1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_inode.h"
13 #include "xfs_trans.h"
14 #include "xfs_inode_item.h"
15 #include "xfs_alloc.h"
16 #include "xfs_error.h"
17 #include "xfs_iomap.h"
18 #include "xfs_trace.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_reflink.h"
23 #include <linux/writeback.h>
26 * structure owned by writepages passed to individual writepage calls
28 struct xfs_writepage_ctx {
29 struct xfs_bmbt_irec imap;
31 struct xfs_ioend *ioend;
35 xfs_find_bdev_for_inode(
38 struct xfs_inode *ip = XFS_I(inode);
39 struct xfs_mount *mp = ip->i_mount;
41 if (XFS_IS_REALTIME_INODE(ip))
42 return mp->m_rtdev_targp->bt_bdev;
44 return mp->m_ddev_targp->bt_bdev;
48 xfs_find_daxdev_for_inode(
51 struct xfs_inode *ip = XFS_I(inode);
52 struct xfs_mount *mp = ip->i_mount;
54 if (XFS_IS_REALTIME_INODE(ip))
55 return mp->m_rtdev_targp->bt_daxdev;
57 return mp->m_ddev_targp->bt_daxdev;
61 xfs_finish_page_writeback(
66 struct iomap_page *iop = to_iomap_page(bvec->bv_page);
69 SetPageError(bvec->bv_page);
70 mapping_set_error(inode->i_mapping, -EIO);
73 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
74 ASSERT(!iop || atomic_read(&iop->write_count) > 0);
76 if (!iop || atomic_dec_and_test(&iop->write_count))
77 end_page_writeback(bvec->bv_page);
81 * We're now finished for good with this ioend structure. Update the page
82 * state, release holds on bios, and finally free up memory. Do not use the
87 struct xfs_ioend *ioend,
90 struct inode *inode = ioend->io_inode;
91 struct bio *bio = &ioend->io_inline_bio;
92 struct bio *last = ioend->io_bio, *next;
93 u64 start = bio->bi_iter.bi_sector;
94 bool quiet = bio_flagged(bio, BIO_QUIET);
96 for (bio = &ioend->io_inline_bio; bio; bio = next) {
101 * For the last bio, bi_private points to the ioend, so we
102 * need to explicitly end the iteration here.
107 next = bio->bi_private;
109 /* walk each page on bio, ending page IO on them */
110 bio_for_each_segment_all(bvec, bio, i)
111 xfs_finish_page_writeback(inode, bvec, error);
115 if (unlikely(error && !quiet)) {
116 xfs_err_ratelimited(XFS_I(inode)->i_mount,
117 "writeback error on sector %llu", start);
122 * Fast and loose check if this write could update the on-disk inode size.
124 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
126 return ioend->io_offset + ioend->io_size >
127 XFS_I(ioend->io_inode)->i_d.di_size;
131 xfs_setfilesize_trans_alloc(
132 struct xfs_ioend *ioend)
134 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
135 struct xfs_trans *tp;
138 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0,
139 XFS_TRANS_NOFS, &tp);
143 ioend->io_append_trans = tp;
146 * We may pass freeze protection with a transaction. So tell lockdep
149 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
151 * We hand off the transaction to the completion thread now, so
152 * clear the flag here.
154 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
159 * Update on-disk file size now that data has been written to disk.
163 struct xfs_inode *ip,
164 struct xfs_trans *tp,
170 xfs_ilock(ip, XFS_ILOCK_EXCL);
171 isize = xfs_new_eof(ip, offset + size);
173 xfs_iunlock(ip, XFS_ILOCK_EXCL);
174 xfs_trans_cancel(tp);
178 trace_xfs_setfilesize(ip, offset, size);
180 ip->i_d.di_size = isize;
181 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
182 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
184 return xfs_trans_commit(tp);
189 struct xfs_inode *ip,
193 struct xfs_mount *mp = ip->i_mount;
194 struct xfs_trans *tp;
197 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
201 return __xfs_setfilesize(ip, tp, offset, size);
205 xfs_setfilesize_ioend(
206 struct xfs_ioend *ioend,
209 struct xfs_inode *ip = XFS_I(ioend->io_inode);
210 struct xfs_trans *tp = ioend->io_append_trans;
213 * The transaction may have been allocated in the I/O submission thread,
214 * thus we need to mark ourselves as being in a transaction manually.
215 * Similarly for freeze protection.
217 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
218 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
220 /* we abort the update if there was an IO error */
222 xfs_trans_cancel(tp);
226 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
230 * IO write completion.
234 struct work_struct *work)
236 struct xfs_ioend *ioend =
237 container_of(work, struct xfs_ioend, io_work);
238 struct xfs_inode *ip = XFS_I(ioend->io_inode);
239 xfs_off_t offset = ioend->io_offset;
240 size_t size = ioend->io_size;
244 * Just clean up the in-memory strutures if the fs has been shut down.
246 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
252 * Clean up any COW blocks on an I/O error.
254 error = blk_status_to_errno(ioend->io_bio->bi_status);
255 if (unlikely(error)) {
256 switch (ioend->io_type) {
258 xfs_reflink_cancel_cow_range(ip, offset, size, true);
266 * Success: commit the COW or unwritten blocks if needed.
268 switch (ioend->io_type) {
270 error = xfs_reflink_end_cow(ip, offset, size);
272 case XFS_IO_UNWRITTEN:
273 /* writeback should never update isize */
274 error = xfs_iomap_write_unwritten(ip, offset, size, false);
277 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
282 if (ioend->io_append_trans)
283 error = xfs_setfilesize_ioend(ioend, error);
284 xfs_destroy_ioend(ioend, error);
291 struct xfs_ioend *ioend = bio->bi_private;
292 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
294 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
295 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
296 else if (ioend->io_append_trans)
297 queue_work(mp->m_data_workqueue, &ioend->io_work);
299 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
304 struct xfs_writepage_ctx *wpc,
308 struct xfs_inode *ip = XFS_I(inode);
309 struct xfs_mount *mp = ip->i_mount;
310 ssize_t count = i_blocksize(inode);
311 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset), end_fsb;
312 struct xfs_bmbt_irec imap;
313 int whichfork = XFS_DATA_FORK;
314 struct xfs_iext_cursor icur;
319 * We have to make sure the cached mapping is within EOF to protect
320 * against eofblocks trimming on file release leaving us with a stale
321 * mapping. Otherwise, a page for a subsequent file extending buffered
322 * write could get picked up by this writeback cycle and written to the
325 * Note that what we really want here is a generic mapping invalidation
326 * mechanism to protect us from arbitrary extent modifying contexts, not
329 xfs_trim_extent_eof(&wpc->imap, ip);
332 * COW fork blocks can overlap data fork blocks even if the blocks
333 * aren't shared. COW I/O always takes precedent, so we must always
334 * check for overlap on reflink inodes unless the mapping is already a
337 imap_valid = offset_fsb >= wpc->imap.br_startoff &&
338 offset_fsb < wpc->imap.br_startoff + wpc->imap.br_blockcount;
340 (!xfs_is_reflink_inode(ip) || wpc->io_type == XFS_IO_COW))
343 if (XFS_FORCED_SHUTDOWN(mp))
347 * If we don't have a valid map, now it's time to get a new one for this
348 * offset. This will convert delayed allocations (including COW ones)
349 * into real extents. If we return without a valid map, it means we
350 * landed in a hole and we skip the block.
352 xfs_ilock(ip, XFS_ILOCK_SHARED);
353 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
354 (ip->i_df.if_flags & XFS_IFEXTENTS));
355 ASSERT(offset <= mp->m_super->s_maxbytes);
357 if (offset > mp->m_super->s_maxbytes - count)
358 count = mp->m_super->s_maxbytes - offset;
359 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
362 * Check if this is offset is covered by a COW extents, and if yes use
363 * it directly instead of looking up anything in the data fork.
365 if (xfs_is_reflink_inode(ip) &&
366 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap) &&
367 imap.br_startoff <= offset_fsb) {
368 xfs_iunlock(ip, XFS_ILOCK_SHARED);
370 * Truncate can race with writeback since writeback doesn't
371 * take the iolock and truncate decreases the file size before
372 * it starts truncating the pages between new_size and old_size.
373 * Therefore, we can end up in the situation where writeback
374 * gets a CoW fork mapping but the truncate makes the mapping
375 * invalid and we end up in here trying to get a new mapping.
376 * bail out here so that we simply never get a valid mapping
377 * and so we drop the write altogether. The page truncation
378 * will kill the contents anyway.
380 if (offset > i_size_read(inode)) {
381 wpc->io_type = XFS_IO_HOLE;
384 whichfork = XFS_COW_FORK;
385 wpc->io_type = XFS_IO_COW;
386 goto allocate_blocks;
390 * Map valid and no COW extent in the way? We're done.
393 xfs_iunlock(ip, XFS_ILOCK_SHARED);
398 * If we don't have a valid map, now it's time to get a new one for this
399 * offset. This will convert delayed allocations (including COW ones)
402 if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
403 imap.br_startoff = end_fsb; /* fake a hole past EOF */
404 xfs_iunlock(ip, XFS_ILOCK_SHARED);
406 if (imap.br_startoff > offset_fsb) {
407 /* landed in a hole or beyond EOF */
408 imap.br_blockcount = imap.br_startoff - offset_fsb;
409 imap.br_startoff = offset_fsb;
410 imap.br_startblock = HOLESTARTBLOCK;
411 wpc->io_type = XFS_IO_HOLE;
413 if (isnullstartblock(imap.br_startblock)) {
414 /* got a delalloc extent */
415 wpc->io_type = XFS_IO_DELALLOC;
416 goto allocate_blocks;
419 if (imap.br_state == XFS_EXT_UNWRITTEN)
420 wpc->io_type = XFS_IO_UNWRITTEN;
422 wpc->io_type = XFS_IO_OVERWRITE;
426 trace_xfs_map_blocks_found(ip, offset, count, wpc->io_type, &imap);
429 error = xfs_iomap_write_allocate(ip, whichfork, offset, &imap);
433 trace_xfs_map_blocks_alloc(ip, offset, count, wpc->io_type, &imap);
438 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
439 * it, and we submit that bio. The ioend may be used for multiple bio
440 * submissions, so we only want to allocate an append transaction for the ioend
441 * once. In the case of multiple bio submission, each bio will take an IO
442 * reference to the ioend to ensure that the ioend completion is only done once
443 * all bios have been submitted and the ioend is really done.
445 * If @fail is non-zero, it means that we have a situation where some part of
446 * the submission process has failed after we have marked paged for writeback
447 * and unlocked them. In this situation, we need to fail the bio and ioend
448 * rather than submit it to IO. This typically only happens on a filesystem
453 struct writeback_control *wbc,
454 struct xfs_ioend *ioend,
457 /* Convert CoW extents to regular */
458 if (!status && ioend->io_type == XFS_IO_COW) {
460 * Yuk. This can do memory allocation, but is not a
461 * transactional operation so everything is done in GFP_KERNEL
462 * context. That can deadlock, because we hold pages in
463 * writeback state and GFP_KERNEL allocations can block on them.
464 * Hence we must operate in nofs conditions here.
468 nofs_flag = memalloc_nofs_save();
469 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
470 ioend->io_offset, ioend->io_size);
471 memalloc_nofs_restore(nofs_flag);
474 /* Reserve log space if we might write beyond the on-disk inode size. */
476 ioend->io_type != XFS_IO_UNWRITTEN &&
477 xfs_ioend_is_append(ioend) &&
478 !ioend->io_append_trans)
479 status = xfs_setfilesize_trans_alloc(ioend);
481 ioend->io_bio->bi_private = ioend;
482 ioend->io_bio->bi_end_io = xfs_end_bio;
483 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
486 * If we are failing the IO now, just mark the ioend with an
487 * error and finish it. This will run IO completion immediately
488 * as there is only one reference to the ioend at this point in
492 ioend->io_bio->bi_status = errno_to_blk_status(status);
493 bio_endio(ioend->io_bio);
497 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
498 submit_bio(ioend->io_bio);
502 static struct xfs_ioend *
507 struct block_device *bdev,
510 struct xfs_ioend *ioend;
513 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset);
514 bio_set_dev(bio, bdev);
515 bio->bi_iter.bi_sector = sector;
517 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
518 INIT_LIST_HEAD(&ioend->io_list);
519 ioend->io_type = type;
520 ioend->io_inode = inode;
522 ioend->io_offset = offset;
523 INIT_WORK(&ioend->io_work, xfs_end_io);
524 ioend->io_append_trans = NULL;
530 * Allocate a new bio, and chain the old bio to the new one.
532 * Note that we have to do perform the chaining in this unintuitive order
533 * so that the bi_private linkage is set up in the right direction for the
534 * traversal in xfs_destroy_ioend().
538 struct xfs_ioend *ioend,
539 struct writeback_control *wbc,
540 struct block_device *bdev,
545 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
546 bio_set_dev(new, bdev);
547 new->bi_iter.bi_sector = sector;
548 bio_chain(ioend->io_bio, new);
549 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
550 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
551 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
552 submit_bio(ioend->io_bio);
557 * Test to see if we have an existing ioend structure that we could append to
558 * first, otherwise finish off the current ioend and start another.
565 struct iomap_page *iop,
566 struct xfs_writepage_ctx *wpc,
567 struct writeback_control *wbc,
568 struct list_head *iolist)
570 struct xfs_inode *ip = XFS_I(inode);
571 struct xfs_mount *mp = ip->i_mount;
572 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
573 unsigned len = i_blocksize(inode);
574 unsigned poff = offset & (PAGE_SIZE - 1);
577 sector = xfs_fsb_to_db(ip, wpc->imap.br_startblock) +
578 ((offset - XFS_FSB_TO_B(mp, wpc->imap.br_startoff)) >> 9);
580 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
581 sector != bio_end_sector(wpc->ioend->io_bio) ||
582 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
584 list_add(&wpc->ioend->io_list, iolist);
585 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset,
589 if (!__bio_try_merge_page(wpc->ioend->io_bio, page, len, poff)) {
591 atomic_inc(&iop->write_count);
592 if (bio_full(wpc->ioend->io_bio))
593 xfs_chain_bio(wpc->ioend, wbc, bdev, sector);
594 __bio_add_page(wpc->ioend->io_bio, page, len, poff);
597 wpc->ioend->io_size += len;
601 xfs_vm_invalidatepage(
606 trace_xfs_invalidatepage(page->mapping->host, page, offset, length);
607 iomap_invalidatepage(page, offset, length);
611 * If the page has delalloc blocks on it, we need to punch them out before we
612 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
613 * inode that can trip up a later direct I/O read operation on the same region.
615 * We prevent this by truncating away the delalloc regions on the page. Because
616 * they are delalloc, we can do this without needing a transaction. Indeed - if
617 * we get ENOSPC errors, we have to be able to do this truncation without a
618 * transaction as there is no space left for block reservation (typically why we
619 * see a ENOSPC in writeback).
622 xfs_aops_discard_page(
625 struct inode *inode = page->mapping->host;
626 struct xfs_inode *ip = XFS_I(inode);
627 struct xfs_mount *mp = ip->i_mount;
628 loff_t offset = page_offset(page);
629 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, offset);
632 if (XFS_FORCED_SHUTDOWN(mp))
636 "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.",
637 page, ip->i_ino, offset);
639 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
640 PAGE_SIZE / i_blocksize(inode));
641 if (error && !XFS_FORCED_SHUTDOWN(mp))
642 xfs_alert(mp, "page discard unable to remove delalloc mapping.");
644 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
648 * We implement an immediate ioend submission policy here to avoid needing to
649 * chain multiple ioends and hence nest mempool allocations which can violate
650 * forward progress guarantees we need to provide. The current ioend we are
651 * adding blocks to is cached on the writepage context, and if the new block
652 * does not append to the cached ioend it will create a new ioend and cache that
655 * If a new ioend is created and cached, the old ioend is returned and queued
656 * locally for submission once the entire page is processed or an error has been
657 * detected. While ioends are submitted immediately after they are completed,
658 * batching optimisations are provided by higher level block plugging.
660 * At the end of a writeback pass, there will be a cached ioend remaining on the
661 * writepage context that the caller will need to submit.
665 struct xfs_writepage_ctx *wpc,
666 struct writeback_control *wbc,
671 LIST_HEAD(submit_list);
672 struct iomap_page *iop = to_iomap_page(page);
673 unsigned len = i_blocksize(inode);
674 struct xfs_ioend *ioend, *next;
675 uint64_t file_offset; /* file offset of page */
676 int error = 0, count = 0, i;
678 ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
679 ASSERT(!iop || atomic_read(&iop->write_count) == 0);
682 * Walk through the page to find areas to write back. If we run off the
683 * end of the current map or find the current map invalid, grab a new
686 for (i = 0, file_offset = page_offset(page);
687 i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
688 i++, file_offset += len) {
689 if (iop && !test_bit(i, iop->uptodate))
692 error = xfs_map_blocks(wpc, inode, file_offset);
695 if (wpc->io_type == XFS_IO_HOLE)
697 xfs_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
702 ASSERT(wpc->ioend || list_empty(&submit_list));
703 ASSERT(PageLocked(page));
704 ASSERT(!PageWriteback(page));
707 * On error, we have to fail the ioend here because we may have set
708 * pages under writeback, we have to make sure we run IO completion to
709 * mark the error state of the IO appropriately, so we can't cancel the
710 * ioend directly here. That means we have to mark this page as under
711 * writeback if we included any blocks from it in the ioend chain so
712 * that completion treats it correctly.
714 * If we didn't include the page in the ioend, the on error we can
715 * simply discard and unlock it as there are no other users of the page
716 * now. The caller will still need to trigger submission of outstanding
717 * ioends on the writepage context so they are treated correctly on
720 if (unlikely(error)) {
722 xfs_aops_discard_page(page);
723 ClearPageUptodate(page);
729 * If the page was not fully cleaned, we need to ensure that the
730 * higher layers come back to it correctly. That means we need
731 * to keep the page dirty, and for WB_SYNC_ALL writeback we need
732 * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed
733 * so another attempt to write this page in this writeback sweep
736 set_page_writeback_keepwrite(page);
738 clear_page_dirty_for_io(page);
739 set_page_writeback(page);
745 * Preserve the original error if there was one, otherwise catch
746 * submission errors here and propagate into subsequent ioend
749 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
752 list_del_init(&ioend->io_list);
753 error2 = xfs_submit_ioend(wbc, ioend, error);
754 if (error2 && !error)
759 * We can end up here with no error and nothing to write only if we race
760 * with a partial page truncate on a sub-page block sized filesystem.
763 end_page_writeback(page);
765 mapping_set_error(page->mapping, error);
770 * Write out a dirty page.
772 * For delalloc space on the page we need to allocate space and flush it.
773 * For unwritten space on the page we need to start the conversion to
774 * regular allocated space.
779 struct writeback_control *wbc,
782 struct xfs_writepage_ctx *wpc = data;
783 struct inode *inode = page->mapping->host;
788 trace_xfs_writepage(inode, page, 0, 0);
791 * Refuse to write the page out if we are called from reclaim context.
793 * This avoids stack overflows when called from deeply used stacks in
794 * random callers for direct reclaim or memcg reclaim. We explicitly
795 * allow reclaim from kswapd as the stack usage there is relatively low.
797 * This should never happen except in the case of a VM regression so
800 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
805 * Given that we do not allow direct reclaim to call us, we should
806 * never be called while in a filesystem transaction.
808 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
812 * Is this page beyond the end of the file?
814 * The page index is less than the end_index, adjust the end_offset
815 * to the highest offset that this page should represent.
816 * -----------------------------------------------------
817 * | file mapping | <EOF> |
818 * -----------------------------------------------------
819 * | Page ... | Page N-2 | Page N-1 | Page N | |
820 * ^--------------------------------^----------|--------
821 * | desired writeback range | see else |
822 * ---------------------------------^------------------|
824 offset = i_size_read(inode);
825 end_index = offset >> PAGE_SHIFT;
826 if (page->index < end_index)
827 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
830 * Check whether the page to write out is beyond or straddles
832 * -------------------------------------------------------
833 * | file mapping | <EOF> |
834 * -------------------------------------------------------
835 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
836 * ^--------------------------------^-----------|---------
838 * ---------------------------------^-----------|--------|
840 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
843 * Skip the page if it is fully outside i_size, e.g. due to a
844 * truncate operation that is in progress. We must redirty the
845 * page so that reclaim stops reclaiming it. Otherwise
846 * xfs_vm_releasepage() is called on it and gets confused.
848 * Note that the end_index is unsigned long, it would overflow
849 * if the given offset is greater than 16TB on 32-bit system
850 * and if we do check the page is fully outside i_size or not
851 * via "if (page->index >= end_index + 1)" as "end_index + 1"
852 * will be evaluated to 0. Hence this page will be redirtied
853 * and be written out repeatedly which would result in an
854 * infinite loop, the user program that perform this operation
855 * will hang. Instead, we can verify this situation by checking
856 * if the page to write is totally beyond the i_size or if it's
857 * offset is just equal to the EOF.
859 if (page->index > end_index ||
860 (page->index == end_index && offset_into_page == 0))
864 * The page straddles i_size. It must be zeroed out on each
865 * and every writepage invocation because it may be mmapped.
866 * "A file is mapped in multiples of the page size. For a file
867 * that is not a multiple of the page size, the remaining
868 * memory is zeroed when mapped, and writes to that region are
869 * not written out to the file."
871 zero_user_segment(page, offset_into_page, PAGE_SIZE);
873 /* Adjust the end_offset to the end of file */
877 return xfs_writepage_map(wpc, wbc, inode, page, end_offset);
880 redirty_page_for_writepage(wbc, page);
888 struct writeback_control *wbc)
890 struct xfs_writepage_ctx wpc = {
891 .io_type = XFS_IO_INVALID,
895 ret = xfs_do_writepage(page, wbc, &wpc);
897 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
903 struct address_space *mapping,
904 struct writeback_control *wbc)
906 struct xfs_writepage_ctx wpc = {
907 .io_type = XFS_IO_INVALID,
911 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
912 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
914 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
920 struct address_space *mapping,
921 struct writeback_control *wbc)
923 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
924 return dax_writeback_mapping_range(mapping,
925 xfs_find_bdev_for_inode(mapping->host), wbc);
933 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
934 return iomap_releasepage(page, gfp_mask);
939 struct address_space *mapping,
942 struct xfs_inode *ip = XFS_I(mapping->host);
944 trace_xfs_vm_bmap(ip);
947 * The swap code (ab-)uses ->bmap to get a block mapping and then
948 * bypasses the file system for actual I/O. We really can't allow
949 * that on reflinks inodes, so we have to skip out here. And yes,
950 * 0 is the magic code for a bmap error.
952 * Since we don't pass back blockdev info, we can't return bmap
953 * information for rt files either.
955 if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip))
957 return iomap_bmap(mapping, block, &xfs_iomap_ops);
965 trace_xfs_vm_readpage(page->mapping->host, 1);
966 return iomap_readpage(page, &xfs_iomap_ops);
972 struct address_space *mapping,
973 struct list_head *pages,
976 trace_xfs_vm_readpages(mapping->host, nr_pages);
977 return iomap_readpages(mapping, pages, nr_pages, &xfs_iomap_ops);
981 xfs_iomap_swapfile_activate(
982 struct swap_info_struct *sis,
983 struct file *swap_file,
986 sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
987 return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops);
990 const struct address_space_operations xfs_address_space_operations = {
991 .readpage = xfs_vm_readpage,
992 .readpages = xfs_vm_readpages,
993 .writepage = xfs_vm_writepage,
994 .writepages = xfs_vm_writepages,
995 .set_page_dirty = iomap_set_page_dirty,
996 .releasepage = xfs_vm_releasepage,
997 .invalidatepage = xfs_vm_invalidatepage,
999 .direct_IO = noop_direct_IO,
1000 .migratepage = iomap_migrate_page,
1001 .is_partially_uptodate = iomap_is_partially_uptodate,
1002 .error_remove_page = generic_error_remove_page,
1003 .swap_activate = xfs_iomap_swapfile_activate,
1006 const struct address_space_operations xfs_dax_aops = {
1007 .writepages = xfs_dax_writepages,
1008 .direct_IO = noop_direct_IO,
1009 .set_page_dirty = noop_set_page_dirty,
1010 .invalidatepage = noop_invalidatepage,
1011 .swap_activate = xfs_iomap_swapfile_activate,