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>
43 * Prime number of hash buckets since address is used as the key.
46 #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
47 static wait_queue_head_t xfs_ioend_wq[NVSYNC];
54 for (i = 0; i < NVSYNC; i++)
55 init_waitqueue_head(&xfs_ioend_wq[i]);
62 wait_queue_head_t *wq = to_ioend_wq(ip);
64 wait_event(*wq, (atomic_read(&ip->i_iocount) == 0));
71 if (atomic_dec_and_test(&ip->i_iocount))
72 wake_up(to_ioend_wq(ip));
81 struct buffer_head *bh, *head;
83 *delalloc = *unwritten = 0;
85 bh = head = page_buffers(page);
87 if (buffer_unwritten(bh))
89 else if (buffer_delay(bh))
91 } while ((bh = bh->b_this_page) != head);
94 STATIC struct block_device *
95 xfs_find_bdev_for_inode(
98 struct xfs_inode *ip = XFS_I(inode);
99 struct xfs_mount *mp = ip->i_mount;
101 if (XFS_IS_REALTIME_INODE(ip))
102 return mp->m_rtdev_targp->bt_bdev;
104 return mp->m_ddev_targp->bt_bdev;
108 * We're now finished for good with this ioend structure.
109 * Update the page state via the associated buffer_heads,
110 * release holds on the inode and bio, and finally free
111 * up memory. Do not use the ioend after this.
117 struct buffer_head *bh, *next;
118 struct xfs_inode *ip = XFS_I(ioend->io_inode);
120 for (bh = ioend->io_buffer_head; bh; bh = next) {
121 next = bh->b_private;
122 bh->b_end_io(bh, !ioend->io_error);
125 if (ioend->io_iocb) {
126 if (ioend->io_isasync)
127 aio_complete(ioend->io_iocb, ioend->io_result, 0);
128 inode_dio_done(ioend->io_inode);
131 mempool_free(ioend, xfs_ioend_pool);
135 * If the end of the current ioend is beyond the current EOF,
136 * return the new EOF value, otherwise zero.
142 xfs_inode_t *ip = XFS_I(ioend->io_inode);
146 bsize = ioend->io_offset + ioend->io_size;
147 isize = MAX(ip->i_size, ip->i_new_size);
148 isize = MIN(isize, bsize);
149 return isize > ip->i_d.di_size ? isize : 0;
153 * Update on-disk file size now that data has been written to disk. The
154 * current in-memory file size is i_size. If a write is beyond eof i_new_size
155 * will be the intended file size until i_size is updated. If this write does
156 * not extend all the way to the valid file size then restrict this update to
157 * the end of the write.
159 * This function does not block as blocking on the inode lock in IO completion
160 * can lead to IO completion order dependency deadlocks.. If it can't get the
161 * inode ilock it will return EAGAIN. Callers must handle this.
167 xfs_inode_t *ip = XFS_I(ioend->io_inode);
170 if (unlikely(ioend->io_error))
173 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
176 isize = xfs_ioend_new_eof(ioend);
178 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
179 ip->i_d.di_size = isize;
180 xfs_mark_inode_dirty(ip);
183 xfs_iunlock(ip, XFS_ILOCK_EXCL);
188 * Schedule IO completion handling on the final put of an ioend.
192 struct xfs_ioend *ioend)
194 if (atomic_dec_and_test(&ioend->io_remaining)) {
195 if (ioend->io_type == IO_UNWRITTEN)
196 queue_work(xfsconvertd_workqueue, &ioend->io_work);
198 queue_work(xfsdatad_workqueue, &ioend->io_work);
203 * IO write completion.
207 struct work_struct *work)
209 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
210 struct xfs_inode *ip = XFS_I(ioend->io_inode);
214 * For unwritten extents we need to issue transactions to convert a
215 * range to normal written extens after the data I/O has finished.
217 if (ioend->io_type == IO_UNWRITTEN &&
218 likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) {
220 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
223 ioend->io_error = error;
227 * We might have to update the on-disk file size after extending
230 error = xfs_setfilesize(ioend);
231 ASSERT(!error || error == EAGAIN);
234 * If we didn't complete processing of the ioend, requeue it to the
235 * tail of the workqueue for another attempt later. Otherwise destroy
238 if (error == EAGAIN) {
239 atomic_inc(&ioend->io_remaining);
240 xfs_finish_ioend(ioend);
241 /* ensure we don't spin on blocked ioends */
244 xfs_destroy_ioend(ioend);
249 * Call IO completion handling in caller context on the final put of an ioend.
252 xfs_finish_ioend_sync(
253 struct xfs_ioend *ioend)
255 if (atomic_dec_and_test(&ioend->io_remaining))
256 xfs_end_io(&ioend->io_work);
260 * Allocate and initialise an IO completion structure.
261 * We need to track unwritten extent write completion here initially.
262 * We'll need to extend this for updating the ondisk inode size later
272 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
275 * Set the count to 1 initially, which will prevent an I/O
276 * completion callback from happening before we have started
277 * all the I/O from calling the completion routine too early.
279 atomic_set(&ioend->io_remaining, 1);
280 ioend->io_isasync = 0;
282 ioend->io_list = NULL;
283 ioend->io_type = type;
284 ioend->io_inode = inode;
285 ioend->io_buffer_head = NULL;
286 ioend->io_buffer_tail = NULL;
287 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
288 ioend->io_offset = 0;
290 ioend->io_iocb = NULL;
291 ioend->io_result = 0;
293 INIT_WORK(&ioend->io_work, xfs_end_io);
301 struct xfs_bmbt_irec *imap,
305 struct xfs_inode *ip = XFS_I(inode);
306 struct xfs_mount *mp = ip->i_mount;
307 ssize_t count = 1 << inode->i_blkbits;
308 xfs_fileoff_t offset_fsb, end_fsb;
310 int bmapi_flags = XFS_BMAPI_ENTIRE;
313 if (XFS_FORCED_SHUTDOWN(mp))
314 return -XFS_ERROR(EIO);
316 if (type == IO_UNWRITTEN)
317 bmapi_flags |= XFS_BMAPI_IGSTATE;
319 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
321 return -XFS_ERROR(EAGAIN);
322 xfs_ilock(ip, XFS_ILOCK_SHARED);
325 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
326 (ip->i_df.if_flags & XFS_IFEXTENTS));
327 ASSERT(offset <= mp->m_maxioffset);
329 if (offset + count > mp->m_maxioffset)
330 count = mp->m_maxioffset - offset;
331 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
332 offset_fsb = XFS_B_TO_FSBT(mp, offset);
333 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
334 bmapi_flags, NULL, 0, imap, &nimaps, NULL);
335 xfs_iunlock(ip, XFS_ILOCK_SHARED);
338 return -XFS_ERROR(error);
340 if (type == IO_DELALLOC &&
341 (!nimaps || isnullstartblock(imap->br_startblock))) {
342 error = xfs_iomap_write_allocate(ip, offset, count, imap);
344 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
345 return -XFS_ERROR(error);
349 if (type == IO_UNWRITTEN) {
351 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
352 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
356 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
363 struct xfs_bmbt_irec *imap,
366 offset >>= inode->i_blkbits;
368 return offset >= imap->br_startoff &&
369 offset < imap->br_startoff + imap->br_blockcount;
373 * BIO completion handler for buffered IO.
380 xfs_ioend_t *ioend = bio->bi_private;
382 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
383 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
385 /* Toss bio and pass work off to an xfsdatad thread */
386 bio->bi_private = NULL;
387 bio->bi_end_io = NULL;
390 xfs_finish_ioend(ioend);
394 xfs_submit_ioend_bio(
395 struct writeback_control *wbc,
399 atomic_inc(&ioend->io_remaining);
400 bio->bi_private = ioend;
401 bio->bi_end_io = xfs_end_bio;
404 * If the I/O is beyond EOF we mark the inode dirty immediately
405 * but don't update the inode size until I/O completion.
407 if (xfs_ioend_new_eof(ioend))
408 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
410 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
415 struct buffer_head *bh)
417 int nvecs = bio_get_nr_vecs(bh->b_bdev);
418 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
420 ASSERT(bio->bi_private == NULL);
421 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
422 bio->bi_bdev = bh->b_bdev;
427 xfs_start_buffer_writeback(
428 struct buffer_head *bh)
430 ASSERT(buffer_mapped(bh));
431 ASSERT(buffer_locked(bh));
432 ASSERT(!buffer_delay(bh));
433 ASSERT(!buffer_unwritten(bh));
435 mark_buffer_async_write(bh);
436 set_buffer_uptodate(bh);
437 clear_buffer_dirty(bh);
441 xfs_start_page_writeback(
446 ASSERT(PageLocked(page));
447 ASSERT(!PageWriteback(page));
449 clear_page_dirty_for_io(page);
450 set_page_writeback(page);
452 /* If no buffers on the page are to be written, finish it here */
454 end_page_writeback(page);
457 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
459 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
463 * Submit all of the bios for all of the ioends we have saved up, covering the
464 * initial writepage page and also any probed pages.
466 * Because we may have multiple ioends spanning a page, we need to start
467 * writeback on all the buffers before we submit them for I/O. If we mark the
468 * buffers as we got, then we can end up with a page that only has buffers
469 * marked async write and I/O complete on can occur before we mark the other
470 * buffers async write.
472 * The end result of this is that we trip a bug in end_page_writeback() because
473 * we call it twice for the one page as the code in end_buffer_async_write()
474 * assumes that all buffers on the page are started at the same time.
476 * The fix is two passes across the ioend list - one to start writeback on the
477 * buffer_heads, and then submit them for I/O on the second pass.
481 struct writeback_control *wbc,
484 xfs_ioend_t *head = ioend;
486 struct buffer_head *bh;
488 sector_t lastblock = 0;
490 /* Pass 1 - start writeback */
492 next = ioend->io_list;
493 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
494 xfs_start_buffer_writeback(bh);
495 } while ((ioend = next) != NULL);
497 /* Pass 2 - submit I/O */
500 next = ioend->io_list;
503 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
507 bio = xfs_alloc_ioend_bio(bh);
508 } else if (bh->b_blocknr != lastblock + 1) {
509 xfs_submit_ioend_bio(wbc, ioend, bio);
513 if (bio_add_buffer(bio, bh) != bh->b_size) {
514 xfs_submit_ioend_bio(wbc, ioend, bio);
518 lastblock = bh->b_blocknr;
521 xfs_submit_ioend_bio(wbc, ioend, bio);
522 xfs_finish_ioend(ioend);
523 } while ((ioend = next) != NULL);
527 * Cancel submission of all buffer_heads so far in this endio.
528 * Toss the endio too. Only ever called for the initial page
529 * in a writepage request, so only ever one page.
536 struct buffer_head *bh, *next_bh;
539 next = ioend->io_list;
540 bh = ioend->io_buffer_head;
542 next_bh = bh->b_private;
543 clear_buffer_async_write(bh);
545 } while ((bh = next_bh) != NULL);
547 xfs_ioend_wake(XFS_I(ioend->io_inode));
548 mempool_free(ioend, xfs_ioend_pool);
549 } while ((ioend = next) != NULL);
553 * Test to see if we've been building up a completion structure for
554 * earlier buffers -- if so, we try to append to this ioend if we
555 * can, otherwise we finish off any current ioend and start another.
556 * Return true if we've finished the given ioend.
561 struct buffer_head *bh,
564 xfs_ioend_t **result,
567 xfs_ioend_t *ioend = *result;
569 if (!ioend || need_ioend || type != ioend->io_type) {
570 xfs_ioend_t *previous = *result;
572 ioend = xfs_alloc_ioend(inode, type);
573 ioend->io_offset = offset;
574 ioend->io_buffer_head = bh;
575 ioend->io_buffer_tail = bh;
577 previous->io_list = ioend;
580 ioend->io_buffer_tail->b_private = bh;
581 ioend->io_buffer_tail = bh;
584 bh->b_private = NULL;
585 ioend->io_size += bh->b_size;
591 struct buffer_head *bh,
592 struct xfs_bmbt_irec *imap,
596 struct xfs_mount *m = XFS_I(inode)->i_mount;
597 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
598 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
600 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
601 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
603 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
604 ((offset - iomap_offset) >> inode->i_blkbits);
606 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
609 set_buffer_mapped(bh);
615 struct buffer_head *bh,
616 struct xfs_bmbt_irec *imap,
619 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
620 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
622 xfs_map_buffer(inode, bh, imap, offset);
623 set_buffer_mapped(bh);
624 clear_buffer_delay(bh);
625 clear_buffer_unwritten(bh);
629 * Test if a given page is suitable for writing as part of an unwritten
630 * or delayed allocate extent.
637 if (PageWriteback(page))
640 if (page->mapping && page_has_buffers(page)) {
641 struct buffer_head *bh, *head;
644 bh = head = page_buffers(page);
646 if (buffer_unwritten(bh))
647 acceptable = (type == IO_UNWRITTEN);
648 else if (buffer_delay(bh))
649 acceptable = (type == IO_DELALLOC);
650 else if (buffer_dirty(bh) && buffer_mapped(bh))
651 acceptable = (type == IO_OVERWRITE);
654 } while ((bh = bh->b_this_page) != head);
664 * Allocate & map buffers for page given the extent map. Write it out.
665 * except for the original page of a writepage, this is called on
666 * delalloc/unwritten pages only, for the original page it is possible
667 * that the page has no mapping at all.
674 struct xfs_bmbt_irec *imap,
675 xfs_ioend_t **ioendp,
676 struct writeback_control *wbc)
678 struct buffer_head *bh, *head;
679 xfs_off_t end_offset;
680 unsigned long p_offset;
683 int count = 0, done = 0, uptodate = 1;
684 xfs_off_t offset = page_offset(page);
686 if (page->index != tindex)
688 if (!trylock_page(page))
690 if (PageWriteback(page))
691 goto fail_unlock_page;
692 if (page->mapping != inode->i_mapping)
693 goto fail_unlock_page;
694 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
695 goto fail_unlock_page;
698 * page_dirty is initially a count of buffers on the page before
699 * EOF and is decremented as we move each into a cleanable state.
703 * End offset is the highest offset that this page should represent.
704 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
705 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
706 * hence give us the correct page_dirty count. On any other page,
707 * it will be zero and in that case we need page_dirty to be the
708 * count of buffers on the page.
710 end_offset = min_t(unsigned long long,
711 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
714 len = 1 << inode->i_blkbits;
715 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
717 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
718 page_dirty = p_offset / len;
720 bh = head = page_buffers(page);
722 if (offset >= end_offset)
724 if (!buffer_uptodate(bh))
726 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
731 if (buffer_unwritten(bh) || buffer_delay(bh) ||
733 if (buffer_unwritten(bh))
735 else if (buffer_delay(bh))
740 if (!xfs_imap_valid(inode, imap, offset)) {
746 if (type != IO_OVERWRITE)
747 xfs_map_at_offset(inode, bh, imap, offset);
748 xfs_add_to_ioend(inode, bh, offset, type,
756 } while (offset += len, (bh = bh->b_this_page) != head);
758 if (uptodate && bh == head)
759 SetPageUptodate(page);
762 if (--wbc->nr_to_write <= 0 &&
763 wbc->sync_mode == WB_SYNC_NONE)
766 xfs_start_page_writeback(page, !page_dirty, count);
776 * Convert & write out a cluster of pages in the same extent as defined
777 * by mp and following the start page.
783 struct xfs_bmbt_irec *imap,
784 xfs_ioend_t **ioendp,
785 struct writeback_control *wbc,
791 pagevec_init(&pvec, 0);
792 while (!done && tindex <= tlast) {
793 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
795 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
798 for (i = 0; i < pagevec_count(&pvec); i++) {
799 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
805 pagevec_release(&pvec);
811 xfs_vm_invalidatepage(
813 unsigned long offset)
815 trace_xfs_invalidatepage(page->mapping->host, page, offset);
816 block_invalidatepage(page, offset);
820 * If the page has delalloc buffers on it, we need to punch them out before we
821 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
822 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
823 * is done on that same region - the delalloc extent is returned when none is
824 * supposed to be there.
826 * We prevent this by truncating away the delalloc regions on the page before
827 * invalidating it. Because they are delalloc, we can do this without needing a
828 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
829 * truncation without a transaction as there is no space left for block
830 * reservation (typically why we see a ENOSPC in writeback).
832 * This is not a performance critical path, so for now just do the punching a
833 * buffer head at a time.
836 xfs_aops_discard_page(
839 struct inode *inode = page->mapping->host;
840 struct xfs_inode *ip = XFS_I(inode);
841 struct buffer_head *bh, *head;
842 loff_t offset = page_offset(page);
844 if (!xfs_is_delayed_page(page, IO_DELALLOC))
847 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
850 xfs_alert(ip->i_mount,
851 "page discard on page %p, inode 0x%llx, offset %llu.",
852 page, ip->i_ino, offset);
854 xfs_ilock(ip, XFS_ILOCK_EXCL);
855 bh = head = page_buffers(page);
858 xfs_fileoff_t start_fsb;
860 if (!buffer_delay(bh))
863 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
864 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
866 /* something screwed, just bail */
867 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
868 xfs_alert(ip->i_mount,
869 "page discard unable to remove delalloc mapping.");
874 offset += 1 << inode->i_blkbits;
876 } while ((bh = bh->b_this_page) != head);
878 xfs_iunlock(ip, XFS_ILOCK_EXCL);
880 xfs_vm_invalidatepage(page, 0);
885 * Write out a dirty page.
887 * For delalloc space on the page we need to allocate space and flush it.
888 * For unwritten space on the page we need to start the conversion to
889 * regular allocated space.
890 * For any other dirty buffer heads on the page we should flush them.
895 struct writeback_control *wbc)
897 struct inode *inode = page->mapping->host;
898 struct buffer_head *bh, *head;
899 struct xfs_bmbt_irec imap;
900 xfs_ioend_t *ioend = NULL, *iohead = NULL;
903 __uint64_t end_offset;
904 pgoff_t end_index, last_index;
906 int err, imap_valid = 0, uptodate = 1;
910 trace_xfs_writepage(inode, page, 0);
912 ASSERT(page_has_buffers(page));
915 * Refuse to write the page out if we are called from reclaim context.
917 * This avoids stack overflows when called from deeply used stacks in
918 * random callers for direct reclaim or memcg reclaim. We explicitly
919 * allow reclaim from kswapd as the stack usage there is relatively low.
921 * This should really be done by the core VM, but until that happens
922 * filesystems like XFS, btrfs and ext4 have to take care of this
925 if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC)
929 * Given that we do not allow direct reclaim to call us, we should
930 * never be called while in a filesystem transaction.
932 if (WARN_ON(current->flags & PF_FSTRANS))
935 /* Is this page beyond the end of the file? */
936 offset = i_size_read(inode);
937 end_index = offset >> PAGE_CACHE_SHIFT;
938 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
939 if (page->index >= end_index) {
940 if ((page->index >= end_index + 1) ||
941 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
947 end_offset = min_t(unsigned long long,
948 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
950 len = 1 << inode->i_blkbits;
952 bh = head = page_buffers(page);
953 offset = page_offset(page);
956 if (wbc->sync_mode == WB_SYNC_NONE)
962 if (offset >= end_offset)
964 if (!buffer_uptodate(bh))
968 * set_page_dirty dirties all buffers in a page, independent
969 * of their state. The dirty state however is entirely
970 * meaningless for holes (!mapped && uptodate), so skip
971 * buffers covering holes here.
973 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
978 if (buffer_unwritten(bh)) {
979 if (type != IO_UNWRITTEN) {
983 } else if (buffer_delay(bh)) {
984 if (type != IO_DELALLOC) {
988 } else if (buffer_uptodate(bh)) {
989 if (type != IO_OVERWRITE) {
994 if (PageUptodate(page)) {
995 ASSERT(buffer_mapped(bh));
1002 imap_valid = xfs_imap_valid(inode, &imap, offset);
1005 * If we didn't have a valid mapping then we need to
1006 * put the new mapping into a separate ioend structure.
1007 * This ensures non-contiguous extents always have
1008 * separate ioends, which is particularly important
1009 * for unwritten extent conversion at I/O completion
1013 err = xfs_map_blocks(inode, offset, &imap, type,
1017 imap_valid = xfs_imap_valid(inode, &imap, offset);
1021 if (type != IO_OVERWRITE)
1022 xfs_map_at_offset(inode, bh, &imap, offset);
1023 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1031 } while (offset += len, ((bh = bh->b_this_page) != head));
1033 if (uptodate && bh == head)
1034 SetPageUptodate(page);
1036 xfs_start_page_writeback(page, 1, count);
1038 if (ioend && imap_valid) {
1039 xfs_off_t end_index;
1041 end_index = imap.br_startoff + imap.br_blockcount;
1044 end_index <<= inode->i_blkbits;
1047 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1049 /* check against file size */
1050 if (end_index > last_index)
1051 end_index = last_index;
1053 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1058 xfs_submit_ioend(wbc, iohead);
1064 xfs_cancel_ioend(iohead);
1069 xfs_aops_discard_page(page);
1070 ClearPageUptodate(page);
1075 redirty_page_for_writepage(wbc, page);
1082 struct address_space *mapping,
1083 struct writeback_control *wbc)
1085 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1086 return generic_writepages(mapping, wbc);
1090 * Called to move a page into cleanable state - and from there
1091 * to be released. The page should already be clean. We always
1092 * have buffer heads in this call.
1094 * Returns 1 if the page is ok to release, 0 otherwise.
1101 int delalloc, unwritten;
1103 trace_xfs_releasepage(page->mapping->host, page, 0);
1105 xfs_count_page_state(page, &delalloc, &unwritten);
1107 if (WARN_ON(delalloc))
1109 if (WARN_ON(unwritten))
1112 return try_to_free_buffers(page);
1117 struct inode *inode,
1119 struct buffer_head *bh_result,
1123 struct xfs_inode *ip = XFS_I(inode);
1124 struct xfs_mount *mp = ip->i_mount;
1125 xfs_fileoff_t offset_fsb, end_fsb;
1128 struct xfs_bmbt_irec imap;
1134 if (XFS_FORCED_SHUTDOWN(mp))
1135 return -XFS_ERROR(EIO);
1137 offset = (xfs_off_t)iblock << inode->i_blkbits;
1138 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1139 size = bh_result->b_size;
1141 if (!create && direct && offset >= i_size_read(inode))
1145 lockmode = XFS_ILOCK_EXCL;
1146 xfs_ilock(ip, lockmode);
1148 lockmode = xfs_ilock_map_shared(ip);
1151 ASSERT(offset <= mp->m_maxioffset);
1152 if (offset + size > mp->m_maxioffset)
1153 size = mp->m_maxioffset - offset;
1154 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1155 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1157 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
1158 XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL);
1164 (imap.br_startblock == HOLESTARTBLOCK ||
1165 imap.br_startblock == DELAYSTARTBLOCK))) {
1167 error = xfs_iomap_write_direct(ip, offset, size,
1170 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1175 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1176 } else if (nimaps) {
1177 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1179 trace_xfs_get_blocks_notfound(ip, offset, size);
1182 xfs_iunlock(ip, lockmode);
1184 if (imap.br_startblock != HOLESTARTBLOCK &&
1185 imap.br_startblock != DELAYSTARTBLOCK) {
1187 * For unwritten extents do not report a disk address on
1188 * the read case (treat as if we're reading into a hole).
1190 if (create || !ISUNWRITTEN(&imap))
1191 xfs_map_buffer(inode, bh_result, &imap, offset);
1192 if (create && ISUNWRITTEN(&imap)) {
1194 bh_result->b_private = inode;
1195 set_buffer_unwritten(bh_result);
1200 * If this is a realtime file, data may be on a different device.
1201 * to that pointed to from the buffer_head b_bdev currently.
1203 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1206 * If we previously allocated a block out beyond eof and we are now
1207 * coming back to use it then we will need to flag it as new even if it
1208 * has a disk address.
1210 * With sub-block writes into unwritten extents we also need to mark
1211 * the buffer as new so that the unwritten parts of the buffer gets
1215 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1216 (offset >= i_size_read(inode)) ||
1217 (new || ISUNWRITTEN(&imap))))
1218 set_buffer_new(bh_result);
1220 if (imap.br_startblock == DELAYSTARTBLOCK) {
1223 set_buffer_uptodate(bh_result);
1224 set_buffer_mapped(bh_result);
1225 set_buffer_delay(bh_result);
1230 * If this is O_DIRECT or the mpage code calling tell them how large
1231 * the mapping is, so that we can avoid repeated get_blocks calls.
1233 if (direct || size > (1 << inode->i_blkbits)) {
1234 xfs_off_t mapping_size;
1236 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1237 mapping_size <<= inode->i_blkbits;
1239 ASSERT(mapping_size > 0);
1240 if (mapping_size > size)
1241 mapping_size = size;
1242 if (mapping_size > LONG_MAX)
1243 mapping_size = LONG_MAX;
1245 bh_result->b_size = mapping_size;
1251 xfs_iunlock(ip, lockmode);
1257 struct inode *inode,
1259 struct buffer_head *bh_result,
1262 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1266 xfs_get_blocks_direct(
1267 struct inode *inode,
1269 struct buffer_head *bh_result,
1272 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1276 * Complete a direct I/O write request.
1278 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1279 * need to issue a transaction to convert the range from unwritten to written
1280 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1281 * to do this and we are done. But in case this was a successful AIO
1282 * request this handler is called from interrupt context, from which we
1283 * can't start transactions. In that case offload the I/O completion to
1284 * the workqueues we also use for buffered I/O completion.
1287 xfs_end_io_direct_write(
1295 struct xfs_ioend *ioend = iocb->private;
1298 * blockdev_direct_IO can return an error even after the I/O
1299 * completion handler was called. Thus we need to protect
1300 * against double-freeing.
1302 iocb->private = NULL;
1304 ioend->io_offset = offset;
1305 ioend->io_size = size;
1306 ioend->io_iocb = iocb;
1307 ioend->io_result = ret;
1308 if (private && size > 0)
1309 ioend->io_type = IO_UNWRITTEN;
1312 ioend->io_isasync = 1;
1313 xfs_finish_ioend(ioend);
1315 xfs_finish_ioend_sync(ioend);
1323 const struct iovec *iov,
1325 unsigned long nr_segs)
1327 struct inode *inode = iocb->ki_filp->f_mapping->host;
1328 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1332 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1334 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1336 xfs_get_blocks_direct,
1337 xfs_end_io_direct_write, NULL, 0);
1338 if (ret != -EIOCBQUEUED && iocb->private)
1339 xfs_destroy_ioend(iocb->private);
1341 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1343 xfs_get_blocks_direct,
1351 xfs_vm_write_failed(
1352 struct address_space *mapping,
1355 struct inode *inode = mapping->host;
1357 if (to > inode->i_size) {
1359 * punch out the delalloc blocks we have already allocated. We
1360 * don't call xfs_setattr() to do this as we may be in the
1361 * middle of a multi-iovec write and so the vfs inode->i_size
1362 * will not match the xfs ip->i_size and so it will zero too
1363 * much. Hence we jus truncate the page cache to zero what is
1364 * necessary and punch the delalloc blocks directly.
1366 struct xfs_inode *ip = XFS_I(inode);
1367 xfs_fileoff_t start_fsb;
1368 xfs_fileoff_t end_fsb;
1371 truncate_pagecache(inode, to, inode->i_size);
1374 * Check if there are any blocks that are outside of i_size
1375 * that need to be trimmed back.
1377 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1378 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1379 if (end_fsb <= start_fsb)
1382 xfs_ilock(ip, XFS_ILOCK_EXCL);
1383 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1384 end_fsb - start_fsb);
1386 /* something screwed, just bail */
1387 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1388 xfs_alert(ip->i_mount,
1389 "xfs_vm_write_failed: unable to clean up ino %lld",
1393 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1400 struct address_space *mapping,
1404 struct page **pagep,
1409 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1410 pagep, xfs_get_blocks);
1412 xfs_vm_write_failed(mapping, pos + len);
1419 struct address_space *mapping,
1428 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1429 if (unlikely(ret < len))
1430 xfs_vm_write_failed(mapping, pos + len);
1436 struct address_space *mapping,
1439 struct inode *inode = (struct inode *)mapping->host;
1440 struct xfs_inode *ip = XFS_I(inode);
1442 trace_xfs_vm_bmap(XFS_I(inode));
1443 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1444 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1445 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1446 return generic_block_bmap(mapping, block, xfs_get_blocks);
1451 struct file *unused,
1454 return mpage_readpage(page, xfs_get_blocks);
1459 struct file *unused,
1460 struct address_space *mapping,
1461 struct list_head *pages,
1464 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1467 const struct address_space_operations xfs_address_space_operations = {
1468 .readpage = xfs_vm_readpage,
1469 .readpages = xfs_vm_readpages,
1470 .writepage = xfs_vm_writepage,
1471 .writepages = xfs_vm_writepages,
1472 .releasepage = xfs_vm_releasepage,
1473 .invalidatepage = xfs_vm_invalidatepage,
1474 .write_begin = xfs_vm_write_begin,
1475 .write_end = xfs_vm_write_end,
1476 .bmap = xfs_vm_bmap,
1477 .direct_IO = xfs_vm_direct_IO,
1478 .migratepage = buffer_migrate_page,
1479 .is_partially_uptodate = block_is_partially_uptodate,
1480 .error_remove_page = generic_error_remove_page,