1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_inode_item.h"
16 #include "xfs_alloc.h"
17 #include "xfs_error.h"
18 #include "xfs_iomap.h"
19 #include "xfs_trace.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_reflink.h"
24 #include <linux/writeback.h>
27 * structure owned by writepages passed to individual writepage calls
29 struct xfs_writepage_ctx {
30 struct xfs_bmbt_irec imap;
32 unsigned int data_seq;
34 struct xfs_ioend *ioend;
38 xfs_find_bdev_for_inode(
41 struct xfs_inode *ip = XFS_I(inode);
42 struct xfs_mount *mp = ip->i_mount;
44 if (XFS_IS_REALTIME_INODE(ip))
45 return mp->m_rtdev_targp->bt_bdev;
47 return mp->m_ddev_targp->bt_bdev;
51 xfs_find_daxdev_for_inode(
54 struct xfs_inode *ip = XFS_I(inode);
55 struct xfs_mount *mp = ip->i_mount;
57 if (XFS_IS_REALTIME_INODE(ip))
58 return mp->m_rtdev_targp->bt_daxdev;
60 return mp->m_ddev_targp->bt_daxdev;
64 xfs_finish_page_writeback(
69 struct iomap_page *iop = to_iomap_page(bvec->bv_page);
72 SetPageError(bvec->bv_page);
73 mapping_set_error(inode->i_mapping, -EIO);
76 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
77 ASSERT(!iop || atomic_read(&iop->write_count) > 0);
79 if (!iop || atomic_dec_and_test(&iop->write_count))
80 end_page_writeback(bvec->bv_page);
84 * We're now finished for good with this ioend structure. Update the page
85 * state, release holds on bios, and finally free up memory. Do not use the
90 struct xfs_ioend *ioend,
93 struct inode *inode = ioend->io_inode;
94 struct bio *bio = &ioend->io_inline_bio;
95 struct bio *last = ioend->io_bio, *next;
96 u64 start = bio->bi_iter.bi_sector;
97 bool quiet = bio_flagged(bio, BIO_QUIET);
99 for (bio = &ioend->io_inline_bio; bio; bio = next) {
100 struct bio_vec *bvec;
104 * For the last bio, bi_private points to the ioend, so we
105 * need to explicitly end the iteration here.
110 next = bio->bi_private;
112 /* walk each page on bio, ending page IO on them */
113 bio_for_each_segment_all(bvec, bio, i)
114 xfs_finish_page_writeback(inode, bvec, error);
118 if (unlikely(error && !quiet)) {
119 xfs_err_ratelimited(XFS_I(inode)->i_mount,
120 "writeback error on sector %llu", start);
125 * Fast and loose check if this write could update the on-disk inode size.
127 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
129 return ioend->io_offset + ioend->io_size >
130 XFS_I(ioend->io_inode)->i_d.di_size;
134 xfs_setfilesize_trans_alloc(
135 struct xfs_ioend *ioend)
137 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
138 struct xfs_trans *tp;
141 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0,
142 XFS_TRANS_NOFS, &tp);
146 ioend->io_append_trans = tp;
149 * We may pass freeze protection with a transaction. So tell lockdep
152 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
154 * We hand off the transaction to the completion thread now, so
155 * clear the flag here.
157 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
162 * Update on-disk file size now that data has been written to disk.
166 struct xfs_inode *ip,
167 struct xfs_trans *tp,
173 xfs_ilock(ip, XFS_ILOCK_EXCL);
174 isize = xfs_new_eof(ip, offset + size);
176 xfs_iunlock(ip, XFS_ILOCK_EXCL);
177 xfs_trans_cancel(tp);
181 trace_xfs_setfilesize(ip, offset, size);
183 ip->i_d.di_size = isize;
184 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
185 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
187 return xfs_trans_commit(tp);
192 struct xfs_inode *ip,
196 struct xfs_mount *mp = ip->i_mount;
197 struct xfs_trans *tp;
200 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
204 return __xfs_setfilesize(ip, tp, offset, size);
208 xfs_setfilesize_ioend(
209 struct xfs_ioend *ioend,
212 struct xfs_inode *ip = XFS_I(ioend->io_inode);
213 struct xfs_trans *tp = ioend->io_append_trans;
216 * The transaction may have been allocated in the I/O submission thread,
217 * thus we need to mark ourselves as being in a transaction manually.
218 * Similarly for freeze protection.
220 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
221 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
223 /* we abort the update if there was an IO error */
225 xfs_trans_cancel(tp);
229 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
233 * IO write completion.
237 struct work_struct *work)
239 struct xfs_ioend *ioend =
240 container_of(work, struct xfs_ioend, io_work);
241 struct xfs_inode *ip = XFS_I(ioend->io_inode);
242 xfs_off_t offset = ioend->io_offset;
243 size_t size = ioend->io_size;
247 * Just clean up the in-memory strutures if the fs has been shut down.
249 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
255 * Clean up any COW blocks on an I/O error.
257 error = blk_status_to_errno(ioend->io_bio->bi_status);
258 if (unlikely(error)) {
259 if (ioend->io_fork == XFS_COW_FORK)
260 xfs_reflink_cancel_cow_range(ip, offset, size, true);
265 * Success: commit the COW or unwritten blocks if needed.
267 if (ioend->io_fork == XFS_COW_FORK)
268 error = xfs_reflink_end_cow(ip, offset, size);
269 else if (ioend->io_state == XFS_EXT_UNWRITTEN)
270 error = xfs_iomap_write_unwritten(ip, offset, size, false);
272 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
275 if (ioend->io_append_trans)
276 error = xfs_setfilesize_ioend(ioend, error);
277 xfs_destroy_ioend(ioend, error);
284 struct xfs_ioend *ioend = bio->bi_private;
285 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
287 if (ioend->io_fork == XFS_COW_FORK ||
288 ioend->io_state == XFS_EXT_UNWRITTEN)
289 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
290 else if (ioend->io_append_trans)
291 queue_work(mp->m_data_workqueue, &ioend->io_work);
293 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
297 * Fast revalidation of the cached writeback mapping. Return true if the current
298 * mapping is valid, false otherwise.
302 struct xfs_writepage_ctx *wpc,
303 struct xfs_inode *ip,
304 xfs_fileoff_t offset_fsb)
306 if (offset_fsb < wpc->imap.br_startoff ||
307 offset_fsb >= wpc->imap.br_startoff + wpc->imap.br_blockcount)
310 * If this is a COW mapping, it is sufficient to check that the mapping
311 * covers the offset. Be careful to check this first because the caller
312 * can revalidate a COW mapping without updating the data seqno.
314 if (wpc->fork == XFS_COW_FORK)
318 * This is not a COW mapping. Check the sequence number of the data fork
319 * because concurrent changes could have invalidated the extent. Check
320 * the COW fork because concurrent changes since the last time we
321 * checked (and found nothing at this offset) could have added
322 * overlapping blocks.
324 if (wpc->data_seq != READ_ONCE(ip->i_df.if_seq))
326 if (xfs_inode_has_cow_data(ip) &&
327 wpc->cow_seq != READ_ONCE(ip->i_cowfp->if_seq))
333 * Pass in a dellalloc extent and convert it to real extents, return the real
334 * extent that maps offset_fsb in wpc->imap.
336 * The current page is held locked so nothing could have removed the block
337 * backing offset_fsb, although it could have moved from the COW to the data
338 * fork by another thread.
342 struct xfs_writepage_ctx *wpc,
343 struct xfs_inode *ip,
344 xfs_fileoff_t offset_fsb)
349 * Attempt to allocate whatever delalloc extent currently backs
350 * offset_fsb and put the result into wpc->imap. Allocate in a loop
351 * because it may take several attempts to allocate real blocks for a
352 * contiguous delalloc extent if free space is sufficiently fragmented.
355 error = xfs_bmapi_convert_delalloc(ip, wpc->fork, offset_fsb,
356 &wpc->imap, wpc->fork == XFS_COW_FORK ?
357 &wpc->cow_seq : &wpc->data_seq);
360 } while (wpc->imap.br_startoff + wpc->imap.br_blockcount <= offset_fsb);
367 struct xfs_writepage_ctx *wpc,
371 struct xfs_inode *ip = XFS_I(inode);
372 struct xfs_mount *mp = ip->i_mount;
373 ssize_t count = i_blocksize(inode);
374 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
375 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
376 xfs_fileoff_t cow_fsb = NULLFILEOFF;
377 struct xfs_bmbt_irec imap;
378 struct xfs_iext_cursor icur;
382 if (XFS_FORCED_SHUTDOWN(mp))
386 * COW fork blocks can overlap data fork blocks even if the blocks
387 * aren't shared. COW I/O always takes precedent, so we must always
388 * check for overlap on reflink inodes unless the mapping is already a
389 * COW one, or the COW fork hasn't changed from the last time we looked
392 * It's safe to check the COW fork if_seq here without the ILOCK because
393 * we've indirectly protected against concurrent updates: writeback has
394 * the page locked, which prevents concurrent invalidations by reflink
395 * and directio and prevents concurrent buffered writes to the same
396 * page. Changes to if_seq always happen under i_lock, which protects
397 * against concurrent updates and provides a memory barrier on the way
398 * out that ensures that we always see the current value.
400 if (xfs_imap_valid(wpc, ip, offset_fsb))
404 * If we don't have a valid map, now it's time to get a new one for this
405 * offset. This will convert delayed allocations (including COW ones)
406 * into real extents. If we return without a valid map, it means we
407 * landed in a hole and we skip the block.
410 xfs_ilock(ip, XFS_ILOCK_SHARED);
411 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
412 (ip->i_df.if_flags & XFS_IFEXTENTS));
415 * Check if this is offset is covered by a COW extents, and if yes use
416 * it directly instead of looking up anything in the data fork.
418 if (xfs_inode_has_cow_data(ip) &&
419 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
420 cow_fsb = imap.br_startoff;
421 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
422 wpc->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
423 xfs_iunlock(ip, XFS_ILOCK_SHARED);
425 wpc->fork = XFS_COW_FORK;
426 goto allocate_blocks;
430 * No COW extent overlap. Revalidate now that we may have updated
431 * ->cow_seq. If the data mapping is still valid, we're done.
433 if (xfs_imap_valid(wpc, ip, offset_fsb)) {
434 xfs_iunlock(ip, XFS_ILOCK_SHARED);
439 * If we don't have a valid map, now it's time to get a new one for this
440 * offset. This will convert delayed allocations (including COW ones)
443 if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
444 imap.br_startoff = end_fsb; /* fake a hole past EOF */
445 wpc->data_seq = READ_ONCE(ip->i_df.if_seq);
446 xfs_iunlock(ip, XFS_ILOCK_SHARED);
448 wpc->fork = XFS_DATA_FORK;
450 /* landed in a hole or beyond EOF? */
451 if (imap.br_startoff > offset_fsb) {
452 imap.br_blockcount = imap.br_startoff - offset_fsb;
453 imap.br_startoff = offset_fsb;
454 imap.br_startblock = HOLESTARTBLOCK;
455 imap.br_state = XFS_EXT_NORM;
459 * Truncate to the next COW extent if there is one. This is the only
460 * opportunity to do this because we can skip COW fork lookups for the
461 * subsequent blocks in the mapping; however, the requirement to treat
462 * the COW range separately remains.
464 if (cow_fsb != NULLFILEOFF &&
465 cow_fsb < imap.br_startoff + imap.br_blockcount)
466 imap.br_blockcount = cow_fsb - imap.br_startoff;
468 /* got a delalloc extent? */
469 if (imap.br_startblock != HOLESTARTBLOCK &&
470 isnullstartblock(imap.br_startblock))
471 goto allocate_blocks;
474 trace_xfs_map_blocks_found(ip, offset, count, wpc->fork, &imap);
477 error = xfs_convert_blocks(wpc, ip, offset_fsb);
480 * If we failed to find the extent in the COW fork we might have
481 * raced with a COW to data fork conversion or truncate.
482 * Restart the lookup to catch the extent in the data fork for
483 * the former case, but prevent additional retries to avoid
484 * looping forever for the latter case.
486 if (error == -EAGAIN && wpc->fork == XFS_COW_FORK && !retries++)
488 ASSERT(error != -EAGAIN);
493 * Due to merging the return real extent might be larger than the
494 * original delalloc one. Trim the return extent to the next COW
495 * boundary again to force a re-lookup.
497 if (wpc->fork != XFS_COW_FORK && cow_fsb != NULLFILEOFF &&
498 cow_fsb < wpc->imap.br_startoff + wpc->imap.br_blockcount)
499 wpc->imap.br_blockcount = cow_fsb - wpc->imap.br_startoff;
501 ASSERT(wpc->imap.br_startoff <= offset_fsb);
502 ASSERT(wpc->imap.br_startoff + wpc->imap.br_blockcount > offset_fsb);
503 trace_xfs_map_blocks_alloc(ip, offset, count, wpc->fork, &imap);
508 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
509 * it, and we submit that bio. The ioend may be used for multiple bio
510 * submissions, so we only want to allocate an append transaction for the ioend
511 * once. In the case of multiple bio submission, each bio will take an IO
512 * reference to the ioend to ensure that the ioend completion is only done once
513 * all bios have been submitted and the ioend is really done.
515 * If @fail is non-zero, it means that we have a situation where some part of
516 * the submission process has failed after we have marked paged for writeback
517 * and unlocked them. In this situation, we need to fail the bio and ioend
518 * rather than submit it to IO. This typically only happens on a filesystem
523 struct writeback_control *wbc,
524 struct xfs_ioend *ioend,
527 /* Convert CoW extents to regular */
528 if (!status && ioend->io_fork == XFS_COW_FORK) {
530 * Yuk. This can do memory allocation, but is not a
531 * transactional operation so everything is done in GFP_KERNEL
532 * context. That can deadlock, because we hold pages in
533 * writeback state and GFP_KERNEL allocations can block on them.
534 * Hence we must operate in nofs conditions here.
538 nofs_flag = memalloc_nofs_save();
539 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
540 ioend->io_offset, ioend->io_size);
541 memalloc_nofs_restore(nofs_flag);
544 /* Reserve log space if we might write beyond the on-disk inode size. */
546 (ioend->io_fork == XFS_COW_FORK ||
547 ioend->io_state != XFS_EXT_UNWRITTEN) &&
548 xfs_ioend_is_append(ioend) &&
549 !ioend->io_append_trans)
550 status = xfs_setfilesize_trans_alloc(ioend);
552 ioend->io_bio->bi_private = ioend;
553 ioend->io_bio->bi_end_io = xfs_end_bio;
554 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
557 * If we are failing the IO now, just mark the ioend with an
558 * error and finish it. This will run IO completion immediately
559 * as there is only one reference to the ioend at this point in
563 ioend->io_bio->bi_status = errno_to_blk_status(status);
564 bio_endio(ioend->io_bio);
568 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
569 submit_bio(ioend->io_bio);
573 static struct xfs_ioend *
579 struct block_device *bdev,
582 struct xfs_ioend *ioend;
585 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset);
586 bio_set_dev(bio, bdev);
587 bio->bi_iter.bi_sector = sector;
589 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
590 INIT_LIST_HEAD(&ioend->io_list);
591 ioend->io_fork = fork;
592 ioend->io_state = state;
593 ioend->io_inode = inode;
595 ioend->io_offset = offset;
596 INIT_WORK(&ioend->io_work, xfs_end_io);
597 ioend->io_append_trans = NULL;
603 * Allocate a new bio, and chain the old bio to the new one.
605 * Note that we have to do perform the chaining in this unintuitive order
606 * so that the bi_private linkage is set up in the right direction for the
607 * traversal in xfs_destroy_ioend().
611 struct xfs_ioend *ioend,
612 struct writeback_control *wbc,
613 struct block_device *bdev,
618 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
619 bio_set_dev(new, bdev);
620 new->bi_iter.bi_sector = sector;
621 bio_chain(ioend->io_bio, new);
622 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
623 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
624 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
625 submit_bio(ioend->io_bio);
630 * Test to see if we have an existing ioend structure that we could append to
631 * first, otherwise finish off the current ioend and start another.
638 struct iomap_page *iop,
639 struct xfs_writepage_ctx *wpc,
640 struct writeback_control *wbc,
641 struct list_head *iolist)
643 struct xfs_inode *ip = XFS_I(inode);
644 struct xfs_mount *mp = ip->i_mount;
645 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
646 unsigned len = i_blocksize(inode);
647 unsigned poff = offset & (PAGE_SIZE - 1);
650 sector = xfs_fsb_to_db(ip, wpc->imap.br_startblock) +
651 ((offset - XFS_FSB_TO_B(mp, wpc->imap.br_startoff)) >> 9);
654 wpc->fork != wpc->ioend->io_fork ||
655 wpc->imap.br_state != wpc->ioend->io_state ||
656 sector != bio_end_sector(wpc->ioend->io_bio) ||
657 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
659 list_add(&wpc->ioend->io_list, iolist);
660 wpc->ioend = xfs_alloc_ioend(inode, wpc->fork,
661 wpc->imap.br_state, offset, bdev, sector);
664 if (!__bio_try_merge_page(wpc->ioend->io_bio, page, len, poff)) {
666 atomic_inc(&iop->write_count);
667 if (bio_full(wpc->ioend->io_bio))
668 xfs_chain_bio(wpc->ioend, wbc, bdev, sector);
669 __bio_add_page(wpc->ioend->io_bio, page, len, poff);
672 wpc->ioend->io_size += len;
676 xfs_vm_invalidatepage(
681 trace_xfs_invalidatepage(page->mapping->host, page, offset, length);
682 iomap_invalidatepage(page, offset, length);
686 * If the page has delalloc blocks on it, we need to punch them out before we
687 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
688 * inode that can trip up a later direct I/O read operation on the same region.
690 * We prevent this by truncating away the delalloc regions on the page. Because
691 * they are delalloc, we can do this without needing a transaction. Indeed - if
692 * we get ENOSPC errors, we have to be able to do this truncation without a
693 * transaction as there is no space left for block reservation (typically why we
694 * see a ENOSPC in writeback).
697 xfs_aops_discard_page(
700 struct inode *inode = page->mapping->host;
701 struct xfs_inode *ip = XFS_I(inode);
702 struct xfs_mount *mp = ip->i_mount;
703 loff_t offset = page_offset(page);
704 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, offset);
707 if (XFS_FORCED_SHUTDOWN(mp))
711 "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.",
712 page, ip->i_ino, offset);
714 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
715 PAGE_SIZE / i_blocksize(inode));
716 if (error && !XFS_FORCED_SHUTDOWN(mp))
717 xfs_alert(mp, "page discard unable to remove delalloc mapping.");
719 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
723 * We implement an immediate ioend submission policy here to avoid needing to
724 * chain multiple ioends and hence nest mempool allocations which can violate
725 * forward progress guarantees we need to provide. The current ioend we are
726 * adding blocks to is cached on the writepage context, and if the new block
727 * does not append to the cached ioend it will create a new ioend and cache that
730 * If a new ioend is created and cached, the old ioend is returned and queued
731 * locally for submission once the entire page is processed or an error has been
732 * detected. While ioends are submitted immediately after they are completed,
733 * batching optimisations are provided by higher level block plugging.
735 * At the end of a writeback pass, there will be a cached ioend remaining on the
736 * writepage context that the caller will need to submit.
740 struct xfs_writepage_ctx *wpc,
741 struct writeback_control *wbc,
746 LIST_HEAD(submit_list);
747 struct iomap_page *iop = to_iomap_page(page);
748 unsigned len = i_blocksize(inode);
749 struct xfs_ioend *ioend, *next;
750 uint64_t file_offset; /* file offset of page */
751 int error = 0, count = 0, i;
753 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
754 ASSERT(!iop || atomic_read(&iop->write_count) == 0);
757 * Walk through the page to find areas to write back. If we run off the
758 * end of the current map or find the current map invalid, grab a new
761 for (i = 0, file_offset = page_offset(page);
762 i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
763 i++, file_offset += len) {
764 if (iop && !test_bit(i, iop->uptodate))
767 error = xfs_map_blocks(wpc, inode, file_offset);
770 if (wpc->imap.br_startblock == HOLESTARTBLOCK)
772 xfs_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
777 ASSERT(wpc->ioend || list_empty(&submit_list));
778 ASSERT(PageLocked(page));
779 ASSERT(!PageWriteback(page));
782 * On error, we have to fail the ioend here because we may have set
783 * pages under writeback, we have to make sure we run IO completion to
784 * mark the error state of the IO appropriately, so we can't cancel the
785 * ioend directly here. That means we have to mark this page as under
786 * writeback if we included any blocks from it in the ioend chain so
787 * that completion treats it correctly.
789 * If we didn't include the page in the ioend, the on error we can
790 * simply discard and unlock it as there are no other users of the page
791 * now. The caller will still need to trigger submission of outstanding
792 * ioends on the writepage context so they are treated correctly on
795 if (unlikely(error)) {
797 xfs_aops_discard_page(page);
798 ClearPageUptodate(page);
804 * If the page was not fully cleaned, we need to ensure that the
805 * higher layers come back to it correctly. That means we need
806 * to keep the page dirty, and for WB_SYNC_ALL writeback we need
807 * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed
808 * so another attempt to write this page in this writeback sweep
811 set_page_writeback_keepwrite(page);
813 clear_page_dirty_for_io(page);
814 set_page_writeback(page);
820 * Preserve the original error if there was one, otherwise catch
821 * submission errors here and propagate into subsequent ioend
824 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
827 list_del_init(&ioend->io_list);
828 error2 = xfs_submit_ioend(wbc, ioend, error);
829 if (error2 && !error)
834 * We can end up here with no error and nothing to write only if we race
835 * with a partial page truncate on a sub-page block sized filesystem.
838 end_page_writeback(page);
840 mapping_set_error(page->mapping, error);
845 * Write out a dirty page.
847 * For delalloc space on the page we need to allocate space and flush it.
848 * For unwritten space on the page we need to start the conversion to
849 * regular allocated space.
854 struct writeback_control *wbc,
857 struct xfs_writepage_ctx *wpc = data;
858 struct inode *inode = page->mapping->host;
863 trace_xfs_writepage(inode, page, 0, 0);
866 * Refuse to write the page out if we are called from reclaim context.
868 * This avoids stack overflows when called from deeply used stacks in
869 * random callers for direct reclaim or memcg reclaim. We explicitly
870 * allow reclaim from kswapd as the stack usage there is relatively low.
872 * This should never happen except in the case of a VM regression so
875 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
880 * Given that we do not allow direct reclaim to call us, we should
881 * never be called while in a filesystem transaction.
883 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
887 * Is this page beyond the end of the file?
889 * The page index is less than the end_index, adjust the end_offset
890 * to the highest offset that this page should represent.
891 * -----------------------------------------------------
892 * | file mapping | <EOF> |
893 * -----------------------------------------------------
894 * | Page ... | Page N-2 | Page N-1 | Page N | |
895 * ^--------------------------------^----------|--------
896 * | desired writeback range | see else |
897 * ---------------------------------^------------------|
899 offset = i_size_read(inode);
900 end_index = offset >> PAGE_SHIFT;
901 if (page->index < end_index)
902 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
905 * Check whether the page to write out is beyond or straddles
907 * -------------------------------------------------------
908 * | file mapping | <EOF> |
909 * -------------------------------------------------------
910 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
911 * ^--------------------------------^-----------|---------
913 * ---------------------------------^-----------|--------|
915 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
918 * Skip the page if it is fully outside i_size, e.g. due to a
919 * truncate operation that is in progress. We must redirty the
920 * page so that reclaim stops reclaiming it. Otherwise
921 * xfs_vm_releasepage() is called on it and gets confused.
923 * Note that the end_index is unsigned long, it would overflow
924 * if the given offset is greater than 16TB on 32-bit system
925 * and if we do check the page is fully outside i_size or not
926 * via "if (page->index >= end_index + 1)" as "end_index + 1"
927 * will be evaluated to 0. Hence this page will be redirtied
928 * and be written out repeatedly which would result in an
929 * infinite loop, the user program that perform this operation
930 * will hang. Instead, we can verify this situation by checking
931 * if the page to write is totally beyond the i_size or if it's
932 * offset is just equal to the EOF.
934 if (page->index > end_index ||
935 (page->index == end_index && offset_into_page == 0))
939 * The page straddles i_size. It must be zeroed out on each
940 * and every writepage invocation because it may be mmapped.
941 * "A file is mapped in multiples of the page size. For a file
942 * that is not a multiple of the page size, the remaining
943 * memory is zeroed when mapped, and writes to that region are
944 * not written out to the file."
946 zero_user_segment(page, offset_into_page, PAGE_SIZE);
948 /* Adjust the end_offset to the end of file */
952 return xfs_writepage_map(wpc, wbc, inode, page, end_offset);
955 redirty_page_for_writepage(wbc, page);
963 struct writeback_control *wbc)
965 struct xfs_writepage_ctx wpc = { };
968 ret = xfs_do_writepage(page, wbc, &wpc);
970 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
976 struct address_space *mapping,
977 struct writeback_control *wbc)
979 struct xfs_writepage_ctx wpc = { };
982 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
983 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
985 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
991 struct address_space *mapping,
992 struct writeback_control *wbc)
994 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
995 return dax_writeback_mapping_range(mapping,
996 xfs_find_bdev_for_inode(mapping->host), wbc);
1004 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1005 return iomap_releasepage(page, gfp_mask);
1010 struct address_space *mapping,
1013 struct xfs_inode *ip = XFS_I(mapping->host);
1015 trace_xfs_vm_bmap(ip);
1018 * The swap code (ab-)uses ->bmap to get a block mapping and then
1019 * bypasses the file system for actual I/O. We really can't allow
1020 * that on reflinks inodes, so we have to skip out here. And yes,
1021 * 0 is the magic code for a bmap error.
1023 * Since we don't pass back blockdev info, we can't return bmap
1024 * information for rt files either.
1026 if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip))
1028 return iomap_bmap(mapping, block, &xfs_iomap_ops);
1033 struct file *unused,
1036 trace_xfs_vm_readpage(page->mapping->host, 1);
1037 return iomap_readpage(page, &xfs_iomap_ops);
1042 struct file *unused,
1043 struct address_space *mapping,
1044 struct list_head *pages,
1047 trace_xfs_vm_readpages(mapping->host, nr_pages);
1048 return iomap_readpages(mapping, pages, nr_pages, &xfs_iomap_ops);
1052 xfs_iomap_swapfile_activate(
1053 struct swap_info_struct *sis,
1054 struct file *swap_file,
1057 sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
1058 return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops);
1061 const struct address_space_operations xfs_address_space_operations = {
1062 .readpage = xfs_vm_readpage,
1063 .readpages = xfs_vm_readpages,
1064 .writepage = xfs_vm_writepage,
1065 .writepages = xfs_vm_writepages,
1066 .set_page_dirty = iomap_set_page_dirty,
1067 .releasepage = xfs_vm_releasepage,
1068 .invalidatepage = xfs_vm_invalidatepage,
1069 .bmap = xfs_vm_bmap,
1070 .direct_IO = noop_direct_IO,
1071 .migratepage = iomap_migrate_page,
1072 .is_partially_uptodate = iomap_is_partially_uptodate,
1073 .error_remove_page = generic_error_remove_page,
1074 .swap_activate = xfs_iomap_swapfile_activate,
1077 const struct address_space_operations xfs_dax_aops = {
1078 .writepages = xfs_dax_writepages,
1079 .direct_IO = noop_direct_IO,
1080 .set_page_dirty = noop_set_page_dirty,
1081 .invalidatepage = noop_invalidatepage,
1082 .swap_activate = xfs_iomap_swapfile_activate,