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/gfp.h>
24 #include <linux/mpage.h>
25 #include <linux/pagevec.h>
26 #include <linux/writeback.h>
29 * structure owned by writepages passed to individual writepage calls
31 struct xfs_writepage_ctx {
32 struct xfs_bmbt_irec imap;
35 struct xfs_ioend *ioend;
45 struct buffer_head *bh, *head;
47 *delalloc = *unwritten = 0;
49 bh = head = page_buffers(page);
51 if (buffer_unwritten(bh))
53 else if (buffer_delay(bh))
55 } while ((bh = bh->b_this_page) != head);
59 xfs_find_bdev_for_inode(
62 struct xfs_inode *ip = XFS_I(inode);
63 struct xfs_mount *mp = ip->i_mount;
65 if (XFS_IS_REALTIME_INODE(ip))
66 return mp->m_rtdev_targp->bt_bdev;
68 return mp->m_ddev_targp->bt_bdev;
72 xfs_find_daxdev_for_inode(
75 struct xfs_inode *ip = XFS_I(inode);
76 struct xfs_mount *mp = ip->i_mount;
78 if (XFS_IS_REALTIME_INODE(ip))
79 return mp->m_rtdev_targp->bt_daxdev;
81 return mp->m_ddev_targp->bt_daxdev;
85 * We're now finished for good with this page. Update the page state via the
86 * associated buffer_heads, paying attention to the start and end offsets that
87 * we need to process on the page.
89 * Note that we open code the action in end_buffer_async_write here so that we
90 * only have to iterate over the buffers attached to the page once. This is not
91 * only more efficient, but also ensures that we only calls end_page_writeback
92 * at the end of the iteration, and thus avoids the pitfall of having the page
93 * and buffers potentially freed after every call to end_buffer_async_write.
96 xfs_finish_page_writeback(
101 struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head;
103 unsigned int off = 0;
106 ASSERT(bvec->bv_offset < PAGE_SIZE);
107 ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0);
108 ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE);
109 ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0);
111 local_irq_save(flags);
112 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
114 if (off >= bvec->bv_offset &&
115 off < bvec->bv_offset + bvec->bv_len) {
116 ASSERT(buffer_async_write(bh));
117 ASSERT(bh->b_end_io == NULL);
120 mark_buffer_write_io_error(bh);
121 clear_buffer_uptodate(bh);
122 SetPageError(bvec->bv_page);
124 set_buffer_uptodate(bh);
126 clear_buffer_async_write(bh);
128 } else if (buffer_async_write(bh)) {
129 ASSERT(buffer_locked(bh));
133 } while ((bh = bh->b_this_page) != head);
134 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
135 local_irq_restore(flags);
138 end_page_writeback(bvec->bv_page);
142 * We're now finished for good with this ioend structure. Update the page
143 * state, release holds on bios, and finally free up memory. Do not use the
148 struct xfs_ioend *ioend,
151 struct inode *inode = ioend->io_inode;
152 struct bio *bio = &ioend->io_inline_bio;
153 struct bio *last = ioend->io_bio, *next;
154 u64 start = bio->bi_iter.bi_sector;
155 bool quiet = bio_flagged(bio, BIO_QUIET);
157 for (bio = &ioend->io_inline_bio; bio; bio = next) {
158 struct bio_vec *bvec;
162 * For the last bio, bi_private points to the ioend, so we
163 * need to explicitly end the iteration here.
168 next = bio->bi_private;
170 /* walk each page on bio, ending page IO on them */
171 bio_for_each_segment_all(bvec, bio, i)
172 xfs_finish_page_writeback(inode, bvec, error);
177 if (unlikely(error && !quiet)) {
178 xfs_err_ratelimited(XFS_I(inode)->i_mount,
179 "writeback error on sector %llu", start);
184 * Fast and loose check if this write could update the on-disk inode size.
186 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
188 return ioend->io_offset + ioend->io_size >
189 XFS_I(ioend->io_inode)->i_d.di_size;
193 xfs_setfilesize_trans_alloc(
194 struct xfs_ioend *ioend)
196 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
197 struct xfs_trans *tp;
200 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0,
201 XFS_TRANS_NOFS, &tp);
205 ioend->io_append_trans = tp;
208 * We may pass freeze protection with a transaction. So tell lockdep
211 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
213 * We hand off the transaction to the completion thread now, so
214 * clear the flag here.
216 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
221 * Update on-disk file size now that data has been written to disk.
225 struct xfs_inode *ip,
226 struct xfs_trans *tp,
232 xfs_ilock(ip, XFS_ILOCK_EXCL);
233 isize = xfs_new_eof(ip, offset + size);
235 xfs_iunlock(ip, XFS_ILOCK_EXCL);
236 xfs_trans_cancel(tp);
240 trace_xfs_setfilesize(ip, offset, size);
242 ip->i_d.di_size = isize;
243 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
244 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
246 return xfs_trans_commit(tp);
251 struct xfs_inode *ip,
255 struct xfs_mount *mp = ip->i_mount;
256 struct xfs_trans *tp;
259 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
263 return __xfs_setfilesize(ip, tp, offset, size);
267 xfs_setfilesize_ioend(
268 struct xfs_ioend *ioend,
271 struct xfs_inode *ip = XFS_I(ioend->io_inode);
272 struct xfs_trans *tp = ioend->io_append_trans;
275 * The transaction may have been allocated in the I/O submission thread,
276 * thus we need to mark ourselves as being in a transaction manually.
277 * Similarly for freeze protection.
279 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
280 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
282 /* we abort the update if there was an IO error */
284 xfs_trans_cancel(tp);
288 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
292 * IO write completion.
296 struct work_struct *work)
298 struct xfs_ioend *ioend =
299 container_of(work, struct xfs_ioend, io_work);
300 struct xfs_inode *ip = XFS_I(ioend->io_inode);
301 xfs_off_t offset = ioend->io_offset;
302 size_t size = ioend->io_size;
306 * Just clean up the in-memory strutures if the fs has been shut down.
308 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
314 * Clean up any COW blocks on an I/O error.
316 error = blk_status_to_errno(ioend->io_bio->bi_status);
317 if (unlikely(error)) {
318 switch (ioend->io_type) {
320 xfs_reflink_cancel_cow_range(ip, offset, size, true);
328 * Success: commit the COW or unwritten blocks if needed.
330 switch (ioend->io_type) {
332 error = xfs_reflink_end_cow(ip, offset, size);
334 case XFS_IO_UNWRITTEN:
335 /* writeback should never update isize */
336 error = xfs_iomap_write_unwritten(ip, offset, size, false);
339 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
344 if (ioend->io_append_trans)
345 error = xfs_setfilesize_ioend(ioend, error);
346 xfs_destroy_ioend(ioend, error);
353 struct xfs_ioend *ioend = bio->bi_private;
354 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
356 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
357 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
358 else if (ioend->io_append_trans)
359 queue_work(mp->m_data_workqueue, &ioend->io_work);
361 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
368 struct xfs_bmbt_irec *imap,
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, end_fsb;
376 int bmapi_flags = XFS_BMAPI_ENTIRE;
379 if (XFS_FORCED_SHUTDOWN(mp))
383 * Truncate can race with writeback since writeback doesn't take the
384 * iolock and truncate decreases the file size before it starts
385 * truncating the pages between new_size and old_size. Therefore, we
386 * can end up in the situation where writeback gets a CoW fork mapping
387 * but the truncate makes the mapping invalid and we end up in here
388 * trying to get a new mapping. Bail out here so that we simply never
389 * get a valid mapping and so we drop the write altogether. The page
390 * truncation will kill the contents anyway.
392 if (type == XFS_IO_COW && offset > i_size_read(inode))
395 ASSERT(type != XFS_IO_COW);
396 if (type == XFS_IO_UNWRITTEN)
397 bmapi_flags |= XFS_BMAPI_IGSTATE;
399 xfs_ilock(ip, XFS_ILOCK_SHARED);
400 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
401 (ip->i_df.if_flags & XFS_IFEXTENTS));
402 ASSERT(offset <= mp->m_super->s_maxbytes);
404 if (offset > mp->m_super->s_maxbytes - count)
405 count = mp->m_super->s_maxbytes - offset;
406 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
407 offset_fsb = XFS_B_TO_FSBT(mp, offset);
408 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
409 imap, &nimaps, bmapi_flags);
411 * Truncate an overwrite extent if there's a pending CoW
412 * reservation before the end of this extent. This forces us
413 * to come back to writepage to take care of the CoW.
415 if (nimaps && type == XFS_IO_OVERWRITE)
416 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
417 xfs_iunlock(ip, XFS_ILOCK_SHARED);
422 if (type == XFS_IO_DELALLOC &&
423 (!nimaps || isnullstartblock(imap->br_startblock))) {
424 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
427 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
432 if (type == XFS_IO_UNWRITTEN) {
434 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
435 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
439 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
446 struct xfs_bmbt_irec *imap,
449 offset >>= inode->i_blkbits;
452 * We have to make sure the cached mapping is within EOF to protect
453 * against eofblocks trimming on file release leaving us with a stale
454 * mapping. Otherwise, a page for a subsequent file extending buffered
455 * write could get picked up by this writeback cycle and written to the
458 * Note that what we really want here is a generic mapping invalidation
459 * mechanism to protect us from arbitrary extent modifying contexts, not
462 xfs_trim_extent_eof(imap, XFS_I(inode));
464 return offset >= imap->br_startoff &&
465 offset < imap->br_startoff + imap->br_blockcount;
469 xfs_start_buffer_writeback(
470 struct buffer_head *bh)
472 ASSERT(buffer_mapped(bh));
473 ASSERT(buffer_locked(bh));
474 ASSERT(!buffer_delay(bh));
475 ASSERT(!buffer_unwritten(bh));
478 set_buffer_async_write(bh);
479 set_buffer_uptodate(bh);
480 clear_buffer_dirty(bh);
484 xfs_start_page_writeback(
488 ASSERT(PageLocked(page));
489 ASSERT(!PageWriteback(page));
492 * if the page was not fully cleaned, we need to ensure that the higher
493 * layers come back to it correctly. That means we need to keep the page
494 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
495 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
496 * write this page in this writeback sweep will be made.
499 clear_page_dirty_for_io(page);
500 set_page_writeback(page);
502 set_page_writeback_keepwrite(page);
507 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
509 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
513 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
514 * it, and we submit that bio. The ioend may be used for multiple bio
515 * submissions, so we only want to allocate an append transaction for the ioend
516 * once. In the case of multiple bio submission, each bio will take an IO
517 * reference to the ioend to ensure that the ioend completion is only done once
518 * all bios have been submitted and the ioend is really done.
520 * If @fail is non-zero, it means that we have a situation where some part of
521 * the submission process has failed after we have marked paged for writeback
522 * and unlocked them. In this situation, we need to fail the bio and ioend
523 * rather than submit it to IO. This typically only happens on a filesystem
528 struct writeback_control *wbc,
529 struct xfs_ioend *ioend,
532 /* Convert CoW extents to regular */
533 if (!status && ioend->io_type == XFS_IO_COW) {
535 * Yuk. This can do memory allocation, but is not a
536 * transactional operation so everything is done in GFP_KERNEL
537 * context. That can deadlock, because we hold pages in
538 * writeback state and GFP_KERNEL allocations can block on them.
539 * Hence we must operate in nofs conditions here.
543 nofs_flag = memalloc_nofs_save();
544 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
545 ioend->io_offset, ioend->io_size);
546 memalloc_nofs_restore(nofs_flag);
549 /* Reserve log space if we might write beyond the on-disk inode size. */
551 ioend->io_type != XFS_IO_UNWRITTEN &&
552 xfs_ioend_is_append(ioend) &&
553 !ioend->io_append_trans)
554 status = xfs_setfilesize_trans_alloc(ioend);
556 ioend->io_bio->bi_private = ioend;
557 ioend->io_bio->bi_end_io = xfs_end_bio;
558 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
561 * If we are failing the IO now, just mark the ioend with an
562 * error and finish it. This will run IO completion immediately
563 * as there is only one reference to the ioend at this point in
567 ioend->io_bio->bi_status = errno_to_blk_status(status);
568 bio_endio(ioend->io_bio);
572 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
573 submit_bio(ioend->io_bio);
578 xfs_init_bio_from_bh(
580 struct buffer_head *bh)
582 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
583 bio_set_dev(bio, bh->b_bdev);
586 static struct xfs_ioend *
591 struct buffer_head *bh)
593 struct xfs_ioend *ioend;
596 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset);
597 xfs_init_bio_from_bh(bio, bh);
599 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
600 INIT_LIST_HEAD(&ioend->io_list);
601 ioend->io_type = type;
602 ioend->io_inode = inode;
604 ioend->io_offset = offset;
605 INIT_WORK(&ioend->io_work, xfs_end_io);
606 ioend->io_append_trans = NULL;
612 * Allocate a new bio, and chain the old bio to the new one.
614 * Note that we have to do perform the chaining in this unintuitive order
615 * so that the bi_private linkage is set up in the right direction for the
616 * traversal in xfs_destroy_ioend().
620 struct xfs_ioend *ioend,
621 struct writeback_control *wbc,
622 struct buffer_head *bh)
626 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
627 xfs_init_bio_from_bh(new, bh);
629 bio_chain(ioend->io_bio, new);
630 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
631 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
632 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
633 submit_bio(ioend->io_bio);
638 * Test to see if we've been building up a completion structure for
639 * earlier buffers -- if so, we try to append to this ioend if we
640 * can, otherwise we finish off any current ioend and start another.
641 * Return the ioend we finished off so that the caller can submit it
642 * once it has finished processing the dirty page.
647 struct buffer_head *bh,
649 struct xfs_writepage_ctx *wpc,
650 struct writeback_control *wbc,
651 struct list_head *iolist)
653 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
654 bh->b_blocknr != wpc->last_block + 1 ||
655 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
657 list_add(&wpc->ioend->io_list, iolist);
658 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
662 * If the buffer doesn't fit into the bio we need to allocate a new
663 * one. This shouldn't happen more than once for a given buffer.
665 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
666 xfs_chain_bio(wpc->ioend, wbc, bh);
668 wpc->ioend->io_size += bh->b_size;
669 wpc->last_block = bh->b_blocknr;
670 xfs_start_buffer_writeback(bh);
676 struct buffer_head *bh,
677 struct xfs_bmbt_irec *imap,
681 struct xfs_mount *m = XFS_I(inode)->i_mount;
682 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
683 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
685 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
686 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
688 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
689 ((offset - iomap_offset) >> inode->i_blkbits);
691 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
694 set_buffer_mapped(bh);
700 struct buffer_head *bh,
701 struct xfs_bmbt_irec *imap,
704 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
705 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
707 xfs_map_buffer(inode, bh, imap, offset);
708 set_buffer_mapped(bh);
709 clear_buffer_delay(bh);
710 clear_buffer_unwritten(bh);
714 * Test if a given page contains at least one buffer of a given @type.
715 * If @check_all_buffers is true, then we walk all the buffers in the page to
716 * try to find one of the type passed in. If it is not set, then the caller only
717 * needs to check the first buffer on the page for a match.
723 bool check_all_buffers)
725 struct buffer_head *bh;
726 struct buffer_head *head;
728 if (PageWriteback(page))
732 if (!page_has_buffers(page))
735 bh = head = page_buffers(page);
737 if (buffer_unwritten(bh)) {
738 if (type == XFS_IO_UNWRITTEN)
740 } else if (buffer_delay(bh)) {
741 if (type == XFS_IO_DELALLOC)
743 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
744 if (type == XFS_IO_OVERWRITE)
748 /* If we are only checking the first buffer, we are done now. */
749 if (!check_all_buffers)
751 } while ((bh = bh->b_this_page) != head);
757 xfs_vm_invalidatepage(
762 trace_xfs_invalidatepage(page->mapping->host, page, offset,
766 * If we are invalidating the entire page, clear the dirty state from it
767 * so that we can check for attempts to release dirty cached pages in
768 * xfs_vm_releasepage().
770 if (offset == 0 && length >= PAGE_SIZE)
771 cancel_dirty_page(page);
772 block_invalidatepage(page, offset, length);
776 * If the page has delalloc buffers on it, we need to punch them out before we
777 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
778 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
779 * is done on that same region - the delalloc extent is returned when none is
780 * supposed to be there.
782 * We prevent this by truncating away the delalloc regions on the page before
783 * invalidating it. Because they are delalloc, we can do this without needing a
784 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
785 * truncation without a transaction as there is no space left for block
786 * reservation (typically why we see a ENOSPC in writeback).
788 * This is not a performance critical path, so for now just do the punching a
789 * buffer head at a time.
792 xfs_aops_discard_page(
795 struct inode *inode = page->mapping->host;
796 struct xfs_inode *ip = XFS_I(inode);
797 struct buffer_head *bh, *head;
798 loff_t offset = page_offset(page);
800 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
803 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
806 xfs_alert(ip->i_mount,
807 "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.",
808 page, ip->i_ino, offset);
810 xfs_ilock(ip, XFS_ILOCK_EXCL);
811 bh = head = page_buffers(page);
814 xfs_fileoff_t start_fsb;
816 if (!buffer_delay(bh))
819 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
820 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
822 /* something screwed, just bail */
823 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
824 xfs_alert(ip->i_mount,
825 "page discard unable to remove delalloc mapping.");
830 offset += i_blocksize(inode);
832 } while ((bh = bh->b_this_page) != head);
834 xfs_iunlock(ip, XFS_ILOCK_EXCL);
836 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
842 struct xfs_writepage_ctx *wpc,
845 unsigned int *new_type)
847 struct xfs_inode *ip = XFS_I(inode);
848 struct xfs_bmbt_irec imap;
853 * If we already have a valid COW mapping keep using it.
855 if (wpc->io_type == XFS_IO_COW) {
856 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
857 if (wpc->imap_valid) {
858 *new_type = XFS_IO_COW;
864 * Else we need to check if there is a COW mapping at this offset.
866 xfs_ilock(ip, XFS_ILOCK_SHARED);
867 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
868 xfs_iunlock(ip, XFS_ILOCK_SHARED);
874 * And if the COW mapping has a delayed extent here we need to
875 * allocate real space for it now.
877 if (isnullstartblock(imap.br_startblock)) {
878 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
884 wpc->io_type = *new_type = XFS_IO_COW;
885 wpc->imap_valid = true;
891 * We implement an immediate ioend submission policy here to avoid needing to
892 * chain multiple ioends and hence nest mempool allocations which can violate
893 * forward progress guarantees we need to provide. The current ioend we are
894 * adding buffers to is cached on the writepage context, and if the new buffer
895 * does not append to the cached ioend it will create a new ioend and cache that
898 * If a new ioend is created and cached, the old ioend is returned and queued
899 * locally for submission once the entire page is processed or an error has been
900 * detected. While ioends are submitted immediately after they are completed,
901 * batching optimisations are provided by higher level block plugging.
903 * At the end of a writeback pass, there will be a cached ioend remaining on the
904 * writepage context that the caller will need to submit.
908 struct xfs_writepage_ctx *wpc,
909 struct writeback_control *wbc,
914 LIST_HEAD(submit_list);
915 struct xfs_ioend *ioend, *next;
916 struct buffer_head *bh, *head;
917 ssize_t len = i_blocksize(inode);
922 unsigned int new_type;
924 bh = head = page_buffers(page);
925 offset = page_offset(page);
927 if (offset >= end_offset)
929 if (!buffer_uptodate(bh))
933 * set_page_dirty dirties all buffers in a page, independent
934 * of their state. The dirty state however is entirely
935 * meaningless for holes (!mapped && uptodate), so skip
936 * buffers covering holes here.
938 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
939 wpc->imap_valid = false;
943 if (buffer_unwritten(bh))
944 new_type = XFS_IO_UNWRITTEN;
945 else if (buffer_delay(bh))
946 new_type = XFS_IO_DELALLOC;
947 else if (buffer_uptodate(bh))
948 new_type = XFS_IO_OVERWRITE;
950 if (PageUptodate(page))
951 ASSERT(buffer_mapped(bh));
953 * This buffer is not uptodate and will not be
954 * written to disk. Ensure that we will put any
955 * subsequent writeable buffers into a new
958 wpc->imap_valid = false;
962 if (xfs_is_reflink_inode(XFS_I(inode))) {
963 error = xfs_map_cow(wpc, inode, offset, &new_type);
968 if (wpc->io_type != new_type) {
969 wpc->io_type = new_type;
970 wpc->imap_valid = false;
974 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
976 if (!wpc->imap_valid) {
977 error = xfs_map_blocks(inode, offset, &wpc->imap,
981 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
984 if (wpc->imap_valid) {
986 if (wpc->io_type != XFS_IO_OVERWRITE)
987 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
988 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
992 } while (offset += len, ((bh = bh->b_this_page) != head));
994 if (uptodate && bh == head)
995 SetPageUptodate(page);
997 ASSERT(wpc->ioend || list_empty(&submit_list));
1001 * On error, we have to fail the ioend here because we have locked
1002 * buffers in the ioend. If we don't do this, we'll deadlock
1003 * invalidating the page as that tries to lock the buffers on the page.
1004 * Also, because we may have set pages under writeback, we have to make
1005 * sure we run IO completion to mark the error state of the IO
1006 * appropriately, so we can't cancel the ioend directly here. That means
1007 * we have to mark this page as under writeback if we included any
1008 * buffers from it in the ioend chain so that completion treats it
1011 * If we didn't include the page in the ioend, the on error we can
1012 * simply discard and unlock it as there are no other users of the page
1013 * or it's buffers right now. The caller will still need to trigger
1014 * submission of outstanding ioends on the writepage context so they are
1015 * treated correctly on error.
1018 xfs_start_page_writeback(page, !error);
1021 * Preserve the original error if there was one, otherwise catch
1022 * submission errors here and propagate into subsequent ioend
1025 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
1028 list_del_init(&ioend->io_list);
1029 error2 = xfs_submit_ioend(wbc, ioend, error);
1030 if (error2 && !error)
1034 xfs_aops_discard_page(page);
1035 ClearPageUptodate(page);
1039 * We can end up here with no error and nothing to write if we
1040 * race with a partial page truncate on a sub-page block sized
1041 * filesystem. In that case we need to mark the page clean.
1043 xfs_start_page_writeback(page, 1);
1044 end_page_writeback(page);
1047 mapping_set_error(page->mapping, error);
1052 * Write out a dirty page.
1054 * For delalloc space on the page we need to allocate space and flush it.
1055 * For unwritten space on the page we need to start the conversion to
1056 * regular allocated space.
1057 * For any other dirty buffer heads on the page we should flush them.
1062 struct writeback_control *wbc,
1065 struct xfs_writepage_ctx *wpc = data;
1066 struct inode *inode = page->mapping->host;
1068 uint64_t end_offset;
1071 trace_xfs_writepage(inode, page, 0, 0);
1073 ASSERT(page_has_buffers(page));
1076 * Refuse to write the page out if we are called from reclaim context.
1078 * This avoids stack overflows when called from deeply used stacks in
1079 * random callers for direct reclaim or memcg reclaim. We explicitly
1080 * allow reclaim from kswapd as the stack usage there is relatively low.
1082 * This should never happen except in the case of a VM regression so
1085 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1090 * Given that we do not allow direct reclaim to call us, we should
1091 * never be called while in a filesystem transaction.
1093 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
1097 * Is this page beyond the end of the file?
1099 * The page index is less than the end_index, adjust the end_offset
1100 * to the highest offset that this page should represent.
1101 * -----------------------------------------------------
1102 * | file mapping | <EOF> |
1103 * -----------------------------------------------------
1104 * | Page ... | Page N-2 | Page N-1 | Page N | |
1105 * ^--------------------------------^----------|--------
1106 * | desired writeback range | see else |
1107 * ---------------------------------^------------------|
1109 offset = i_size_read(inode);
1110 end_index = offset >> PAGE_SHIFT;
1111 if (page->index < end_index)
1112 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1115 * Check whether the page to write out is beyond or straddles
1117 * -------------------------------------------------------
1118 * | file mapping | <EOF> |
1119 * -------------------------------------------------------
1120 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1121 * ^--------------------------------^-----------|---------
1123 * ---------------------------------^-----------|--------|
1125 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1128 * Skip the page if it is fully outside i_size, e.g. due to a
1129 * truncate operation that is in progress. We must redirty the
1130 * page so that reclaim stops reclaiming it. Otherwise
1131 * xfs_vm_releasepage() is called on it and gets confused.
1133 * Note that the end_index is unsigned long, it would overflow
1134 * if the given offset is greater than 16TB on 32-bit system
1135 * and if we do check the page is fully outside i_size or not
1136 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1137 * will be evaluated to 0. Hence this page will be redirtied
1138 * and be written out repeatedly which would result in an
1139 * infinite loop, the user program that perform this operation
1140 * will hang. Instead, we can verify this situation by checking
1141 * if the page to write is totally beyond the i_size or if it's
1142 * offset is just equal to the EOF.
1144 if (page->index > end_index ||
1145 (page->index == end_index && offset_into_page == 0))
1149 * The page straddles i_size. It must be zeroed out on each
1150 * and every writepage invocation because it may be mmapped.
1151 * "A file is mapped in multiples of the page size. For a file
1152 * that is not a multiple of the page size, the remaining
1153 * memory is zeroed when mapped, and writes to that region are
1154 * not written out to the file."
1156 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1158 /* Adjust the end_offset to the end of file */
1159 end_offset = offset;
1162 return xfs_writepage_map(wpc, wbc, inode, page, end_offset);
1165 redirty_page_for_writepage(wbc, page);
1173 struct writeback_control *wbc)
1175 struct xfs_writepage_ctx wpc = {
1176 .io_type = XFS_IO_INVALID,
1180 ret = xfs_do_writepage(page, wbc, &wpc);
1182 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1188 struct address_space *mapping,
1189 struct writeback_control *wbc)
1191 struct xfs_writepage_ctx wpc = {
1192 .io_type = XFS_IO_INVALID,
1196 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1197 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1199 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1205 struct address_space *mapping,
1206 struct writeback_control *wbc)
1208 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1209 return dax_writeback_mapping_range(mapping,
1210 xfs_find_bdev_for_inode(mapping->host), wbc);
1214 * Called to move a page into cleanable state - and from there
1215 * to be released. The page should already be clean. We always
1216 * have buffer heads in this call.
1218 * Returns 1 if the page is ok to release, 0 otherwise.
1225 int delalloc, unwritten;
1227 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1230 * mm accommodates an old ext3 case where clean pages might not have had
1231 * the dirty bit cleared. Thus, it can send actual dirty pages to
1232 * ->releasepage() via shrink_active_list(). Conversely,
1233 * block_invalidatepage() can send pages that are still marked dirty but
1234 * otherwise have invalidated buffers.
1236 * We want to release the latter to avoid unnecessary buildup of the
1237 * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages
1238 * that are entirely invalidated and need to be released. Hence the
1239 * only time we should get dirty pages here is through
1240 * shrink_active_list() and so we can simply skip those now.
1242 * warn if we've left any lingering delalloc/unwritten buffers on clean
1243 * or invalidated pages we are about to release.
1245 if (PageDirty(page))
1248 xfs_count_page_state(page, &delalloc, &unwritten);
1250 if (WARN_ON_ONCE(delalloc))
1252 if (WARN_ON_ONCE(unwritten))
1255 return try_to_free_buffers(page);
1259 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1260 * is, so that we can avoid repeated get_blocks calls.
1262 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1263 * for blocks beyond EOF must be marked new so that sub block regions can be
1264 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1265 * was just allocated or is unwritten, otherwise the callers would overwrite
1266 * existing data with zeros. Hence we have to split the mapping into a range up
1267 * to and including EOF, and a second mapping for beyond EOF.
1271 struct inode *inode,
1273 struct buffer_head *bh_result,
1274 struct xfs_bmbt_irec *imap,
1278 xfs_off_t mapping_size;
1280 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1281 mapping_size <<= inode->i_blkbits;
1283 ASSERT(mapping_size > 0);
1284 if (mapping_size > size)
1285 mapping_size = size;
1286 if (offset < i_size_read(inode) &&
1287 (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) {
1288 /* limit mapping to block that spans EOF */
1289 mapping_size = roundup_64(i_size_read(inode) - offset,
1290 i_blocksize(inode));
1292 if (mapping_size > LONG_MAX)
1293 mapping_size = LONG_MAX;
1295 bh_result->b_size = mapping_size;
1300 struct inode *inode,
1302 struct buffer_head *bh_result,
1305 struct xfs_inode *ip = XFS_I(inode);
1306 struct xfs_mount *mp = ip->i_mount;
1307 xfs_fileoff_t offset_fsb, end_fsb;
1310 struct xfs_bmbt_irec imap;
1317 if (XFS_FORCED_SHUTDOWN(mp))
1320 offset = (xfs_off_t)iblock << inode->i_blkbits;
1321 ASSERT(bh_result->b_size >= i_blocksize(inode));
1322 size = bh_result->b_size;
1324 if (offset >= i_size_read(inode))
1328 * Direct I/O is usually done on preallocated files, so try getting
1329 * a block mapping without an exclusive lock first.
1331 lockmode = xfs_ilock_data_map_shared(ip);
1333 ASSERT(offset <= mp->m_super->s_maxbytes);
1334 if (offset > mp->m_super->s_maxbytes - size)
1335 size = mp->m_super->s_maxbytes - offset;
1336 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1337 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1339 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1344 trace_xfs_get_blocks_notfound(ip, offset, size);
1348 trace_xfs_get_blocks_found(ip, offset, size,
1349 imap.br_state == XFS_EXT_UNWRITTEN ?
1350 XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap);
1351 xfs_iunlock(ip, lockmode);
1353 /* trim mapping down to size requested */
1354 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1357 * For unwritten extents do not report a disk address in the buffered
1358 * read case (treat as if we're reading into a hole).
1360 if (xfs_bmap_is_real_extent(&imap))
1361 xfs_map_buffer(inode, bh_result, &imap, offset);
1364 * If this is a realtime file, data may be on a different device.
1365 * to that pointed to from the buffer_head b_bdev currently.
1367 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1371 xfs_iunlock(ip, lockmode);
1377 struct address_space *mapping,
1380 struct xfs_inode *ip = XFS_I(mapping->host);
1382 trace_xfs_vm_bmap(ip);
1385 * The swap code (ab-)uses ->bmap to get a block mapping and then
1386 * bypasses the file system for actual I/O. We really can't allow
1387 * that on reflinks inodes, so we have to skip out here. And yes,
1388 * 0 is the magic code for a bmap error.
1390 * Since we don't pass back blockdev info, we can't return bmap
1391 * information for rt files either.
1393 if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip))
1395 return iomap_bmap(mapping, block, &xfs_iomap_ops);
1400 struct file *unused,
1403 trace_xfs_vm_readpage(page->mapping->host, 1);
1404 return mpage_readpage(page, xfs_get_blocks);
1409 struct file *unused,
1410 struct address_space *mapping,
1411 struct list_head *pages,
1414 trace_xfs_vm_readpages(mapping->host, nr_pages);
1415 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1419 * This is basically a copy of __set_page_dirty_buffers() with one
1420 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1421 * dirty, we'll never be able to clean them because we don't write buffers
1422 * beyond EOF, and that means we can't invalidate pages that span EOF
1423 * that have been marked dirty. Further, the dirty state can leak into
1424 * the file interior if the file is extended, resulting in all sorts of
1425 * bad things happening as the state does not match the underlying data.
1427 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1428 * this only exist because of bufferheads and how the generic code manages them.
1431 xfs_vm_set_page_dirty(
1434 struct address_space *mapping = page->mapping;
1435 struct inode *inode = mapping->host;
1440 if (unlikely(!mapping))
1441 return !TestSetPageDirty(page);
1443 end_offset = i_size_read(inode);
1444 offset = page_offset(page);
1446 spin_lock(&mapping->private_lock);
1447 if (page_has_buffers(page)) {
1448 struct buffer_head *head = page_buffers(page);
1449 struct buffer_head *bh = head;
1452 if (offset < end_offset)
1453 set_buffer_dirty(bh);
1454 bh = bh->b_this_page;
1455 offset += i_blocksize(inode);
1456 } while (bh != head);
1459 * Lock out page->mem_cgroup migration to keep PageDirty
1460 * synchronized with per-memcg dirty page counters.
1462 lock_page_memcg(page);
1463 newly_dirty = !TestSetPageDirty(page);
1464 spin_unlock(&mapping->private_lock);
1467 __set_page_dirty(page, mapping, 1);
1468 unlock_page_memcg(page);
1470 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1475 xfs_iomap_swapfile_activate(
1476 struct swap_info_struct *sis,
1477 struct file *swap_file,
1480 sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
1481 return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops);
1484 const struct address_space_operations xfs_address_space_operations = {
1485 .readpage = xfs_vm_readpage,
1486 .readpages = xfs_vm_readpages,
1487 .writepage = xfs_vm_writepage,
1488 .writepages = xfs_vm_writepages,
1489 .set_page_dirty = xfs_vm_set_page_dirty,
1490 .releasepage = xfs_vm_releasepage,
1491 .invalidatepage = xfs_vm_invalidatepage,
1492 .bmap = xfs_vm_bmap,
1493 .direct_IO = noop_direct_IO,
1494 .migratepage = buffer_migrate_page,
1495 .is_partially_uptodate = block_is_partially_uptodate,
1496 .error_remove_page = generic_error_remove_page,
1497 .swap_activate = xfs_iomap_swapfile_activate,
1500 const struct address_space_operations xfs_dax_aops = {
1501 .writepages = xfs_dax_writepages,
1502 .direct_IO = noop_direct_IO,
1503 .set_page_dirty = noop_set_page_dirty,
1504 .invalidatepage = noop_invalidatepage,
1505 .swap_activate = xfs_iomap_swapfile_activate,