2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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.
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.
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
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"
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>
41 * structure owned by writepages passed to individual writepage calls
43 struct xfs_writepage_ctx {
44 struct xfs_bmbt_irec imap;
47 struct xfs_ioend *ioend;
57 struct buffer_head *bh, *head;
59 *delalloc = *unwritten = 0;
61 bh = head = page_buffers(page);
63 if (buffer_unwritten(bh))
65 else if (buffer_delay(bh))
67 } while ((bh = bh->b_this_page) != head);
71 xfs_find_bdev_for_inode(
74 struct xfs_inode *ip = XFS_I(inode);
75 struct xfs_mount *mp = ip->i_mount;
77 if (XFS_IS_REALTIME_INODE(ip))
78 return mp->m_rtdev_targp->bt_bdev;
80 return mp->m_ddev_targp->bt_bdev;
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.
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
95 xfs_finish_page_writeback(
100 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
101 struct buffer_head *head, *bh, *next;
102 unsigned int off = 0;
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);
110 bh = head = page_buffers(bvec->bv_page);
114 next = bh->b_this_page;
115 if (off < bvec->bv_offset)
119 bh->b_end_io(bh, !error);
122 } while ((bh = next) != head);
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
132 struct xfs_ioend *ioend,
135 struct inode *inode = ioend->io_inode;
136 struct bio *last = ioend->io_bio;
137 struct bio *bio, *next;
139 for (bio = &ioend->io_inline_bio; bio; bio = next) {
140 struct bio_vec *bvec;
144 * For the last bio, bi_private points to the ioend, so we
145 * need to explicitly end the iteration here.
150 next = bio->bi_private;
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);
161 * Fast and loose check if this write could update the on-disk inode size.
163 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
165 return ioend->io_offset + ioend->io_size >
166 XFS_I(ioend->io_inode)->i_d.di_size;
170 xfs_setfilesize_trans_alloc(
171 struct xfs_ioend *ioend)
173 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
174 struct xfs_trans *tp;
177 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
181 ioend->io_append_trans = tp;
184 * We may pass freeze protection with a transaction. So tell lockdep
187 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
189 * We hand off the transaction to the completion thread now, so
190 * clear the flag here.
192 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
197 * Update on-disk file size now that data has been written to disk.
201 struct xfs_inode *ip,
202 struct xfs_trans *tp,
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209 isize = xfs_new_eof(ip, offset + size);
211 xfs_iunlock(ip, XFS_ILOCK_EXCL);
212 xfs_trans_cancel(tp);
216 trace_xfs_setfilesize(ip, offset, size);
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);
222 return xfs_trans_commit(tp);
227 struct xfs_inode *ip,
231 struct xfs_mount *mp = ip->i_mount;
232 struct xfs_trans *tp;
235 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
239 return __xfs_setfilesize(ip, tp, offset, size);
243 xfs_setfilesize_ioend(
244 struct xfs_ioend *ioend,
247 struct xfs_inode *ip = XFS_I(ioend->io_inode);
248 struct xfs_trans *tp = ioend->io_append_trans;
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.
255 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
256 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
258 /* we abort the update if there was an IO error */
260 xfs_trans_cancel(tp);
264 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
268 * IO write completion.
272 struct work_struct *work)
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 int error = ioend->io_bio->bi_error;
280 * Set an error if the mount has shut down and proceed with end I/O
281 * processing so it can perform whatever cleanups are necessary.
283 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
287 * For a CoW extent, we need to move the mapping from the CoW fork
288 * to the data fork. If instead an error happened, just dump the
291 if (ioend->io_type == XFS_IO_COW) {
294 if (ioend->io_bio->bi_error) {
295 error = xfs_reflink_cancel_cow_range(ip,
296 ioend->io_offset, ioend->io_size);
299 error = xfs_reflink_end_cow(ip, ioend->io_offset,
306 * For unwritten extents we need to issue transactions to convert a
307 * range to normal written extens after the data I/O has finished.
308 * Detecting and handling completion IO errors is done individually
309 * for each case as different cleanup operations need to be performed
312 if (ioend->io_type == XFS_IO_UNWRITTEN) {
315 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
317 } else if (ioend->io_append_trans) {
318 error = xfs_setfilesize_ioend(ioend, error);
320 ASSERT(!xfs_ioend_is_append(ioend) ||
321 ioend->io_type == XFS_IO_COW);
325 xfs_destroy_ioend(ioend, error);
332 struct xfs_ioend *ioend = bio->bi_private;
333 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
335 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
336 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
337 else if (ioend->io_append_trans)
338 queue_work(mp->m_data_workqueue, &ioend->io_work);
340 xfs_destroy_ioend(ioend, bio->bi_error);
347 struct xfs_bmbt_irec *imap,
350 struct xfs_inode *ip = XFS_I(inode);
351 struct xfs_mount *mp = ip->i_mount;
352 ssize_t count = 1 << inode->i_blkbits;
353 xfs_fileoff_t offset_fsb, end_fsb;
355 int bmapi_flags = XFS_BMAPI_ENTIRE;
358 if (XFS_FORCED_SHUTDOWN(mp))
361 ASSERT(type != XFS_IO_COW);
362 if (type == XFS_IO_UNWRITTEN)
363 bmapi_flags |= XFS_BMAPI_IGSTATE;
365 xfs_ilock(ip, XFS_ILOCK_SHARED);
366 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
367 (ip->i_df.if_flags & XFS_IFEXTENTS));
368 ASSERT(offset <= mp->m_super->s_maxbytes);
370 if (offset + count > mp->m_super->s_maxbytes)
371 count = mp->m_super->s_maxbytes - offset;
372 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
373 offset_fsb = XFS_B_TO_FSBT(mp, offset);
374 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
375 imap, &nimaps, bmapi_flags);
377 * Truncate an overwrite extent if there's a pending CoW
378 * reservation before the end of this extent. This forces us
379 * to come back to writepage to take care of the CoW.
381 if (nimaps && type == XFS_IO_OVERWRITE)
382 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
383 xfs_iunlock(ip, XFS_ILOCK_SHARED);
388 if (type == XFS_IO_DELALLOC &&
389 (!nimaps || isnullstartblock(imap->br_startblock))) {
390 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
393 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
398 if (type == XFS_IO_UNWRITTEN) {
400 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
401 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
405 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
412 struct xfs_bmbt_irec *imap,
415 offset >>= inode->i_blkbits;
417 return offset >= imap->br_startoff &&
418 offset < imap->br_startoff + imap->br_blockcount;
422 xfs_start_buffer_writeback(
423 struct buffer_head *bh)
425 ASSERT(buffer_mapped(bh));
426 ASSERT(buffer_locked(bh));
427 ASSERT(!buffer_delay(bh));
428 ASSERT(!buffer_unwritten(bh));
430 mark_buffer_async_write(bh);
431 set_buffer_uptodate(bh);
432 clear_buffer_dirty(bh);
436 xfs_start_page_writeback(
440 ASSERT(PageLocked(page));
441 ASSERT(!PageWriteback(page));
444 * if the page was not fully cleaned, we need to ensure that the higher
445 * layers come back to it correctly. That means we need to keep the page
446 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
447 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
448 * write this page in this writeback sweep will be made.
451 clear_page_dirty_for_io(page);
452 set_page_writeback(page);
454 set_page_writeback_keepwrite(page);
459 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
461 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
465 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
466 * it, and we submit that bio. The ioend may be used for multiple bio
467 * submissions, so we only want to allocate an append transaction for the ioend
468 * once. In the case of multiple bio submission, each bio will take an IO
469 * reference to the ioend to ensure that the ioend completion is only done once
470 * all bios have been submitted and the ioend is really done.
472 * If @fail is non-zero, it means that we have a situation where some part of
473 * the submission process has failed after we have marked paged for writeback
474 * and unlocked them. In this situation, we need to fail the bio and ioend
475 * rather than submit it to IO. This typically only happens on a filesystem
480 struct writeback_control *wbc,
481 struct xfs_ioend *ioend,
484 /* Reserve log space if we might write beyond the on-disk inode size. */
486 ioend->io_type != XFS_IO_UNWRITTEN &&
487 xfs_ioend_is_append(ioend) &&
488 !ioend->io_append_trans)
489 status = xfs_setfilesize_trans_alloc(ioend);
491 ioend->io_bio->bi_private = ioend;
492 ioend->io_bio->bi_end_io = xfs_end_bio;
493 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
496 * If we are failing the IO now, just mark the ioend with an
497 * error and finish it. This will run IO completion immediately
498 * as there is only one reference to the ioend at this point in
502 ioend->io_bio->bi_error = status;
503 bio_endio(ioend->io_bio);
507 submit_bio(ioend->io_bio);
512 xfs_init_bio_from_bh(
514 struct buffer_head *bh)
516 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
517 bio->bi_bdev = bh->b_bdev;
520 static struct xfs_ioend *
525 struct buffer_head *bh)
527 struct xfs_ioend *ioend;
530 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
531 xfs_init_bio_from_bh(bio, bh);
533 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
534 INIT_LIST_HEAD(&ioend->io_list);
535 ioend->io_type = type;
536 ioend->io_inode = inode;
538 ioend->io_offset = offset;
539 INIT_WORK(&ioend->io_work, xfs_end_io);
540 ioend->io_append_trans = NULL;
546 * Allocate a new bio, and chain the old bio to the new one.
548 * Note that we have to do perform the chaining in this unintuitive order
549 * so that the bi_private linkage is set up in the right direction for the
550 * traversal in xfs_destroy_ioend().
554 struct xfs_ioend *ioend,
555 struct writeback_control *wbc,
556 struct buffer_head *bh)
560 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
561 xfs_init_bio_from_bh(new, bh);
563 bio_chain(ioend->io_bio, new);
564 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
565 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
566 submit_bio(ioend->io_bio);
571 * Test to see if we've been building up a completion structure for
572 * earlier buffers -- if so, we try to append to this ioend if we
573 * can, otherwise we finish off any current ioend and start another.
574 * Return the ioend we finished off so that the caller can submit it
575 * once it has finished processing the dirty page.
580 struct buffer_head *bh,
582 struct xfs_writepage_ctx *wpc,
583 struct writeback_control *wbc,
584 struct list_head *iolist)
586 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
587 bh->b_blocknr != wpc->last_block + 1 ||
588 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
590 list_add(&wpc->ioend->io_list, iolist);
591 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
595 * If the buffer doesn't fit into the bio we need to allocate a new
596 * one. This shouldn't happen more than once for a given buffer.
598 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
599 xfs_chain_bio(wpc->ioend, wbc, bh);
601 wpc->ioend->io_size += bh->b_size;
602 wpc->last_block = bh->b_blocknr;
603 xfs_start_buffer_writeback(bh);
609 struct buffer_head *bh,
610 struct xfs_bmbt_irec *imap,
614 struct xfs_mount *m = XFS_I(inode)->i_mount;
615 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
616 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
618 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
619 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
621 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
622 ((offset - iomap_offset) >> inode->i_blkbits);
624 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
627 set_buffer_mapped(bh);
633 struct buffer_head *bh,
634 struct xfs_bmbt_irec *imap,
637 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
638 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
640 xfs_map_buffer(inode, bh, imap, offset);
641 set_buffer_mapped(bh);
642 clear_buffer_delay(bh);
643 clear_buffer_unwritten(bh);
647 * Test if a given page contains at least one buffer of a given @type.
648 * If @check_all_buffers is true, then we walk all the buffers in the page to
649 * try to find one of the type passed in. If it is not set, then the caller only
650 * needs to check the first buffer on the page for a match.
656 bool check_all_buffers)
658 struct buffer_head *bh;
659 struct buffer_head *head;
661 if (PageWriteback(page))
665 if (!page_has_buffers(page))
668 bh = head = page_buffers(page);
670 if (buffer_unwritten(bh)) {
671 if (type == XFS_IO_UNWRITTEN)
673 } else if (buffer_delay(bh)) {
674 if (type == XFS_IO_DELALLOC)
676 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
677 if (type == XFS_IO_OVERWRITE)
681 /* If we are only checking the first buffer, we are done now. */
682 if (!check_all_buffers)
684 } while ((bh = bh->b_this_page) != head);
690 xfs_vm_invalidatepage(
695 trace_xfs_invalidatepage(page->mapping->host, page, offset,
697 block_invalidatepage(page, offset, length);
701 * If the page has delalloc buffers on it, we need to punch them out before we
702 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
703 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
704 * is done on that same region - the delalloc extent is returned when none is
705 * supposed to be there.
707 * We prevent this by truncating away the delalloc regions on the page before
708 * invalidating it. Because they are delalloc, we can do this without needing a
709 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
710 * truncation without a transaction as there is no space left for block
711 * reservation (typically why we see a ENOSPC in writeback).
713 * This is not a performance critical path, so for now just do the punching a
714 * buffer head at a time.
717 xfs_aops_discard_page(
720 struct inode *inode = page->mapping->host;
721 struct xfs_inode *ip = XFS_I(inode);
722 struct buffer_head *bh, *head;
723 loff_t offset = page_offset(page);
725 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
728 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
731 xfs_alert(ip->i_mount,
732 "page discard on page %p, inode 0x%llx, offset %llu.",
733 page, ip->i_ino, offset);
735 xfs_ilock(ip, XFS_ILOCK_EXCL);
736 bh = head = page_buffers(page);
739 xfs_fileoff_t start_fsb;
741 if (!buffer_delay(bh))
744 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
745 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
747 /* something screwed, just bail */
748 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
749 xfs_alert(ip->i_mount,
750 "page discard unable to remove delalloc mapping.");
755 offset += 1 << inode->i_blkbits;
757 } while ((bh = bh->b_this_page) != head);
759 xfs_iunlock(ip, XFS_ILOCK_EXCL);
761 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
767 struct xfs_writepage_ctx *wpc,
770 unsigned int *new_type)
772 struct xfs_inode *ip = XFS_I(inode);
773 struct xfs_bmbt_irec imap;
778 * If we already have a valid COW mapping keep using it.
780 if (wpc->io_type == XFS_IO_COW) {
781 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
782 if (wpc->imap_valid) {
783 *new_type = XFS_IO_COW;
789 * Else we need to check if there is a COW mapping at this offset.
791 xfs_ilock(ip, XFS_ILOCK_SHARED);
792 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
793 xfs_iunlock(ip, XFS_ILOCK_SHARED);
799 * And if the COW mapping has a delayed extent here we need to
800 * allocate real space for it now.
802 if (isnullstartblock(imap.br_startblock)) {
803 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
809 wpc->io_type = *new_type = XFS_IO_COW;
810 wpc->imap_valid = true;
816 * We implement an immediate ioend submission policy here to avoid needing to
817 * chain multiple ioends and hence nest mempool allocations which can violate
818 * forward progress guarantees we need to provide. The current ioend we are
819 * adding buffers to is cached on the writepage context, and if the new buffer
820 * does not append to the cached ioend it will create a new ioend and cache that
823 * If a new ioend is created and cached, the old ioend is returned and queued
824 * locally for submission once the entire page is processed or an error has been
825 * detected. While ioends are submitted immediately after they are completed,
826 * batching optimisations are provided by higher level block plugging.
828 * At the end of a writeback pass, there will be a cached ioend remaining on the
829 * writepage context that the caller will need to submit.
833 struct xfs_writepage_ctx *wpc,
834 struct writeback_control *wbc,
838 __uint64_t end_offset)
840 LIST_HEAD(submit_list);
841 struct xfs_ioend *ioend, *next;
842 struct buffer_head *bh, *head;
843 ssize_t len = 1 << inode->i_blkbits;
847 unsigned int new_type;
849 bh = head = page_buffers(page);
850 offset = page_offset(page);
852 if (offset >= end_offset)
854 if (!buffer_uptodate(bh))
858 * set_page_dirty dirties all buffers in a page, independent
859 * of their state. The dirty state however is entirely
860 * meaningless for holes (!mapped && uptodate), so skip
861 * buffers covering holes here.
863 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
864 wpc->imap_valid = false;
868 if (buffer_unwritten(bh))
869 new_type = XFS_IO_UNWRITTEN;
870 else if (buffer_delay(bh))
871 new_type = XFS_IO_DELALLOC;
872 else if (buffer_uptodate(bh))
873 new_type = XFS_IO_OVERWRITE;
875 if (PageUptodate(page))
876 ASSERT(buffer_mapped(bh));
878 * This buffer is not uptodate and will not be
879 * written to disk. Ensure that we will put any
880 * subsequent writeable buffers into a new
883 wpc->imap_valid = false;
887 if (xfs_is_reflink_inode(XFS_I(inode))) {
888 error = xfs_map_cow(wpc, inode, offset, &new_type);
893 if (wpc->io_type != new_type) {
894 wpc->io_type = new_type;
895 wpc->imap_valid = false;
899 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
901 if (!wpc->imap_valid) {
902 error = xfs_map_blocks(inode, offset, &wpc->imap,
906 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
909 if (wpc->imap_valid) {
911 if (wpc->io_type != XFS_IO_OVERWRITE)
912 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
913 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
917 } while (offset += len, ((bh = bh->b_this_page) != head));
919 if (uptodate && bh == head)
920 SetPageUptodate(page);
922 ASSERT(wpc->ioend || list_empty(&submit_list));
926 * On error, we have to fail the ioend here because we have locked
927 * buffers in the ioend. If we don't do this, we'll deadlock
928 * invalidating the page as that tries to lock the buffers on the page.
929 * Also, because we may have set pages under writeback, we have to make
930 * sure we run IO completion to mark the error state of the IO
931 * appropriately, so we can't cancel the ioend directly here. That means
932 * we have to mark this page as under writeback if we included any
933 * buffers from it in the ioend chain so that completion treats it
936 * If we didn't include the page in the ioend, the on error we can
937 * simply discard and unlock it as there are no other users of the page
938 * or it's buffers right now. The caller will still need to trigger
939 * submission of outstanding ioends on the writepage context so they are
940 * treated correctly on error.
943 xfs_start_page_writeback(page, !error);
946 * Preserve the original error if there was one, otherwise catch
947 * submission errors here and propagate into subsequent ioend
950 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
953 list_del_init(&ioend->io_list);
954 error2 = xfs_submit_ioend(wbc, ioend, error);
955 if (error2 && !error)
959 xfs_aops_discard_page(page);
960 ClearPageUptodate(page);
964 * We can end up here with no error and nothing to write if we
965 * race with a partial page truncate on a sub-page block sized
966 * filesystem. In that case we need to mark the page clean.
968 xfs_start_page_writeback(page, 1);
969 end_page_writeback(page);
972 mapping_set_error(page->mapping, error);
977 * Write out a dirty page.
979 * For delalloc space on the page we need to allocate space and flush it.
980 * For unwritten space on the page we need to start the conversion to
981 * regular allocated space.
982 * For any other dirty buffer heads on the page we should flush them.
987 struct writeback_control *wbc,
990 struct xfs_writepage_ctx *wpc = data;
991 struct inode *inode = page->mapping->host;
993 __uint64_t end_offset;
996 trace_xfs_writepage(inode, page, 0, 0);
998 ASSERT(page_has_buffers(page));
1001 * Refuse to write the page out if we are called from reclaim context.
1003 * This avoids stack overflows when called from deeply used stacks in
1004 * random callers for direct reclaim or memcg reclaim. We explicitly
1005 * allow reclaim from kswapd as the stack usage there is relatively low.
1007 * This should never happen except in the case of a VM regression so
1010 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1015 * Given that we do not allow direct reclaim to call us, we should
1016 * never be called while in a filesystem transaction.
1018 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
1022 * Is this page beyond the end of the file?
1024 * The page index is less than the end_index, adjust the end_offset
1025 * to the highest offset that this page should represent.
1026 * -----------------------------------------------------
1027 * | file mapping | <EOF> |
1028 * -----------------------------------------------------
1029 * | Page ... | Page N-2 | Page N-1 | Page N | |
1030 * ^--------------------------------^----------|--------
1031 * | desired writeback range | see else |
1032 * ---------------------------------^------------------|
1034 offset = i_size_read(inode);
1035 end_index = offset >> PAGE_SHIFT;
1036 if (page->index < end_index)
1037 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1040 * Check whether the page to write out is beyond or straddles
1042 * -------------------------------------------------------
1043 * | file mapping | <EOF> |
1044 * -------------------------------------------------------
1045 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1046 * ^--------------------------------^-----------|---------
1048 * ---------------------------------^-----------|--------|
1050 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1053 * Skip the page if it is fully outside i_size, e.g. due to a
1054 * truncate operation that is in progress. We must redirty the
1055 * page so that reclaim stops reclaiming it. Otherwise
1056 * xfs_vm_releasepage() is called on it and gets confused.
1058 * Note that the end_index is unsigned long, it would overflow
1059 * if the given offset is greater than 16TB on 32-bit system
1060 * and if we do check the page is fully outside i_size or not
1061 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1062 * will be evaluated to 0. Hence this page will be redirtied
1063 * and be written out repeatedly which would result in an
1064 * infinite loop, the user program that perform this operation
1065 * will hang. Instead, we can verify this situation by checking
1066 * if the page to write is totally beyond the i_size or if it's
1067 * offset is just equal to the EOF.
1069 if (page->index > end_index ||
1070 (page->index == end_index && offset_into_page == 0))
1074 * The page straddles i_size. It must be zeroed out on each
1075 * and every writepage invocation because it may be mmapped.
1076 * "A file is mapped in multiples of the page size. For a file
1077 * that is not a multiple of the page size, the remaining
1078 * memory is zeroed when mapped, and writes to that region are
1079 * not written out to the file."
1081 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1083 /* Adjust the end_offset to the end of file */
1084 end_offset = offset;
1087 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1090 redirty_page_for_writepage(wbc, page);
1098 struct writeback_control *wbc)
1100 struct xfs_writepage_ctx wpc = {
1101 .io_type = XFS_IO_INVALID,
1105 ret = xfs_do_writepage(page, wbc, &wpc);
1107 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1113 struct address_space *mapping,
1114 struct writeback_control *wbc)
1116 struct xfs_writepage_ctx wpc = {
1117 .io_type = XFS_IO_INVALID,
1121 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1122 if (dax_mapping(mapping))
1123 return dax_writeback_mapping_range(mapping,
1124 xfs_find_bdev_for_inode(mapping->host), wbc);
1126 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1128 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1133 * Called to move a page into cleanable state - and from there
1134 * to be released. The page should already be clean. We always
1135 * have buffer heads in this call.
1137 * Returns 1 if the page is ok to release, 0 otherwise.
1144 int delalloc, unwritten;
1146 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1149 * mm accommodates an old ext3 case where clean pages might not have had
1150 * the dirty bit cleared. Thus, it can send actual dirty pages to
1151 * ->releasepage() via shrink_active_list(). Conversely,
1152 * block_invalidatepage() can send pages that are still marked dirty
1153 * but otherwise have invalidated buffers.
1155 * We've historically freed buffers on the latter. Instead, quietly
1156 * filter out all dirty pages to avoid spurious buffer state warnings.
1157 * This can likely be removed once shrink_active_list() is fixed.
1159 if (PageDirty(page))
1162 xfs_count_page_state(page, &delalloc, &unwritten);
1164 if (WARN_ON_ONCE(delalloc))
1166 if (WARN_ON_ONCE(unwritten))
1169 return try_to_free_buffers(page);
1173 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1174 * is, so that we can avoid repeated get_blocks calls.
1176 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1177 * for blocks beyond EOF must be marked new so that sub block regions can be
1178 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1179 * was just allocated or is unwritten, otherwise the callers would overwrite
1180 * existing data with zeros. Hence we have to split the mapping into a range up
1181 * to and including EOF, and a second mapping for beyond EOF.
1185 struct inode *inode,
1187 struct buffer_head *bh_result,
1188 struct xfs_bmbt_irec *imap,
1192 xfs_off_t mapping_size;
1194 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1195 mapping_size <<= inode->i_blkbits;
1197 ASSERT(mapping_size > 0);
1198 if (mapping_size > size)
1199 mapping_size = size;
1200 if (offset < i_size_read(inode) &&
1201 offset + mapping_size >= i_size_read(inode)) {
1202 /* limit mapping to block that spans EOF */
1203 mapping_size = roundup_64(i_size_read(inode) - offset,
1204 1 << inode->i_blkbits);
1206 if (mapping_size > LONG_MAX)
1207 mapping_size = LONG_MAX;
1209 bh_result->b_size = mapping_size;
1214 struct inode *inode,
1216 struct buffer_head *bh_result,
1219 struct xfs_inode *ip = XFS_I(inode);
1220 struct xfs_mount *mp = ip->i_mount;
1221 xfs_fileoff_t offset_fsb, end_fsb;
1224 struct xfs_bmbt_irec imap;
1231 if (XFS_FORCED_SHUTDOWN(mp))
1234 offset = (xfs_off_t)iblock << inode->i_blkbits;
1235 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1236 size = bh_result->b_size;
1238 if (offset >= i_size_read(inode))
1242 * Direct I/O is usually done on preallocated files, so try getting
1243 * a block mapping without an exclusive lock first.
1245 lockmode = xfs_ilock_data_map_shared(ip);
1247 ASSERT(offset <= mp->m_super->s_maxbytes);
1248 if (offset + size > mp->m_super->s_maxbytes)
1249 size = mp->m_super->s_maxbytes - offset;
1250 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1251 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1253 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1254 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1259 trace_xfs_get_blocks_found(ip, offset, size,
1260 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1261 : XFS_IO_OVERWRITE, &imap);
1262 xfs_iunlock(ip, lockmode);
1264 trace_xfs_get_blocks_notfound(ip, offset, size);
1268 /* trim mapping down to size requested */
1269 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1272 * For unwritten extents do not report a disk address in the buffered
1273 * read case (treat as if we're reading into a hole).
1275 if (imap.br_startblock != HOLESTARTBLOCK &&
1276 imap.br_startblock != DELAYSTARTBLOCK &&
1277 !ISUNWRITTEN(&imap))
1278 xfs_map_buffer(inode, bh_result, &imap, offset);
1281 * If this is a realtime file, data may be on a different device.
1282 * to that pointed to from the buffer_head b_bdev currently.
1284 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1288 xfs_iunlock(ip, lockmode);
1295 struct iov_iter *iter)
1298 * We just need the method present so that open/fcntl allow direct I/O.
1305 struct address_space *mapping,
1308 struct inode *inode = (struct inode *)mapping->host;
1309 struct xfs_inode *ip = XFS_I(inode);
1311 trace_xfs_vm_bmap(XFS_I(inode));
1314 * The swap code (ab-)uses ->bmap to get a block mapping and then
1315 * bypasseѕ the file system for actual I/O. We really can't allow
1316 * that on reflinks inodes, so we have to skip out here. And yes,
1317 * 0 is the magic code for a bmap error..
1319 if (xfs_is_reflink_inode(ip))
1322 filemap_write_and_wait(mapping);
1323 return generic_block_bmap(mapping, block, xfs_get_blocks);
1328 struct file *unused,
1331 trace_xfs_vm_readpage(page->mapping->host, 1);
1332 return mpage_readpage(page, xfs_get_blocks);
1337 struct file *unused,
1338 struct address_space *mapping,
1339 struct list_head *pages,
1342 trace_xfs_vm_readpages(mapping->host, nr_pages);
1343 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1347 * This is basically a copy of __set_page_dirty_buffers() with one
1348 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1349 * dirty, we'll never be able to clean them because we don't write buffers
1350 * beyond EOF, and that means we can't invalidate pages that span EOF
1351 * that have been marked dirty. Further, the dirty state can leak into
1352 * the file interior if the file is extended, resulting in all sorts of
1353 * bad things happening as the state does not match the underlying data.
1355 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1356 * this only exist because of bufferheads and how the generic code manages them.
1359 xfs_vm_set_page_dirty(
1362 struct address_space *mapping = page->mapping;
1363 struct inode *inode = mapping->host;
1368 if (unlikely(!mapping))
1369 return !TestSetPageDirty(page);
1371 end_offset = i_size_read(inode);
1372 offset = page_offset(page);
1374 spin_lock(&mapping->private_lock);
1375 if (page_has_buffers(page)) {
1376 struct buffer_head *head = page_buffers(page);
1377 struct buffer_head *bh = head;
1380 if (offset < end_offset)
1381 set_buffer_dirty(bh);
1382 bh = bh->b_this_page;
1383 offset += 1 << inode->i_blkbits;
1384 } while (bh != head);
1387 * Lock out page->mem_cgroup migration to keep PageDirty
1388 * synchronized with per-memcg dirty page counters.
1390 lock_page_memcg(page);
1391 newly_dirty = !TestSetPageDirty(page);
1392 spin_unlock(&mapping->private_lock);
1395 /* sigh - __set_page_dirty() is static, so copy it here, too */
1396 unsigned long flags;
1398 spin_lock_irqsave(&mapping->tree_lock, flags);
1399 if (page->mapping) { /* Race with truncate? */
1400 WARN_ON_ONCE(!PageUptodate(page));
1401 account_page_dirtied(page, mapping);
1402 radix_tree_tag_set(&mapping->page_tree,
1403 page_index(page), PAGECACHE_TAG_DIRTY);
1405 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1407 unlock_page_memcg(page);
1409 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1413 const struct address_space_operations xfs_address_space_operations = {
1414 .readpage = xfs_vm_readpage,
1415 .readpages = xfs_vm_readpages,
1416 .writepage = xfs_vm_writepage,
1417 .writepages = xfs_vm_writepages,
1418 .set_page_dirty = xfs_vm_set_page_dirty,
1419 .releasepage = xfs_vm_releasepage,
1420 .invalidatepage = xfs_vm_invalidatepage,
1421 .bmap = xfs_vm_bmap,
1422 .direct_IO = xfs_vm_direct_IO,
1423 .migratepage = buffer_migrate_page,
1424 .is_partially_uptodate = block_is_partially_uptodate,
1425 .error_remove_page = generic_error_remove_page,