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 <linux/mpage.h>
41 #include <linux/pagevec.h>
42 #include <linux/writeback.h>
51 struct buffer_head *bh, *head;
53 *delalloc = *unmapped = *unwritten = 0;
55 bh = head = page_buffers(page);
57 if (buffer_uptodate(bh) && !buffer_mapped(bh))
59 else if (buffer_unwritten(bh) && !buffer_delay(bh))
60 clear_buffer_unwritten(bh);
61 else if (buffer_unwritten(bh))
63 else if (buffer_delay(bh))
65 } while ((bh = bh->b_this_page) != head);
68 #if defined(XFS_RW_TRACE)
77 bhv_vnode_t *vp = vn_from_inode(inode);
78 loff_t isize = i_size_read(inode);
79 loff_t offset = page_offset(page);
80 int delalloc = -1, unmapped = -1, unwritten = -1;
82 if (page_has_buffers(page))
83 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
89 ktrace_enter(ip->i_rwtrace,
90 (void *)((unsigned long)tag),
94 (void *)((unsigned long)mask),
95 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
96 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
97 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(isize & 0xffffffff)),
99 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
100 (void *)((unsigned long)(offset & 0xffffffff)),
101 (void *)((unsigned long)delalloc),
102 (void *)((unsigned long)unmapped),
103 (void *)((unsigned long)unwritten),
104 (void *)((unsigned long)current_pid()),
108 #define xfs_page_trace(tag, inode, page, mask)
112 * Schedule IO completion handling on a xfsdatad if this was
113 * the final hold on this ioend.
119 if (atomic_dec_and_test(&ioend->io_remaining))
120 queue_work(xfsdatad_workqueue, &ioend->io_work);
124 * We're now finished for good with this ioend structure.
125 * Update the page state via the associated buffer_heads,
126 * release holds on the inode and bio, and finally free
127 * up memory. Do not use the ioend after this.
133 struct buffer_head *bh, *next;
135 for (bh = ioend->io_buffer_head; bh; bh = next) {
136 next = bh->b_private;
137 bh->b_end_io(bh, !ioend->io_error);
139 if (unlikely(ioend->io_error))
140 vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__);
141 vn_iowake(ioend->io_vnode);
142 mempool_free(ioend, xfs_ioend_pool);
146 * Buffered IO write completion for delayed allocate extents.
147 * TODO: Update ondisk isize now that we know the file data
148 * has been flushed (i.e. the notorious "NULL file" problem).
151 xfs_end_bio_delalloc(
154 xfs_ioend_t *ioend = data;
156 xfs_destroy_ioend(ioend);
160 * Buffered IO write completion for regular, written extents.
166 xfs_ioend_t *ioend = data;
168 xfs_destroy_ioend(ioend);
172 * IO write completion for unwritten extents.
174 * Issue transactions to convert a buffer range from unwritten
175 * to written extents.
178 xfs_end_bio_unwritten(
181 xfs_ioend_t *ioend = data;
182 bhv_vnode_t *vp = ioend->io_vnode;
183 xfs_off_t offset = ioend->io_offset;
184 size_t size = ioend->io_size;
186 if (likely(!ioend->io_error))
187 bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL);
188 xfs_destroy_ioend(ioend);
192 * Allocate and initialise an IO completion structure.
193 * We need to track unwritten extent write completion here initially.
194 * We'll need to extend this for updating the ondisk inode size later
204 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
207 * Set the count to 1 initially, which will prevent an I/O
208 * completion callback from happening before we have started
209 * all the I/O from calling the completion routine too early.
211 atomic_set(&ioend->io_remaining, 1);
213 ioend->io_list = NULL;
214 ioend->io_type = type;
215 ioend->io_vnode = vn_from_inode(inode);
216 ioend->io_buffer_head = NULL;
217 ioend->io_buffer_tail = NULL;
218 atomic_inc(&ioend->io_vnode->v_iocount);
219 ioend->io_offset = 0;
222 if (type == IOMAP_UNWRITTEN)
223 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
224 else if (type == IOMAP_DELAY)
225 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend);
227 INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend);
240 bhv_vnode_t *vp = vn_from_inode(inode);
241 int error, nmaps = 1;
243 error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps);
244 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
254 return offset >= iomapp->iomap_offset &&
255 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
259 * BIO completion handler for buffered IO.
264 unsigned int bytes_done,
267 xfs_ioend_t *ioend = bio->bi_private;
272 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
273 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
275 /* Toss bio and pass work off to an xfsdatad thread */
276 bio->bi_private = NULL;
277 bio->bi_end_io = NULL;
280 xfs_finish_ioend(ioend);
285 xfs_submit_ioend_bio(
289 atomic_inc(&ioend->io_remaining);
291 bio->bi_private = ioend;
292 bio->bi_end_io = xfs_end_bio;
294 submit_bio(WRITE, bio);
295 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
301 struct buffer_head *bh)
304 int nvecs = bio_get_nr_vecs(bh->b_bdev);
307 bio = bio_alloc(GFP_NOIO, nvecs);
311 ASSERT(bio->bi_private == NULL);
312 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
313 bio->bi_bdev = bh->b_bdev;
319 xfs_start_buffer_writeback(
320 struct buffer_head *bh)
322 ASSERT(buffer_mapped(bh));
323 ASSERT(buffer_locked(bh));
324 ASSERT(!buffer_delay(bh));
325 ASSERT(!buffer_unwritten(bh));
327 mark_buffer_async_write(bh);
328 set_buffer_uptodate(bh);
329 clear_buffer_dirty(bh);
333 xfs_start_page_writeback(
335 struct writeback_control *wbc,
339 ASSERT(PageLocked(page));
340 ASSERT(!PageWriteback(page));
341 set_page_writeback(page);
343 clear_page_dirty(page);
346 end_page_writeback(page);
347 wbc->pages_skipped++; /* We didn't write this page */
351 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
353 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
357 * Submit all of the bios for all of the ioends we have saved up, covering the
358 * initial writepage page and also any probed pages.
360 * Because we may have multiple ioends spanning a page, we need to start
361 * writeback on all the buffers before we submit them for I/O. If we mark the
362 * buffers as we got, then we can end up with a page that only has buffers
363 * marked async write and I/O complete on can occur before we mark the other
364 * buffers async write.
366 * The end result of this is that we trip a bug in end_page_writeback() because
367 * we call it twice for the one page as the code in end_buffer_async_write()
368 * assumes that all buffers on the page are started at the same time.
370 * The fix is two passes across the ioend list - one to start writeback on the
371 * buffer_heads, and then submit them for I/O on the second pass.
377 xfs_ioend_t *head = ioend;
379 struct buffer_head *bh;
381 sector_t lastblock = 0;
383 /* Pass 1 - start writeback */
385 next = ioend->io_list;
386 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
387 xfs_start_buffer_writeback(bh);
389 } while ((ioend = next) != NULL);
391 /* Pass 2 - submit I/O */
394 next = ioend->io_list;
397 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
401 bio = xfs_alloc_ioend_bio(bh);
402 } else if (bh->b_blocknr != lastblock + 1) {
403 xfs_submit_ioend_bio(ioend, bio);
407 if (bio_add_buffer(bio, bh) != bh->b_size) {
408 xfs_submit_ioend_bio(ioend, bio);
412 lastblock = bh->b_blocknr;
415 xfs_submit_ioend_bio(ioend, bio);
416 xfs_finish_ioend(ioend);
417 } while ((ioend = next) != NULL);
421 * Cancel submission of all buffer_heads so far in this endio.
422 * Toss the endio too. Only ever called for the initial page
423 * in a writepage request, so only ever one page.
430 struct buffer_head *bh, *next_bh;
433 next = ioend->io_list;
434 bh = ioend->io_buffer_head;
436 next_bh = bh->b_private;
437 clear_buffer_async_write(bh);
439 } while ((bh = next_bh) != NULL);
441 vn_iowake(ioend->io_vnode);
442 mempool_free(ioend, xfs_ioend_pool);
443 } while ((ioend = next) != NULL);
447 * Test to see if we've been building up a completion structure for
448 * earlier buffers -- if so, we try to append to this ioend if we
449 * can, otherwise we finish off any current ioend and start another.
450 * Return true if we've finished the given ioend.
455 struct buffer_head *bh,
458 xfs_ioend_t **result,
461 xfs_ioend_t *ioend = *result;
463 if (!ioend || need_ioend || type != ioend->io_type) {
464 xfs_ioend_t *previous = *result;
466 ioend = xfs_alloc_ioend(inode, type);
467 ioend->io_offset = offset;
468 ioend->io_buffer_head = bh;
469 ioend->io_buffer_tail = bh;
471 previous->io_list = ioend;
474 ioend->io_buffer_tail->b_private = bh;
475 ioend->io_buffer_tail = bh;
478 bh->b_private = NULL;
479 ioend->io_size += bh->b_size;
484 struct buffer_head *bh,
491 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
493 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
494 ((offset - mp->iomap_offset) >> block_bits);
496 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
499 set_buffer_mapped(bh);
504 struct buffer_head *bh,
509 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
510 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
513 xfs_map_buffer(bh, iomapp, offset, block_bits);
514 bh->b_bdev = iomapp->iomap_target->bt_bdev;
515 set_buffer_mapped(bh);
516 clear_buffer_delay(bh);
517 clear_buffer_unwritten(bh);
521 * Look for a page at index that is suitable for clustering.
526 unsigned int pg_offset,
531 if (PageWriteback(page))
534 if (page->mapping && PageDirty(page)) {
535 if (page_has_buffers(page)) {
536 struct buffer_head *bh, *head;
538 bh = head = page_buffers(page);
540 if (!buffer_uptodate(bh))
542 if (mapped != buffer_mapped(bh))
545 if (ret >= pg_offset)
547 } while ((bh = bh->b_this_page) != head);
549 ret = mapped ? 0 : PAGE_CACHE_SIZE;
558 struct page *startpage,
559 struct buffer_head *bh,
560 struct buffer_head *head,
564 pgoff_t tindex, tlast, tloff;
568 /* First sum forwards in this page */
570 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
573 } while ((bh = bh->b_this_page) != head);
575 /* if we reached the end of the page, sum forwards in following pages */
576 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
577 tindex = startpage->index + 1;
579 /* Prune this back to avoid pathological behavior */
580 tloff = min(tlast, startpage->index + 64);
582 pagevec_init(&pvec, 0);
583 while (!done && tindex <= tloff) {
584 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
586 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
589 for (i = 0; i < pagevec_count(&pvec); i++) {
590 struct page *page = pvec.pages[i];
591 size_t pg_offset, len = 0;
593 if (tindex == tlast) {
595 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
601 pg_offset = PAGE_CACHE_SIZE;
603 if (page->index == tindex && !TestSetPageLocked(page)) {
604 len = xfs_probe_page(page, pg_offset, mapped);
617 pagevec_release(&pvec);
625 * Test if a given page is suitable for writing as part of an unwritten
626 * or delayed allocate extent.
633 if (PageWriteback(page))
636 if (page->mapping && page_has_buffers(page)) {
637 struct buffer_head *bh, *head;
640 bh = head = page_buffers(page);
642 if (buffer_unwritten(bh))
643 acceptable = (type == IOMAP_UNWRITTEN);
644 else if (buffer_delay(bh))
645 acceptable = (type == IOMAP_DELAY);
646 else if (buffer_dirty(bh) && buffer_mapped(bh))
647 acceptable = (type == 0);
650 } while ((bh = bh->b_this_page) != head);
660 * Allocate & map buffers for page given the extent map. Write it out.
661 * except for the original page of a writepage, this is called on
662 * delalloc/unwritten pages only, for the original page it is possible
663 * that the page has no mapping at all.
671 xfs_ioend_t **ioendp,
672 struct writeback_control *wbc,
676 struct buffer_head *bh, *head;
677 xfs_off_t end_offset;
678 unsigned long p_offset;
680 int bbits = inode->i_blkbits;
682 int count = 0, done = 0, uptodate = 1;
683 xfs_off_t offset = page_offset(page);
685 if (page->index != tindex)
687 if (TestSetPageLocked(page))
689 if (PageWriteback(page))
690 goto fail_unlock_page;
691 if (page->mapping != inode->i_mapping)
692 goto fail_unlock_page;
693 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
694 goto fail_unlock_page;
697 * page_dirty is initially a count of buffers on the page before
698 * EOF and is decremented as we move each into a cleanable state.
702 * End offset is the highest offset that this page should represent.
703 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
704 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
705 * hence give us the correct page_dirty count. On any other page,
706 * it will be zero and in that case we need page_dirty to be the
707 * count of buffers on the page.
709 end_offset = min_t(unsigned long long,
710 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
713 len = 1 << inode->i_blkbits;
714 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
716 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
717 page_dirty = p_offset / len;
719 bh = head = page_buffers(page);
721 if (offset >= end_offset)
723 if (!buffer_uptodate(bh))
725 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
730 if (buffer_unwritten(bh) || buffer_delay(bh)) {
731 if (buffer_unwritten(bh))
732 type = IOMAP_UNWRITTEN;
736 if (!xfs_iomap_valid(mp, offset)) {
741 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
742 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
744 xfs_map_at_offset(bh, offset, bbits, mp);
746 xfs_add_to_ioend(inode, bh, offset,
749 set_buffer_dirty(bh);
751 mark_buffer_dirty(bh);
757 if (buffer_mapped(bh) && all_bh && startio) {
759 xfs_add_to_ioend(inode, bh, offset,
767 } while (offset += len, (bh = bh->b_this_page) != head);
769 if (uptodate && bh == head)
770 SetPageUptodate(page);
774 struct backing_dev_info *bdi;
776 bdi = inode->i_mapping->backing_dev_info;
778 if (bdi_write_congested(bdi)) {
779 wbc->encountered_congestion = 1;
781 } else if (wbc->nr_to_write <= 0) {
785 xfs_start_page_writeback(page, wbc, !page_dirty, count);
796 * Convert & write out a cluster of pages in the same extent as defined
797 * by mp and following the start page.
804 xfs_ioend_t **ioendp,
805 struct writeback_control *wbc,
813 pagevec_init(&pvec, 0);
814 while (!done && tindex <= tlast) {
815 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
817 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
820 for (i = 0; i < pagevec_count(&pvec); i++) {
821 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
822 iomapp, ioendp, wbc, startio, all_bh);
827 pagevec_release(&pvec);
833 * Calling this without startio set means we are being asked to make a dirty
834 * page ready for freeing it's buffers. When called with startio set then
835 * we are coming from writepage.
837 * When called with startio set it is important that we write the WHOLE
839 * The bh->b_state's cannot know if any of the blocks or which block for
840 * that matter are dirty due to mmap writes, and therefore bh uptodate is
841 * only valid if the page itself isn't completely uptodate. Some layers
842 * may clear the page dirty flag prior to calling write page, under the
843 * assumption the entire page will be written out; by not writing out the
844 * whole page the page can be reused before all valid dirty data is
845 * written out. Note: in the case of a page that has been dirty'd by
846 * mapwrite and but partially setup by block_prepare_write the
847 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
848 * valid state, thus the whole page must be written out thing.
852 xfs_page_state_convert(
855 struct writeback_control *wbc,
857 int unmapped) /* also implies page uptodate */
859 struct buffer_head *bh, *head;
861 xfs_ioend_t *ioend = NULL, *iohead = NULL;
863 unsigned long p_offset = 0;
865 __uint64_t end_offset;
866 pgoff_t end_index, last_index, tlast;
868 int flags, err, iomap_valid = 0, uptodate = 1;
869 int page_dirty, count = 0;
871 int all_bh = unmapped;
874 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
875 trylock |= BMAPI_TRYLOCK;
878 /* Is this page beyond the end of the file? */
879 offset = i_size_read(inode);
880 end_index = offset >> PAGE_CACHE_SHIFT;
881 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
882 if (page->index >= end_index) {
883 if ((page->index >= end_index + 1) ||
884 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
892 * page_dirty is initially a count of buffers on the page before
893 * EOF and is decremented as we move each into a cleanable state.
897 * End offset is the highest offset that this page should represent.
898 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
899 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
900 * hence give us the correct page_dirty count. On any other page,
901 * it will be zero and in that case we need page_dirty to be the
902 * count of buffers on the page.
904 end_offset = min_t(unsigned long long,
905 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
906 len = 1 << inode->i_blkbits;
907 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
909 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
910 page_dirty = p_offset / len;
912 bh = head = page_buffers(page);
913 offset = page_offset(page);
917 /* TODO: cleanup count and page_dirty */
920 if (offset >= end_offset)
922 if (!buffer_uptodate(bh))
924 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
926 * the iomap is actually still valid, but the ioend
927 * isn't. shouldn't happen too often.
934 iomap_valid = xfs_iomap_valid(&iomap, offset);
937 * First case, map an unwritten extent and prepare for
938 * extent state conversion transaction on completion.
940 * Second case, allocate space for a delalloc buffer.
941 * We can return EAGAIN here in the release page case.
943 * Third case, an unmapped buffer was found, and we are
944 * in a path where we need to write the whole page out.
946 if (buffer_unwritten(bh) || buffer_delay(bh) ||
947 ((buffer_uptodate(bh) || PageUptodate(page)) &&
948 !buffer_mapped(bh) && (unmapped || startio))) {
950 * Make sure we don't use a read-only iomap
952 if (flags == BMAPI_READ)
955 if (buffer_unwritten(bh)) {
956 type = IOMAP_UNWRITTEN;
957 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
958 } else if (buffer_delay(bh)) {
960 flags = BMAPI_ALLOCATE | trylock;
963 flags = BMAPI_WRITE | BMAPI_MMAP;
967 if (type == IOMAP_NEW) {
968 size = xfs_probe_cluster(inode,
974 err = xfs_map_blocks(inode, offset, size,
978 iomap_valid = xfs_iomap_valid(&iomap, offset);
981 xfs_map_at_offset(bh, offset,
982 inode->i_blkbits, &iomap);
984 xfs_add_to_ioend(inode, bh, offset,
988 set_buffer_dirty(bh);
990 mark_buffer_dirty(bh);
995 } else if (buffer_uptodate(bh) && startio) {
997 * we got here because the buffer is already mapped.
998 * That means it must already have extents allocated
999 * underneath it. Map the extent by reading it.
1001 if (!iomap_valid || type != 0) {
1003 size = xfs_probe_cluster(inode, page, bh,
1005 err = xfs_map_blocks(inode, offset, size,
1009 iomap_valid = xfs_iomap_valid(&iomap, offset);
1013 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1014 ASSERT(buffer_mapped(bh));
1017 xfs_add_to_ioend(inode, bh, offset, type,
1018 &ioend, !iomap_valid);
1024 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1025 (unmapped || startio)) {
1032 } while (offset += len, ((bh = bh->b_this_page) != head));
1034 if (uptodate && bh == head)
1035 SetPageUptodate(page);
1038 xfs_start_page_writeback(page, wbc, 1, count);
1040 if (ioend && iomap_valid) {
1041 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1043 tlast = min_t(pgoff_t, offset, last_index);
1044 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1045 wbc, startio, all_bh, tlast);
1049 xfs_submit_ioend(iohead);
1055 xfs_cancel_ioend(iohead);
1058 * If it's delalloc and we have nowhere to put it,
1059 * throw it away, unless the lower layers told
1062 if (err != -EAGAIN) {
1064 block_invalidatepage(page, 0);
1065 ClearPageUptodate(page);
1071 * writepage: Called from one of two places:
1073 * 1. we are flushing a delalloc buffer head.
1075 * 2. we are writing out a dirty page. Typically the page dirty
1076 * state is cleared before we get here. In this case is it
1077 * conceivable we have no buffer heads.
1079 * For delalloc space on the page we need to allocate space and
1080 * flush it. For unmapped buffer heads on the page we should
1081 * allocate space if the page is uptodate. For any other dirty
1082 * buffer heads on the page we should flush them.
1084 * If we detect that a transaction would be required to flush
1085 * the page, we have to check the process flags first, if we
1086 * are already in a transaction or disk I/O during allocations
1087 * is off, we need to fail the writepage and redirty the page.
1093 struct writeback_control *wbc)
1097 int delalloc, unmapped, unwritten;
1098 struct inode *inode = page->mapping->host;
1100 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1103 * We need a transaction if:
1104 * 1. There are delalloc buffers on the page
1105 * 2. The page is uptodate and we have unmapped buffers
1106 * 3. The page is uptodate and we have no buffers
1107 * 4. There are unwritten buffers on the page
1110 if (!page_has_buffers(page)) {
1114 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1115 if (!PageUptodate(page))
1117 need_trans = delalloc + unmapped + unwritten;
1121 * If we need a transaction and the process flags say
1122 * we are already in a transaction, or no IO is allowed
1123 * then mark the page dirty again and leave the page
1126 if (current_test_flags(PF_FSTRANS) && need_trans)
1130 * Delay hooking up buffer heads until we have
1131 * made our go/no-go decision.
1133 if (!page_has_buffers(page))
1134 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1137 * Convert delayed allocate, unwritten or unmapped space
1138 * to real space and flush out to disk.
1140 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1141 if (error == -EAGAIN)
1143 if (unlikely(error < 0))
1149 redirty_page_for_writepage(wbc, page);
1159 struct address_space *mapping,
1160 struct writeback_control *wbc)
1162 struct bhv_vnode *vp = vn_from_inode(mapping->host);
1166 return generic_writepages(mapping, wbc);
1170 * Called to move a page into cleanable state - and from there
1171 * to be released. Possibly the page is already clean. We always
1172 * have buffer heads in this call.
1174 * Returns 0 if the page is ok to release, 1 otherwise.
1176 * Possible scenarios are:
1178 * 1. We are being called to release a page which has been written
1179 * to via regular I/O. buffer heads will be dirty and possibly
1180 * delalloc. If no delalloc buffer heads in this case then we
1181 * can just return zero.
1183 * 2. We are called to release a page which has been written via
1184 * mmap, all we need to do is ensure there is no delalloc
1185 * state in the buffer heads, if not we can let the caller
1186 * free them and we should come back later via writepage.
1193 struct inode *inode = page->mapping->host;
1194 int dirty, delalloc, unmapped, unwritten;
1195 struct writeback_control wbc = {
1196 .sync_mode = WB_SYNC_ALL,
1200 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1202 if (!page_has_buffers(page))
1205 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1206 if (!delalloc && !unwritten)
1209 if (!(gfp_mask & __GFP_FS))
1212 /* If we are already inside a transaction or the thread cannot
1213 * do I/O, we cannot release this page.
1215 if (current_test_flags(PF_FSTRANS))
1219 * Convert delalloc space to real space, do not flush the
1220 * data out to disk, that will be done by the caller.
1221 * Never need to allocate space here - we will always
1222 * come back to writepage in that case.
1224 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1225 if (dirty == 0 && !unwritten)
1230 return try_to_free_buffers(page);
1235 struct inode *inode,
1237 struct buffer_head *bh_result,
1240 bmapi_flags_t flags)
1242 bhv_vnode_t *vp = vn_from_inode(inode);
1249 offset = (xfs_off_t)iblock << inode->i_blkbits;
1250 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1251 size = bh_result->b_size;
1252 error = bhv_vop_bmap(vp, offset, size,
1253 create ? flags : BMAPI_READ, &iomap, &niomap);
1259 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1261 * For unwritten extents do not report a disk address on
1262 * the read case (treat as if we're reading into a hole).
1264 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1265 xfs_map_buffer(bh_result, &iomap, offset,
1268 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1270 bh_result->b_private = inode;
1271 set_buffer_unwritten(bh_result);
1272 set_buffer_delay(bh_result);
1277 * If this is a realtime file, data may be on a different device.
1278 * to that pointed to from the buffer_head b_bdev currently.
1280 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1283 * If we previously allocated a block out beyond eof and we are
1284 * now coming back to use it then we will need to flag it as new
1285 * even if it has a disk address.
1288 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1289 (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
1290 set_buffer_new(bh_result);
1292 if (iomap.iomap_flags & IOMAP_DELAY) {
1295 set_buffer_uptodate(bh_result);
1296 set_buffer_mapped(bh_result);
1297 set_buffer_delay(bh_result);
1301 if (direct || size > (1 << inode->i_blkbits)) {
1302 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1303 offset = min_t(xfs_off_t,
1304 iomap.iomap_bsize - iomap.iomap_delta, size);
1305 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1313 struct inode *inode,
1315 struct buffer_head *bh_result,
1318 return __xfs_get_blocks(inode, iblock,
1319 bh_result, create, 0, BMAPI_WRITE);
1323 xfs_get_blocks_direct(
1324 struct inode *inode,
1326 struct buffer_head *bh_result,
1329 return __xfs_get_blocks(inode, iblock,
1330 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1340 xfs_ioend_t *ioend = iocb->private;
1343 * Non-NULL private data means we need to issue a transaction to
1344 * convert a range from unwritten to written extents. This needs
1345 * to happen from process context but aio+dio I/O completion
1346 * happens from irq context so we need to defer it to a workqueue.
1347 * This is not necessary for synchronous direct I/O, but we do
1348 * it anyway to keep the code uniform and simpler.
1350 * The core direct I/O code might be changed to always call the
1351 * completion handler in the future, in which case all this can
1354 if (private && size > 0) {
1355 ioend->io_offset = offset;
1356 ioend->io_size = size;
1357 xfs_finish_ioend(ioend);
1360 xfs_destroy_ioend(ioend);
1364 * blockdev_direct_IO can return an error even after the I/O
1365 * completion handler was called. Thus we need to protect
1366 * against double-freeing.
1368 iocb->private = NULL;
1375 const struct iovec *iov,
1377 unsigned long nr_segs)
1379 struct file *file = iocb->ki_filp;
1380 struct inode *inode = file->f_mapping->host;
1381 bhv_vnode_t *vp = vn_from_inode(inode);
1387 error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps);
1391 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1393 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1394 iomap.iomap_target->bt_bdev,
1395 iov, offset, nr_segs,
1396 xfs_get_blocks_direct,
1399 if (unlikely(ret <= 0 && iocb->private))
1400 xfs_destroy_ioend(iocb->private);
1405 xfs_vm_prepare_write(
1411 return block_prepare_write(page, from, to, xfs_get_blocks);
1416 struct address_space *mapping,
1419 struct inode *inode = (struct inode *)mapping->host;
1420 bhv_vnode_t *vp = vn_from_inode(inode);
1422 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1423 bhv_vop_rwlock(vp, VRWLOCK_READ);
1424 bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
1425 bhv_vop_rwunlock(vp, VRWLOCK_READ);
1426 return generic_block_bmap(mapping, block, xfs_get_blocks);
1431 struct file *unused,
1434 return mpage_readpage(page, xfs_get_blocks);
1439 struct file *unused,
1440 struct address_space *mapping,
1441 struct list_head *pages,
1444 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1448 xfs_vm_invalidatepage(
1450 unsigned long offset)
1452 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1453 page->mapping->host, page, offset);
1454 block_invalidatepage(page, offset);
1457 struct address_space_operations xfs_address_space_operations = {
1458 .readpage = xfs_vm_readpage,
1459 .readpages = xfs_vm_readpages,
1460 .writepage = xfs_vm_writepage,
1461 .writepages = xfs_vm_writepages,
1462 .sync_page = block_sync_page,
1463 .releasepage = xfs_vm_releasepage,
1464 .invalidatepage = xfs_vm_invalidatepage,
1465 .prepare_write = xfs_vm_prepare_write,
1466 .commit_write = generic_commit_write,
1467 .bmap = xfs_vm_bmap,
1468 .direct_IO = xfs_vm_direct_IO,
1469 .migratepage = buffer_migrate_page,