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 bhv_vnode_t *vp = vn_from_inode(inode);
77 loff_t isize = i_size_read(inode);
78 loff_t offset = page_offset(page);
79 int delalloc = -1, unmapped = -1, unwritten = -1;
81 if (page_has_buffers(page))
82 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
88 ktrace_enter(ip->i_rwtrace,
89 (void *)((unsigned long)tag),
94 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
95 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
96 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
97 (void *)((unsigned long)(isize & 0xffffffff)),
98 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
99 (void *)((unsigned long)(offset & 0xffffffff)),
100 (void *)((unsigned long)delalloc),
101 (void *)((unsigned long)unmapped),
102 (void *)((unsigned long)unwritten),
103 (void *)((unsigned long)current_pid()),
107 #define xfs_page_trace(tag, inode, page, pgoff)
110 STATIC struct block_device *
111 xfs_find_bdev_for_inode(
112 struct xfs_inode *ip)
114 struct xfs_mount *mp = ip->i_mount;
116 if (XFS_IS_REALTIME_INODE(ip))
117 return mp->m_rtdev_targp->bt_bdev;
119 return mp->m_ddev_targp->bt_bdev;
123 * Schedule IO completion handling on a xfsdatad if this was
124 * the final hold on this ioend. If we are asked to wait,
125 * flush the workqueue.
132 if (atomic_dec_and_test(&ioend->io_remaining)) {
133 queue_work(xfsdatad_workqueue, &ioend->io_work);
135 flush_workqueue(xfsdatad_workqueue);
140 * We're now finished for good with this ioend structure.
141 * Update the page state via the associated buffer_heads,
142 * release holds on the inode and bio, and finally free
143 * up memory. Do not use the ioend after this.
149 struct buffer_head *bh, *next;
151 for (bh = ioend->io_buffer_head; bh; bh = next) {
152 next = bh->b_private;
153 bh->b_end_io(bh, !ioend->io_error);
155 if (unlikely(ioend->io_error)) {
156 vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
159 vn_iowake(XFS_I(ioend->io_inode));
160 mempool_free(ioend, xfs_ioend_pool);
164 * Update on-disk file size now that data has been written to disk.
165 * The current in-memory file size is i_size. If a write is beyond
166 * eof i_new_size will be the intended file size until i_size is
167 * updated. If this write does not extend all the way to the valid
168 * file size then restrict this update to the end of the write.
174 xfs_inode_t *ip = XFS_I(ioend->io_inode);
178 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
179 ASSERT(ioend->io_type != IOMAP_READ);
181 if (unlikely(ioend->io_error))
184 bsize = ioend->io_offset + ioend->io_size;
186 xfs_ilock(ip, XFS_ILOCK_EXCL);
188 isize = MAX(ip->i_size, ip->i_new_size);
189 isize = MIN(isize, bsize);
191 if (ip->i_d.di_size < isize) {
192 ip->i_d.di_size = isize;
193 ip->i_update_core = 1;
194 ip->i_update_size = 1;
195 mark_inode_dirty_sync(ioend->io_inode);
198 xfs_iunlock(ip, XFS_ILOCK_EXCL);
202 * Buffered IO write completion for delayed allocate extents.
205 xfs_end_bio_delalloc(
206 struct work_struct *work)
209 container_of(work, xfs_ioend_t, io_work);
211 xfs_setfilesize(ioend);
212 xfs_destroy_ioend(ioend);
216 * Buffered IO write completion for regular, written extents.
220 struct work_struct *work)
223 container_of(work, xfs_ioend_t, io_work);
225 xfs_setfilesize(ioend);
226 xfs_destroy_ioend(ioend);
230 * IO write completion for unwritten extents.
232 * Issue transactions to convert a buffer range from unwritten
233 * to written extents.
236 xfs_end_bio_unwritten(
237 struct work_struct *work)
240 container_of(work, xfs_ioend_t, io_work);
241 struct xfs_inode *ip = XFS_I(ioend->io_inode);
242 xfs_off_t offset = ioend->io_offset;
243 size_t size = ioend->io_size;
245 if (likely(!ioend->io_error)) {
246 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
248 error = xfs_iomap_write_unwritten(ip, offset, size);
250 ioend->io_error = error;
252 xfs_setfilesize(ioend);
254 xfs_destroy_ioend(ioend);
258 * IO read completion for regular, written extents.
262 struct work_struct *work)
265 container_of(work, xfs_ioend_t, io_work);
267 xfs_destroy_ioend(ioend);
271 * Allocate and initialise an IO completion structure.
272 * We need to track unwritten extent write completion here initially.
273 * We'll need to extend this for updating the ondisk inode size later
283 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
286 * Set the count to 1 initially, which will prevent an I/O
287 * completion callback from happening before we have started
288 * all the I/O from calling the completion routine too early.
290 atomic_set(&ioend->io_remaining, 1);
292 ioend->io_list = NULL;
293 ioend->io_type = type;
294 ioend->io_inode = inode;
295 ioend->io_buffer_head = NULL;
296 ioend->io_buffer_tail = NULL;
297 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
298 ioend->io_offset = 0;
301 if (type == IOMAP_UNWRITTEN)
302 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
303 else if (type == IOMAP_DELAY)
304 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
305 else if (type == IOMAP_READ)
306 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
308 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
321 xfs_inode_t *ip = XFS_I(inode);
322 int error, nmaps = 1;
324 error = xfs_iomap(ip, offset, count,
325 flags, mapp, &nmaps);
326 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
327 xfs_iflags_set(ip, XFS_IMODIFIED);
336 return offset >= iomapp->iomap_offset &&
337 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
341 * BIO completion handler for buffered IO.
348 xfs_ioend_t *ioend = bio->bi_private;
350 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
351 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
353 /* Toss bio and pass work off to an xfsdatad thread */
354 bio->bi_private = NULL;
355 bio->bi_end_io = NULL;
358 xfs_finish_ioend(ioend, 0);
362 xfs_submit_ioend_bio(
366 atomic_inc(&ioend->io_remaining);
368 bio->bi_private = ioend;
369 bio->bi_end_io = xfs_end_bio;
371 submit_bio(WRITE, bio);
372 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
378 struct buffer_head *bh)
381 int nvecs = bio_get_nr_vecs(bh->b_bdev);
384 bio = bio_alloc(GFP_NOIO, nvecs);
388 ASSERT(bio->bi_private == NULL);
389 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
390 bio->bi_bdev = bh->b_bdev;
396 xfs_start_buffer_writeback(
397 struct buffer_head *bh)
399 ASSERT(buffer_mapped(bh));
400 ASSERT(buffer_locked(bh));
401 ASSERT(!buffer_delay(bh));
402 ASSERT(!buffer_unwritten(bh));
404 mark_buffer_async_write(bh);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
410 xfs_start_page_writeback(
415 ASSERT(PageLocked(page));
416 ASSERT(!PageWriteback(page));
418 clear_page_dirty_for_io(page);
419 set_page_writeback(page);
421 /* If no buffers on the page are to be written, finish it here */
423 end_page_writeback(page);
426 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
428 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
432 * Submit all of the bios for all of the ioends we have saved up, covering the
433 * initial writepage page and also any probed pages.
435 * Because we may have multiple ioends spanning a page, we need to start
436 * writeback on all the buffers before we submit them for I/O. If we mark the
437 * buffers as we got, then we can end up with a page that only has buffers
438 * marked async write and I/O complete on can occur before we mark the other
439 * buffers async write.
441 * The end result of this is that we trip a bug in end_page_writeback() because
442 * we call it twice for the one page as the code in end_buffer_async_write()
443 * assumes that all buffers on the page are started at the same time.
445 * The fix is two passes across the ioend list - one to start writeback on the
446 * buffer_heads, and then submit them for I/O on the second pass.
452 xfs_ioend_t *head = ioend;
454 struct buffer_head *bh;
456 sector_t lastblock = 0;
458 /* Pass 1 - start writeback */
460 next = ioend->io_list;
461 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
462 xfs_start_buffer_writeback(bh);
464 } while ((ioend = next) != NULL);
466 /* Pass 2 - submit I/O */
469 next = ioend->io_list;
472 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
476 bio = xfs_alloc_ioend_bio(bh);
477 } else if (bh->b_blocknr != lastblock + 1) {
478 xfs_submit_ioend_bio(ioend, bio);
482 if (bio_add_buffer(bio, bh) != bh->b_size) {
483 xfs_submit_ioend_bio(ioend, bio);
487 lastblock = bh->b_blocknr;
490 xfs_submit_ioend_bio(ioend, bio);
491 xfs_finish_ioend(ioend, 0);
492 } while ((ioend = next) != NULL);
496 * Cancel submission of all buffer_heads so far in this endio.
497 * Toss the endio too. Only ever called for the initial page
498 * in a writepage request, so only ever one page.
505 struct buffer_head *bh, *next_bh;
508 next = ioend->io_list;
509 bh = ioend->io_buffer_head;
511 next_bh = bh->b_private;
512 clear_buffer_async_write(bh);
514 } while ((bh = next_bh) != NULL);
516 vn_iowake(XFS_I(ioend->io_inode));
517 mempool_free(ioend, xfs_ioend_pool);
518 } while ((ioend = next) != NULL);
522 * Test to see if we've been building up a completion structure for
523 * earlier buffers -- if so, we try to append to this ioend if we
524 * can, otherwise we finish off any current ioend and start another.
525 * Return true if we've finished the given ioend.
530 struct buffer_head *bh,
533 xfs_ioend_t **result,
536 xfs_ioend_t *ioend = *result;
538 if (!ioend || need_ioend || type != ioend->io_type) {
539 xfs_ioend_t *previous = *result;
541 ioend = xfs_alloc_ioend(inode, type);
542 ioend->io_offset = offset;
543 ioend->io_buffer_head = bh;
544 ioend->io_buffer_tail = bh;
546 previous->io_list = ioend;
549 ioend->io_buffer_tail->b_private = bh;
550 ioend->io_buffer_tail = bh;
553 bh->b_private = NULL;
554 ioend->io_size += bh->b_size;
559 struct buffer_head *bh,
566 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
568 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
569 ((offset - mp->iomap_offset) >> block_bits);
571 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
574 set_buffer_mapped(bh);
579 struct buffer_head *bh,
584 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
585 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
588 xfs_map_buffer(bh, iomapp, offset, block_bits);
589 bh->b_bdev = iomapp->iomap_target->bt_bdev;
590 set_buffer_mapped(bh);
591 clear_buffer_delay(bh);
592 clear_buffer_unwritten(bh);
596 * Look for a page at index that is suitable for clustering.
601 unsigned int pg_offset,
606 if (PageWriteback(page))
609 if (page->mapping && PageDirty(page)) {
610 if (page_has_buffers(page)) {
611 struct buffer_head *bh, *head;
613 bh = head = page_buffers(page);
615 if (!buffer_uptodate(bh))
617 if (mapped != buffer_mapped(bh))
620 if (ret >= pg_offset)
622 } while ((bh = bh->b_this_page) != head);
624 ret = mapped ? 0 : PAGE_CACHE_SIZE;
633 struct page *startpage,
634 struct buffer_head *bh,
635 struct buffer_head *head,
639 pgoff_t tindex, tlast, tloff;
643 /* First sum forwards in this page */
645 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
648 } while ((bh = bh->b_this_page) != head);
650 /* if we reached the end of the page, sum forwards in following pages */
651 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
652 tindex = startpage->index + 1;
654 /* Prune this back to avoid pathological behavior */
655 tloff = min(tlast, startpage->index + 64);
657 pagevec_init(&pvec, 0);
658 while (!done && tindex <= tloff) {
659 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
661 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
664 for (i = 0; i < pagevec_count(&pvec); i++) {
665 struct page *page = pvec.pages[i];
666 size_t pg_offset, pg_len = 0;
668 if (tindex == tlast) {
670 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
676 pg_offset = PAGE_CACHE_SIZE;
678 if (page->index == tindex && !TestSetPageLocked(page)) {
679 pg_len = xfs_probe_page(page, pg_offset, mapped);
692 pagevec_release(&pvec);
700 * Test if a given page is suitable for writing as part of an unwritten
701 * or delayed allocate extent.
708 if (PageWriteback(page))
711 if (page->mapping && page_has_buffers(page)) {
712 struct buffer_head *bh, *head;
715 bh = head = page_buffers(page);
717 if (buffer_unwritten(bh))
718 acceptable = (type == IOMAP_UNWRITTEN);
719 else if (buffer_delay(bh))
720 acceptable = (type == IOMAP_DELAY);
721 else if (buffer_dirty(bh) && buffer_mapped(bh))
722 acceptable = (type == IOMAP_NEW);
725 } while ((bh = bh->b_this_page) != head);
735 * Allocate & map buffers for page given the extent map. Write it out.
736 * except for the original page of a writepage, this is called on
737 * delalloc/unwritten pages only, for the original page it is possible
738 * that the page has no mapping at all.
746 xfs_ioend_t **ioendp,
747 struct writeback_control *wbc,
751 struct buffer_head *bh, *head;
752 xfs_off_t end_offset;
753 unsigned long p_offset;
755 int bbits = inode->i_blkbits;
757 int count = 0, done = 0, uptodate = 1;
758 xfs_off_t offset = page_offset(page);
760 if (page->index != tindex)
762 if (TestSetPageLocked(page))
764 if (PageWriteback(page))
765 goto fail_unlock_page;
766 if (page->mapping != inode->i_mapping)
767 goto fail_unlock_page;
768 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
769 goto fail_unlock_page;
772 * page_dirty is initially a count of buffers on the page before
773 * EOF and is decremented as we move each into a cleanable state.
777 * End offset is the highest offset that this page should represent.
778 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
779 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
780 * hence give us the correct page_dirty count. On any other page,
781 * it will be zero and in that case we need page_dirty to be the
782 * count of buffers on the page.
784 end_offset = min_t(unsigned long long,
785 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
788 len = 1 << inode->i_blkbits;
789 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
791 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
792 page_dirty = p_offset / len;
794 bh = head = page_buffers(page);
796 if (offset >= end_offset)
798 if (!buffer_uptodate(bh))
800 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
805 if (buffer_unwritten(bh) || buffer_delay(bh)) {
806 if (buffer_unwritten(bh))
807 type = IOMAP_UNWRITTEN;
811 if (!xfs_iomap_valid(mp, offset)) {
816 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
817 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
819 xfs_map_at_offset(bh, offset, bbits, mp);
821 xfs_add_to_ioend(inode, bh, offset,
824 set_buffer_dirty(bh);
826 mark_buffer_dirty(bh);
832 if (buffer_mapped(bh) && all_bh && startio) {
834 xfs_add_to_ioend(inode, bh, offset,
842 } while (offset += len, (bh = bh->b_this_page) != head);
844 if (uptodate && bh == head)
845 SetPageUptodate(page);
849 struct backing_dev_info *bdi;
851 bdi = inode->i_mapping->backing_dev_info;
853 if (bdi_write_congested(bdi)) {
854 wbc->encountered_congestion = 1;
856 } else if (wbc->nr_to_write <= 0) {
860 xfs_start_page_writeback(page, !page_dirty, count);
871 * Convert & write out a cluster of pages in the same extent as defined
872 * by mp and following the start page.
879 xfs_ioend_t **ioendp,
880 struct writeback_control *wbc,
888 pagevec_init(&pvec, 0);
889 while (!done && tindex <= tlast) {
890 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
892 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
895 for (i = 0; i < pagevec_count(&pvec); i++) {
896 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
897 iomapp, ioendp, wbc, startio, all_bh);
902 pagevec_release(&pvec);
908 * Calling this without startio set means we are being asked to make a dirty
909 * page ready for freeing it's buffers. When called with startio set then
910 * we are coming from writepage.
912 * When called with startio set it is important that we write the WHOLE
914 * The bh->b_state's cannot know if any of the blocks or which block for
915 * that matter are dirty due to mmap writes, and therefore bh uptodate is
916 * only valid if the page itself isn't completely uptodate. Some layers
917 * may clear the page dirty flag prior to calling write page, under the
918 * assumption the entire page will be written out; by not writing out the
919 * whole page the page can be reused before all valid dirty data is
920 * written out. Note: in the case of a page that has been dirty'd by
921 * mapwrite and but partially setup by block_prepare_write the
922 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
923 * valid state, thus the whole page must be written out thing.
927 xfs_page_state_convert(
930 struct writeback_control *wbc,
932 int unmapped) /* also implies page uptodate */
934 struct buffer_head *bh, *head;
936 xfs_ioend_t *ioend = NULL, *iohead = NULL;
938 unsigned long p_offset = 0;
940 __uint64_t end_offset;
941 pgoff_t end_index, last_index, tlast;
943 int flags, err, iomap_valid = 0, uptodate = 1;
944 int page_dirty, count = 0;
946 int all_bh = unmapped;
949 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
950 trylock |= BMAPI_TRYLOCK;
953 /* Is this page beyond the end of the file? */
954 offset = i_size_read(inode);
955 end_index = offset >> PAGE_CACHE_SHIFT;
956 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
957 if (page->index >= end_index) {
958 if ((page->index >= end_index + 1) ||
959 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
967 * page_dirty is initially a count of buffers on the page before
968 * EOF and is decremented as we move each into a cleanable state.
972 * End offset is the highest offset that this page should represent.
973 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
974 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
975 * hence give us the correct page_dirty count. On any other page,
976 * it will be zero and in that case we need page_dirty to be the
977 * count of buffers on the page.
979 end_offset = min_t(unsigned long long,
980 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
981 len = 1 << inode->i_blkbits;
982 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
984 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
985 page_dirty = p_offset / len;
987 bh = head = page_buffers(page);
988 offset = page_offset(page);
992 /* TODO: cleanup count and page_dirty */
995 if (offset >= end_offset)
997 if (!buffer_uptodate(bh))
999 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
1001 * the iomap is actually still valid, but the ioend
1002 * isn't. shouldn't happen too often.
1009 iomap_valid = xfs_iomap_valid(&iomap, offset);
1012 * First case, map an unwritten extent and prepare for
1013 * extent state conversion transaction on completion.
1015 * Second case, allocate space for a delalloc buffer.
1016 * We can return EAGAIN here in the release page case.
1018 * Third case, an unmapped buffer was found, and we are
1019 * in a path where we need to write the whole page out.
1021 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1022 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1023 !buffer_mapped(bh) && (unmapped || startio))) {
1027 * Make sure we don't use a read-only iomap
1029 if (flags == BMAPI_READ)
1032 if (buffer_unwritten(bh)) {
1033 type = IOMAP_UNWRITTEN;
1034 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1035 } else if (buffer_delay(bh)) {
1037 flags = BMAPI_ALLOCATE | trylock;
1040 flags = BMAPI_WRITE | BMAPI_MMAP;
1045 * if we didn't have a valid mapping then we
1046 * need to ensure that we put the new mapping
1047 * in a new ioend structure. This needs to be
1048 * done to ensure that the ioends correctly
1049 * reflect the block mappings at io completion
1050 * for unwritten extent conversion.
1053 if (type == IOMAP_NEW) {
1054 size = xfs_probe_cluster(inode,
1060 err = xfs_map_blocks(inode, offset, size,
1064 iomap_valid = xfs_iomap_valid(&iomap, offset);
1067 xfs_map_at_offset(bh, offset,
1068 inode->i_blkbits, &iomap);
1070 xfs_add_to_ioend(inode, bh, offset,
1074 set_buffer_dirty(bh);
1076 mark_buffer_dirty(bh);
1081 } else if (buffer_uptodate(bh) && startio) {
1083 * we got here because the buffer is already mapped.
1084 * That means it must already have extents allocated
1085 * underneath it. Map the extent by reading it.
1087 if (!iomap_valid || flags != BMAPI_READ) {
1089 size = xfs_probe_cluster(inode, page, bh,
1091 err = xfs_map_blocks(inode, offset, size,
1095 iomap_valid = xfs_iomap_valid(&iomap, offset);
1099 * We set the type to IOMAP_NEW in case we are doing a
1100 * small write at EOF that is extending the file but
1101 * without needing an allocation. We need to update the
1102 * file size on I/O completion in this case so it is
1103 * the same case as having just allocated a new extent
1104 * that we are writing into for the first time.
1107 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1108 ASSERT(buffer_mapped(bh));
1111 xfs_add_to_ioend(inode, bh, offset, type,
1112 &ioend, !iomap_valid);
1118 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1119 (unmapped || startio)) {
1126 } while (offset += len, ((bh = bh->b_this_page) != head));
1128 if (uptodate && bh == head)
1129 SetPageUptodate(page);
1132 xfs_start_page_writeback(page, 1, count);
1134 if (ioend && iomap_valid) {
1135 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1137 tlast = min_t(pgoff_t, offset, last_index);
1138 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1139 wbc, startio, all_bh, tlast);
1143 xfs_submit_ioend(iohead);
1149 xfs_cancel_ioend(iohead);
1152 * If it's delalloc and we have nowhere to put it,
1153 * throw it away, unless the lower layers told
1156 if (err != -EAGAIN) {
1158 block_invalidatepage(page, 0);
1159 ClearPageUptodate(page);
1165 * writepage: Called from one of two places:
1167 * 1. we are flushing a delalloc buffer head.
1169 * 2. we are writing out a dirty page. Typically the page dirty
1170 * state is cleared before we get here. In this case is it
1171 * conceivable we have no buffer heads.
1173 * For delalloc space on the page we need to allocate space and
1174 * flush it. For unmapped buffer heads on the page we should
1175 * allocate space if the page is uptodate. For any other dirty
1176 * buffer heads on the page we should flush them.
1178 * If we detect that a transaction would be required to flush
1179 * the page, we have to check the process flags first, if we
1180 * are already in a transaction or disk I/O during allocations
1181 * is off, we need to fail the writepage and redirty the page.
1187 struct writeback_control *wbc)
1191 int delalloc, unmapped, unwritten;
1192 struct inode *inode = page->mapping->host;
1194 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1197 * We need a transaction if:
1198 * 1. There are delalloc buffers on the page
1199 * 2. The page is uptodate and we have unmapped buffers
1200 * 3. The page is uptodate and we have no buffers
1201 * 4. There are unwritten buffers on the page
1204 if (!page_has_buffers(page)) {
1208 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1209 if (!PageUptodate(page))
1211 need_trans = delalloc + unmapped + unwritten;
1215 * If we need a transaction and the process flags say
1216 * we are already in a transaction, or no IO is allowed
1217 * then mark the page dirty again and leave the page
1220 if (current_test_flags(PF_FSTRANS) && need_trans)
1224 * Delay hooking up buffer heads until we have
1225 * made our go/no-go decision.
1227 if (!page_has_buffers(page))
1228 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1231 * Convert delayed allocate, unwritten or unmapped space
1232 * to real space and flush out to disk.
1234 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1235 if (error == -EAGAIN)
1237 if (unlikely(error < 0))
1243 redirty_page_for_writepage(wbc, page);
1253 struct address_space *mapping,
1254 struct writeback_control *wbc)
1256 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1257 return generic_writepages(mapping, wbc);
1261 * Called to move a page into cleanable state - and from there
1262 * to be released. Possibly the page is already clean. We always
1263 * have buffer heads in this call.
1265 * Returns 0 if the page is ok to release, 1 otherwise.
1267 * Possible scenarios are:
1269 * 1. We are being called to release a page which has been written
1270 * to via regular I/O. buffer heads will be dirty and possibly
1271 * delalloc. If no delalloc buffer heads in this case then we
1272 * can just return zero.
1274 * 2. We are called to release a page which has been written via
1275 * mmap, all we need to do is ensure there is no delalloc
1276 * state in the buffer heads, if not we can let the caller
1277 * free them and we should come back later via writepage.
1284 struct inode *inode = page->mapping->host;
1285 int dirty, delalloc, unmapped, unwritten;
1286 struct writeback_control wbc = {
1287 .sync_mode = WB_SYNC_ALL,
1291 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1293 if (!page_has_buffers(page))
1296 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1297 if (!delalloc && !unwritten)
1300 if (!(gfp_mask & __GFP_FS))
1303 /* If we are already inside a transaction or the thread cannot
1304 * do I/O, we cannot release this page.
1306 if (current_test_flags(PF_FSTRANS))
1310 * Convert delalloc space to real space, do not flush the
1311 * data out to disk, that will be done by the caller.
1312 * Never need to allocate space here - we will always
1313 * come back to writepage in that case.
1315 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1316 if (dirty == 0 && !unwritten)
1321 return try_to_free_buffers(page);
1326 struct inode *inode,
1328 struct buffer_head *bh_result,
1331 bmapi_flags_t flags)
1339 offset = (xfs_off_t)iblock << inode->i_blkbits;
1340 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1341 size = bh_result->b_size;
1342 error = xfs_iomap(XFS_I(inode), offset, size,
1343 create ? flags : BMAPI_READ, &iomap, &niomap);
1349 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1351 * For unwritten extents do not report a disk address on
1352 * the read case (treat as if we're reading into a hole).
1354 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1355 xfs_map_buffer(bh_result, &iomap, offset,
1358 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1360 bh_result->b_private = inode;
1361 set_buffer_unwritten(bh_result);
1366 * If this is a realtime file, data may be on a different device.
1367 * to that pointed to from the buffer_head b_bdev currently.
1369 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1372 * If we previously allocated a block out beyond eof and we are now
1373 * coming back to use it then we will need to flag it as new even if it
1374 * has a disk address.
1376 * With sub-block writes into unwritten extents we also need to mark
1377 * the buffer as new so that the unwritten parts of the buffer gets
1381 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1382 (offset >= i_size_read(inode)) ||
1383 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1384 set_buffer_new(bh_result);
1386 if (iomap.iomap_flags & IOMAP_DELAY) {
1389 set_buffer_uptodate(bh_result);
1390 set_buffer_mapped(bh_result);
1391 set_buffer_delay(bh_result);
1395 if (direct || size > (1 << inode->i_blkbits)) {
1396 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1397 offset = min_t(xfs_off_t,
1398 iomap.iomap_bsize - iomap.iomap_delta, size);
1399 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1407 struct inode *inode,
1409 struct buffer_head *bh_result,
1412 return __xfs_get_blocks(inode, iblock,
1413 bh_result, create, 0, BMAPI_WRITE);
1417 xfs_get_blocks_direct(
1418 struct inode *inode,
1420 struct buffer_head *bh_result,
1423 return __xfs_get_blocks(inode, iblock,
1424 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1434 xfs_ioend_t *ioend = iocb->private;
1437 * Non-NULL private data means we need to issue a transaction to
1438 * convert a range from unwritten to written extents. This needs
1439 * to happen from process context but aio+dio I/O completion
1440 * happens from irq context so we need to defer it to a workqueue.
1441 * This is not necessary for synchronous direct I/O, but we do
1442 * it anyway to keep the code uniform and simpler.
1444 * Well, if only it were that simple. Because synchronous direct I/O
1445 * requires extent conversion to occur *before* we return to userspace,
1446 * we have to wait for extent conversion to complete. Look at the
1447 * iocb that has been passed to us to determine if this is AIO or
1448 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1449 * workqueue and wait for it to complete.
1451 * The core direct I/O code might be changed to always call the
1452 * completion handler in the future, in which case all this can
1455 ioend->io_offset = offset;
1456 ioend->io_size = size;
1457 if (ioend->io_type == IOMAP_READ) {
1458 xfs_finish_ioend(ioend, 0);
1459 } else if (private && size > 0) {
1460 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1463 * A direct I/O write ioend starts it's life in unwritten
1464 * state in case they map an unwritten extent. This write
1465 * didn't map an unwritten extent so switch it's completion
1468 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1469 xfs_finish_ioend(ioend, 0);
1473 * blockdev_direct_IO can return an error even after the I/O
1474 * completion handler was called. Thus we need to protect
1475 * against double-freeing.
1477 iocb->private = NULL;
1484 const struct iovec *iov,
1486 unsigned long nr_segs)
1488 struct file *file = iocb->ki_filp;
1489 struct inode *inode = file->f_mapping->host;
1490 struct block_device *bdev;
1493 bdev = xfs_find_bdev_for_inode(XFS_I(inode));
1496 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1497 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1498 bdev, iov, offset, nr_segs,
1499 xfs_get_blocks_direct,
1502 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1503 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1504 bdev, iov, offset, nr_segs,
1505 xfs_get_blocks_direct,
1509 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1510 xfs_destroy_ioend(iocb->private);
1517 struct address_space *mapping,
1521 struct page **pagep,
1525 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1531 struct address_space *mapping,
1534 struct inode *inode = (struct inode *)mapping->host;
1535 struct xfs_inode *ip = XFS_I(inode);
1537 xfs_itrace_entry(XFS_I(inode));
1538 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1539 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1540 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1541 return generic_block_bmap(mapping, block, xfs_get_blocks);
1546 struct file *unused,
1549 return mpage_readpage(page, xfs_get_blocks);
1554 struct file *unused,
1555 struct address_space *mapping,
1556 struct list_head *pages,
1559 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1563 xfs_vm_invalidatepage(
1565 unsigned long offset)
1567 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1568 page->mapping->host, page, offset);
1569 block_invalidatepage(page, offset);
1572 const struct address_space_operations xfs_address_space_operations = {
1573 .readpage = xfs_vm_readpage,
1574 .readpages = xfs_vm_readpages,
1575 .writepage = xfs_vm_writepage,
1576 .writepages = xfs_vm_writepages,
1577 .sync_page = block_sync_page,
1578 .releasepage = xfs_vm_releasepage,
1579 .invalidatepage = xfs_vm_invalidatepage,
1580 .write_begin = xfs_vm_write_begin,
1581 .write_end = generic_write_end,
1582 .bmap = xfs_vm_bmap,
1583 .direct_IO = xfs_vm_direct_IO,
1584 .migratepage = buffer_migrate_page,