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
25 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_error.h"
39 #include "xfs_iomap.h"
40 #include "xfs_vnodeops.h"
41 #include <linux/mpage.h>
42 #include <linux/pagevec.h>
43 #include <linux/writeback.h>
52 struct buffer_head *bh, *head;
54 *delalloc = *unmapped = *unwritten = 0;
56 bh = head = page_buffers(page);
58 if (buffer_uptodate(bh) && !buffer_mapped(bh))
60 else if (buffer_unwritten(bh))
62 else if (buffer_delay(bh))
64 } while ((bh = bh->b_this_page) != head);
67 #if defined(XFS_RW_TRACE)
76 loff_t isize = i_size_read(inode);
77 loff_t offset = page_offset(page);
78 int delalloc = -1, unmapped = -1, unwritten = -1;
80 if (page_has_buffers(page))
81 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
87 ktrace_enter(ip->i_rwtrace,
88 (void *)((unsigned long)tag),
93 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
94 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
95 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
96 (void *)((unsigned long)(isize & 0xffffffff)),
97 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(offset & 0xffffffff)),
99 (void *)((unsigned long)delalloc),
100 (void *)((unsigned long)unmapped),
101 (void *)((unsigned long)unwritten),
102 (void *)((unsigned long)current_pid()),
106 #define xfs_page_trace(tag, inode, page, pgoff)
109 STATIC struct block_device *
110 xfs_find_bdev_for_inode(
111 struct xfs_inode *ip)
113 struct xfs_mount *mp = ip->i_mount;
115 if (XFS_IS_REALTIME_INODE(ip))
116 return mp->m_rtdev_targp->bt_bdev;
118 return mp->m_ddev_targp->bt_bdev;
122 * Schedule IO completion handling on a xfsdatad if this was
123 * the final hold on this ioend. If we are asked to wait,
124 * flush the workqueue.
131 if (atomic_dec_and_test(&ioend->io_remaining)) {
132 queue_work(xfsdatad_workqueue, &ioend->io_work);
134 flush_workqueue(xfsdatad_workqueue);
139 * We're now finished for good with this ioend structure.
140 * Update the page state via the associated buffer_heads,
141 * release holds on the inode and bio, and finally free
142 * up memory. Do not use the ioend after this.
148 struct buffer_head *bh, *next;
150 for (bh = ioend->io_buffer_head; bh; bh = next) {
151 next = bh->b_private;
152 bh->b_end_io(bh, !ioend->io_error);
154 if (unlikely(ioend->io_error)) {
155 vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
158 vn_iowake(XFS_I(ioend->io_inode));
159 mempool_free(ioend, xfs_ioend_pool);
163 * Update on-disk file size now that data has been written to disk.
164 * The current in-memory file size is i_size. If a write is beyond
165 * eof i_new_size will be the intended file size until i_size is
166 * updated. If this write does not extend all the way to the valid
167 * file size then restrict this update to the end of the write.
173 xfs_inode_t *ip = XFS_I(ioend->io_inode);
177 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
178 ASSERT(ioend->io_type != IOMAP_READ);
180 if (unlikely(ioend->io_error))
183 bsize = ioend->io_offset + ioend->io_size;
185 xfs_ilock(ip, XFS_ILOCK_EXCL);
187 isize = MAX(ip->i_size, ip->i_new_size);
188 isize = MIN(isize, bsize);
190 if (ip->i_d.di_size < isize) {
191 ip->i_d.di_size = isize;
192 ip->i_update_core = 1;
193 ip->i_update_size = 1;
194 mark_inode_dirty_sync(ioend->io_inode);
197 xfs_iunlock(ip, XFS_ILOCK_EXCL);
201 * Buffered IO write completion for delayed allocate extents.
204 xfs_end_bio_delalloc(
205 struct work_struct *work)
208 container_of(work, xfs_ioend_t, io_work);
210 xfs_setfilesize(ioend);
211 xfs_destroy_ioend(ioend);
215 * Buffered IO write completion for regular, written extents.
219 struct work_struct *work)
222 container_of(work, xfs_ioend_t, io_work);
224 xfs_setfilesize(ioend);
225 xfs_destroy_ioend(ioend);
229 * IO write completion for unwritten extents.
231 * Issue transactions to convert a buffer range from unwritten
232 * to written extents.
235 xfs_end_bio_unwritten(
236 struct work_struct *work)
239 container_of(work, xfs_ioend_t, io_work);
240 struct xfs_inode *ip = XFS_I(ioend->io_inode);
241 xfs_off_t offset = ioend->io_offset;
242 size_t size = ioend->io_size;
244 if (likely(!ioend->io_error)) {
245 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
247 error = xfs_iomap_write_unwritten(ip, offset, size);
249 ioend->io_error = error;
251 xfs_setfilesize(ioend);
253 xfs_destroy_ioend(ioend);
257 * IO read completion for regular, written extents.
261 struct work_struct *work)
264 container_of(work, xfs_ioend_t, io_work);
266 xfs_destroy_ioend(ioend);
270 * Allocate and initialise an IO completion structure.
271 * We need to track unwritten extent write completion here initially.
272 * We'll need to extend this for updating the ondisk inode size later
282 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
285 * Set the count to 1 initially, which will prevent an I/O
286 * completion callback from happening before we have started
287 * all the I/O from calling the completion routine too early.
289 atomic_set(&ioend->io_remaining, 1);
291 ioend->io_list = NULL;
292 ioend->io_type = type;
293 ioend->io_inode = inode;
294 ioend->io_buffer_head = NULL;
295 ioend->io_buffer_tail = NULL;
296 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
297 ioend->io_offset = 0;
300 if (type == IOMAP_UNWRITTEN)
301 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
302 else if (type == IOMAP_DELAY)
303 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
304 else if (type == IOMAP_READ)
305 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
307 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
320 xfs_inode_t *ip = XFS_I(inode);
321 int error, nmaps = 1;
323 error = xfs_iomap(ip, offset, count,
324 flags, mapp, &nmaps);
325 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
326 xfs_iflags_set(ip, XFS_IMODIFIED);
335 return offset >= iomapp->iomap_offset &&
336 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
340 * BIO completion handler for buffered IO.
347 xfs_ioend_t *ioend = bio->bi_private;
349 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
350 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
352 /* Toss bio and pass work off to an xfsdatad thread */
353 bio->bi_private = NULL;
354 bio->bi_end_io = NULL;
357 xfs_finish_ioend(ioend, 0);
361 xfs_submit_ioend_bio(
365 atomic_inc(&ioend->io_remaining);
367 bio->bi_private = ioend;
368 bio->bi_end_io = xfs_end_bio;
370 submit_bio(WRITE, bio);
371 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
377 struct buffer_head *bh)
380 int nvecs = bio_get_nr_vecs(bh->b_bdev);
383 bio = bio_alloc(GFP_NOIO, nvecs);
387 ASSERT(bio->bi_private == NULL);
388 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
389 bio->bi_bdev = bh->b_bdev;
395 xfs_start_buffer_writeback(
396 struct buffer_head *bh)
398 ASSERT(buffer_mapped(bh));
399 ASSERT(buffer_locked(bh));
400 ASSERT(!buffer_delay(bh));
401 ASSERT(!buffer_unwritten(bh));
403 mark_buffer_async_write(bh);
404 set_buffer_uptodate(bh);
405 clear_buffer_dirty(bh);
409 xfs_start_page_writeback(
414 ASSERT(PageLocked(page));
415 ASSERT(!PageWriteback(page));
417 clear_page_dirty_for_io(page);
418 set_page_writeback(page);
420 /* If no buffers on the page are to be written, finish it here */
422 end_page_writeback(page);
425 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
427 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
431 * Submit all of the bios for all of the ioends we have saved up, covering the
432 * initial writepage page and also any probed pages.
434 * Because we may have multiple ioends spanning a page, we need to start
435 * writeback on all the buffers before we submit them for I/O. If we mark the
436 * buffers as we got, then we can end up with a page that only has buffers
437 * marked async write and I/O complete on can occur before we mark the other
438 * buffers async write.
440 * The end result of this is that we trip a bug in end_page_writeback() because
441 * we call it twice for the one page as the code in end_buffer_async_write()
442 * assumes that all buffers on the page are started at the same time.
444 * The fix is two passes across the ioend list - one to start writeback on the
445 * buffer_heads, and then submit them for I/O on the second pass.
451 xfs_ioend_t *head = ioend;
453 struct buffer_head *bh;
455 sector_t lastblock = 0;
457 /* Pass 1 - start writeback */
459 next = ioend->io_list;
460 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
461 xfs_start_buffer_writeback(bh);
463 } while ((ioend = next) != NULL);
465 /* Pass 2 - submit I/O */
468 next = ioend->io_list;
471 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
475 bio = xfs_alloc_ioend_bio(bh);
476 } else if (bh->b_blocknr != lastblock + 1) {
477 xfs_submit_ioend_bio(ioend, bio);
481 if (bio_add_buffer(bio, bh) != bh->b_size) {
482 xfs_submit_ioend_bio(ioend, bio);
486 lastblock = bh->b_blocknr;
489 xfs_submit_ioend_bio(ioend, bio);
490 xfs_finish_ioend(ioend, 0);
491 } while ((ioend = next) != NULL);
495 * Cancel submission of all buffer_heads so far in this endio.
496 * Toss the endio too. Only ever called for the initial page
497 * in a writepage request, so only ever one page.
504 struct buffer_head *bh, *next_bh;
507 next = ioend->io_list;
508 bh = ioend->io_buffer_head;
510 next_bh = bh->b_private;
511 clear_buffer_async_write(bh);
513 } while ((bh = next_bh) != NULL);
515 vn_iowake(XFS_I(ioend->io_inode));
516 mempool_free(ioend, xfs_ioend_pool);
517 } while ((ioend = next) != NULL);
521 * Test to see if we've been building up a completion structure for
522 * earlier buffers -- if so, we try to append to this ioend if we
523 * can, otherwise we finish off any current ioend and start another.
524 * Return true if we've finished the given ioend.
529 struct buffer_head *bh,
532 xfs_ioend_t **result,
535 xfs_ioend_t *ioend = *result;
537 if (!ioend || need_ioend || type != ioend->io_type) {
538 xfs_ioend_t *previous = *result;
540 ioend = xfs_alloc_ioend(inode, type);
541 ioend->io_offset = offset;
542 ioend->io_buffer_head = bh;
543 ioend->io_buffer_tail = bh;
545 previous->io_list = ioend;
548 ioend->io_buffer_tail->b_private = bh;
549 ioend->io_buffer_tail = bh;
552 bh->b_private = NULL;
553 ioend->io_size += bh->b_size;
558 struct buffer_head *bh,
565 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
567 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
568 ((offset - mp->iomap_offset) >> block_bits);
570 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
573 set_buffer_mapped(bh);
578 struct buffer_head *bh,
583 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
584 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
587 xfs_map_buffer(bh, iomapp, offset, block_bits);
588 bh->b_bdev = iomapp->iomap_target->bt_bdev;
589 set_buffer_mapped(bh);
590 clear_buffer_delay(bh);
591 clear_buffer_unwritten(bh);
595 * Look for a page at index that is suitable for clustering.
600 unsigned int pg_offset,
605 if (PageWriteback(page))
608 if (page->mapping && PageDirty(page)) {
609 if (page_has_buffers(page)) {
610 struct buffer_head *bh, *head;
612 bh = head = page_buffers(page);
614 if (!buffer_uptodate(bh))
616 if (mapped != buffer_mapped(bh))
619 if (ret >= pg_offset)
621 } while ((bh = bh->b_this_page) != head);
623 ret = mapped ? 0 : PAGE_CACHE_SIZE;
632 struct page *startpage,
633 struct buffer_head *bh,
634 struct buffer_head *head,
638 pgoff_t tindex, tlast, tloff;
642 /* First sum forwards in this page */
644 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
647 } while ((bh = bh->b_this_page) != head);
649 /* if we reached the end of the page, sum forwards in following pages */
650 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
651 tindex = startpage->index + 1;
653 /* Prune this back to avoid pathological behavior */
654 tloff = min(tlast, startpage->index + 64);
656 pagevec_init(&pvec, 0);
657 while (!done && tindex <= tloff) {
658 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
660 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
663 for (i = 0; i < pagevec_count(&pvec); i++) {
664 struct page *page = pvec.pages[i];
665 size_t pg_offset, pg_len = 0;
667 if (tindex == tlast) {
669 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
675 pg_offset = PAGE_CACHE_SIZE;
677 if (page->index == tindex && trylock_page(page)) {
678 pg_len = xfs_probe_page(page, pg_offset, mapped);
691 pagevec_release(&pvec);
699 * Test if a given page is suitable for writing as part of an unwritten
700 * or delayed allocate extent.
707 if (PageWriteback(page))
710 if (page->mapping && page_has_buffers(page)) {
711 struct buffer_head *bh, *head;
714 bh = head = page_buffers(page);
716 if (buffer_unwritten(bh))
717 acceptable = (type == IOMAP_UNWRITTEN);
718 else if (buffer_delay(bh))
719 acceptable = (type == IOMAP_DELAY);
720 else if (buffer_dirty(bh) && buffer_mapped(bh))
721 acceptable = (type == IOMAP_NEW);
724 } while ((bh = bh->b_this_page) != head);
734 * Allocate & map buffers for page given the extent map. Write it out.
735 * except for the original page of a writepage, this is called on
736 * delalloc/unwritten pages only, for the original page it is possible
737 * that the page has no mapping at all.
745 xfs_ioend_t **ioendp,
746 struct writeback_control *wbc,
750 struct buffer_head *bh, *head;
751 xfs_off_t end_offset;
752 unsigned long p_offset;
754 int bbits = inode->i_blkbits;
756 int count = 0, done = 0, uptodate = 1;
757 xfs_off_t offset = page_offset(page);
759 if (page->index != tindex)
761 if (!trylock_page(page))
763 if (PageWriteback(page))
764 goto fail_unlock_page;
765 if (page->mapping != inode->i_mapping)
766 goto fail_unlock_page;
767 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
768 goto fail_unlock_page;
771 * page_dirty is initially a count of buffers on the page before
772 * EOF and is decremented as we move each into a cleanable state.
776 * End offset is the highest offset that this page should represent.
777 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
778 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
779 * hence give us the correct page_dirty count. On any other page,
780 * it will be zero and in that case we need page_dirty to be the
781 * count of buffers on the page.
783 end_offset = min_t(unsigned long long,
784 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
787 len = 1 << inode->i_blkbits;
788 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
790 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
791 page_dirty = p_offset / len;
793 bh = head = page_buffers(page);
795 if (offset >= end_offset)
797 if (!buffer_uptodate(bh))
799 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
804 if (buffer_unwritten(bh) || buffer_delay(bh)) {
805 if (buffer_unwritten(bh))
806 type = IOMAP_UNWRITTEN;
810 if (!xfs_iomap_valid(mp, offset)) {
815 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
816 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
818 xfs_map_at_offset(bh, offset, bbits, mp);
820 xfs_add_to_ioend(inode, bh, offset,
823 set_buffer_dirty(bh);
825 mark_buffer_dirty(bh);
831 if (buffer_mapped(bh) && all_bh && startio) {
833 xfs_add_to_ioend(inode, bh, offset,
841 } while (offset += len, (bh = bh->b_this_page) != head);
843 if (uptodate && bh == head)
844 SetPageUptodate(page);
848 struct backing_dev_info *bdi;
850 bdi = inode->i_mapping->backing_dev_info;
852 if (bdi_write_congested(bdi)) {
853 wbc->encountered_congestion = 1;
855 } else if (wbc->nr_to_write <= 0) {
859 xfs_start_page_writeback(page, !page_dirty, count);
870 * Convert & write out a cluster of pages in the same extent as defined
871 * by mp and following the start page.
878 xfs_ioend_t **ioendp,
879 struct writeback_control *wbc,
887 pagevec_init(&pvec, 0);
888 while (!done && tindex <= tlast) {
889 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
891 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
894 for (i = 0; i < pagevec_count(&pvec); i++) {
895 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
896 iomapp, ioendp, wbc, startio, all_bh);
901 pagevec_release(&pvec);
907 * Calling this without startio set means we are being asked to make a dirty
908 * page ready for freeing it's buffers. When called with startio set then
909 * we are coming from writepage.
911 * When called with startio set it is important that we write the WHOLE
913 * The bh->b_state's cannot know if any of the blocks or which block for
914 * that matter are dirty due to mmap writes, and therefore bh uptodate is
915 * only valid if the page itself isn't completely uptodate. Some layers
916 * may clear the page dirty flag prior to calling write page, under the
917 * assumption the entire page will be written out; by not writing out the
918 * whole page the page can be reused before all valid dirty data is
919 * written out. Note: in the case of a page that has been dirty'd by
920 * mapwrite and but partially setup by block_prepare_write the
921 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
922 * valid state, thus the whole page must be written out thing.
926 xfs_page_state_convert(
929 struct writeback_control *wbc,
931 int unmapped) /* also implies page uptodate */
933 struct buffer_head *bh, *head;
935 xfs_ioend_t *ioend = NULL, *iohead = NULL;
937 unsigned long p_offset = 0;
939 __uint64_t end_offset;
940 pgoff_t end_index, last_index, tlast;
942 int flags, err, iomap_valid = 0, uptodate = 1;
943 int page_dirty, count = 0;
945 int all_bh = unmapped;
948 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
949 trylock |= BMAPI_TRYLOCK;
952 /* Is this page beyond the end of the file? */
953 offset = i_size_read(inode);
954 end_index = offset >> PAGE_CACHE_SHIFT;
955 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
956 if (page->index >= end_index) {
957 if ((page->index >= end_index + 1) ||
958 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
966 * page_dirty is initially a count of buffers on the page before
967 * EOF and is decremented as we move each into a cleanable state.
971 * End offset is the highest offset that this page should represent.
972 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
973 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
974 * hence give us the correct page_dirty count. On any other page,
975 * it will be zero and in that case we need page_dirty to be the
976 * count of buffers on the page.
978 end_offset = min_t(unsigned long long,
979 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
980 len = 1 << inode->i_blkbits;
981 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
983 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
984 page_dirty = p_offset / len;
986 bh = head = page_buffers(page);
987 offset = page_offset(page);
991 /* TODO: cleanup count and page_dirty */
994 if (offset >= end_offset)
996 if (!buffer_uptodate(bh))
998 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
1000 * the iomap is actually still valid, but the ioend
1001 * isn't. shouldn't happen too often.
1008 iomap_valid = xfs_iomap_valid(&iomap, offset);
1011 * First case, map an unwritten extent and prepare for
1012 * extent state conversion transaction on completion.
1014 * Second case, allocate space for a delalloc buffer.
1015 * We can return EAGAIN here in the release page case.
1017 * Third case, an unmapped buffer was found, and we are
1018 * in a path where we need to write the whole page out.
1020 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1021 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1022 !buffer_mapped(bh) && (unmapped || startio))) {
1026 * Make sure we don't use a read-only iomap
1028 if (flags == BMAPI_READ)
1031 if (buffer_unwritten(bh)) {
1032 type = IOMAP_UNWRITTEN;
1033 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1034 } else if (buffer_delay(bh)) {
1036 flags = BMAPI_ALLOCATE | trylock;
1039 flags = BMAPI_WRITE | BMAPI_MMAP;
1044 * if we didn't have a valid mapping then we
1045 * need to ensure that we put the new mapping
1046 * in a new ioend structure. This needs to be
1047 * done to ensure that the ioends correctly
1048 * reflect the block mappings at io completion
1049 * for unwritten extent conversion.
1052 if (type == IOMAP_NEW) {
1053 size = xfs_probe_cluster(inode,
1059 err = xfs_map_blocks(inode, offset, size,
1063 iomap_valid = xfs_iomap_valid(&iomap, offset);
1066 xfs_map_at_offset(bh, offset,
1067 inode->i_blkbits, &iomap);
1069 xfs_add_to_ioend(inode, bh, offset,
1073 set_buffer_dirty(bh);
1075 mark_buffer_dirty(bh);
1080 } else if (buffer_uptodate(bh) && startio) {
1082 * we got here because the buffer is already mapped.
1083 * That means it must already have extents allocated
1084 * underneath it. Map the extent by reading it.
1086 if (!iomap_valid || flags != BMAPI_READ) {
1088 size = xfs_probe_cluster(inode, page, bh,
1090 err = xfs_map_blocks(inode, offset, size,
1094 iomap_valid = xfs_iomap_valid(&iomap, offset);
1098 * We set the type to IOMAP_NEW in case we are doing a
1099 * small write at EOF that is extending the file but
1100 * without needing an allocation. We need to update the
1101 * file size on I/O completion in this case so it is
1102 * the same case as having just allocated a new extent
1103 * that we are writing into for the first time.
1106 if (trylock_buffer(bh)) {
1107 ASSERT(buffer_mapped(bh));
1110 xfs_add_to_ioend(inode, bh, offset, type,
1111 &ioend, !iomap_valid);
1117 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1118 (unmapped || startio)) {
1125 } while (offset += len, ((bh = bh->b_this_page) != head));
1127 if (uptodate && bh == head)
1128 SetPageUptodate(page);
1131 xfs_start_page_writeback(page, 1, count);
1133 if (ioend && iomap_valid) {
1134 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1136 tlast = min_t(pgoff_t, offset, last_index);
1137 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1138 wbc, startio, all_bh, tlast);
1142 xfs_submit_ioend(iohead);
1148 xfs_cancel_ioend(iohead);
1151 * If it's delalloc and we have nowhere to put it,
1152 * throw it away, unless the lower layers told
1155 if (err != -EAGAIN) {
1157 block_invalidatepage(page, 0);
1158 ClearPageUptodate(page);
1164 * writepage: Called from one of two places:
1166 * 1. we are flushing a delalloc buffer head.
1168 * 2. we are writing out a dirty page. Typically the page dirty
1169 * state is cleared before we get here. In this case is it
1170 * conceivable we have no buffer heads.
1172 * For delalloc space on the page we need to allocate space and
1173 * flush it. For unmapped buffer heads on the page we should
1174 * allocate space if the page is uptodate. For any other dirty
1175 * buffer heads on the page we should flush them.
1177 * If we detect that a transaction would be required to flush
1178 * the page, we have to check the process flags first, if we
1179 * are already in a transaction or disk I/O during allocations
1180 * is off, we need to fail the writepage and redirty the page.
1186 struct writeback_control *wbc)
1190 int delalloc, unmapped, unwritten;
1191 struct inode *inode = page->mapping->host;
1193 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1196 * We need a transaction if:
1197 * 1. There are delalloc buffers on the page
1198 * 2. The page is uptodate and we have unmapped buffers
1199 * 3. The page is uptodate and we have no buffers
1200 * 4. There are unwritten buffers on the page
1203 if (!page_has_buffers(page)) {
1207 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1208 if (!PageUptodate(page))
1210 need_trans = delalloc + unmapped + unwritten;
1214 * If we need a transaction and the process flags say
1215 * we are already in a transaction, or no IO is allowed
1216 * then mark the page dirty again and leave the page
1219 if (current_test_flags(PF_FSTRANS) && need_trans)
1223 * Delay hooking up buffer heads until we have
1224 * made our go/no-go decision.
1226 if (!page_has_buffers(page))
1227 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1230 * Convert delayed allocate, unwritten or unmapped space
1231 * to real space and flush out to disk.
1233 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1234 if (error == -EAGAIN)
1236 if (unlikely(error < 0))
1242 redirty_page_for_writepage(wbc, page);
1252 struct address_space *mapping,
1253 struct writeback_control *wbc)
1255 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1256 return generic_writepages(mapping, wbc);
1260 * Called to move a page into cleanable state - and from there
1261 * to be released. Possibly the page is already clean. We always
1262 * have buffer heads in this call.
1264 * Returns 0 if the page is ok to release, 1 otherwise.
1266 * Possible scenarios are:
1268 * 1. We are being called to release a page which has been written
1269 * to via regular I/O. buffer heads will be dirty and possibly
1270 * delalloc. If no delalloc buffer heads in this case then we
1271 * can just return zero.
1273 * 2. We are called to release a page which has been written via
1274 * mmap, all we need to do is ensure there is no delalloc
1275 * state in the buffer heads, if not we can let the caller
1276 * free them and we should come back later via writepage.
1283 struct inode *inode = page->mapping->host;
1284 int dirty, delalloc, unmapped, unwritten;
1285 struct writeback_control wbc = {
1286 .sync_mode = WB_SYNC_ALL,
1290 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1292 if (!page_has_buffers(page))
1295 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1296 if (!delalloc && !unwritten)
1299 if (!(gfp_mask & __GFP_FS))
1302 /* If we are already inside a transaction or the thread cannot
1303 * do I/O, we cannot release this page.
1305 if (current_test_flags(PF_FSTRANS))
1309 * Convert delalloc space to real space, do not flush the
1310 * data out to disk, that will be done by the caller.
1311 * Never need to allocate space here - we will always
1312 * come back to writepage in that case.
1314 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1315 if (dirty == 0 && !unwritten)
1320 return try_to_free_buffers(page);
1325 struct inode *inode,
1327 struct buffer_head *bh_result,
1330 bmapi_flags_t flags)
1338 offset = (xfs_off_t)iblock << inode->i_blkbits;
1339 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1340 size = bh_result->b_size;
1342 if (!create && direct && offset >= i_size_read(inode))
1345 error = xfs_iomap(XFS_I(inode), offset, size,
1346 create ? flags : BMAPI_READ, &iomap, &niomap);
1352 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1354 * For unwritten extents do not report a disk address on
1355 * the read case (treat as if we're reading into a hole).
1357 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1358 xfs_map_buffer(bh_result, &iomap, offset,
1361 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1363 bh_result->b_private = inode;
1364 set_buffer_unwritten(bh_result);
1369 * If this is a realtime file, data may be on a different device.
1370 * to that pointed to from the buffer_head b_bdev currently.
1372 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1375 * If we previously allocated a block out beyond eof and we are now
1376 * coming back to use it then we will need to flag it as new even if it
1377 * has a disk address.
1379 * With sub-block writes into unwritten extents we also need to mark
1380 * the buffer as new so that the unwritten parts of the buffer gets
1384 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1385 (offset >= i_size_read(inode)) ||
1386 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1387 set_buffer_new(bh_result);
1389 if (iomap.iomap_flags & IOMAP_DELAY) {
1392 set_buffer_uptodate(bh_result);
1393 set_buffer_mapped(bh_result);
1394 set_buffer_delay(bh_result);
1398 if (direct || size > (1 << inode->i_blkbits)) {
1399 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1400 offset = min_t(xfs_off_t,
1401 iomap.iomap_bsize - iomap.iomap_delta, size);
1402 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1410 struct inode *inode,
1412 struct buffer_head *bh_result,
1415 return __xfs_get_blocks(inode, iblock,
1416 bh_result, create, 0, BMAPI_WRITE);
1420 xfs_get_blocks_direct(
1421 struct inode *inode,
1423 struct buffer_head *bh_result,
1426 return __xfs_get_blocks(inode, iblock,
1427 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1437 xfs_ioend_t *ioend = iocb->private;
1440 * Non-NULL private data means we need to issue a transaction to
1441 * convert a range from unwritten to written extents. This needs
1442 * to happen from process context but aio+dio I/O completion
1443 * happens from irq context so we need to defer it to a workqueue.
1444 * This is not necessary for synchronous direct I/O, but we do
1445 * it anyway to keep the code uniform and simpler.
1447 * Well, if only it were that simple. Because synchronous direct I/O
1448 * requires extent conversion to occur *before* we return to userspace,
1449 * we have to wait for extent conversion to complete. Look at the
1450 * iocb that has been passed to us to determine if this is AIO or
1451 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1452 * workqueue and wait for it to complete.
1454 * The core direct I/O code might be changed to always call the
1455 * completion handler in the future, in which case all this can
1458 ioend->io_offset = offset;
1459 ioend->io_size = size;
1460 if (ioend->io_type == IOMAP_READ) {
1461 xfs_finish_ioend(ioend, 0);
1462 } else if (private && size > 0) {
1463 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1466 * A direct I/O write ioend starts it's life in unwritten
1467 * state in case they map an unwritten extent. This write
1468 * didn't map an unwritten extent so switch it's completion
1471 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1472 xfs_finish_ioend(ioend, 0);
1476 * blockdev_direct_IO can return an error even after the I/O
1477 * completion handler was called. Thus we need to protect
1478 * against double-freeing.
1480 iocb->private = NULL;
1487 const struct iovec *iov,
1489 unsigned long nr_segs)
1491 struct file *file = iocb->ki_filp;
1492 struct inode *inode = file->f_mapping->host;
1493 struct block_device *bdev;
1496 bdev = xfs_find_bdev_for_inode(XFS_I(inode));
1499 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1500 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1501 bdev, iov, offset, nr_segs,
1502 xfs_get_blocks_direct,
1505 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1506 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1507 bdev, iov, offset, nr_segs,
1508 xfs_get_blocks_direct,
1512 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1513 xfs_destroy_ioend(iocb->private);
1520 struct address_space *mapping,
1524 struct page **pagep,
1528 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1534 struct address_space *mapping,
1537 struct inode *inode = (struct inode *)mapping->host;
1538 struct xfs_inode *ip = XFS_I(inode);
1540 xfs_itrace_entry(XFS_I(inode));
1541 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1542 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1543 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1544 return generic_block_bmap(mapping, block, xfs_get_blocks);
1549 struct file *unused,
1552 return mpage_readpage(page, xfs_get_blocks);
1557 struct file *unused,
1558 struct address_space *mapping,
1559 struct list_head *pages,
1562 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1566 xfs_vm_invalidatepage(
1568 unsigned long offset)
1570 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1571 page->mapping->host, page, offset);
1572 block_invalidatepage(page, offset);
1575 const struct address_space_operations xfs_address_space_operations = {
1576 .readpage = xfs_vm_readpage,
1577 .readpages = xfs_vm_readpages,
1578 .writepage = xfs_vm_writepage,
1579 .writepages = xfs_vm_writepages,
1580 .sync_page = block_sync_page,
1581 .releasepage = xfs_vm_releasepage,
1582 .invalidatepage = xfs_vm_invalidatepage,
1583 .write_begin = xfs_vm_write_begin,
1584 .write_end = generic_write_end,
1585 .bmap = xfs_vm_bmap,
1586 .direct_IO = xfs_vm_direct_IO,
1587 .migratepage = buffer_migrate_page,