2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
32 #include "xfs_iomap.h"
33 #include "xfs_vnodeops.h"
34 #include "xfs_trace.h"
36 #include <linux/gfp.h>
37 #include <linux/mpage.h>
38 #include <linux/pagevec.h>
39 #include <linux/writeback.h>
47 struct buffer_head *bh, *head;
49 *delalloc = *unwritten = 0;
51 bh = head = page_buffers(page);
53 if (buffer_unwritten(bh))
55 else if (buffer_delay(bh))
57 } while ((bh = bh->b_this_page) != head);
60 STATIC struct block_device *
61 xfs_find_bdev_for_inode(
64 struct xfs_inode *ip = XFS_I(inode);
65 struct xfs_mount *mp = ip->i_mount;
67 if (XFS_IS_REALTIME_INODE(ip))
68 return mp->m_rtdev_targp->bt_bdev;
70 return mp->m_ddev_targp->bt_bdev;
74 * We're now finished for good with this ioend structure.
75 * Update the page state via the associated buffer_heads,
76 * release holds on the inode and bio, and finally free
77 * up memory. Do not use the ioend after this.
83 struct buffer_head *bh, *next;
85 for (bh = ioend->io_buffer_head; bh; bh = next) {
87 bh->b_end_io(bh, !ioend->io_error);
91 if (ioend->io_isasync) {
92 aio_complete(ioend->io_iocb, ioend->io_error ?
93 ioend->io_error : ioend->io_result, 0);
95 inode_dio_done(ioend->io_inode);
98 mempool_free(ioend, xfs_ioend_pool);
102 * If the end of the current ioend is beyond the current EOF,
103 * return the new EOF value, otherwise zero.
109 xfs_inode_t *ip = XFS_I(ioend->io_inode);
113 bsize = ioend->io_offset + ioend->io_size;
114 isize = MAX(ip->i_size, ip->i_new_size);
115 isize = MIN(isize, bsize);
116 return isize > ip->i_d.di_size ? isize : 0;
120 * Fast and loose check if this write could update the on-disk inode size.
122 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
124 return ioend->io_offset + ioend->io_size >
125 XFS_I(ioend->io_inode)->i_d.di_size;
129 * Update on-disk file size now that data has been written to disk. The
130 * current in-memory file size is i_size. If a write is beyond eof i_new_size
131 * will be the intended file size until i_size is updated. If this write does
132 * not extend all the way to the valid file size then restrict this update to
133 * the end of the write.
135 * This function does not block as blocking on the inode lock in IO completion
136 * can lead to IO completion order dependency deadlocks.. If it can't get the
137 * inode ilock it will return EAGAIN. Callers must handle this.
143 xfs_inode_t *ip = XFS_I(ioend->io_inode);
146 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
149 isize = xfs_ioend_new_eof(ioend);
151 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
152 ip->i_d.di_size = isize;
153 xfs_mark_inode_dirty(ip);
156 xfs_iunlock(ip, XFS_ILOCK_EXCL);
161 * Schedule IO completion handling on the final put of an ioend.
163 * If there is no work to do we might as well call it a day and free the
168 struct xfs_ioend *ioend)
170 if (atomic_dec_and_test(&ioend->io_remaining)) {
171 if (ioend->io_type == IO_UNWRITTEN)
172 queue_work(xfsconvertd_workqueue, &ioend->io_work);
173 else if (xfs_ioend_is_append(ioend))
174 queue_work(xfsdatad_workqueue, &ioend->io_work);
176 xfs_destroy_ioend(ioend);
181 * IO write completion.
185 struct work_struct *work)
187 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
188 struct xfs_inode *ip = XFS_I(ioend->io_inode);
191 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
192 ioend->io_error = -EIO;
199 * For unwritten extents we need to issue transactions to convert a
200 * range to normal written extens after the data I/O has finished.
202 if (ioend->io_type == IO_UNWRITTEN) {
203 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
206 ioend->io_error = -error;
212 * We might have to update the on-disk file size after extending
215 error = xfs_setfilesize(ioend);
216 ASSERT(!error || error == EAGAIN);
220 * If we didn't complete processing of the ioend, requeue it to the
221 * tail of the workqueue for another attempt later. Otherwise destroy
224 if (error == EAGAIN) {
225 atomic_inc(&ioend->io_remaining);
226 xfs_finish_ioend(ioend);
227 /* ensure we don't spin on blocked ioends */
230 xfs_destroy_ioend(ioend);
235 * Call IO completion handling in caller context on the final put of an ioend.
238 xfs_finish_ioend_sync(
239 struct xfs_ioend *ioend)
241 if (atomic_dec_and_test(&ioend->io_remaining))
242 xfs_end_io(&ioend->io_work);
246 * Allocate and initialise an IO completion structure.
247 * We need to track unwritten extent write completion here initially.
248 * We'll need to extend this for updating the ondisk inode size later
258 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
261 * Set the count to 1 initially, which will prevent an I/O
262 * completion callback from happening before we have started
263 * all the I/O from calling the completion routine too early.
265 atomic_set(&ioend->io_remaining, 1);
266 ioend->io_isasync = 0;
268 ioend->io_list = NULL;
269 ioend->io_type = type;
270 ioend->io_inode = inode;
271 ioend->io_buffer_head = NULL;
272 ioend->io_buffer_tail = NULL;
273 ioend->io_offset = 0;
275 ioend->io_iocb = NULL;
276 ioend->io_result = 0;
278 INIT_WORK(&ioend->io_work, xfs_end_io);
286 struct xfs_bmbt_irec *imap,
290 struct xfs_inode *ip = XFS_I(inode);
291 struct xfs_mount *mp = ip->i_mount;
292 ssize_t count = 1 << inode->i_blkbits;
293 xfs_fileoff_t offset_fsb, end_fsb;
295 int bmapi_flags = XFS_BMAPI_ENTIRE;
298 if (XFS_FORCED_SHUTDOWN(mp))
299 return -XFS_ERROR(EIO);
301 if (type == IO_UNWRITTEN)
302 bmapi_flags |= XFS_BMAPI_IGSTATE;
304 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
306 return -XFS_ERROR(EAGAIN);
307 xfs_ilock(ip, XFS_ILOCK_SHARED);
310 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
311 (ip->i_df.if_flags & XFS_IFEXTENTS));
312 ASSERT(offset <= mp->m_maxioffset);
314 if (offset + count > mp->m_maxioffset)
315 count = mp->m_maxioffset - offset;
316 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
317 offset_fsb = XFS_B_TO_FSBT(mp, offset);
318 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
319 imap, &nimaps, bmapi_flags);
320 xfs_iunlock(ip, XFS_ILOCK_SHARED);
323 return -XFS_ERROR(error);
325 if (type == IO_DELALLOC &&
326 (!nimaps || isnullstartblock(imap->br_startblock))) {
327 error = xfs_iomap_write_allocate(ip, offset, count, imap);
329 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
330 return -XFS_ERROR(error);
334 if (type == IO_UNWRITTEN) {
336 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
337 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
341 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
348 struct xfs_bmbt_irec *imap,
351 offset >>= inode->i_blkbits;
353 return offset >= imap->br_startoff &&
354 offset < imap->br_startoff + imap->br_blockcount;
358 * BIO completion handler for buffered IO.
365 xfs_ioend_t *ioend = bio->bi_private;
367 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
368 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
370 /* Toss bio and pass work off to an xfsdatad thread */
371 bio->bi_private = NULL;
372 bio->bi_end_io = NULL;
375 xfs_finish_ioend(ioend);
379 xfs_submit_ioend_bio(
380 struct writeback_control *wbc,
384 atomic_inc(&ioend->io_remaining);
385 bio->bi_private = ioend;
386 bio->bi_end_io = xfs_end_bio;
389 * If the I/O is beyond EOF we mark the inode dirty immediately
390 * but don't update the inode size until I/O completion.
392 if (xfs_ioend_new_eof(ioend))
393 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
395 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
400 struct buffer_head *bh)
402 int nvecs = bio_get_nr_vecs(bh->b_bdev);
403 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
405 ASSERT(bio->bi_private == NULL);
406 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
407 bio->bi_bdev = bh->b_bdev;
412 xfs_start_buffer_writeback(
413 struct buffer_head *bh)
415 ASSERT(buffer_mapped(bh));
416 ASSERT(buffer_locked(bh));
417 ASSERT(!buffer_delay(bh));
418 ASSERT(!buffer_unwritten(bh));
420 mark_buffer_async_write(bh);
421 set_buffer_uptodate(bh);
422 clear_buffer_dirty(bh);
426 xfs_start_page_writeback(
431 ASSERT(PageLocked(page));
432 ASSERT(!PageWriteback(page));
434 clear_page_dirty_for_io(page);
435 set_page_writeback(page);
437 /* If no buffers on the page are to be written, finish it here */
439 end_page_writeback(page);
442 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
444 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
448 * Submit all of the bios for all of the ioends we have saved up, covering the
449 * initial writepage page and also any probed pages.
451 * Because we may have multiple ioends spanning a page, we need to start
452 * writeback on all the buffers before we submit them for I/O. If we mark the
453 * buffers as we got, then we can end up with a page that only has buffers
454 * marked async write and I/O complete on can occur before we mark the other
455 * buffers async write.
457 * The end result of this is that we trip a bug in end_page_writeback() because
458 * we call it twice for the one page as the code in end_buffer_async_write()
459 * assumes that all buffers on the page are started at the same time.
461 * The fix is two passes across the ioend list - one to start writeback on the
462 * buffer_heads, and then submit them for I/O on the second pass.
466 struct writeback_control *wbc,
469 xfs_ioend_t *head = ioend;
471 struct buffer_head *bh;
473 sector_t lastblock = 0;
475 /* Pass 1 - start writeback */
477 next = ioend->io_list;
478 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
479 xfs_start_buffer_writeback(bh);
480 } while ((ioend = next) != NULL);
482 /* Pass 2 - submit I/O */
485 next = ioend->io_list;
488 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
492 bio = xfs_alloc_ioend_bio(bh);
493 } else if (bh->b_blocknr != lastblock + 1) {
494 xfs_submit_ioend_bio(wbc, ioend, bio);
498 if (bio_add_buffer(bio, bh) != bh->b_size) {
499 xfs_submit_ioend_bio(wbc, ioend, bio);
503 lastblock = bh->b_blocknr;
506 xfs_submit_ioend_bio(wbc, ioend, bio);
507 xfs_finish_ioend(ioend);
508 } while ((ioend = next) != NULL);
512 * Cancel submission of all buffer_heads so far in this endio.
513 * Toss the endio too. Only ever called for the initial page
514 * in a writepage request, so only ever one page.
521 struct buffer_head *bh, *next_bh;
524 next = ioend->io_list;
525 bh = ioend->io_buffer_head;
527 next_bh = bh->b_private;
528 clear_buffer_async_write(bh);
530 } while ((bh = next_bh) != NULL);
532 mempool_free(ioend, xfs_ioend_pool);
533 } while ((ioend = next) != NULL);
537 * Test to see if we've been building up a completion structure for
538 * earlier buffers -- if so, we try to append to this ioend if we
539 * can, otherwise we finish off any current ioend and start another.
540 * Return true if we've finished the given ioend.
545 struct buffer_head *bh,
548 xfs_ioend_t **result,
551 xfs_ioend_t *ioend = *result;
553 if (!ioend || need_ioend || type != ioend->io_type) {
554 xfs_ioend_t *previous = *result;
556 ioend = xfs_alloc_ioend(inode, type);
557 ioend->io_offset = offset;
558 ioend->io_buffer_head = bh;
559 ioend->io_buffer_tail = bh;
561 previous->io_list = ioend;
564 ioend->io_buffer_tail->b_private = bh;
565 ioend->io_buffer_tail = bh;
568 bh->b_private = NULL;
569 ioend->io_size += bh->b_size;
575 struct buffer_head *bh,
576 struct xfs_bmbt_irec *imap,
580 struct xfs_mount *m = XFS_I(inode)->i_mount;
581 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
582 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
584 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
585 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
587 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
588 ((offset - iomap_offset) >> inode->i_blkbits);
590 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
593 set_buffer_mapped(bh);
599 struct buffer_head *bh,
600 struct xfs_bmbt_irec *imap,
603 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
604 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
606 xfs_map_buffer(inode, bh, imap, offset);
607 set_buffer_mapped(bh);
608 clear_buffer_delay(bh);
609 clear_buffer_unwritten(bh);
613 * Test if a given page is suitable for writing as part of an unwritten
614 * or delayed allocate extent.
621 if (PageWriteback(page))
624 if (page->mapping && page_has_buffers(page)) {
625 struct buffer_head *bh, *head;
628 bh = head = page_buffers(page);
630 if (buffer_unwritten(bh))
631 acceptable = (type == IO_UNWRITTEN);
632 else if (buffer_delay(bh))
633 acceptable = (type == IO_DELALLOC);
634 else if (buffer_dirty(bh) && buffer_mapped(bh))
635 acceptable = (type == IO_OVERWRITE);
638 } while ((bh = bh->b_this_page) != head);
648 * Allocate & map buffers for page given the extent map. Write it out.
649 * except for the original page of a writepage, this is called on
650 * delalloc/unwritten pages only, for the original page it is possible
651 * that the page has no mapping at all.
658 struct xfs_bmbt_irec *imap,
659 xfs_ioend_t **ioendp,
660 struct writeback_control *wbc)
662 struct buffer_head *bh, *head;
663 xfs_off_t end_offset;
664 unsigned long p_offset;
667 int count = 0, done = 0, uptodate = 1;
668 xfs_off_t offset = page_offset(page);
670 if (page->index != tindex)
672 if (!trylock_page(page))
674 if (PageWriteback(page))
675 goto fail_unlock_page;
676 if (page->mapping != inode->i_mapping)
677 goto fail_unlock_page;
678 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
679 goto fail_unlock_page;
682 * page_dirty is initially a count of buffers on the page before
683 * EOF and is decremented as we move each into a cleanable state.
687 * End offset is the highest offset that this page should represent.
688 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
689 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
690 * hence give us the correct page_dirty count. On any other page,
691 * it will be zero and in that case we need page_dirty to be the
692 * count of buffers on the page.
694 end_offset = min_t(unsigned long long,
695 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
698 len = 1 << inode->i_blkbits;
699 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
701 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
702 page_dirty = p_offset / len;
704 bh = head = page_buffers(page);
706 if (offset >= end_offset)
708 if (!buffer_uptodate(bh))
710 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
715 if (buffer_unwritten(bh) || buffer_delay(bh) ||
717 if (buffer_unwritten(bh))
719 else if (buffer_delay(bh))
724 if (!xfs_imap_valid(inode, imap, offset)) {
730 if (type != IO_OVERWRITE)
731 xfs_map_at_offset(inode, bh, imap, offset);
732 xfs_add_to_ioend(inode, bh, offset, type,
740 } while (offset += len, (bh = bh->b_this_page) != head);
742 if (uptodate && bh == head)
743 SetPageUptodate(page);
746 if (--wbc->nr_to_write <= 0 &&
747 wbc->sync_mode == WB_SYNC_NONE)
750 xfs_start_page_writeback(page, !page_dirty, count);
760 * Convert & write out a cluster of pages in the same extent as defined
761 * by mp and following the start page.
767 struct xfs_bmbt_irec *imap,
768 xfs_ioend_t **ioendp,
769 struct writeback_control *wbc,
775 pagevec_init(&pvec, 0);
776 while (!done && tindex <= tlast) {
777 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
779 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
782 for (i = 0; i < pagevec_count(&pvec); i++) {
783 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
789 pagevec_release(&pvec);
795 xfs_vm_invalidatepage(
797 unsigned long offset)
799 trace_xfs_invalidatepage(page->mapping->host, page, offset);
800 block_invalidatepage(page, offset);
804 * If the page has delalloc buffers on it, we need to punch them out before we
805 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
806 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
807 * is done on that same region - the delalloc extent is returned when none is
808 * supposed to be there.
810 * We prevent this by truncating away the delalloc regions on the page before
811 * invalidating it. Because they are delalloc, we can do this without needing a
812 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
813 * truncation without a transaction as there is no space left for block
814 * reservation (typically why we see a ENOSPC in writeback).
816 * This is not a performance critical path, so for now just do the punching a
817 * buffer head at a time.
820 xfs_aops_discard_page(
823 struct inode *inode = page->mapping->host;
824 struct xfs_inode *ip = XFS_I(inode);
825 struct buffer_head *bh, *head;
826 loff_t offset = page_offset(page);
828 if (!xfs_is_delayed_page(page, IO_DELALLOC))
831 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
834 xfs_alert(ip->i_mount,
835 "page discard on page %p, inode 0x%llx, offset %llu.",
836 page, ip->i_ino, offset);
838 xfs_ilock(ip, XFS_ILOCK_EXCL);
839 bh = head = page_buffers(page);
842 xfs_fileoff_t start_fsb;
844 if (!buffer_delay(bh))
847 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
848 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
850 /* something screwed, just bail */
851 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
852 xfs_alert(ip->i_mount,
853 "page discard unable to remove delalloc mapping.");
858 offset += 1 << inode->i_blkbits;
860 } while ((bh = bh->b_this_page) != head);
862 xfs_iunlock(ip, XFS_ILOCK_EXCL);
864 xfs_vm_invalidatepage(page, 0);
869 * Write out a dirty page.
871 * For delalloc space on the page we need to allocate space and flush it.
872 * For unwritten space on the page we need to start the conversion to
873 * regular allocated space.
874 * For any other dirty buffer heads on the page we should flush them.
879 struct writeback_control *wbc)
881 struct inode *inode = page->mapping->host;
882 struct buffer_head *bh, *head;
883 struct xfs_bmbt_irec imap;
884 xfs_ioend_t *ioend = NULL, *iohead = NULL;
887 __uint64_t end_offset;
888 pgoff_t end_index, last_index;
890 int err, imap_valid = 0, uptodate = 1;
894 trace_xfs_writepage(inode, page, 0);
896 ASSERT(page_has_buffers(page));
899 * Refuse to write the page out if we are called from reclaim context.
901 * This avoids stack overflows when called from deeply used stacks in
902 * random callers for direct reclaim or memcg reclaim. We explicitly
903 * allow reclaim from kswapd as the stack usage there is relatively low.
905 * This should never happen except in the case of a VM regression so
908 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
913 * Given that we do not allow direct reclaim to call us, we should
914 * never be called while in a filesystem transaction.
916 if (WARN_ON(current->flags & PF_FSTRANS))
919 /* Is this page beyond the end of the file? */
920 offset = i_size_read(inode);
921 end_index = offset >> PAGE_CACHE_SHIFT;
922 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
923 if (page->index >= end_index) {
924 if ((page->index >= end_index + 1) ||
925 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
931 end_offset = min_t(unsigned long long,
932 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
934 len = 1 << inode->i_blkbits;
936 bh = head = page_buffers(page);
937 offset = page_offset(page);
940 if (wbc->sync_mode == WB_SYNC_NONE)
946 if (offset >= end_offset)
948 if (!buffer_uptodate(bh))
952 * set_page_dirty dirties all buffers in a page, independent
953 * of their state. The dirty state however is entirely
954 * meaningless for holes (!mapped && uptodate), so skip
955 * buffers covering holes here.
957 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
962 if (buffer_unwritten(bh)) {
963 if (type != IO_UNWRITTEN) {
967 } else if (buffer_delay(bh)) {
968 if (type != IO_DELALLOC) {
972 } else if (buffer_uptodate(bh)) {
973 if (type != IO_OVERWRITE) {
978 if (PageUptodate(page)) {
979 ASSERT(buffer_mapped(bh));
986 imap_valid = xfs_imap_valid(inode, &imap, offset);
989 * If we didn't have a valid mapping then we need to
990 * put the new mapping into a separate ioend structure.
991 * This ensures non-contiguous extents always have
992 * separate ioends, which is particularly important
993 * for unwritten extent conversion at I/O completion
997 err = xfs_map_blocks(inode, offset, &imap, type,
1001 imap_valid = xfs_imap_valid(inode, &imap, offset);
1005 if (type != IO_OVERWRITE)
1006 xfs_map_at_offset(inode, bh, &imap, offset);
1007 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1015 } while (offset += len, ((bh = bh->b_this_page) != head));
1017 if (uptodate && bh == head)
1018 SetPageUptodate(page);
1020 xfs_start_page_writeback(page, 1, count);
1022 if (ioend && imap_valid) {
1023 xfs_off_t end_index;
1025 end_index = imap.br_startoff + imap.br_blockcount;
1028 end_index <<= inode->i_blkbits;
1031 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1033 /* check against file size */
1034 if (end_index > last_index)
1035 end_index = last_index;
1037 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1042 xfs_submit_ioend(wbc, iohead);
1048 xfs_cancel_ioend(iohead);
1053 xfs_aops_discard_page(page);
1054 ClearPageUptodate(page);
1059 redirty_page_for_writepage(wbc, page);
1066 struct address_space *mapping,
1067 struct writeback_control *wbc)
1069 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1070 return generic_writepages(mapping, wbc);
1074 * Called to move a page into cleanable state - and from there
1075 * to be released. The page should already be clean. We always
1076 * have buffer heads in this call.
1078 * Returns 1 if the page is ok to release, 0 otherwise.
1085 int delalloc, unwritten;
1087 trace_xfs_releasepage(page->mapping->host, page, 0);
1089 xfs_count_page_state(page, &delalloc, &unwritten);
1091 if (WARN_ON(delalloc))
1093 if (WARN_ON(unwritten))
1096 return try_to_free_buffers(page);
1101 struct inode *inode,
1103 struct buffer_head *bh_result,
1107 struct xfs_inode *ip = XFS_I(inode);
1108 struct xfs_mount *mp = ip->i_mount;
1109 xfs_fileoff_t offset_fsb, end_fsb;
1112 struct xfs_bmbt_irec imap;
1118 if (XFS_FORCED_SHUTDOWN(mp))
1119 return -XFS_ERROR(EIO);
1121 offset = (xfs_off_t)iblock << inode->i_blkbits;
1122 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1123 size = bh_result->b_size;
1125 if (!create && direct && offset >= i_size_read(inode))
1129 lockmode = XFS_ILOCK_EXCL;
1130 xfs_ilock(ip, lockmode);
1132 lockmode = xfs_ilock_map_shared(ip);
1135 ASSERT(offset <= mp->m_maxioffset);
1136 if (offset + size > mp->m_maxioffset)
1137 size = mp->m_maxioffset - offset;
1138 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1139 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1141 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1142 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1148 (imap.br_startblock == HOLESTARTBLOCK ||
1149 imap.br_startblock == DELAYSTARTBLOCK))) {
1151 error = xfs_iomap_write_direct(ip, offset, size,
1154 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1159 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1160 } else if (nimaps) {
1161 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1163 trace_xfs_get_blocks_notfound(ip, offset, size);
1166 xfs_iunlock(ip, lockmode);
1168 if (imap.br_startblock != HOLESTARTBLOCK &&
1169 imap.br_startblock != DELAYSTARTBLOCK) {
1171 * For unwritten extents do not report a disk address on
1172 * the read case (treat as if we're reading into a hole).
1174 if (create || !ISUNWRITTEN(&imap))
1175 xfs_map_buffer(inode, bh_result, &imap, offset);
1176 if (create && ISUNWRITTEN(&imap)) {
1178 bh_result->b_private = inode;
1179 set_buffer_unwritten(bh_result);
1184 * If this is a realtime file, data may be on a different device.
1185 * to that pointed to from the buffer_head b_bdev currently.
1187 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1190 * If we previously allocated a block out beyond eof and we are now
1191 * coming back to use it then we will need to flag it as new even if it
1192 * has a disk address.
1194 * With sub-block writes into unwritten extents we also need to mark
1195 * the buffer as new so that the unwritten parts of the buffer gets
1199 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1200 (offset >= i_size_read(inode)) ||
1201 (new || ISUNWRITTEN(&imap))))
1202 set_buffer_new(bh_result);
1204 if (imap.br_startblock == DELAYSTARTBLOCK) {
1207 set_buffer_uptodate(bh_result);
1208 set_buffer_mapped(bh_result);
1209 set_buffer_delay(bh_result);
1214 * If this is O_DIRECT or the mpage code calling tell them how large
1215 * the mapping is, so that we can avoid repeated get_blocks calls.
1217 if (direct || size > (1 << inode->i_blkbits)) {
1218 xfs_off_t mapping_size;
1220 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1221 mapping_size <<= inode->i_blkbits;
1223 ASSERT(mapping_size > 0);
1224 if (mapping_size > size)
1225 mapping_size = size;
1226 if (mapping_size > LONG_MAX)
1227 mapping_size = LONG_MAX;
1229 bh_result->b_size = mapping_size;
1235 xfs_iunlock(ip, lockmode);
1241 struct inode *inode,
1243 struct buffer_head *bh_result,
1246 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1250 xfs_get_blocks_direct(
1251 struct inode *inode,
1253 struct buffer_head *bh_result,
1256 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1260 * Complete a direct I/O write request.
1262 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1263 * need to issue a transaction to convert the range from unwritten to written
1264 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1265 * to do this and we are done. But in case this was a successful AIO
1266 * request this handler is called from interrupt context, from which we
1267 * can't start transactions. In that case offload the I/O completion to
1268 * the workqueues we also use for buffered I/O completion.
1271 xfs_end_io_direct_write(
1279 struct xfs_ioend *ioend = iocb->private;
1282 * blockdev_direct_IO can return an error even after the I/O
1283 * completion handler was called. Thus we need to protect
1284 * against double-freeing.
1286 iocb->private = NULL;
1288 ioend->io_offset = offset;
1289 ioend->io_size = size;
1290 ioend->io_iocb = iocb;
1291 ioend->io_result = ret;
1292 if (private && size > 0)
1293 ioend->io_type = IO_UNWRITTEN;
1296 ioend->io_isasync = 1;
1297 xfs_finish_ioend(ioend);
1299 xfs_finish_ioend_sync(ioend);
1307 const struct iovec *iov,
1309 unsigned long nr_segs)
1311 struct inode *inode = iocb->ki_filp->f_mapping->host;
1312 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1316 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1318 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1320 xfs_get_blocks_direct,
1321 xfs_end_io_direct_write, NULL, 0);
1322 if (ret != -EIOCBQUEUED && iocb->private)
1323 xfs_destroy_ioend(iocb->private);
1325 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1327 xfs_get_blocks_direct,
1335 xfs_vm_write_failed(
1336 struct address_space *mapping,
1339 struct inode *inode = mapping->host;
1341 if (to > inode->i_size) {
1343 * punch out the delalloc blocks we have already allocated. We
1344 * don't call xfs_setattr() to do this as we may be in the
1345 * middle of a multi-iovec write and so the vfs inode->i_size
1346 * will not match the xfs ip->i_size and so it will zero too
1347 * much. Hence we jus truncate the page cache to zero what is
1348 * necessary and punch the delalloc blocks directly.
1350 struct xfs_inode *ip = XFS_I(inode);
1351 xfs_fileoff_t start_fsb;
1352 xfs_fileoff_t end_fsb;
1355 truncate_pagecache(inode, to, inode->i_size);
1358 * Check if there are any blocks that are outside of i_size
1359 * that need to be trimmed back.
1361 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1362 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1363 if (end_fsb <= start_fsb)
1366 xfs_ilock(ip, XFS_ILOCK_EXCL);
1367 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1368 end_fsb - start_fsb);
1370 /* something screwed, just bail */
1371 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1372 xfs_alert(ip->i_mount,
1373 "xfs_vm_write_failed: unable to clean up ino %lld",
1377 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1384 struct address_space *mapping,
1388 struct page **pagep,
1393 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1394 pagep, xfs_get_blocks);
1396 xfs_vm_write_failed(mapping, pos + len);
1403 struct address_space *mapping,
1412 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1413 if (unlikely(ret < len))
1414 xfs_vm_write_failed(mapping, pos + len);
1420 struct address_space *mapping,
1423 struct inode *inode = (struct inode *)mapping->host;
1424 struct xfs_inode *ip = XFS_I(inode);
1426 trace_xfs_vm_bmap(XFS_I(inode));
1427 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1428 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1429 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1430 return generic_block_bmap(mapping, block, xfs_get_blocks);
1435 struct file *unused,
1438 return mpage_readpage(page, xfs_get_blocks);
1443 struct file *unused,
1444 struct address_space *mapping,
1445 struct list_head *pages,
1448 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1451 const struct address_space_operations xfs_address_space_operations = {
1452 .readpage = xfs_vm_readpage,
1453 .readpages = xfs_vm_readpages,
1454 .writepage = xfs_vm_writepage,
1455 .writepages = xfs_vm_writepages,
1456 .releasepage = xfs_vm_releasepage,
1457 .invalidatepage = xfs_vm_invalidatepage,
1458 .write_begin = xfs_vm_write_begin,
1459 .write_end = xfs_vm_write_end,
1460 .bmap = xfs_vm_bmap,
1461 .direct_IO = xfs_vm_direct_IO,
1462 .migratepage = buffer_migrate_page,
1463 .is_partially_uptodate = block_is_partially_uptodate,
1464 .error_remove_page = generic_error_remove_page,