writepage(page, wbc, &wpc);
1107 if (wpc.ioend)
1108 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1109 return ret;
1110}
1111
1112STATIC int
1113xfs_vm_writepages(
1114 struct address_space *mapping,
1115 struct writeback_control *wbc)
1116{
1117 struct xfs_writepage_ctx wpc = {
1118 .io_type = XFS_IO_INVALID,
1119 };
1120 int ret;
1121
1122 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1123 if (dax_mapping(mapping))
1124 return dax_writeback_mapping_range(mapping,
1125 xfs_find_bdev_for_inode(mapping->host), wbc);
1126
1127 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1128 if (wpc.ioend)
1129 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1130 return ret;
1131}
1132
1133/*
1134 * Called to move a page into cleanable state - and from there
1135 * to be released. The page should already be clean. We always
1136 * have buffer heads in this call.
1137 *
1138 * Returns 1 if the page is ok to release, 0 otherwise.
1139 */
1140STATIC int
1141xfs_vm_releasepage(
1142 struct page *page,
1143 gfp_t gfp_mask)
1144{
1145 int delalloc, unwritten;
1146
1147 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1148
1149 /*
1150 * mm accommodates an old ext3 case where clean pages might not have had
1151 * the dirty bit cleared. Thus, it can send actual dirty pages to
1152 * ->releasepage() via shrink_active_list(). Conversely,
1153 * block_invalidatepage() can send pages that are still marked dirty
1154 * but otherwise have invalidated buffers.
1155 *
1156 * We want to release the latter to avoid unnecessary buildup of the
1157 * LRU, skip the former and warn if we've left any lingering
1158 * delalloc/unwritten buffers on clean pages. Skip pages with delalloc
1159 * or unwritten buffers and warn if the page is not dirty. Otherwise
1160 * try to release the buffers.
1161 */
1162 xfs_count_page_state(page, &delalloc, &unwritten);
1163
1164 if (delalloc) {
1165 WARN_ON_ONCE(!PageDirty(page));
1166 return 0;
1167 }
1168 if (unwritten) {
1169 WARN_ON_ONCE(!PageDirty(page));
1170 return 0;
1171 }
1172
1173 return try_to_free_buffers(page);
1174}
1175
1176/*
1177 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1178 * is, so that we can avoid repeated get_blocks calls.
1179 *
1180 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1181 * for blocks beyond EOF must be marked new so that sub block regions can be
1182 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1183 * was just allocated or is unwritten, otherwise the callers would overwrite
1184 * existing data with zeros. Hence we have to split the mapping into a range up
1185 * to and including EOF, and a second mapping for beyond EOF.
1186 */
1187static void
1188xfs_map_trim_size(
1189 struct inode *inode,
1190 sector_t iblock,
1191 struct buffer_head *bh_result,
1192 struct xfs_bmbt_irec *imap,
1193 xfs_off_t offset,
1194 ssize_t size)
1195{
1196 xfs_off_t mapping_size;
1197
1198 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1199 mapping_size <<= inode->i_blkbits;
1200
1201 ASSERT(mapping_size > 0);
1202 if (mapping_size > size)
1203 mapping_size = size;
1204 if (offset < i_size_read(inode) &&
1205 offset + mapping_size >= i_size_read(inode)) {
1206 /* limit mapping to block that spans EOF */
1207 mapping_size = roundup_64(i_size_read(inode) - offset,
1208 1 << inode->i_blkbits);
1209 }
1210 if (mapping_size > LONG_MAX)
1211 mapping_size = LONG_MAX;
1212
1213 bh_result->b_size = mapping_size;
1214}
1215
1216static int
1217xfs_get_blocks(
1218 struct inode *inode,
1219 sector_t iblock,
1220 struct buffer_head *bh_result,
1221 int create)
1222{
1223 struct xfs_inode *ip = XFS_I(inode);
1224 struct xfs_mount *mp = ip->i_mount;
1225 xfs_fileoff_t offset_fsb, end_fsb;
1226 int error = 0;
1227 int lockmode = 0;
1228 struct xfs_bmbt_irec imap;
1229 int nimaps = 1;
1230 xfs_off_t offset;
1231 ssize_t size;
1232
1233 BUG_ON(create);
1234
1235 if (XFS_FORCED_SHUTDOWN(mp))
1236 return -EIO;
1237
1238 offset = (xfs_off_t)iblock << inode->i_blkbits;
1239 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1240 size = bh_result->b_size;
1241
1242 if (offset >= i_size_read(inode))
1243 return 0;
1244
1245 /*
1246 * Direct I/O is usually done on preallocated files, so try getting
1247 * a block mapping without an exclusive lock first.
1248 */
1249 lockmode = xfs_ilock_data_map_shared(ip);
1250
1251 ASSERT(offset <= mp->m_super->s_maxbytes);
1252 if (offset + size > mp->m_super->s_maxbytes)
1253 size = mp->m_super->s_maxbytes - offset;
1254 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1255 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1256
1257 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1258 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1259 if (error)
1260 goto out_unlock;
1261
1262 if (nimaps) {
1263 trace_xfs_get_blocks_found(ip, offset, size,
1264 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1265 : XFS_IO_OVERWRITE, &imap);
1266 xfs_iunlock(ip, lockmode);
1267 } else {
1268 trace_xfs_get_blocks_notfound(ip, offset, size);
1269 goto out_unlock;
1270 }
1271
1272 /* trim mapping down to size requested */
1273 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1274
1275 /*
1276 * For unwritten extents do not report a disk address in the buffered
1277 * read case (treat as if we're reading into a hole).
1278 */
1279 if (imap.br_startblock != HOLESTARTBLOCK &&
1280 imap.br_startblock != DELAYSTARTBLOCK &&
1281 !ISUNWRITTEN(&imap))
1282 xfs_map_buffer(inode, bh_result, &imap, offset);
1283
1284 /*
1285 * If this is a realtime file, data may be on a different device.
1286 * to that pointed to from the buffer_head b_bdev currently.
1287 */
1288 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1289 return 0;
1290
1291out_unlock:
1292 xfs_iunlock(ip, lockmode);
1293 return error;
1294}
1295
1296STATIC ssize_t
1297xfs_vm_direct_IO(
1298 struct kiocb *iocb,
1299 struct iov_iter *iter)
1300{
1301 /*
1302 * We just need the method present so that open/fcntl allow direct I/O.
1303 */
1304 return -EINVAL;
1305}
1306
1307STATIC sector_t
1308xfs_vm_bmap(
1309 struct address_space *mapping,
1310 sector_t block)
1311{
1312 struct inode *inode = (struct inode *)mapping->host;
1313 struct xfs_inode *ip = XFS_I(inode);
1314
1315 trace_xfs_vm_bmap(XFS_I(inode));
1316
1317 /*
1318 * The swap code (ab-)uses ->bmap to get a block mapping and then
1319 * bypasseѕ the file system for actual I/O. We really can't allow
1320 * that on reflinks inodes, so we have to skip out here. And yes,
1321 * 0 is the magic code for a bmap error..
1322 */
1323 if (xfs_is_reflink_inode(ip))
1324 return 0;
1325
1326 filemap_write_and_wait(mapping);
1327 return generic_block_bmap(mapping, block, xfs_get_blocks);
1328}
1329
1330STATIC int
1331xfs_vm_readpage(
1332 struct file *unused,
1333 struct page *page)
1334{
1335 trace_xfs_vm_readpage(page->mapping->host, 1);
1336 return mpage_readpage(page, xfs_get_blocks);
1337}
1338
1339STATIC int
1340xfs_vm_readpages(
1341 struct file *unused,
1342 struct address_space *mapping,
1343 struct list_head *pages,
1344 unsigned nr_pages)
1345{
1346 trace_xfs_vm_readpages(mapping->host, nr_pages);
1347 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1348}
1349
1350/*
1351 * This is basically a copy of __set_page_dirty_buffers() with one
1352 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1353 * dirty, we'll never be able to clean them because we don't write buffers
1354 * beyond EOF, and that means we can't invalidate pages that span EOF
1355 * that have been marked dirty. Further, the dirty state can leak into
1356 * the file interior if the file is extended, resulting in all sorts of
1357 * bad things happening as the state does not match the underlying data.
1358 *
1359 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1360 * this only exist because of bufferheads and how the generic code manages them.
1361 */
1362STATIC int
1363xfs_vm_set_page_dirty(
1364 struct page *page)
1365{
1366 struct address_space *mapping = page->mapping;
1367 struct inode *inode = mapping->host;
1368 loff_t end_offset;
1369 loff_t offset;
1370 int newly_dirty;
1371
1372 if (unlikely(!mapping))
1373 return !TestSetPageDirty(page);
1374
1375 end_offset = i_size_read(inode);
1376 offset = page_offset(page);
1377
1378 spin_lock(&mapping->private_lock);
1379 if (page_has_buffers(page)) {
1380 struct buffer_head *head = page_buffers(page);
1381 struct buffer_head *bh = head;
1382
1383 do {
1384 if (offset < end_offset)
1385 set_buffer_dirty(bh);
1386 bh = bh->b_this_page;
1387 offset += 1 << inode->i_blkbits;
1388 } while (bh != head);
1389 }
1390 /*
1391 * Lock out page->mem_cgroup migration to keep PageDirty
1392 * synchronized with per-memcg dirty page counters.
1393 */
1394 lock_page_memcg(page);
1395 newly_dirty = !TestSetPageDirty(page);
1396 spin_unlock(&mapping->private_lock);
1397
1398 if (newly_dirty) {
1399 /* sigh - __set_page_dirty() is static, so copy it here, too */
1400 unsigned long flags;
1401
1402 spin_lock_irqsave(&mapping->tree_lock, flags);
1403 if (page->mapping) { /* Race with truncate? */
1404 WARN_ON_ONCE(!PageUptodate(page));
1405 account_page_dirtied(page, mapping);
1406 radix_tree_tag_set(&mapping->page_tree,
1407 page_index(page), PAGECACHE_TAG_DIRTY);
1408 }
1409 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1410 }
1411 unlock_page_memcg(page);
1412 if (newly_dirty)
1413 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1414 return newly_dirty;
1415}
1416
1417const struct address_space_operations xfs_address_space_operations = {
1418 .readpage = xfs_vm_readpage,
1419 .readpages = xfs_vm_readpages,
1420 .writepage = xfs_vm_writepage,
1421 .writepages = xfs_vm_writepages,
1422 .set_page_dirty = xfs_vm_set_page_dirty,
1423 .releasepage = xfs_vm_releasepage,
1424 .invalidatepage = xfs_vm_invalidatepage,
1425 .bmap = xfs_vm_bmap,
1426 .direct_IO = xfs_vm_direct_IO,
1427 .migratepage = buffer_migrate_page,
1428 .is_partially_uptodate = block_is_partially_uptodate,
1429 .error_remove_page = generic_error_remove_page,
1430};