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(
412 struct writeback_control *wbc,
416 ASSERT(PageLocked(page));
417 ASSERT(!PageWriteback(page));
419 clear_page_dirty_for_io(page);
420 set_page_writeback(page);
422 /* If no buffers on the page are to be written, finish it here */
424 end_page_writeback(page);
427 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
429 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
433 * Submit all of the bios for all of the ioends we have saved up, covering the
434 * initial writepage page and also any probed pages.
436 * Because we may have multiple ioends spanning a page, we need to start
437 * writeback on all the buffers before we submit them for I/O. If we mark the
438 * buffers as we got, then we can end up with a page that only has buffers
439 * marked async write and I/O complete on can occur before we mark the other
440 * buffers async write.
442 * The end result of this is that we trip a bug in end_page_writeback() because
443 * we call it twice for the one page as the code in end_buffer_async_write()
444 * assumes that all buffers on the page are started at the same time.
446 * The fix is two passes across the ioend list - one to start writeback on the
447 * buffer_heads, and then submit them for I/O on the second pass.
453 xfs_ioend_t *head = ioend;
455 struct buffer_head *bh;
457 sector_t lastblock = 0;
459 /* Pass 1 - start writeback */
461 next = ioend->io_list;
462 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
463 xfs_start_buffer_writeback(bh);
465 } while ((ioend = next) != NULL);
467 /* Pass 2 - submit I/O */
470 next = ioend->io_list;
473 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
477 bio = xfs_alloc_ioend_bio(bh);
478 } else if (bh->b_blocknr != lastblock + 1) {
479 xfs_submit_ioend_bio(ioend, bio);
483 if (bio_add_buffer(bio, bh) != bh->b_size) {
484 xfs_submit_ioend_bio(ioend, bio);
488 lastblock = bh->b_blocknr;
491 xfs_submit_ioend_bio(ioend, bio);
492 xfs_finish_ioend(ioend, 0);
493 } while ((ioend = next) != NULL);
497 * Cancel submission of all buffer_heads so far in this endio.
498 * Toss the endio too. Only ever called for the initial page
499 * in a writepage request, so only ever one page.
506 struct buffer_head *bh, *next_bh;
509 next = ioend->io_list;
510 bh = ioend->io_buffer_head;
512 next_bh = bh->b_private;
513 clear_buffer_async_write(bh);
515 } while ((bh = next_bh) != NULL);
517 vn_iowake(XFS_I(ioend->io_inode));
518 mempool_free(ioend, xfs_ioend_pool);
519 } while ((ioend = next) != NULL);
523 * Test to see if we've been building up a completion structure for
524 * earlier buffers -- if so, we try to append to this ioend if we
525 * can, otherwise we finish off any current ioend and start another.
526 * Return true if we've finished the given ioend.
531 struct buffer_head *bh,
534 xfs_ioend_t **result,
537 xfs_ioend_t *ioend = *result;
539 if (!ioend || need_ioend || type != ioend->io_type) {
540 xfs_ioend_t *previous = *result;
542 ioend = xfs_alloc_ioend(inode, type);
543 ioend->io_offset = offset;
544 ioend->io_buffer_head = bh;
545 ioend->io_buffer_tail = bh;
547 previous->io_list = ioend;
550 ioend->io_buffer_tail->b_private = bh;
551 ioend->io_buffer_tail = bh;
554 bh->b_private = NULL;
555 ioend->io_size += bh->b_size;
560 struct buffer_head *bh,
567 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
569 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
570 ((offset - mp->iomap_offset) >> block_bits);
572 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
575 set_buffer_mapped(bh);
580 struct buffer_head *bh,
585 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
586 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
589 xfs_map_buffer(bh, iomapp, offset, block_bits);
590 bh->b_bdev = iomapp->iomap_target->bt_bdev;
591 set_buffer_mapped(bh);
592 clear_buffer_delay(bh);
593 clear_buffer_unwritten(bh);
597 * Look for a page at index that is suitable for clustering.
602 unsigned int pg_offset,
607 if (PageWriteback(page))
610 if (page->mapping && PageDirty(page)) {
611 if (page_has_buffers(page)) {
612 struct buffer_head *bh, *head;
614 bh = head = page_buffers(page);
616 if (!buffer_uptodate(bh))
618 if (mapped != buffer_mapped(bh))
621 if (ret >= pg_offset)
623 } while ((bh = bh->b_this_page) != head);
625 ret = mapped ? 0 : PAGE_CACHE_SIZE;
634 struct page *startpage,
635 struct buffer_head *bh,
636 struct buffer_head *head,
640 pgoff_t tindex, tlast, tloff;
644 /* First sum forwards in this page */
646 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
649 } while ((bh = bh->b_this_page) != head);
651 /* if we reached the end of the page, sum forwards in following pages */
652 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
653 tindex = startpage->index + 1;
655 /* Prune this back to avoid pathological behavior */
656 tloff = min(tlast, startpage->index + 64);
658 pagevec_init(&pvec, 0);
659 while (!done && tindex <= tloff) {
660 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
662 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
665 for (i = 0; i < pagevec_count(&pvec); i++) {
666 struct page *page = pvec.pages[i];
667 size_t pg_offset, pg_len = 0;
669 if (tindex == tlast) {
671 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
677 pg_offset = PAGE_CACHE_SIZE;
679 if (page->index == tindex && !TestSetPageLocked(page)) {
680 pg_len = xfs_probe_page(page, pg_offset, mapped);
693 pagevec_release(&pvec);
701 * Test if a given page is suitable for writing as part of an unwritten
702 * or delayed allocate extent.
709 if (PageWriteback(page))
712 if (page->mapping && page_has_buffers(page)) {
713 struct buffer_head *bh, *head;
716 bh = head = page_buffers(page);
718 if (buffer_unwritten(bh))
719 acceptable = (type == IOMAP_UNWRITTEN);
720 else if (buffer_delay(bh))
721 acceptable = (type == IOMAP_DELAY);
722 else if (buffer_dirty(bh) && buffer_mapped(bh))
723 acceptable = (type == IOMAP_NEW);
726 } while ((bh = bh->b_this_page) != head);
736 * Allocate & map buffers for page given the extent map. Write it out.
737 * except for the original page of a writepage, this is called on
738 * delalloc/unwritten pages only, for the original page it is possible
739 * that the page has no mapping at all.
747 xfs_ioend_t **ioendp,
748 struct writeback_control *wbc,
752 struct buffer_head *bh, *head;
753 xfs_off_t end_offset;
754 unsigned long p_offset;
756 int bbits = inode->i_blkbits;
758 int count = 0, done = 0, uptodate = 1;
759 xfs_off_t offset = page_offset(page);
761 if (page->index != tindex)
763 if (TestSetPageLocked(page))
765 if (PageWriteback(page))
766 goto fail_unlock_page;
767 if (page->mapping != inode->i_mapping)
768 goto fail_unlock_page;
769 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
770 goto fail_unlock_page;
773 * page_dirty is initially a count of buffers on the page before
774 * EOF and is decremented as we move each into a cleanable state.
778 * End offset is the highest offset that this page should represent.
779 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
780 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
781 * hence give us the correct page_dirty count. On any other page,
782 * it will be zero and in that case we need page_dirty to be the
783 * count of buffers on the page.
785 end_offset = min_t(unsigned long long,
786 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
789 len = 1 << inode->i_blkbits;
790 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
792 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
793 page_dirty = p_offset / len;
795 bh = head = page_buffers(page);
797 if (offset >= end_offset)
799 if (!buffer_uptodate(bh))
801 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
806 if (buffer_unwritten(bh) || buffer_delay(bh)) {
807 if (buffer_unwritten(bh))
808 type = IOMAP_UNWRITTEN;
812 if (!xfs_iomap_valid(mp, offset)) {
817 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
818 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
820 xfs_map_at_offset(bh, offset, bbits, mp);
822 xfs_add_to_ioend(inode, bh, offset,
825 set_buffer_dirty(bh);
827 mark_buffer_dirty(bh);
833 if (buffer_mapped(bh) && all_bh && startio) {
835 xfs_add_to_ioend(inode, bh, offset,
843 } while (offset += len, (bh = bh->b_this_page) != head);
845 if (uptodate && bh == head)
846 SetPageUptodate(page);
850 struct backing_dev_info *bdi;
852 bdi = inode->i_mapping->backing_dev_info;
854 if (bdi_write_congested(bdi)) {
855 wbc->encountered_congestion = 1;
857 } else if (wbc->nr_to_write <= 0) {
861 xfs_start_page_writeback(page, wbc, !page_dirty, count);
872 * Convert & write out a cluster of pages in the same extent as defined
873 * by mp and following the start page.
880 xfs_ioend_t **ioendp,
881 struct writeback_control *wbc,
889 pagevec_init(&pvec, 0);
890 while (!done && tindex <= tlast) {
891 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
893 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
896 for (i = 0; i < pagevec_count(&pvec); i++) {
897 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
898 iomapp, ioendp, wbc, startio, all_bh);
903 pagevec_release(&pvec);
909 * Calling this without startio set means we are being asked to make a dirty
910 * page ready for freeing it's buffers. When called with startio set then
911 * we are coming from writepage.
913 * When called with startio set it is important that we write the WHOLE
915 * The bh->b_state's cannot know if any of the blocks or which block for
916 * that matter are dirty due to mmap writes, and therefore bh uptodate is
917 * only valid if the page itself isn't completely uptodate. Some layers
918 * may clear the page dirty flag prior to calling write page, under the
919 * assumption the entire page will be written out; by not writing out the
920 * whole page the page can be reused before all valid dirty data is
921 * written out. Note: in the case of a page that has been dirty'd by
922 * mapwrite and but partially setup by block_prepare_write the
923 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
924 * valid state, thus the whole page must be written out thing.
928 xfs_page_state_convert(
931 struct writeback_control *wbc,
933 int unmapped) /* also implies page uptodate */
935 struct buffer_head *bh, *head;
937 xfs_ioend_t *ioend = NULL, *iohead = NULL;
939 unsigned long p_offset = 0;
941 __uint64_t end_offset;
942 pgoff_t end_index, last_index, tlast;
944 int flags, err, iomap_valid = 0, uptodate = 1;
945 int page_dirty, count = 0;
947 int all_bh = unmapped;
950 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
951 trylock |= BMAPI_TRYLOCK;
954 /* Is this page beyond the end of the file? */
955 offset = i_size_read(inode);
956 end_index = offset >> PAGE_CACHE_SHIFT;
957 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
958 if (page->index >= end_index) {
959 if ((page->index >= end_index + 1) ||
960 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
968 * page_dirty is initially a count of buffers on the page before
969 * EOF and is decremented as we move each into a cleanable state.
973 * End offset is the highest offset that this page should represent.
974 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
975 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
976 * hence give us the correct page_dirty count. On any other page,
977 * it will be zero and in that case we need page_dirty to be the
978 * count of buffers on the page.
980 end_offset = min_t(unsigned long long,
981 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
982 len = 1 << inode->i_blkbits;
983 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
985 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
986 page_dirty = p_offset / len;
988 bh = head = page_buffers(page);
989 offset = page_offset(page);
993 /* TODO: cleanup count and page_dirty */
996 if (offset >= end_offset)
998 if (!buffer_uptodate(bh))
1000 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
1002 * the iomap is actually still valid, but the ioend
1003 * isn't. shouldn't happen too often.
1010 iomap_valid = xfs_iomap_valid(&iomap, offset);
1013 * First case, map an unwritten extent and prepare for
1014 * extent state conversion transaction on completion.
1016 * Second case, allocate space for a delalloc buffer.
1017 * We can return EAGAIN here in the release page case.
1019 * Third case, an unmapped buffer was found, and we are
1020 * in a path where we need to write the whole page out.
1022 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1023 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1024 !buffer_mapped(bh) && (unmapped || startio))) {
1028 * Make sure we don't use a read-only iomap
1030 if (flags == BMAPI_READ)
1033 if (buffer_unwritten(bh)) {
1034 type = IOMAP_UNWRITTEN;
1035 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1036 } else if (buffer_delay(bh)) {
1038 flags = BMAPI_ALLOCATE | trylock;
1041 flags = BMAPI_WRITE | BMAPI_MMAP;
1046 * if we didn't have a valid mapping then we
1047 * need to ensure that we put the new mapping
1048 * in a new ioend structure. This needs to be
1049 * done to ensure that the ioends correctly
1050 * reflect the block mappings at io completion
1051 * for unwritten extent conversion.
1054 if (type == IOMAP_NEW) {
1055 size = xfs_probe_cluster(inode,
1061 err = xfs_map_blocks(inode, offset, size,
1065 iomap_valid = xfs_iomap_valid(&iomap, offset);
1068 xfs_map_at_offset(bh, offset,
1069 inode->i_blkbits, &iomap);
1071 xfs_add_to_ioend(inode, bh, offset,
1075 set_buffer_dirty(bh);
1077 mark_buffer_dirty(bh);
1082 } else if (buffer_uptodate(bh) && startio) {
1084 * we got here because the buffer is already mapped.
1085 * That means it must already have extents allocated
1086 * underneath it. Map the extent by reading it.
1088 if (!iomap_valid || flags != BMAPI_READ) {
1090 size = xfs_probe_cluster(inode, page, bh,
1092 err = xfs_map_blocks(inode, offset, size,
1096 iomap_valid = xfs_iomap_valid(&iomap, offset);
1100 * We set the type to IOMAP_NEW in case we are doing a
1101 * small write at EOF that is extending the file but
1102 * without needing an allocation. We need to update the
1103 * file size on I/O completion in this case so it is
1104 * the same case as having just allocated a new extent
1105 * that we are writing into for the first time.
1108 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1109 ASSERT(buffer_mapped(bh));
1112 xfs_add_to_ioend(inode, bh, offset, type,
1113 &ioend, !iomap_valid);
1119 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1120 (unmapped || startio)) {
1127 } while (offset += len, ((bh = bh->b_this_page) != head));
1129 if (uptodate && bh == head)
1130 SetPageUptodate(page);
1133 xfs_start_page_writeback(page, wbc, 1, count);
1135 if (ioend && iomap_valid) {
1136 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1138 tlast = min_t(pgoff_t, offset, last_index);
1139 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1140 wbc, startio, all_bh, tlast);
1144 xfs_submit_ioend(iohead);
1150 xfs_cancel_ioend(iohead);
1153 * If it's delalloc and we have nowhere to put it,
1154 * throw it away, unless the lower layers told
1157 if (err != -EAGAIN) {
1159 block_invalidatepage(page, 0);
1160 ClearPageUptodate(page);
1166 * writepage: Called from one of two places:
1168 * 1. we are flushing a delalloc buffer head.
1170 * 2. we are writing out a dirty page. Typically the page dirty
1171 * state is cleared before we get here. In this case is it
1172 * conceivable we have no buffer heads.
1174 * For delalloc space on the page we need to allocate space and
1175 * flush it. For unmapped buffer heads on the page we should
1176 * allocate space if the page is uptodate. For any other dirty
1177 * buffer heads on the page we should flush them.
1179 * If we detect that a transaction would be required to flush
1180 * the page, we have to check the process flags first, if we
1181 * are already in a transaction or disk I/O during allocations
1182 * is off, we need to fail the writepage and redirty the page.
1188 struct writeback_control *wbc)
1192 int delalloc, unmapped, unwritten;
1193 struct inode *inode = page->mapping->host;
1195 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1198 * We need a transaction if:
1199 * 1. There are delalloc buffers on the page
1200 * 2. The page is uptodate and we have unmapped buffers
1201 * 3. The page is uptodate and we have no buffers
1202 * 4. There are unwritten buffers on the page
1205 if (!page_has_buffers(page)) {
1209 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1210 if (!PageUptodate(page))
1212 need_trans = delalloc + unmapped + unwritten;
1216 * If we need a transaction and the process flags say
1217 * we are already in a transaction, or no IO is allowed
1218 * then mark the page dirty again and leave the page
1221 if (current_test_flags(PF_FSTRANS) && need_trans)
1225 * Delay hooking up buffer heads until we have
1226 * made our go/no-go decision.
1228 if (!page_has_buffers(page))
1229 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1232 * Convert delayed allocate, unwritten or unmapped space
1233 * to real space and flush out to disk.
1235 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1236 if (error == -EAGAIN)
1238 if (unlikely(error < 0))
1244 redirty_page_for_writepage(wbc, page);
1254 struct address_space *mapping,
1255 struct writeback_control *wbc)
1257 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1258 return generic_writepages(mapping, wbc);
1262 * Called to move a page into cleanable state - and from there
1263 * to be released. Possibly the page is already clean. We always
1264 * have buffer heads in this call.
1266 * Returns 0 if the page is ok to release, 1 otherwise.
1268 * Possible scenarios are:
1270 * 1. We are being called to release a page which has been written
1271 * to via regular I/O. buffer heads will be dirty and possibly
1272 * delalloc. If no delalloc buffer heads in this case then we
1273 * can just return zero.
1275 * 2. We are called to release a page which has been written via
1276 * mmap, all we need to do is ensure there is no delalloc
1277 * state in the buffer heads, if not we can let the caller
1278 * free them and we should come back later via writepage.
1285 struct inode *inode = page->mapping->host;
1286 int dirty, delalloc, unmapped, unwritten;
1287 struct writeback_control wbc = {
1288 .sync_mode = WB_SYNC_ALL,
1292 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1294 if (!page_has_buffers(page))
1297 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1298 if (!delalloc && !unwritten)
1301 if (!(gfp_mask & __GFP_FS))
1304 /* If we are already inside a transaction or the thread cannot
1305 * do I/O, we cannot release this page.
1307 if (current_test_flags(PF_FSTRANS))
1311 * Convert delalloc space to real space, do not flush the
1312 * data out to disk, that will be done by the caller.
1313 * Never need to allocate space here - we will always
1314 * come back to writepage in that case.
1316 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1317 if (dirty == 0 && !unwritten)
1322 return try_to_free_buffers(page);
1327 struct inode *inode,
1329 struct buffer_head *bh_result,
1332 bmapi_flags_t flags)
1340 offset = (xfs_off_t)iblock << inode->i_blkbits;
1341 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1342 size = bh_result->b_size;
1343 error = xfs_iomap(XFS_I(inode), offset, size,
1344 create ? flags : BMAPI_READ, &iomap, &niomap);
1350 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1352 * For unwritten extents do not report a disk address on
1353 * the read case (treat as if we're reading into a hole).
1355 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1356 xfs_map_buffer(bh_result, &iomap, offset,
1359 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1361 bh_result->b_private = inode;
1362 set_buffer_unwritten(bh_result);
1367 * If this is a realtime file, data may be on a different device.
1368 * to that pointed to from the buffer_head b_bdev currently.
1370 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1373 * If we previously allocated a block out beyond eof and we are now
1374 * coming back to use it then we will need to flag it as new even if it
1375 * has a disk address.
1377 * With sub-block writes into unwritten extents we also need to mark
1378 * the buffer as new so that the unwritten parts of the buffer gets
1382 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1383 (offset >= i_size_read(inode)) ||
1384 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1385 set_buffer_new(bh_result);
1387 if (iomap.iomap_flags & IOMAP_DELAY) {
1390 set_buffer_uptodate(bh_result);
1391 set_buffer_mapped(bh_result);
1392 set_buffer_delay(bh_result);
1396 if (direct || size > (1 << inode->i_blkbits)) {
1397 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1398 offset = min_t(xfs_off_t,
1399 iomap.iomap_bsize - iomap.iomap_delta, size);
1400 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1408 struct inode *inode,
1410 struct buffer_head *bh_result,
1413 return __xfs_get_blocks(inode, iblock,
1414 bh_result, create, 0, BMAPI_WRITE);
1418 xfs_get_blocks_direct(
1419 struct inode *inode,
1421 struct buffer_head *bh_result,
1424 return __xfs_get_blocks(inode, iblock,
1425 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1435 xfs_ioend_t *ioend = iocb->private;
1438 * Non-NULL private data means we need to issue a transaction to
1439 * convert a range from unwritten to written extents. This needs
1440 * to happen from process context but aio+dio I/O completion
1441 * happens from irq context so we need to defer it to a workqueue.
1442 * This is not necessary for synchronous direct I/O, but we do
1443 * it anyway to keep the code uniform and simpler.
1445 * Well, if only it were that simple. Because synchronous direct I/O
1446 * requires extent conversion to occur *before* we return to userspace,
1447 * we have to wait for extent conversion to complete. Look at the
1448 * iocb that has been passed to us to determine if this is AIO or
1449 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1450 * workqueue and wait for it to complete.
1452 * The core direct I/O code might be changed to always call the
1453 * completion handler in the future, in which case all this can
1456 ioend->io_offset = offset;
1457 ioend->io_size = size;
1458 if (ioend->io_type == IOMAP_READ) {
1459 xfs_finish_ioend(ioend, 0);
1460 } else if (private && size > 0) {
1461 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1464 * A direct I/O write ioend starts it's life in unwritten
1465 * state in case they map an unwritten extent. This write
1466 * didn't map an unwritten extent so switch it's completion
1469 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1470 xfs_finish_ioend(ioend, 0);
1474 * blockdev_direct_IO can return an error even after the I/O
1475 * completion handler was called. Thus we need to protect
1476 * against double-freeing.
1478 iocb->private = NULL;
1485 const struct iovec *iov,
1487 unsigned long nr_segs)
1489 struct file *file = iocb->ki_filp;
1490 struct inode *inode = file->f_mapping->host;
1491 struct block_device *bdev;
1494 bdev = xfs_find_bdev_for_inode(XFS_I(inode));
1497 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1498 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1499 bdev, iov, offset, nr_segs,
1500 xfs_get_blocks_direct,
1503 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1504 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1505 bdev, iov, offset, nr_segs,
1506 xfs_get_blocks_direct,
1510 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1511 xfs_destroy_ioend(iocb->private);
1518 struct address_space *mapping,
1522 struct page **pagep,
1526 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1532 struct address_space *mapping,
1535 struct inode *inode = (struct inode *)mapping->host;
1536 struct xfs_inode *ip = XFS_I(inode);
1538 xfs_itrace_entry(XFS_I(inode));
1539 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1540 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1541 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1542 return generic_block_bmap(mapping, block, xfs_get_blocks);
1547 struct file *unused,
1550 return mpage_readpage(page, xfs_get_blocks);
1555 struct file *unused,
1556 struct address_space *mapping,
1557 struct list_head *pages,
1560 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1564 xfs_vm_invalidatepage(
1566 unsigned long offset)
1568 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1569 page->mapping->host, page, offset);
1570 block_invalidatepage(page, offset);
1573 const struct address_space_operations xfs_address_space_operations = {
1574 .readpage = xfs_vm_readpage,
1575 .readpages = xfs_vm_readpages,
1576 .writepage = xfs_vm_writepage,
1577 .writepages = xfs_vm_writepages,
1578 .sync_page = block_sync_page,
1579 .releasepage = xfs_vm_releasepage,
1580 .invalidatepage = xfs_vm_invalidatepage,
1581 .write_begin = xfs_vm_write_begin,
1582 .write_end = generic_write_end,
1583 .bmap = xfs_vm_bmap,
1584 .direct_IO = xfs_vm_direct_IO,
1585 .migratepage = buffer_migrate_page,