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
26 #include "xfs_trans.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir_sf.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_alloc.h"
38 #include "xfs_btree.h"
39 #include "xfs_error.h"
41 #include "xfs_iomap.h"
42 #include <linux/mpage.h>
43 #include <linux/pagevec.h>
44 #include <linux/writeback.h>
46 STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
48 #if defined(XFS_RW_TRACE)
57 vnode_t *vp = LINVFS_GET_VP(inode);
58 loff_t isize = i_size_read(inode);
59 loff_t offset = page_offset(page);
60 int delalloc = -1, unmapped = -1, unwritten = -1;
62 if (page_has_buffers(page))
63 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
69 ktrace_enter(ip->i_rwtrace,
70 (void *)((unsigned long)tag),
74 (void *)((unsigned long)mask),
75 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
76 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
77 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
78 (void *)((unsigned long)(isize & 0xffffffff)),
79 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
80 (void *)((unsigned long)(offset & 0xffffffff)),
81 (void *)((unsigned long)delalloc),
82 (void *)((unsigned long)unmapped),
83 (void *)((unsigned long)unwritten),
88 #define xfs_page_trace(tag, inode, page, mask)
92 * Schedule IO completion handling on a xfsdatad if this was
93 * the final hold on this ioend.
99 if (atomic_dec_and_test(&ioend->io_remaining))
100 queue_work(xfsdatad_workqueue, &ioend->io_work);
104 * We're now finished for good with this ioend structure.
105 * Update the page state via the associated buffer_heads,
106 * release holds on the inode and bio, and finally free
107 * up memory. Do not use the ioend after this.
113 struct buffer_head *bh, *next;
115 for (bh = ioend->io_buffer_head; bh; bh = next) {
116 next = bh->b_private;
117 bh->b_end_io(bh, ioend->io_uptodate);
120 vn_iowake(ioend->io_vnode);
121 mempool_free(ioend, xfs_ioend_pool);
125 * Buffered IO write completion for delayed allocate extents.
126 * TODO: Update ondisk isize now that we know the file data
127 * has been flushed (i.e. the notorious "NULL file" problem).
130 xfs_end_bio_delalloc(
133 xfs_ioend_t *ioend = data;
135 xfs_destroy_ioend(ioend);
139 * Buffered IO write completion for regular, written extents.
145 xfs_ioend_t *ioend = data;
147 xfs_destroy_ioend(ioend);
151 * IO write completion for unwritten extents.
153 * Issue transactions to convert a buffer range from unwritten
154 * to written extents.
157 xfs_end_bio_unwritten(
160 xfs_ioend_t *ioend = data;
161 vnode_t *vp = ioend->io_vnode;
162 xfs_off_t offset = ioend->io_offset;
163 size_t size = ioend->io_size;
166 if (ioend->io_uptodate)
167 VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
168 xfs_destroy_ioend(ioend);
172 * Allocate and initialise an IO completion structure.
173 * We need to track unwritten extent write completion here initially.
174 * We'll need to extend this for updating the ondisk inode size later
184 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
187 * Set the count to 1 initially, which will prevent an I/O
188 * completion callback from happening before we have started
189 * all the I/O from calling the completion routine too early.
191 atomic_set(&ioend->io_remaining, 1);
192 ioend->io_uptodate = 1; /* cleared if any I/O fails */
193 ioend->io_list = NULL;
194 ioend->io_type = type;
195 ioend->io_vnode = LINVFS_GET_VP(inode);
196 ioend->io_buffer_head = NULL;
197 ioend->io_buffer_tail = NULL;
198 atomic_inc(&ioend->io_vnode->v_iocount);
199 ioend->io_offset = 0;
202 if (type == IOMAP_UNWRITTEN)
203 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
204 else if (type == IOMAP_DELAY)
205 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend);
207 INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend);
220 vnode_t *vp = LINVFS_GET_VP(inode);
221 int error, nmaps = 1;
223 VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
224 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
234 return offset >= iomapp->iomap_offset &&
235 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
239 * BIO completion handler for buffered IO.
244 unsigned int bytes_done,
247 xfs_ioend_t *ioend = bio->bi_private;
253 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
255 /* Toss bio and pass work off to an xfsdatad thread */
256 if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
257 ioend->io_uptodate = 0;
258 bio->bi_private = NULL;
259 bio->bi_end_io = NULL;
262 xfs_finish_ioend(ioend);
267 xfs_submit_ioend_bio(
271 atomic_inc(&ioend->io_remaining);
273 bio->bi_private = ioend;
274 bio->bi_end_io = xfs_end_bio;
276 submit_bio(WRITE, bio);
277 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
283 struct buffer_head *bh)
286 int nvecs = bio_get_nr_vecs(bh->b_bdev);
289 bio = bio_alloc(GFP_NOIO, nvecs);
293 ASSERT(bio->bi_private == NULL);
294 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
295 bio->bi_bdev = bh->b_bdev;
301 xfs_start_buffer_writeback(
302 struct buffer_head *bh)
304 ASSERT(buffer_mapped(bh));
305 ASSERT(buffer_locked(bh));
306 ASSERT(!buffer_delay(bh));
307 ASSERT(!buffer_unwritten(bh));
309 mark_buffer_async_write(bh);
310 set_buffer_uptodate(bh);
311 clear_buffer_dirty(bh);
315 xfs_start_page_writeback(
317 struct writeback_control *wbc,
321 ASSERT(PageLocked(page));
322 ASSERT(!PageWriteback(page));
323 set_page_writeback(page);
325 clear_page_dirty(page);
328 end_page_writeback(page);
329 wbc->pages_skipped++; /* We didn't write this page */
333 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
335 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
339 * Submit all of the bios for all of the ioends we have saved up, covering the
340 * initial writepage page and also any probed pages.
342 * Because we may have multiple ioends spanning a page, we need to start
343 * writeback on all the buffers before we submit them for I/O. If we mark the
344 * buffers as we got, then we can end up with a page that only has buffers
345 * marked async write and I/O complete on can occur before we mark the other
346 * buffers async write.
348 * The end result of this is that we trip a bug in end_page_writeback() because
349 * we call it twice for the one page as the code in end_buffer_async_write()
350 * assumes that all buffers on the page are started at the same time.
352 * The fix is two passes across the ioend list - one to start writeback on the
353 * bufferheads, and then the second one submit them for I/O.
359 xfs_ioend_t *head = ioend;
361 struct buffer_head *bh;
363 sector_t lastblock = 0;
365 /* Pass 1 - start writeback */
367 next = ioend->io_list;
368 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
369 xfs_start_buffer_writeback(bh);
371 } while ((ioend = next) != NULL);
373 /* Pass 2 - submit I/O */
376 next = ioend->io_list;
379 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
383 bio = xfs_alloc_ioend_bio(bh);
384 } else if (bh->b_blocknr != lastblock + 1) {
385 xfs_submit_ioend_bio(ioend, bio);
389 if (bio_add_buffer(bio, bh) != bh->b_size) {
390 xfs_submit_ioend_bio(ioend, bio);
394 lastblock = bh->b_blocknr;
397 xfs_submit_ioend_bio(ioend, bio);
398 xfs_finish_ioend(ioend);
399 } while ((ioend = next) != NULL);
403 * Cancel submission of all buffer_heads so far in this endio.
404 * Toss the endio too. Only ever called for the initial page
405 * in a writepage request, so only ever one page.
412 struct buffer_head *bh, *next_bh;
415 next = ioend->io_list;
416 bh = ioend->io_buffer_head;
418 next_bh = bh->b_private;
419 clear_buffer_async_write(bh);
421 } while ((bh = next_bh) != NULL);
423 vn_iowake(ioend->io_vnode);
424 mempool_free(ioend, xfs_ioend_pool);
425 } while ((ioend = next) != NULL);
429 * Test to see if we've been building up a completion structure for
430 * earlier buffers -- if so, we try to append to this ioend if we
431 * can, otherwise we finish off any current ioend and start another.
432 * Return true if we've finished the given ioend.
437 struct buffer_head *bh,
440 xfs_ioend_t **result,
443 xfs_ioend_t *ioend = *result;
445 if (!ioend || need_ioend || type != ioend->io_type) {
446 xfs_ioend_t *previous = *result;
448 ioend = xfs_alloc_ioend(inode, type);
449 ioend->io_offset = offset;
450 ioend->io_buffer_head = bh;
451 ioend->io_buffer_tail = bh;
453 previous->io_list = ioend;
456 ioend->io_buffer_tail->b_private = bh;
457 ioend->io_buffer_tail = bh;
460 bh->b_private = NULL;
461 ioend->io_size += bh->b_size;
466 struct buffer_head *bh,
474 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
475 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
476 ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);
478 sector_shift = block_bits - BBSHIFT;
479 bn = (iomapp->iomap_bn >> sector_shift) +
480 ((offset - iomapp->iomap_offset) >> block_bits);
482 ASSERT(bn || (iomapp->iomap_flags & IOMAP_REALTIME));
483 ASSERT((bn << sector_shift) >= iomapp->iomap_bn);
487 bh->b_bdev = iomapp->iomap_target->bt_bdev;
488 set_buffer_mapped(bh);
489 clear_buffer_delay(bh);
490 clear_buffer_unwritten(bh);
494 * Look for a page at index that is suitable for clustering.
499 unsigned int pg_offset,
504 if (PageWriteback(page))
507 if (page->mapping && PageDirty(page)) {
508 if (page_has_buffers(page)) {
509 struct buffer_head *bh, *head;
511 bh = head = page_buffers(page);
513 if (!buffer_uptodate(bh))
515 if (mapped != buffer_mapped(bh))
518 if (ret >= pg_offset)
520 } while ((bh = bh->b_this_page) != head);
522 ret = mapped ? 0 : PAGE_CACHE_SIZE;
531 struct page *startpage,
532 struct buffer_head *bh,
533 struct buffer_head *head,
537 pgoff_t tindex, tlast, tloff;
541 /* First sum forwards in this page */
543 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
546 } while ((bh = bh->b_this_page) != head);
548 /* if we reached the end of the page, sum forwards in following pages */
549 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
550 tindex = startpage->index + 1;
552 /* Prune this back to avoid pathological behavior */
553 tloff = min(tlast, startpage->index + 64);
555 pagevec_init(&pvec, 0);
556 while (!done && tindex <= tloff) {
557 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
559 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
562 for (i = 0; i < pagevec_count(&pvec); i++) {
563 struct page *page = pvec.pages[i];
564 size_t pg_offset, len = 0;
566 if (tindex == tlast) {
568 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
574 pg_offset = PAGE_CACHE_SIZE;
576 if (page->index == tindex && !TestSetPageLocked(page)) {
577 len = xfs_probe_page(page, pg_offset, mapped);
590 pagevec_release(&pvec);
598 * Test if a given page is suitable for writing as part of an unwritten
599 * or delayed allocate extent.
606 if (PageWriteback(page))
609 if (page->mapping && page_has_buffers(page)) {
610 struct buffer_head *bh, *head;
613 bh = head = page_buffers(page);
615 if (buffer_unwritten(bh))
616 acceptable = (type == IOMAP_UNWRITTEN);
617 else if (buffer_delay(bh))
618 acceptable = (type == IOMAP_DELAY);
619 else if (buffer_mapped(bh))
620 acceptable = (type == 0);
623 } while ((bh = bh->b_this_page) != head);
633 * Allocate & map buffers for page given the extent map. Write it out.
634 * except for the original page of a writepage, this is called on
635 * delalloc/unwritten pages only, for the original page it is possible
636 * that the page has no mapping at all.
644 xfs_ioend_t **ioendp,
645 struct writeback_control *wbc,
649 struct buffer_head *bh, *head;
650 xfs_off_t end_offset;
651 unsigned long p_offset;
653 int bbits = inode->i_blkbits;
655 int count = 0, done = 0, uptodate = 1;
656 xfs_off_t offset = page_offset(page);
658 if (page->index != tindex)
660 if (TestSetPageLocked(page))
662 if (PageWriteback(page))
663 goto fail_unlock_page;
664 if (page->mapping != inode->i_mapping)
665 goto fail_unlock_page;
666 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
667 goto fail_unlock_page;
670 * page_dirty is initially a count of buffers on the page before
671 * EOF and is decrememted as we move each into a cleanable state.
675 * End offset is the highest offset that this page should represent.
676 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
677 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
678 * hence give us the correct page_dirty count. On any other page,
679 * it will be zero and in that case we need page_dirty to be the
680 * count of buffers on the page.
682 end_offset = min_t(unsigned long long,
683 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
686 len = 1 << inode->i_blkbits;
687 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
689 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
690 page_dirty = p_offset / len;
692 bh = head = page_buffers(page);
694 if (offset >= end_offset)
696 if (!buffer_uptodate(bh))
698 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
703 if (buffer_unwritten(bh) || buffer_delay(bh)) {
704 if (buffer_unwritten(bh))
705 type = IOMAP_UNWRITTEN;
709 if (!xfs_iomap_valid(mp, offset)) {
714 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
715 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
717 xfs_map_at_offset(bh, offset, bbits, mp);
719 xfs_add_to_ioend(inode, bh, offset,
722 set_buffer_dirty(bh);
724 mark_buffer_dirty(bh);
730 if (buffer_mapped(bh) && all_bh && startio) {
732 xfs_add_to_ioend(inode, bh, offset,
740 } while (offset += len, (bh = bh->b_this_page) != head);
742 if (uptodate && bh == head)
743 SetPageUptodate(page);
747 struct backing_dev_info *bdi;
749 bdi = inode->i_mapping->backing_dev_info;
751 if (bdi_write_congested(bdi)) {
752 wbc->encountered_congestion = 1;
754 } else if (wbc->nr_to_write <= 0) {
758 xfs_start_page_writeback(page, wbc, !page_dirty, count);
769 * Convert & write out a cluster of pages in the same extent as defined
770 * by mp and following the start page.
777 xfs_ioend_t **ioendp,
778 struct writeback_control *wbc,
786 pagevec_init(&pvec, 0);
787 while (!done && tindex <= tlast) {
788 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
790 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
793 for (i = 0; i < pagevec_count(&pvec); i++) {
794 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
795 iomapp, ioendp, wbc, startio, all_bh);
800 pagevec_release(&pvec);
806 * Calling this without startio set means we are being asked to make a dirty
807 * page ready for freeing it's buffers. When called with startio set then
808 * we are coming from writepage.
810 * When called with startio set it is important that we write the WHOLE
812 * The bh->b_state's cannot know if any of the blocks or which block for
813 * that matter are dirty due to mmap writes, and therefore bh uptodate is
814 * only vaild if the page itself isn't completely uptodate. Some layers
815 * may clear the page dirty flag prior to calling write page, under the
816 * assumption the entire page will be written out; by not writing out the
817 * whole page the page can be reused before all valid dirty data is
818 * written out. Note: in the case of a page that has been dirty'd by
819 * mapwrite and but partially setup by block_prepare_write the
820 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
821 * valid state, thus the whole page must be written out thing.
825 xfs_page_state_convert(
828 struct writeback_control *wbc,
830 int unmapped) /* also implies page uptodate */
832 struct buffer_head *bh, *head;
834 xfs_ioend_t *ioend = NULL, *iohead = NULL;
836 unsigned long p_offset = 0;
838 __uint64_t end_offset;
839 pgoff_t end_index, last_index, tlast;
841 int flags, err, iomap_valid = 0, uptodate = 1;
842 int page_dirty, count = 0, trylock_flag = 0;
843 int all_bh = unmapped;
845 /* wait for other IO threads? */
846 if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking))
847 trylock_flag |= BMAPI_TRYLOCK;
849 /* Is this page beyond the end of the file? */
850 offset = i_size_read(inode);
851 end_index = offset >> PAGE_CACHE_SHIFT;
852 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
853 if (page->index >= end_index) {
854 if ((page->index >= end_index + 1) ||
855 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
863 * page_dirty is initially a count of buffers on the page before
864 * EOF and is decrememted as we move each into a cleanable state.
868 * End offset is the highest offset that this page should represent.
869 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
870 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
871 * hence give us the correct page_dirty count. On any other page,
872 * it will be zero and in that case we need page_dirty to be the
873 * count of buffers on the page.
875 end_offset = min_t(unsigned long long,
876 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
877 len = 1 << inode->i_blkbits;
878 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
880 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
881 page_dirty = p_offset / len;
883 bh = head = page_buffers(page);
884 offset = page_offset(page);
888 /* TODO: cleanup count and page_dirty */
891 if (offset >= end_offset)
893 if (!buffer_uptodate(bh))
895 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
897 * the iomap is actually still valid, but the ioend
898 * isn't. shouldn't happen too often.
905 iomap_valid = xfs_iomap_valid(&iomap, offset);
908 * First case, map an unwritten extent and prepare for
909 * extent state conversion transaction on completion.
911 * Second case, allocate space for a delalloc buffer.
912 * We can return EAGAIN here in the release page case.
914 * Third case, an unmapped buffer was found, and we are
915 * in a path where we need to write the whole page out.
917 if (buffer_unwritten(bh) || buffer_delay(bh) ||
918 ((buffer_uptodate(bh) || PageUptodate(page)) &&
919 !buffer_mapped(bh) && (unmapped || startio))) {
921 * Make sure we don't use a read-only iomap
923 if (flags == BMAPI_READ)
926 if (buffer_unwritten(bh)) {
927 type = IOMAP_UNWRITTEN;
928 flags = BMAPI_WRITE|BMAPI_IGNSTATE;
929 } else if (buffer_delay(bh)) {
931 flags = BMAPI_ALLOCATE;
933 flags |= trylock_flag;
936 flags = BMAPI_WRITE|BMAPI_MMAP;
940 if (type == IOMAP_NEW) {
941 size = xfs_probe_cluster(inode,
947 err = xfs_map_blocks(inode, offset, size,
951 iomap_valid = xfs_iomap_valid(&iomap, offset);
954 xfs_map_at_offset(bh, offset,
955 inode->i_blkbits, &iomap);
957 xfs_add_to_ioend(inode, bh, offset,
961 set_buffer_dirty(bh);
963 mark_buffer_dirty(bh);
968 } else if (buffer_uptodate(bh) && startio) {
970 * we got here because the buffer is already mapped.
971 * That means it must already have extents allocated
972 * underneath it. Map the extent by reading it.
974 if (!iomap_valid || type != 0) {
976 size = xfs_probe_cluster(inode, page, bh,
978 err = xfs_map_blocks(inode, offset, size,
982 iomap_valid = xfs_iomap_valid(&iomap, offset);
986 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
987 ASSERT(buffer_mapped(bh));
990 xfs_add_to_ioend(inode, bh, offset, type,
991 &ioend, !iomap_valid);
997 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
998 (unmapped || startio)) {
1005 } while (offset += len, ((bh = bh->b_this_page) != head));
1007 if (uptodate && bh == head)
1008 SetPageUptodate(page);
1011 xfs_start_page_writeback(page, wbc, 1, count);
1013 if (ioend && iomap_valid) {
1014 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1016 tlast = min_t(pgoff_t, offset, last_index);
1017 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1018 wbc, startio, all_bh, tlast);
1022 xfs_submit_ioend(iohead);
1028 xfs_cancel_ioend(iohead);
1031 * If it's delalloc and we have nowhere to put it,
1032 * throw it away, unless the lower layers told
1035 if (err != -EAGAIN) {
1037 block_invalidatepage(page, 0);
1038 ClearPageUptodate(page);
1045 struct inode *inode,
1047 unsigned long blocks,
1048 struct buffer_head *bh_result,
1051 bmapi_flags_t flags)
1053 vnode_t *vp = LINVFS_GET_VP(inode);
1060 offset = (xfs_off_t)iblock << inode->i_blkbits;
1062 size = (ssize_t) min_t(xfs_off_t, LONG_MAX,
1063 (xfs_off_t)blocks << inode->i_blkbits);
1065 size = 1 << inode->i_blkbits;
1067 VOP_BMAP(vp, offset, size,
1068 create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
1075 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1079 /* For unwritten extents do not report a disk address on
1080 * the read case (treat as if we're reading into a hole).
1082 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1083 delta = offset - iomap.iomap_offset;
1084 delta >>= inode->i_blkbits;
1086 bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
1088 BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
1089 bh_result->b_blocknr = bn;
1090 set_buffer_mapped(bh_result);
1092 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1094 bh_result->b_private = inode;
1095 set_buffer_unwritten(bh_result);
1096 set_buffer_delay(bh_result);
1100 /* If this is a realtime file, data might be on a new device */
1101 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1103 /* If we previously allocated a block out beyond eof and
1104 * we are now coming back to use it then we will need to
1105 * flag it as new even if it has a disk address.
1108 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1109 (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
1110 set_buffer_new(bh_result);
1112 if (iomap.iomap_flags & IOMAP_DELAY) {
1115 set_buffer_uptodate(bh_result);
1116 set_buffer_mapped(bh_result);
1117 set_buffer_delay(bh_result);
1122 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1123 offset = min_t(xfs_off_t,
1124 iomap.iomap_bsize - iomap.iomap_delta,
1125 (xfs_off_t)blocks << inode->i_blkbits);
1126 bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset);
1134 struct inode *inode,
1136 struct buffer_head *bh_result,
1139 return __linvfs_get_block(inode, iblock, 0, bh_result,
1140 create, 0, BMAPI_WRITE);
1144 linvfs_get_blocks_direct(
1145 struct inode *inode,
1147 unsigned long max_blocks,
1148 struct buffer_head *bh_result,
1151 return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
1152 create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1156 linvfs_end_io_direct(
1162 xfs_ioend_t *ioend = iocb->private;
1165 * Non-NULL private data means we need to issue a transaction to
1166 * convert a range from unwritten to written extents. This needs
1167 * to happen from process contect but aio+dio I/O completion
1168 * happens from irq context so we need to defer it to a workqueue.
1169 * This is not nessecary for synchronous direct I/O, but we do
1170 * it anyway to keep the code uniform and simpler.
1172 * The core direct I/O code might be changed to always call the
1173 * completion handler in the future, in which case all this can
1176 if (private && size > 0) {
1177 ioend->io_offset = offset;
1178 ioend->io_size = size;
1179 xfs_finish_ioend(ioend);
1182 xfs_destroy_ioend(ioend);
1186 * blockdev_direct_IO can return an error even afer the I/O
1187 * completion handler was called. Thus we need to protect
1188 * against double-freeing.
1190 iocb->private = NULL;
1197 const struct iovec *iov,
1199 unsigned long nr_segs)
1201 struct file *file = iocb->ki_filp;
1202 struct inode *inode = file->f_mapping->host;
1203 vnode_t *vp = LINVFS_GET_VP(inode);
1209 VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
1213 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1215 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1216 iomap.iomap_target->bt_bdev,
1217 iov, offset, nr_segs,
1218 linvfs_get_blocks_direct,
1219 linvfs_end_io_direct);
1221 if (unlikely(ret <= 0 && iocb->private))
1222 xfs_destroy_ioend(iocb->private);
1229 struct address_space *mapping,
1232 struct inode *inode = (struct inode *)mapping->host;
1233 vnode_t *vp = LINVFS_GET_VP(inode);
1236 vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);
1238 VOP_RWLOCK(vp, VRWLOCK_READ);
1239 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
1240 VOP_RWUNLOCK(vp, VRWLOCK_READ);
1241 return generic_block_bmap(mapping, block, linvfs_get_block);
1246 struct file *unused,
1249 return mpage_readpage(page, linvfs_get_block);
1254 struct file *unused,
1255 struct address_space *mapping,
1256 struct list_head *pages,
1259 return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
1263 xfs_count_page_state(
1269 struct buffer_head *bh, *head;
1271 *delalloc = *unmapped = *unwritten = 0;
1273 bh = head = page_buffers(page);
1275 if (buffer_uptodate(bh) && !buffer_mapped(bh))
1277 else if (buffer_unwritten(bh) && !buffer_delay(bh))
1278 clear_buffer_unwritten(bh);
1279 else if (buffer_unwritten(bh))
1281 else if (buffer_delay(bh))
1283 } while ((bh = bh->b_this_page) != head);
1288 * writepage: Called from one of two places:
1290 * 1. we are flushing a delalloc buffer head.
1292 * 2. we are writing out a dirty page. Typically the page dirty
1293 * state is cleared before we get here. In this case is it
1294 * conceivable we have no buffer heads.
1296 * For delalloc space on the page we need to allocate space and
1297 * flush it. For unmapped buffer heads on the page we should
1298 * allocate space if the page is uptodate. For any other dirty
1299 * buffer heads on the page we should flush them.
1301 * If we detect that a transaction would be required to flush
1302 * the page, we have to check the process flags first, if we
1303 * are already in a transaction or disk I/O during allocations
1304 * is off, we need to fail the writepage and redirty the page.
1310 struct writeback_control *wbc)
1314 int delalloc, unmapped, unwritten;
1315 struct inode *inode = page->mapping->host;
1317 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1320 * We need a transaction if:
1321 * 1. There are delalloc buffers on the page
1322 * 2. The page is uptodate and we have unmapped buffers
1323 * 3. The page is uptodate and we have no buffers
1324 * 4. There are unwritten buffers on the page
1327 if (!page_has_buffers(page)) {
1331 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1332 if (!PageUptodate(page))
1334 need_trans = delalloc + unmapped + unwritten;
1338 * If we need a transaction and the process flags say
1339 * we are already in a transaction, or no IO is allowed
1340 * then mark the page dirty again and leave the page
1343 if (PFLAGS_TEST_FSTRANS() && need_trans)
1347 * Delay hooking up buffer heads until we have
1348 * made our go/no-go decision.
1350 if (!page_has_buffers(page))
1351 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1354 * Convert delayed allocate, unwritten or unmapped space
1355 * to real space and flush out to disk.
1357 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1358 if (error == -EAGAIN)
1360 if (unlikely(error < 0))
1366 redirty_page_for_writepage(wbc, page);
1375 linvfs_invalidate_page(
1377 unsigned long offset)
1379 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1380 page->mapping->host, page, offset);
1381 return block_invalidatepage(page, offset);
1385 * Called to move a page into cleanable state - and from there
1386 * to be released. Possibly the page is already clean. We always
1387 * have buffer heads in this call.
1389 * Returns 0 if the page is ok to release, 1 otherwise.
1391 * Possible scenarios are:
1393 * 1. We are being called to release a page which has been written
1394 * to via regular I/O. buffer heads will be dirty and possibly
1395 * delalloc. If no delalloc buffer heads in this case then we
1396 * can just return zero.
1398 * 2. We are called to release a page which has been written via
1399 * mmap, all we need to do is ensure there is no delalloc
1400 * state in the buffer heads, if not we can let the caller
1401 * free them and we should come back later via writepage.
1404 linvfs_release_page(
1408 struct inode *inode = page->mapping->host;
1409 int dirty, delalloc, unmapped, unwritten;
1410 struct writeback_control wbc = {
1411 .sync_mode = WB_SYNC_ALL,
1415 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1417 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1418 if (!delalloc && !unwritten)
1421 if (!(gfp_mask & __GFP_FS))
1424 /* If we are already inside a transaction or the thread cannot
1425 * do I/O, we cannot release this page.
1427 if (PFLAGS_TEST_FSTRANS())
1431 * Convert delalloc space to real space, do not flush the
1432 * data out to disk, that will be done by the caller.
1433 * Never need to allocate space here - we will always
1434 * come back to writepage in that case.
1436 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1437 if (dirty == 0 && !unwritten)
1442 return try_to_free_buffers(page);
1446 linvfs_prepare_write(
1452 return block_prepare_write(page, from, to, linvfs_get_block);
1455 struct address_space_operations linvfs_aops = {
1456 .readpage = linvfs_readpage,
1457 .readpages = linvfs_readpages,
1458 .writepage = linvfs_writepage,
1459 .sync_page = block_sync_page,
1460 .releasepage = linvfs_release_page,
1461 .invalidatepage = linvfs_invalidate_page,
1462 .prepare_write = linvfs_prepare_write,
1463 .commit_write = generic_commit_write,
1464 .bmap = linvfs_bmap,
1465 .direct_IO = linvfs_direct_IO,
1466 .migratepage = buffer_migrate_page,