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;
33 struct xfs_ioend *ioend;
37 xfs_find_bdev_for_inode(
40 struct xfs_inode *ip = XFS_I(inode);
41 struct xfs_mount *mp = ip->i_mount;
43 if (XFS_IS_REALTIME_INODE(ip))
44 return mp->m_rtdev_targp->bt_bdev;
46 return mp->m_ddev_targp->bt_bdev;
50 xfs_find_daxdev_for_inode(
53 struct xfs_inode *ip = XFS_I(inode);
54 struct xfs_mount *mp = ip->i_mount;
56 if (XFS_IS_REALTIME_INODE(ip))
57 return mp->m_rtdev_targp->bt_daxdev;
59 return mp->m_ddev_targp->bt_daxdev;
63 xfs_finish_page_writeback(
68 struct iomap_page *iop = to_iomap_page(bvec->bv_page);
71 SetPageError(bvec->bv_page);
72 mapping_set_error(inode->i_mapping, -EIO);
75 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
76 ASSERT(!iop || atomic_read(&iop->write_count) > 0);
78 if (!iop || atomic_dec_and_test(&iop->write_count))
79 end_page_writeback(bvec->bv_page);
83 * We're now finished for good with this ioend structure. Update the page
84 * state, release holds on bios, and finally free up memory. Do not use the
89 struct xfs_ioend *ioend,
92 struct inode *inode = ioend->io_inode;
93 struct bio *bio = &ioend->io_inline_bio;
94 struct bio *last = ioend->io_bio, *next;
95 u64 start = bio->bi_iter.bi_sector;
96 bool quiet = bio_flagged(bio, BIO_QUIET);
98 for (bio = &ioend->io_inline_bio; bio; bio = next) {
103 * For the last bio, bi_private points to the ioend, so we
104 * need to explicitly end the iteration here.
109 next = bio->bi_private;
111 /* walk each page on bio, ending page IO on them */
112 bio_for_each_segment_all(bvec, bio, i)
113 xfs_finish_page_writeback(inode, bvec, error);
117 if (unlikely(error && !quiet)) {
118 xfs_err_ratelimited(XFS_I(inode)->i_mount,
119 "writeback error on sector %llu", start);
124 * Fast and loose check if this write could update the on-disk inode size.
126 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
128 return ioend->io_offset + ioend->io_size >
129 XFS_I(ioend->io_inode)->i_d.di_size;
133 xfs_setfilesize_trans_alloc(
134 struct xfs_ioend *ioend)
136 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
137 struct xfs_trans *tp;
140 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0,
141 XFS_TRANS_NOFS, &tp);
145 ioend->io_append_trans = tp;
148 * We may pass freeze protection with a transaction. So tell lockdep
151 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
153 * We hand off the transaction to the completion thread now, so
154 * clear the flag here.
156 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
161 * Update on-disk file size now that data has been written to disk.
165 struct xfs_inode *ip,
166 struct xfs_trans *tp,
172 xfs_ilock(ip, XFS_ILOCK_EXCL);
173 isize = xfs_new_eof(ip, offset + size);
175 xfs_iunlock(ip, XFS_ILOCK_EXCL);
176 xfs_trans_cancel(tp);
180 trace_xfs_setfilesize(ip, offset, size);
182 ip->i_d.di_size = isize;
183 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
184 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
186 return xfs_trans_commit(tp);
191 struct xfs_inode *ip,
195 struct xfs_mount *mp = ip->i_mount;
196 struct xfs_trans *tp;
199 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
203 return __xfs_setfilesize(ip, tp, offset, size);
207 xfs_setfilesize_ioend(
208 struct xfs_ioend *ioend,
211 struct xfs_inode *ip = XFS_I(ioend->io_inode);
212 struct xfs_trans *tp = ioend->io_append_trans;
215 * The transaction may have been allocated in the I/O submission thread,
216 * thus we need to mark ourselves as being in a transaction manually.
217 * Similarly for freeze protection.
219 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
220 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
222 /* we abort the update if there was an IO error */
224 xfs_trans_cancel(tp);
228 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
232 * IO write completion.
236 struct work_struct *work)
238 struct xfs_ioend *ioend =
239 container_of(work, struct xfs_ioend, io_work);
240 struct xfs_inode *ip = XFS_I(ioend->io_inode);
241 xfs_off_t offset = ioend->io_offset;
242 size_t size = ioend->io_size;
246 * Just clean up the in-memory strutures if the fs has been shut down.
248 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
254 * Clean up any COW blocks on an I/O error.
256 error = blk_status_to_errno(ioend->io_bio->bi_status);
257 if (unlikely(error)) {
258 switch (ioend->io_type) {
260 xfs_reflink_cancel_cow_range(ip, offset, size, true);
268 * Success: commit the COW or unwritten blocks if needed.
270 switch (ioend->io_type) {
272 error = xfs_reflink_end_cow(ip, offset, size);
274 case XFS_IO_UNWRITTEN:
275 /* writeback should never update isize */
276 error = xfs_iomap_write_unwritten(ip, offset, size, false);
279 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
284 if (ioend->io_append_trans)
285 error = xfs_setfilesize_ioend(ioend, error);
286 xfs_destroy_ioend(ioend, error);
293 struct xfs_ioend *ioend = bio->bi_private;
294 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
296 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
297 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
298 else if (ioend->io_append_trans)
299 queue_work(mp->m_data_workqueue, &ioend->io_work);
301 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
306 struct xfs_writepage_ctx *wpc,
310 struct xfs_inode *ip = XFS_I(inode);
311 struct xfs_mount *mp = ip->i_mount;
312 ssize_t count = i_blocksize(inode);
313 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset), end_fsb;
314 xfs_fileoff_t cow_fsb = NULLFILEOFF;
315 struct xfs_bmbt_irec imap;
316 int whichfork = XFS_DATA_FORK;
317 struct xfs_iext_cursor icur;
322 * We have to make sure the cached mapping is within EOF to protect
323 * against eofblocks trimming on file release leaving us with a stale
324 * mapping. Otherwise, a page for a subsequent file extending buffered
325 * write could get picked up by this writeback cycle and written to the
328 * Note that what we really want here is a generic mapping invalidation
329 * mechanism to protect us from arbitrary extent modifying contexts, not
332 xfs_trim_extent_eof(&wpc->imap, ip);
335 * COW fork blocks can overlap data fork blocks even if the blocks
336 * aren't shared. COW I/O always takes precedent, so we must always
337 * check for overlap on reflink inodes unless the mapping is already a
338 * COW one, or the COW fork hasn't changed from the last time we looked
341 * It's safe to check the COW fork if_seq here without the ILOCK because
342 * we've indirectly protected against concurrent updates: writeback has
343 * the page locked, which prevents concurrent invalidations by reflink
344 * and directio and prevents concurrent buffered writes to the same
345 * page. Changes to if_seq always happen under i_lock, which protects
346 * against concurrent updates and provides a memory barrier on the way
347 * out that ensures that we always see the current value.
349 imap_valid = offset_fsb >= wpc->imap.br_startoff &&
350 offset_fsb < wpc->imap.br_startoff + wpc->imap.br_blockcount;
352 (!xfs_inode_has_cow_data(ip) ||
353 wpc->io_type == XFS_IO_COW ||
354 wpc->cow_seq == READ_ONCE(ip->i_cowfp->if_seq)))
357 if (XFS_FORCED_SHUTDOWN(mp))
361 * If we don't have a valid map, now it's time to get a new one for this
362 * offset. This will convert delayed allocations (including COW ones)
363 * into real extents. If we return without a valid map, it means we
364 * landed in a hole and we skip the block.
366 xfs_ilock(ip, XFS_ILOCK_SHARED);
367 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
368 (ip->i_df.if_flags & XFS_IFEXTENTS));
369 ASSERT(offset <= mp->m_super->s_maxbytes);
371 if (offset > mp->m_super->s_maxbytes - count)
372 count = mp->m_super->s_maxbytes - offset;
373 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
376 * Check if this is offset is covered by a COW extents, and if yes use
377 * it directly instead of looking up anything in the data fork.
379 if (xfs_inode_has_cow_data(ip) &&
380 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
381 cow_fsb = imap.br_startoff;
382 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
383 wpc->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
384 xfs_iunlock(ip, XFS_ILOCK_SHARED);
386 * Truncate can race with writeback since writeback doesn't
387 * take the iolock and truncate decreases the file size before
388 * it starts truncating the pages between new_size and old_size.
389 * Therefore, we can end up in the situation where writeback
390 * gets a CoW fork mapping but the truncate makes the mapping
391 * invalid and we end up in here trying to get a new mapping.
392 * bail out here so that we simply never get a valid mapping
393 * and so we drop the write altogether. The page truncation
394 * will kill the contents anyway.
396 if (offset > i_size_read(inode)) {
397 wpc->io_type = XFS_IO_HOLE;
400 whichfork = XFS_COW_FORK;
401 wpc->io_type = XFS_IO_COW;
402 goto allocate_blocks;
406 * Map valid and no COW extent in the way? We're done.
409 xfs_iunlock(ip, XFS_ILOCK_SHARED);
414 * If we don't have a valid map, now it's time to get a new one for this
415 * offset. This will convert delayed allocations (including COW ones)
418 if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
419 imap.br_startoff = end_fsb; /* fake a hole past EOF */
420 xfs_iunlock(ip, XFS_ILOCK_SHARED);
422 if (imap.br_startoff > offset_fsb) {
423 /* landed in a hole or beyond EOF */
424 imap.br_blockcount = imap.br_startoff - offset_fsb;
425 imap.br_startoff = offset_fsb;
426 imap.br_startblock = HOLESTARTBLOCK;
427 wpc->io_type = XFS_IO_HOLE;
430 * Truncate to the next COW extent if there is one. This is the
431 * only opportunity to do this because we can skip COW fork
432 * lookups for the subsequent blocks in the mapping; however,
433 * the requirement to treat the COW range separately remains.
435 if (cow_fsb != NULLFILEOFF &&
436 cow_fsb < imap.br_startoff + imap.br_blockcount)
437 imap.br_blockcount = cow_fsb - imap.br_startoff;
439 if (isnullstartblock(imap.br_startblock)) {
440 /* got a delalloc extent */
441 wpc->io_type = XFS_IO_DELALLOC;
442 goto allocate_blocks;
445 if (imap.br_state == XFS_EXT_UNWRITTEN)
446 wpc->io_type = XFS_IO_UNWRITTEN;
448 wpc->io_type = XFS_IO_OVERWRITE;
452 xfs_trim_extent_eof(&wpc->imap, ip);
453 trace_xfs_map_blocks_found(ip, offset, count, wpc->io_type, &imap);
456 error = xfs_iomap_write_allocate(ip, whichfork, offset, &imap,
460 ASSERT(whichfork == XFS_COW_FORK || cow_fsb == NULLFILEOFF ||
461 imap.br_startoff + imap.br_blockcount <= cow_fsb);
463 xfs_trim_extent_eof(&wpc->imap, ip);
464 trace_xfs_map_blocks_alloc(ip, offset, count, wpc->io_type, &imap);
469 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
470 * it, and we submit that bio. The ioend may be used for multiple bio
471 * submissions, so we only want to allocate an append transaction for the ioend
472 * once. In the case of multiple bio submission, each bio will take an IO
473 * reference to the ioend to ensure that the ioend completion is only done once
474 * all bios have been submitted and the ioend is really done.
476 * If @fail is non-zero, it means that we have a situation where some part of
477 * the submission process has failed after we have marked paged for writeback
478 * and unlocked them. In this situation, we need to fail the bio and ioend
479 * rather than submit it to IO. This typically only happens on a filesystem
484 struct writeback_control *wbc,
485 struct xfs_ioend *ioend,
488 /* Convert CoW extents to regular */
489 if (!status && ioend->io_type == XFS_IO_COW) {
491 * Yuk. This can do memory allocation, but is not a
492 * transactional operation so everything is done in GFP_KERNEL
493 * context. That can deadlock, because we hold pages in
494 * writeback state and GFP_KERNEL allocations can block on them.
495 * Hence we must operate in nofs conditions here.
499 nofs_flag = memalloc_nofs_save();
500 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
501 ioend->io_offset, ioend->io_size);
502 memalloc_nofs_restore(nofs_flag);
505 /* Reserve log space if we might write beyond the on-disk inode size. */
507 ioend->io_type != XFS_IO_UNWRITTEN &&
508 xfs_ioend_is_append(ioend) &&
509 !ioend->io_append_trans)
510 status = xfs_setfilesize_trans_alloc(ioend);
512 ioend->io_bio->bi_private = ioend;
513 ioend->io_bio->bi_end_io = xfs_end_bio;
514 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
517 * If we are failing the IO now, just mark the ioend with an
518 * error and finish it. This will run IO completion immediately
519 * as there is only one reference to the ioend at this point in
523 ioend->io_bio->bi_status = errno_to_blk_status(status);
524 bio_endio(ioend->io_bio);
528 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
529 submit_bio(ioend->io_bio);
533 static struct xfs_ioend *
538 struct block_device *bdev,
541 struct xfs_ioend *ioend;
544 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset);
545 bio_set_dev(bio, bdev);
546 bio->bi_iter.bi_sector = sector;
548 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
549 INIT_LIST_HEAD(&ioend->io_list);
550 ioend->io_type = type;
551 ioend->io_inode = inode;
553 ioend->io_offset = offset;
554 INIT_WORK(&ioend->io_work, xfs_end_io);
555 ioend->io_append_trans = NULL;
561 * Allocate a new bio, and chain the old bio to the new one.
563 * Note that we have to do perform the chaining in this unintuitive order
564 * so that the bi_private linkage is set up in the right direction for the
565 * traversal in xfs_destroy_ioend().
569 struct xfs_ioend *ioend,
570 struct writeback_control *wbc,
571 struct block_device *bdev,
576 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
577 bio_set_dev(new, bdev);
578 new->bi_iter.bi_sector = sector;
579 bio_chain(ioend->io_bio, new);
580 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
581 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
582 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
583 submit_bio(ioend->io_bio);
588 * Test to see if we have an existing ioend structure that we could append to
589 * first, otherwise finish off the current ioend and start another.
596 struct iomap_page *iop,
597 struct xfs_writepage_ctx *wpc,
598 struct writeback_control *wbc,
599 struct list_head *iolist)
601 struct xfs_inode *ip = XFS_I(inode);
602 struct xfs_mount *mp = ip->i_mount;
603 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
604 unsigned len = i_blocksize(inode);
605 unsigned poff = offset & (PAGE_SIZE - 1);
608 sector = xfs_fsb_to_db(ip, wpc->imap.br_startblock) +
609 ((offset - XFS_FSB_TO_B(mp, wpc->imap.br_startoff)) >> 9);
611 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
612 sector != bio_end_sector(wpc->ioend->io_bio) ||
613 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
615 list_add(&wpc->ioend->io_list, iolist);
616 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset,
620 if (!__bio_try_merge_page(wpc->ioend->io_bio, page, len, poff)) {
622 atomic_inc(&iop->write_count);
623 if (bio_full(wpc->ioend->io_bio))
624 xfs_chain_bio(wpc->ioend, wbc, bdev, sector);
625 __bio_add_page(wpc->ioend->io_bio, page, len, poff);
628 wpc->ioend->io_size += len;
632 xfs_vm_invalidatepage(
637 trace_xfs_invalidatepage(page->mapping->host, page, offset, length);
638 iomap_invalidatepage(page, offset, length);
642 * If the page has delalloc blocks on it, we need to punch them out before we
643 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
644 * inode that can trip up a later direct I/O read operation on the same region.
646 * We prevent this by truncating away the delalloc regions on the page. Because
647 * they are delalloc, we can do this without needing a transaction. Indeed - if
648 * we get ENOSPC errors, we have to be able to do this truncation without a
649 * transaction as there is no space left for block reservation (typically why we
650 * see a ENOSPC in writeback).
653 xfs_aops_discard_page(
656 struct inode *inode = page->mapping->host;
657 struct xfs_inode *ip = XFS_I(inode);
658 struct xfs_mount *mp = ip->i_mount;
659 loff_t offset = page_offset(page);
660 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, offset);
663 if (XFS_FORCED_SHUTDOWN(mp))
667 "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.",
668 page, ip->i_ino, offset);
670 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
671 PAGE_SIZE / i_blocksize(inode));
672 if (error && !XFS_FORCED_SHUTDOWN(mp))
673 xfs_alert(mp, "page discard unable to remove delalloc mapping.");
675 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
679 * We implement an immediate ioend submission policy here to avoid needing to
680 * chain multiple ioends and hence nest mempool allocations which can violate
681 * forward progress guarantees we need to provide. The current ioend we are
682 * adding blocks to is cached on the writepage context, and if the new block
683 * does not append to the cached ioend it will create a new ioend and cache that
686 * If a new ioend is created and cached, the old ioend is returned and queued
687 * locally for submission once the entire page is processed or an error has been
688 * detected. While ioends are submitted immediately after they are completed,
689 * batching optimisations are provided by higher level block plugging.
691 * At the end of a writeback pass, there will be a cached ioend remaining on the
692 * writepage context that the caller will need to submit.
696 struct xfs_writepage_ctx *wpc,
697 struct writeback_control *wbc,
702 LIST_HEAD(submit_list);
703 struct iomap_page *iop = to_iomap_page(page);
704 unsigned len = i_blocksize(inode);
705 struct xfs_ioend *ioend, *next;
706 uint64_t file_offset; /* file offset of page */
707 int error = 0, count = 0, i;
709 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
710 ASSERT(!iop || atomic_read(&iop->write_count) == 0);
713 * Walk through the page to find areas to write back. If we run off the
714 * end of the current map or find the current map invalid, grab a new
717 for (i = 0, file_offset = page_offset(page);
718 i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
719 i++, file_offset += len) {
720 if (iop && !test_bit(i, iop->uptodate))
723 error = xfs_map_blocks(wpc, inode, file_offset);
726 if (wpc->io_type == XFS_IO_HOLE)
728 xfs_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
733 ASSERT(wpc->ioend || list_empty(&submit_list));
734 ASSERT(PageLocked(page));
735 ASSERT(!PageWriteback(page));
738 * On error, we have to fail the ioend here because we may have set
739 * pages under writeback, we have to make sure we run IO completion to
740 * mark the error state of the IO appropriately, so we can't cancel the
741 * ioend directly here. That means we have to mark this page as under
742 * writeback if we included any blocks from it in the ioend chain so
743 * that completion treats it correctly.
745 * If we didn't include the page in the ioend, the on error we can
746 * simply discard and unlock it as there are no other users of the page
747 * now. The caller will still need to trigger submission of outstanding
748 * ioends on the writepage context so they are treated correctly on
751 if (unlikely(error)) {
753 xfs_aops_discard_page(page);
754 ClearPageUptodate(page);
760 * If the page was not fully cleaned, we need to ensure that the
761 * higher layers come back to it correctly. That means we need
762 * to keep the page dirty, and for WB_SYNC_ALL writeback we need
763 * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed
764 * so another attempt to write this page in this writeback sweep
767 set_page_writeback_keepwrite(page);
769 clear_page_dirty_for_io(page);
770 set_page_writeback(page);
776 * Preserve the original error if there was one, otherwise catch
777 * submission errors here and propagate into subsequent ioend
780 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
783 list_del_init(&ioend->io_list);
784 error2 = xfs_submit_ioend(wbc, ioend, error);
785 if (error2 && !error)
790 * We can end up here with no error and nothing to write only if we race
791 * with a partial page truncate on a sub-page block sized filesystem.
794 end_page_writeback(page);
796 mapping_set_error(page->mapping, error);
801 * Write out a dirty page.
803 * For delalloc space on the page we need to allocate space and flush it.
804 * For unwritten space on the page we need to start the conversion to
805 * regular allocated space.
810 struct writeback_control *wbc,
813 struct xfs_writepage_ctx *wpc = data;
814 struct inode *inode = page->mapping->host;
819 trace_xfs_writepage(inode, page, 0, 0);
822 * Refuse to write the page out if we are called from reclaim context.
824 * This avoids stack overflows when called from deeply used stacks in
825 * random callers for direct reclaim or memcg reclaim. We explicitly
826 * allow reclaim from kswapd as the stack usage there is relatively low.
828 * This should never happen except in the case of a VM regression so
831 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
836 * Given that we do not allow direct reclaim to call us, we should
837 * never be called while in a filesystem transaction.
839 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
843 * Is this page beyond the end of the file?
845 * The page index is less than the end_index, adjust the end_offset
846 * to the highest offset that this page should represent.
847 * -----------------------------------------------------
848 * | file mapping | <EOF> |
849 * -----------------------------------------------------
850 * | Page ... | Page N-2 | Page N-1 | Page N | |
851 * ^--------------------------------^----------|--------
852 * | desired writeback range | see else |
853 * ---------------------------------^------------------|
855 offset = i_size_read(inode);
856 end_index = offset >> PAGE_SHIFT;
857 if (page->index < end_index)
858 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
861 * Check whether the page to write out is beyond or straddles
863 * -------------------------------------------------------
864 * | file mapping | <EOF> |
865 * -------------------------------------------------------
866 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
867 * ^--------------------------------^-----------|---------
869 * ---------------------------------^-----------|--------|
871 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
874 * Skip the page if it is fully outside i_size, e.g. due to a
875 * truncate operation that is in progress. We must redirty the
876 * page so that reclaim stops reclaiming it. Otherwise
877 * xfs_vm_releasepage() is called on it and gets confused.
879 * Note that the end_index is unsigned long, it would overflow
880 * if the given offset is greater than 16TB on 32-bit system
881 * and if we do check the page is fully outside i_size or not
882 * via "if (page->index >= end_index + 1)" as "end_index + 1"
883 * will be evaluated to 0. Hence this page will be redirtied
884 * and be written out repeatedly which would result in an
885 * infinite loop, the user program that perform this operation
886 * will hang. Instead, we can verify this situation by checking
887 * if the page to write is totally beyond the i_size or if it's
888 * offset is just equal to the EOF.
890 if (page->index > end_index ||
891 (page->index == end_index && offset_into_page == 0))
895 * The page straddles i_size. It must be zeroed out on each
896 * and every writepage invocation because it may be mmapped.
897 * "A file is mapped in multiples of the page size. For a file
898 * that is not a multiple of the page size, the remaining
899 * memory is zeroed when mapped, and writes to that region are
900 * not written out to the file."
902 zero_user_segment(page, offset_into_page, PAGE_SIZE);
904 /* Adjust the end_offset to the end of file */
908 return xfs_writepage_map(wpc, wbc, inode, page, end_offset);
911 redirty_page_for_writepage(wbc, page);
919 struct writeback_control *wbc)
921 struct xfs_writepage_ctx wpc = {
922 .io_type = XFS_IO_HOLE,
926 ret = xfs_do_writepage(page, wbc, &wpc);
928 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
934 struct address_space *mapping,
935 struct writeback_control *wbc)
937 struct xfs_writepage_ctx wpc = {
938 .io_type = XFS_IO_HOLE,
942 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
943 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
945 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
951 struct address_space *mapping,
952 struct writeback_control *wbc)
954 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
955 return dax_writeback_mapping_range(mapping,
956 xfs_find_bdev_for_inode(mapping->host), wbc);
964 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
965 return iomap_releasepage(page, gfp_mask);
970 struct address_space *mapping,
973 struct xfs_inode *ip = XFS_I(mapping->host);
975 trace_xfs_vm_bmap(ip);
978 * The swap code (ab-)uses ->bmap to get a block mapping and then
979 * bypasses the file system for actual I/O. We really can't allow
980 * that on reflinks inodes, so we have to skip out here. And yes,
981 * 0 is the magic code for a bmap error.
983 * Since we don't pass back blockdev info, we can't return bmap
984 * information for rt files either.
986 if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip))
988 return iomap_bmap(mapping, block, &xfs_iomap_ops);
996 trace_xfs_vm_readpage(page->mapping->host, 1);
997 return iomap_readpage(page, &xfs_iomap_ops);
1002 struct file *unused,
1003 struct address_space *mapping,
1004 struct list_head *pages,
1007 trace_xfs_vm_readpages(mapping->host, nr_pages);
1008 return iomap_readpages(mapping, pages, nr_pages, &xfs_iomap_ops);
1012 xfs_iomap_swapfile_activate(
1013 struct swap_info_struct *sis,
1014 struct file *swap_file,
1017 sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
1018 return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops);
1021 const struct address_space_operations xfs_address_space_operations = {
1022 .readpage = xfs_vm_readpage,
1023 .readpages = xfs_vm_readpages,
1024 .writepage = xfs_vm_writepage,
1025 .writepages = xfs_vm_writepages,
1026 .set_page_dirty = iomap_set_page_dirty,
1027 .releasepage = xfs_vm_releasepage,
1028 .invalidatepage = xfs_vm_invalidatepage,
1029 .bmap = xfs_vm_bmap,
1030 .direct_IO = noop_direct_IO,
1031 .migratepage = iomap_migrate_page,
1032 .is_partially_uptodate = iomap_is_partially_uptodate,
1033 .error_remove_page = generic_error_remove_page,
1034 .swap_activate = xfs_iomap_swapfile_activate,
1037 const struct address_space_operations xfs_dax_aops = {
1038 .writepages = xfs_dax_writepages,
1039 .direct_IO = noop_direct_IO,
1040 .set_page_dirty = noop_set_page_dirty,
1041 .invalidatepage = noop_invalidatepage,
1042 .swap_activate = xfs_iomap_swapfile_activate,