2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_error.h"
39 #include "xfs_iomap.h"
40 #include "xfs_vnodeops.h"
41 #include <linux/mpage.h>
42 #include <linux/pagevec.h>
43 #include <linux/writeback.h>
52 struct buffer_head *bh, *head;
54 *delalloc = *unmapped = *unwritten = 0;
56 bh = head = page_buffers(page);
58 if (buffer_uptodate(bh) && !buffer_mapped(bh))
60 else if (buffer_unwritten(bh))
62 else if (buffer_delay(bh))
64 } while ((bh = bh->b_this_page) != head);
67 #if defined(XFS_RW_TRACE)
76 loff_t isize = i_size_read(inode);
77 loff_t offset = page_offset(page);
78 int delalloc = -1, unmapped = -1, unwritten = -1;
80 if (page_has_buffers(page))
81 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
87 ktrace_enter(ip->i_rwtrace,
88 (void *)((unsigned long)tag),
93 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
94 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
95 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
96 (void *)((unsigned long)(isize & 0xffffffff)),
97 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(offset & 0xffffffff)),
99 (void *)((unsigned long)delalloc),
100 (void *)((unsigned long)unmapped),
101 (void *)((unsigned long)unwritten),
102 (void *)((unsigned long)current_pid()),
106 #define xfs_page_trace(tag, inode, page, pgoff)
109 STATIC struct block_device *
110 xfs_find_bdev_for_inode(
111 struct xfs_inode *ip)
113 struct xfs_mount *mp = ip->i_mount;
115 if (XFS_IS_REALTIME_INODE(ip))
116 return mp->m_rtdev_targp->bt_bdev;
118 return mp->m_ddev_targp->bt_bdev;
122 * Schedule IO completion handling on a xfsdatad if this was
123 * the final hold on this ioend. If we are asked to wait,
124 * flush the workqueue.
131 if (atomic_dec_and_test(&ioend->io_remaining)) {
132 queue_work(xfsdatad_workqueue, &ioend->io_work);
134 flush_workqueue(xfsdatad_workqueue);
139 * We're now finished for good with this ioend structure.
140 * Update the page state via the associated buffer_heads,
141 * release holds on the inode and bio, and finally free
142 * up memory. Do not use the ioend after this.
148 struct buffer_head *bh, *next;
150 for (bh = ioend->io_buffer_head; bh; bh = next) {
151 next = bh->b_private;
152 bh->b_end_io(bh, !ioend->io_error);
154 if (unlikely(ioend->io_error)) {
155 vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
158 vn_iowake(XFS_I(ioend->io_inode));
159 mempool_free(ioend, xfs_ioend_pool);
163 * Update on-disk file size now that data has been written to disk.
164 * The current in-memory file size is i_size. If a write is beyond
165 * eof i_new_size will be the intended file size until i_size is
166 * updated. If this write does not extend all the way to the valid
167 * file size then restrict this update to the end of the write.
173 xfs_inode_t *ip = XFS_I(ioend->io_inode);
177 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
178 ASSERT(ioend->io_type != IOMAP_READ);
180 if (unlikely(ioend->io_error))
183 bsize = ioend->io_offset + ioend->io_size;
185 xfs_ilock(ip, XFS_ILOCK_EXCL);
187 isize = MAX(ip->i_size, ip->i_new_size);
188 isize = MIN(isize, bsize);
190 if (ip->i_d.di_size < isize) {
191 ip->i_d.di_size = isize;
192 ip->i_update_core = 1;
193 ip->i_update_size = 1;
194 xfs_mark_inode_dirty_sync(ip);
197 xfs_iunlock(ip, XFS_ILOCK_EXCL);
201 * Buffered IO write completion for delayed allocate extents.
204 xfs_end_bio_delalloc(
205 struct work_struct *work)
208 container_of(work, xfs_ioend_t, io_work);
210 xfs_setfilesize(ioend);
211 xfs_destroy_ioend(ioend);
215 * Buffered IO write completion for regular, written extents.
219 struct work_struct *work)
222 container_of(work, xfs_ioend_t, io_work);
224 xfs_setfilesize(ioend);
225 xfs_destroy_ioend(ioend);
229 * IO write completion for unwritten extents.
231 * Issue transactions to convert a buffer range from unwritten
232 * to written extents.
235 xfs_end_bio_unwritten(
236 struct work_struct *work)
239 container_of(work, xfs_ioend_t, io_work);
240 struct xfs_inode *ip = XFS_I(ioend->io_inode);
241 xfs_off_t offset = ioend->io_offset;
242 size_t size = ioend->io_size;
244 if (likely(!ioend->io_error)) {
245 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
247 error = xfs_iomap_write_unwritten(ip, offset, size);
249 ioend->io_error = error;
251 xfs_setfilesize(ioend);
253 xfs_destroy_ioend(ioend);
257 * IO read completion for regular, written extents.
261 struct work_struct *work)
264 container_of(work, xfs_ioend_t, io_work);
266 xfs_destroy_ioend(ioend);
270 * Allocate and initialise an IO completion structure.
271 * We need to track unwritten extent write completion here initially.
272 * We'll need to extend this for updating the ondisk inode size later
282 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
285 * Set the count to 1 initially, which will prevent an I/O
286 * completion callback from happening before we have started
287 * all the I/O from calling the completion routine too early.
289 atomic_set(&ioend->io_remaining, 1);
291 ioend->io_list = NULL;
292 ioend->io_type = type;
293 ioend->io_inode = inode;
294 ioend->io_buffer_head = NULL;
295 ioend->io_buffer_tail = NULL;
296 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
297 ioend->io_offset = 0;
300 if (type == IOMAP_UNWRITTEN)
301 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
302 else if (type == IOMAP_DELAY)
303 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
304 else if (type == IOMAP_READ)
305 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
307 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
322 return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps);
330 return offset >= iomapp->iomap_offset &&
331 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
335 * BIO completion handler for buffered IO.
342 xfs_ioend_t *ioend = bio->bi_private;
344 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
345 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
347 /* Toss bio and pass work off to an xfsdatad thread */
348 bio->bi_private = NULL;
349 bio->bi_end_io = NULL;
352 xfs_finish_ioend(ioend, 0);
356 xfs_submit_ioend_bio(
360 atomic_inc(&ioend->io_remaining);
362 bio->bi_private = ioend;
363 bio->bi_end_io = xfs_end_bio;
365 submit_bio(WRITE, bio);
366 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
372 struct buffer_head *bh)
375 int nvecs = bio_get_nr_vecs(bh->b_bdev);
378 bio = bio_alloc(GFP_NOIO, nvecs);
382 ASSERT(bio->bi_private == NULL);
383 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
384 bio->bi_bdev = bh->b_bdev;
390 xfs_start_buffer_writeback(
391 struct buffer_head *bh)
393 ASSERT(buffer_mapped(bh));
394 ASSERT(buffer_locked(bh));
395 ASSERT(!buffer_delay(bh));
396 ASSERT(!buffer_unwritten(bh));
398 mark_buffer_async_write(bh);
399 set_buffer_uptodate(bh);
400 clear_buffer_dirty(bh);
404 xfs_start_page_writeback(
409 ASSERT(PageLocked(page));
410 ASSERT(!PageWriteback(page));
412 clear_page_dirty_for_io(page);
413 set_page_writeback(page);
415 /* If no buffers on the page are to be written, finish it here */
417 end_page_writeback(page);
420 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
422 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
426 * Submit all of the bios for all of the ioends we have saved up, covering the
427 * initial writepage page and also any probed pages.
429 * Because we may have multiple ioends spanning a page, we need to start
430 * writeback on all the buffers before we submit them for I/O. If we mark the
431 * buffers as we got, then we can end up with a page that only has buffers
432 * marked async write and I/O complete on can occur before we mark the other
433 * buffers async write.
435 * The end result of this is that we trip a bug in end_page_writeback() because
436 * we call it twice for the one page as the code in end_buffer_async_write()
437 * assumes that all buffers on the page are started at the same time.
439 * The fix is two passes across the ioend list - one to start writeback on the
440 * buffer_heads, and then submit them for I/O on the second pass.
446 xfs_ioend_t *head = ioend;
448 struct buffer_head *bh;
450 sector_t lastblock = 0;
452 /* Pass 1 - start writeback */
454 next = ioend->io_list;
455 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
456 xfs_start_buffer_writeback(bh);
458 } while ((ioend = next) != NULL);
460 /* Pass 2 - submit I/O */
463 next = ioend->io_list;
466 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
470 bio = xfs_alloc_ioend_bio(bh);
471 } else if (bh->b_blocknr != lastblock + 1) {
472 xfs_submit_ioend_bio(ioend, bio);
476 if (bio_add_buffer(bio, bh) != bh->b_size) {
477 xfs_submit_ioend_bio(ioend, bio);
481 lastblock = bh->b_blocknr;
484 xfs_submit_ioend_bio(ioend, bio);
485 xfs_finish_ioend(ioend, 0);
486 } while ((ioend = next) != NULL);
490 * Cancel submission of all buffer_heads so far in this endio.
491 * Toss the endio too. Only ever called for the initial page
492 * in a writepage request, so only ever one page.
499 struct buffer_head *bh, *next_bh;
502 next = ioend->io_list;
503 bh = ioend->io_buffer_head;
505 next_bh = bh->b_private;
506 clear_buffer_async_write(bh);
508 } while ((bh = next_bh) != NULL);
510 vn_iowake(XFS_I(ioend->io_inode));
511 mempool_free(ioend, xfs_ioend_pool);
512 } while ((ioend = next) != NULL);
516 * Test to see if we've been building up a completion structure for
517 * earlier buffers -- if so, we try to append to this ioend if we
518 * can, otherwise we finish off any current ioend and start another.
519 * Return true if we've finished the given ioend.
524 struct buffer_head *bh,
527 xfs_ioend_t **result,
530 xfs_ioend_t *ioend = *result;
532 if (!ioend || need_ioend || type != ioend->io_type) {
533 xfs_ioend_t *previous = *result;
535 ioend = xfs_alloc_ioend(inode, type);
536 ioend->io_offset = offset;
537 ioend->io_buffer_head = bh;
538 ioend->io_buffer_tail = bh;
540 previous->io_list = ioend;
543 ioend->io_buffer_tail->b_private = bh;
544 ioend->io_buffer_tail = bh;
547 bh->b_private = NULL;
548 ioend->io_size += bh->b_size;
553 struct buffer_head *bh,
560 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
562 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
563 ((offset - mp->iomap_offset) >> block_bits);
565 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
568 set_buffer_mapped(bh);
573 struct buffer_head *bh,
578 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
579 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
582 xfs_map_buffer(bh, iomapp, offset, block_bits);
583 bh->b_bdev = iomapp->iomap_target->bt_bdev;
584 set_buffer_mapped(bh);
585 clear_buffer_delay(bh);
586 clear_buffer_unwritten(bh);
590 * Look for a page at index that is suitable for clustering.
595 unsigned int pg_offset,
600 if (PageWriteback(page))
603 if (page->mapping && PageDirty(page)) {
604 if (page_has_buffers(page)) {
605 struct buffer_head *bh, *head;
607 bh = head = page_buffers(page);
609 if (!buffer_uptodate(bh))
611 if (mapped != buffer_mapped(bh))
614 if (ret >= pg_offset)
616 } while ((bh = bh->b_this_page) != head);
618 ret = mapped ? 0 : PAGE_CACHE_SIZE;
627 struct page *startpage,
628 struct buffer_head *bh,
629 struct buffer_head *head,
633 pgoff_t tindex, tlast, tloff;
637 /* First sum forwards in this page */
639 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
642 } while ((bh = bh->b_this_page) != head);
644 /* if we reached the end of the page, sum forwards in following pages */
645 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
646 tindex = startpage->index + 1;
648 /* Prune this back to avoid pathological behavior */
649 tloff = min(tlast, startpage->index + 64);
651 pagevec_init(&pvec, 0);
652 while (!done && tindex <= tloff) {
653 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
655 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
658 for (i = 0; i < pagevec_count(&pvec); i++) {
659 struct page *page = pvec.pages[i];
660 size_t pg_offset, pg_len = 0;
662 if (tindex == tlast) {
664 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
670 pg_offset = PAGE_CACHE_SIZE;
672 if (page->index == tindex && trylock_page(page)) {
673 pg_len = xfs_probe_page(page, pg_offset, mapped);
686 pagevec_release(&pvec);
694 * Test if a given page is suitable for writing as part of an unwritten
695 * or delayed allocate extent.
702 if (PageWriteback(page))
705 if (page->mapping && page_has_buffers(page)) {
706 struct buffer_head *bh, *head;
709 bh = head = page_buffers(page);
711 if (buffer_unwritten(bh))
712 acceptable = (type == IOMAP_UNWRITTEN);
713 else if (buffer_delay(bh))
714 acceptable = (type == IOMAP_DELAY);
715 else if (buffer_dirty(bh) && buffer_mapped(bh))
716 acceptable = (type == IOMAP_NEW);
719 } while ((bh = bh->b_this_page) != head);
729 * Allocate & map buffers for page given the extent map. Write it out.
730 * except for the original page of a writepage, this is called on
731 * delalloc/unwritten pages only, for the original page it is possible
732 * that the page has no mapping at all.
740 xfs_ioend_t **ioendp,
741 struct writeback_control *wbc,
745 struct buffer_head *bh, *head;
746 xfs_off_t end_offset;
747 unsigned long p_offset;
749 int bbits = inode->i_blkbits;
751 int count = 0, done = 0, uptodate = 1;
752 xfs_off_t offset = page_offset(page);
754 if (page->index != tindex)
756 if (!trylock_page(page))
758 if (PageWriteback(page))
759 goto fail_unlock_page;
760 if (page->mapping != inode->i_mapping)
761 goto fail_unlock_page;
762 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
763 goto fail_unlock_page;
766 * page_dirty is initially a count of buffers on the page before
767 * EOF and is decremented as we move each into a cleanable state.
771 * End offset is the highest offset that this page should represent.
772 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
773 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
774 * hence give us the correct page_dirty count. On any other page,
775 * it will be zero and in that case we need page_dirty to be the
776 * count of buffers on the page.
778 end_offset = min_t(unsigned long long,
779 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
782 len = 1 << inode->i_blkbits;
783 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
785 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
786 page_dirty = p_offset / len;
788 bh = head = page_buffers(page);
790 if (offset >= end_offset)
792 if (!buffer_uptodate(bh))
794 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
799 if (buffer_unwritten(bh) || buffer_delay(bh)) {
800 if (buffer_unwritten(bh))
801 type = IOMAP_UNWRITTEN;
805 if (!xfs_iomap_valid(mp, offset)) {
810 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
811 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
813 xfs_map_at_offset(bh, offset, bbits, mp);
815 xfs_add_to_ioend(inode, bh, offset,
818 set_buffer_dirty(bh);
820 mark_buffer_dirty(bh);
826 if (buffer_mapped(bh) && all_bh && startio) {
828 xfs_add_to_ioend(inode, bh, offset,
836 } while (offset += len, (bh = bh->b_this_page) != head);
838 if (uptodate && bh == head)
839 SetPageUptodate(page);
843 struct backing_dev_info *bdi;
845 bdi = inode->i_mapping->backing_dev_info;
847 if (bdi_write_congested(bdi)) {
848 wbc->encountered_congestion = 1;
850 } else if (wbc->nr_to_write <= 0) {
854 xfs_start_page_writeback(page, !page_dirty, count);
865 * Convert & write out a cluster of pages in the same extent as defined
866 * by mp and following the start page.
873 xfs_ioend_t **ioendp,
874 struct writeback_control *wbc,
882 pagevec_init(&pvec, 0);
883 while (!done && tindex <= tlast) {
884 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
886 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
889 for (i = 0; i < pagevec_count(&pvec); i++) {
890 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
891 iomapp, ioendp, wbc, startio, all_bh);
896 pagevec_release(&pvec);
902 * Calling this without startio set means we are being asked to make a dirty
903 * page ready for freeing it's buffers. When called with startio set then
904 * we are coming from writepage.
906 * When called with startio set it is important that we write the WHOLE
908 * The bh->b_state's cannot know if any of the blocks or which block for
909 * that matter are dirty due to mmap writes, and therefore bh uptodate is
910 * only valid if the page itself isn't completely uptodate. Some layers
911 * may clear the page dirty flag prior to calling write page, under the
912 * assumption the entire page will be written out; by not writing out the
913 * whole page the page can be reused before all valid dirty data is
914 * written out. Note: in the case of a page that has been dirty'd by
915 * mapwrite and but partially setup by block_prepare_write the
916 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
917 * valid state, thus the whole page must be written out thing.
921 xfs_page_state_convert(
924 struct writeback_control *wbc,
926 int unmapped) /* also implies page uptodate */
928 struct buffer_head *bh, *head;
930 xfs_ioend_t *ioend = NULL, *iohead = NULL;
932 unsigned long p_offset = 0;
934 __uint64_t end_offset;
935 pgoff_t end_index, last_index, tlast;
937 int flags, err, iomap_valid = 0, uptodate = 1;
938 int page_dirty, count = 0;
940 int all_bh = unmapped;
943 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
944 trylock |= BMAPI_TRYLOCK;
947 /* Is this page beyond the end of the file? */
948 offset = i_size_read(inode);
949 end_index = offset >> PAGE_CACHE_SHIFT;
950 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
951 if (page->index >= end_index) {
952 if ((page->index >= end_index + 1) ||
953 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
961 * page_dirty is initially a count of buffers on the page before
962 * EOF and is decremented as we move each into a cleanable state.
966 * End offset is the highest offset that this page should represent.
967 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
968 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
969 * hence give us the correct page_dirty count. On any other page,
970 * it will be zero and in that case we need page_dirty to be the
971 * count of buffers on the page.
973 end_offset = min_t(unsigned long long,
974 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
975 len = 1 << inode->i_blkbits;
976 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
978 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
979 page_dirty = p_offset / len;
981 bh = head = page_buffers(page);
982 offset = page_offset(page);
986 /* TODO: cleanup count and page_dirty */
989 if (offset >= end_offset)
991 if (!buffer_uptodate(bh))
993 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
995 * the iomap is actually still valid, but the ioend
996 * isn't. shouldn't happen too often.
1003 iomap_valid = xfs_iomap_valid(&iomap, offset);
1006 * First case, map an unwritten extent and prepare for
1007 * extent state conversion transaction on completion.
1009 * Second case, allocate space for a delalloc buffer.
1010 * We can return EAGAIN here in the release page case.
1012 * Third case, an unmapped buffer was found, and we are
1013 * in a path where we need to write the whole page out.
1015 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1016 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1017 !buffer_mapped(bh) && (unmapped || startio))) {
1021 * Make sure we don't use a read-only iomap
1023 if (flags == BMAPI_READ)
1026 if (buffer_unwritten(bh)) {
1027 type = IOMAP_UNWRITTEN;
1028 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1029 } else if (buffer_delay(bh)) {
1031 flags = BMAPI_ALLOCATE | trylock;
1034 flags = BMAPI_WRITE | BMAPI_MMAP;
1039 * if we didn't have a valid mapping then we
1040 * need to ensure that we put the new mapping
1041 * in a new ioend structure. This needs to be
1042 * done to ensure that the ioends correctly
1043 * reflect the block mappings at io completion
1044 * for unwritten extent conversion.
1047 if (type == IOMAP_NEW) {
1048 size = xfs_probe_cluster(inode,
1054 err = xfs_map_blocks(inode, offset, size,
1058 iomap_valid = xfs_iomap_valid(&iomap, offset);
1061 xfs_map_at_offset(bh, offset,
1062 inode->i_blkbits, &iomap);
1064 xfs_add_to_ioend(inode, bh, offset,
1068 set_buffer_dirty(bh);
1070 mark_buffer_dirty(bh);
1075 } else if (buffer_uptodate(bh) && startio) {
1077 * we got here because the buffer is already mapped.
1078 * That means it must already have extents allocated
1079 * underneath it. Map the extent by reading it.
1081 if (!iomap_valid || flags != BMAPI_READ) {
1083 size = xfs_probe_cluster(inode, page, bh,
1085 err = xfs_map_blocks(inode, offset, size,
1089 iomap_valid = xfs_iomap_valid(&iomap, offset);
1093 * We set the type to IOMAP_NEW in case we are doing a
1094 * small write at EOF that is extending the file but
1095 * without needing an allocation. We need to update the
1096 * file size on I/O completion in this case so it is
1097 * the same case as having just allocated a new extent
1098 * that we are writing into for the first time.
1101 if (trylock_buffer(bh)) {
1102 ASSERT(buffer_mapped(bh));
1105 xfs_add_to_ioend(inode, bh, offset, type,
1106 &ioend, !iomap_valid);
1112 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1113 (unmapped || startio)) {
1120 } while (offset += len, ((bh = bh->b_this_page) != head));
1122 if (uptodate && bh == head)
1123 SetPageUptodate(page);
1126 xfs_start_page_writeback(page, 1, count);
1128 if (ioend && iomap_valid) {
1129 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1131 tlast = min_t(pgoff_t, offset, last_index);
1132 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1133 wbc, startio, all_bh, tlast);
1137 xfs_submit_ioend(iohead);
1143 xfs_cancel_ioend(iohead);
1146 * If it's delalloc and we have nowhere to put it,
1147 * throw it away, unless the lower layers told
1150 if (err != -EAGAIN) {
1152 block_invalidatepage(page, 0);
1153 ClearPageUptodate(page);
1159 * writepage: Called from one of two places:
1161 * 1. we are flushing a delalloc buffer head.
1163 * 2. we are writing out a dirty page. Typically the page dirty
1164 * state is cleared before we get here. In this case is it
1165 * conceivable we have no buffer heads.
1167 * For delalloc space on the page we need to allocate space and
1168 * flush it. For unmapped buffer heads on the page we should
1169 * allocate space if the page is uptodate. For any other dirty
1170 * buffer heads on the page we should flush them.
1172 * If we detect that a transaction would be required to flush
1173 * the page, we have to check the process flags first, if we
1174 * are already in a transaction or disk I/O during allocations
1175 * is off, we need to fail the writepage and redirty the page.
1181 struct writeback_control *wbc)
1185 int delalloc, unmapped, unwritten;
1186 struct inode *inode = page->mapping->host;
1188 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1191 * We need a transaction if:
1192 * 1. There are delalloc buffers on the page
1193 * 2. The page is uptodate and we have unmapped buffers
1194 * 3. The page is uptodate and we have no buffers
1195 * 4. There are unwritten buffers on the page
1198 if (!page_has_buffers(page)) {
1202 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1203 if (!PageUptodate(page))
1205 need_trans = delalloc + unmapped + unwritten;
1209 * If we need a transaction and the process flags say
1210 * we are already in a transaction, or no IO is allowed
1211 * then mark the page dirty again and leave the page
1214 if (current_test_flags(PF_FSTRANS) && need_trans)
1218 * Delay hooking up buffer heads until we have
1219 * made our go/no-go decision.
1221 if (!page_has_buffers(page))
1222 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1225 * Convert delayed allocate, unwritten or unmapped space
1226 * to real space and flush out to disk.
1228 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1229 if (error == -EAGAIN)
1231 if (unlikely(error < 0))
1237 redirty_page_for_writepage(wbc, page);
1247 struct address_space *mapping,
1248 struct writeback_control *wbc)
1250 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1251 return generic_writepages(mapping, wbc);
1255 * Called to move a page into cleanable state - and from there
1256 * to be released. Possibly the page is already clean. We always
1257 * have buffer heads in this call.
1259 * Returns 0 if the page is ok to release, 1 otherwise.
1261 * Possible scenarios are:
1263 * 1. We are being called to release a page which has been written
1264 * to via regular I/O. buffer heads will be dirty and possibly
1265 * delalloc. If no delalloc buffer heads in this case then we
1266 * can just return zero.
1268 * 2. We are called to release a page which has been written via
1269 * mmap, all we need to do is ensure there is no delalloc
1270 * state in the buffer heads, if not we can let the caller
1271 * free them and we should come back later via writepage.
1278 struct inode *inode = page->mapping->host;
1279 int dirty, delalloc, unmapped, unwritten;
1280 struct writeback_control wbc = {
1281 .sync_mode = WB_SYNC_ALL,
1285 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1287 if (!page_has_buffers(page))
1290 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1291 if (!delalloc && !unwritten)
1294 if (!(gfp_mask & __GFP_FS))
1297 /* If we are already inside a transaction or the thread cannot
1298 * do I/O, we cannot release this page.
1300 if (current_test_flags(PF_FSTRANS))
1304 * Convert delalloc space to real space, do not flush the
1305 * data out to disk, that will be done by the caller.
1306 * Never need to allocate space here - we will always
1307 * come back to writepage in that case.
1309 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1310 if (dirty == 0 && !unwritten)
1315 return try_to_free_buffers(page);
1320 struct inode *inode,
1322 struct buffer_head *bh_result,
1325 bmapi_flags_t flags)
1333 offset = (xfs_off_t)iblock << inode->i_blkbits;
1334 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1335 size = bh_result->b_size;
1337 if (!create && direct && offset >= i_size_read(inode))
1340 error = xfs_iomap(XFS_I(inode), offset, size,
1341 create ? flags : BMAPI_READ, &iomap, &niomap);
1347 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1349 * For unwritten extents do not report a disk address on
1350 * the read case (treat as if we're reading into a hole).
1352 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1353 xfs_map_buffer(bh_result, &iomap, offset,
1356 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1358 bh_result->b_private = inode;
1359 set_buffer_unwritten(bh_result);
1364 * If this is a realtime file, data may be on a different device.
1365 * to that pointed to from the buffer_head b_bdev currently.
1367 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1370 * If we previously allocated a block out beyond eof and we are now
1371 * coming back to use it then we will need to flag it as new even if it
1372 * has a disk address.
1374 * With sub-block writes into unwritten extents we also need to mark
1375 * the buffer as new so that the unwritten parts of the buffer gets
1379 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1380 (offset >= i_size_read(inode)) ||
1381 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1382 set_buffer_new(bh_result);
1384 if (iomap.iomap_flags & IOMAP_DELAY) {
1387 set_buffer_uptodate(bh_result);
1388 set_buffer_mapped(bh_result);
1389 set_buffer_delay(bh_result);
1393 if (direct || size > (1 << inode->i_blkbits)) {
1394 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1395 offset = min_t(xfs_off_t,
1396 iomap.iomap_bsize - iomap.iomap_delta, size);
1397 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1405 struct inode *inode,
1407 struct buffer_head *bh_result,
1410 return __xfs_get_blocks(inode, iblock,
1411 bh_result, create, 0, BMAPI_WRITE);
1415 xfs_get_blocks_direct(
1416 struct inode *inode,
1418 struct buffer_head *bh_result,
1421 return __xfs_get_blocks(inode, iblock,
1422 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1432 xfs_ioend_t *ioend = iocb->private;
1435 * Non-NULL private data means we need to issue a transaction to
1436 * convert a range from unwritten to written extents. This needs
1437 * to happen from process context but aio+dio I/O completion
1438 * happens from irq context so we need to defer it to a workqueue.
1439 * This is not necessary for synchronous direct I/O, but we do
1440 * it anyway to keep the code uniform and simpler.
1442 * Well, if only it were that simple. Because synchronous direct I/O
1443 * requires extent conversion to occur *before* we return to userspace,
1444 * we have to wait for extent conversion to complete. Look at the
1445 * iocb that has been passed to us to determine if this is AIO or
1446 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1447 * workqueue and wait for it to complete.
1449 * The core direct I/O code might be changed to always call the
1450 * completion handler in the future, in which case all this can
1453 ioend->io_offset = offset;
1454 ioend->io_size = size;
1455 if (ioend->io_type == IOMAP_READ) {
1456 xfs_finish_ioend(ioend, 0);
1457 } else if (private && size > 0) {
1458 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1461 * A direct I/O write ioend starts it's life in unwritten
1462 * state in case they map an unwritten extent. This write
1463 * didn't map an unwritten extent so switch it's completion
1466 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1467 xfs_finish_ioend(ioend, 0);
1471 * blockdev_direct_IO can return an error even after the I/O
1472 * completion handler was called. Thus we need to protect
1473 * against double-freeing.
1475 iocb->private = NULL;
1482 const struct iovec *iov,
1484 unsigned long nr_segs)
1486 struct file *file = iocb->ki_filp;
1487 struct inode *inode = file->f_mapping->host;
1488 struct block_device *bdev;
1491 bdev = xfs_find_bdev_for_inode(XFS_I(inode));
1494 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1495 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1496 bdev, iov, offset, nr_segs,
1497 xfs_get_blocks_direct,
1500 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1501 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1502 bdev, iov, offset, nr_segs,
1503 xfs_get_blocks_direct,
1507 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1508 xfs_destroy_ioend(iocb->private);
1515 struct address_space *mapping,
1519 struct page **pagep,
1523 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1529 struct address_space *mapping,
1532 struct inode *inode = (struct inode *)mapping->host;
1533 struct xfs_inode *ip = XFS_I(inode);
1535 xfs_itrace_entry(XFS_I(inode));
1536 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1537 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1538 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1539 return generic_block_bmap(mapping, block, xfs_get_blocks);
1544 struct file *unused,
1547 return mpage_readpage(page, xfs_get_blocks);
1552 struct file *unused,
1553 struct address_space *mapping,
1554 struct list_head *pages,
1557 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1561 xfs_vm_invalidatepage(
1563 unsigned long offset)
1565 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1566 page->mapping->host, page, offset);
1567 block_invalidatepage(page, offset);
1570 const struct address_space_operations xfs_address_space_operations = {
1571 .readpage = xfs_vm_readpage,
1572 .readpages = xfs_vm_readpages,
1573 .writepage = xfs_vm_writepage,
1574 .writepages = xfs_vm_writepages,
1575 .sync_page = block_sync_page,
1576 .releasepage = xfs_vm_releasepage,
1577 .invalidatepage = xfs_vm_invalidatepage,
1578 .write_begin = xfs_vm_write_begin,
1579 .write_end = generic_write_end,
1580 .bmap = xfs_vm_bmap,
1581 .direct_IO = xfs_vm_direct_IO,
1582 .migratepage = buffer_migrate_page,