2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
33 * fs/xfs/linux/xfs_lrw.c (Linux Read Write stuff)
42 #include "xfs_trans.h"
47 #include "xfs_alloc.h"
48 #include "xfs_dmapi.h"
49 #include "xfs_quota.h"
50 #include "xfs_mount.h"
51 #include "xfs_alloc_btree.h"
52 #include "xfs_bmap_btree.h"
53 #include "xfs_ialloc_btree.h"
54 #include "xfs_btree.h"
55 #include "xfs_ialloc.h"
56 #include "xfs_attr_sf.h"
57 #include "xfs_dir_sf.h"
58 #include "xfs_dir2_sf.h"
59 #include "xfs_dinode.h"
60 #include "xfs_inode.h"
63 #include "xfs_rtalloc.h"
64 #include "xfs_error.h"
65 #include "xfs_itable.h"
71 #include "xfs_inode_item.h"
72 #include "xfs_buf_item.h"
73 #include "xfs_utils.h"
74 #include "xfs_iomap.h"
76 #include <linux/capability.h>
77 #include <linux/writeback.h>
80 #if defined(XFS_RW_TRACE)
90 xfs_inode_t *ip = XFS_IO_INODE(io);
92 if (ip->i_rwtrace == NULL)
94 ktrace_enter(ip->i_rwtrace,
95 (void *)(unsigned long)tag,
97 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
100 (void *)((unsigned long)segs),
101 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
102 (void *)((unsigned long)(offset & 0xffffffff)),
103 (void *)((unsigned long)ioflags),
104 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
105 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
114 xfs_inval_cached_trace(
121 xfs_inode_t *ip = XFS_IO_INODE(io);
123 if (ip->i_rwtrace == NULL)
125 ktrace_enter(ip->i_rwtrace,
126 (void *)(__psint_t)XFS_INVAL_CACHED,
128 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
129 (void *)((unsigned long)(offset & 0xffffffff)),
130 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
131 (void *)((unsigned long)(len & 0xffffffff)),
132 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
133 (void *)((unsigned long)(first & 0xffffffff)),
134 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
135 (void *)((unsigned long)(last & 0xffffffff)),
148 * xfs_iozero clears the specified range of buffer supplied,
149 * and marks all the affected blocks as valid and modified. If
150 * an affected block is not allocated, it will be allocated. If
151 * an affected block is not completely overwritten, and is not
152 * valid before the operation, it will be read from disk before
153 * being partially zeroed.
157 struct inode *ip, /* inode */
158 loff_t pos, /* offset in file */
159 size_t count, /* size of data to zero */
160 loff_t end_size) /* max file size to set */
164 struct address_space *mapping;
168 mapping = ip->i_mapping;
170 unsigned long index, offset;
172 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
173 index = pos >> PAGE_CACHE_SHIFT;
174 bytes = PAGE_CACHE_SIZE - offset;
179 page = grab_cache_page(mapping, index);
184 status = mapping->a_ops->prepare_write(NULL, page, offset,
190 memset((void *) (kaddr + offset), 0, bytes);
191 flush_dcache_page(page);
192 status = mapping->a_ops->commit_write(NULL, page, offset,
197 if (pos > i_size_read(ip))
198 i_size_write(ip, pos < end_size ? pos : end_size);
204 page_cache_release(page);
213 * xfs_inval_cached_pages
215 * This routine is responsible for keeping direct I/O and buffered I/O
216 * somewhat coherent. From here we make sure that we're at least
217 * temporarily holding the inode I/O lock exclusively and then call
218 * the page cache to flush and invalidate any cached pages. If there
219 * are no cached pages this routine will be very quick.
222 xfs_inval_cached_pages(
230 xfs_inval_cached_trace(io, offset, -1, ctooff(offtoct(offset)), -1);
231 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), -1, FI_REMAPF_LOCKED);
236 ssize_t /* bytes read, or (-) error */
240 const struct iovec *iovp,
246 struct file *file = iocb->ki_filp;
247 struct inode *inode = file->f_mapping->host;
256 ip = XFS_BHVTOI(bdp);
257 vp = BHV_TO_VNODE(bdp);
260 XFS_STATS_INC(xs_read_calls);
262 /* START copy & waste from filemap.c */
263 for (seg = 0; seg < segs; seg++) {
264 const struct iovec *iv = &iovp[seg];
267 * If any segment has a negative length, or the cumulative
268 * length ever wraps negative then return -EINVAL.
271 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
272 return XFS_ERROR(-EINVAL);
274 /* END copy & waste from filemap.c */
276 if (unlikely(ioflags & IO_ISDIRECT)) {
277 xfs_buftarg_t *target =
278 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
279 mp->m_rtdev_targp : mp->m_ddev_targp;
280 if ((*offset & target->pbr_smask) ||
281 (size & target->pbr_smask)) {
282 if (*offset == ip->i_d.di_size) {
285 return -XFS_ERROR(EINVAL);
289 n = XFS_MAXIOFFSET(mp) - *offset;
290 if ((n <= 0) || (size == 0))
296 if (XFS_FORCED_SHUTDOWN(mp)) {
300 if (unlikely(ioflags & IO_ISDIRECT))
302 xfs_ilock(ip, XFS_IOLOCK_SHARED);
304 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
305 !(ioflags & IO_INVIS)) {
306 vrwlock_t locktype = VRWLOCK_READ;
308 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
309 BHV_TO_VNODE(bdp), *offset, size,
310 FILP_DELAY_FLAG(file), &locktype);
312 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
317 xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
318 (void *)iovp, segs, *offset, ioflags);
319 ret = __generic_file_aio_read(iocb, iovp, segs, offset);
320 if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
321 ret = wait_on_sync_kiocb(iocb);
323 XFS_STATS_ADD(xs_read_bytes, ret);
325 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
327 if (likely(!(ioflags & IO_INVIS)))
328 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
331 if (unlikely(ioflags & IO_ISDIRECT))
353 ip = XFS_BHVTOI(bdp);
354 vp = BHV_TO_VNODE(bdp);
357 XFS_STATS_INC(xs_read_calls);
359 n = XFS_MAXIOFFSET(mp) - *offset;
360 if ((n <= 0) || (count == 0))
366 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
369 xfs_ilock(ip, XFS_IOLOCK_SHARED);
371 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
372 (!(ioflags & IO_INVIS))) {
373 vrwlock_t locktype = VRWLOCK_READ;
376 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp), *offset, count,
377 FILP_DELAY_FLAG(filp), &locktype);
379 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
383 xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
384 (void *)(unsigned long)target, count, *offset, ioflags);
385 ret = generic_file_sendfile(filp, offset, count, actor, target);
387 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
390 XFS_STATS_ADD(xs_read_bytes, ret);
392 if (likely(!(ioflags & IO_INVIS)))
393 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
399 * This routine is called to handle zeroing any space in the last
400 * block of the file that is beyond the EOF. We do this since the
401 * size is being increased without writing anything to that block
402 * and we don't want anyone to read the garbage on the disk.
404 STATIC int /* error (positive) */
410 xfs_fsize_t end_size)
412 xfs_fileoff_t last_fsb;
417 int isize_fsb_offset;
419 xfs_bmbt_irec_t imap;
423 ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
424 ASSERT(offset > isize);
428 isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
429 if (isize_fsb_offset == 0) {
431 * There are no extra bytes in the last block on disk to
437 last_fsb = XFS_B_TO_FSBT(mp, isize);
439 error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
446 * If the block underlying isize is just a hole, then there
447 * is nothing to zero.
449 if (imap.br_startblock == HOLESTARTBLOCK) {
453 * Zero the part of the last block beyond the EOF, and write it
454 * out sync. We need to drop the ilock while we do this so we
455 * don't deadlock when the buffer cache calls back to us.
457 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
458 loff = XFS_FSB_TO_B(mp, last_fsb);
459 lsize = XFS_FSB_TO_B(mp, 1);
461 zero_offset = isize_fsb_offset;
462 zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
464 error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
466 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
472 * Zero any on disk space between the current EOF and the new,
473 * larger EOF. This handles the normal case of zeroing the remainder
474 * of the last block in the file and the unusual case of zeroing blocks
475 * out beyond the size of the file. This second case only happens
476 * with fixed size extents and when the system crashes before the inode
477 * size was updated but after blocks were allocated. If fill is set,
478 * then any holes in the range are filled and zeroed. If not, the holes
479 * are left alone as holes.
482 int /* error (positive) */
486 xfs_off_t offset, /* starting I/O offset */
487 xfs_fsize_t isize, /* current inode size */
488 xfs_fsize_t end_size) /* terminal inode size */
490 struct inode *ip = LINVFS_GET_IP(vp);
491 xfs_fileoff_t start_zero_fsb;
492 xfs_fileoff_t end_zero_fsb;
493 xfs_fileoff_t prev_zero_fsb;
494 xfs_fileoff_t zero_count_fsb;
495 xfs_fileoff_t last_fsb;
496 xfs_extlen_t buf_len_fsb;
497 xfs_extlen_t prev_zero_count;
501 xfs_bmbt_irec_t imap;
505 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
506 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
511 * First handle zeroing the block on which isize resides.
512 * We only zero a part of that block so it is handled specially.
514 error = xfs_zero_last_block(ip, io, offset, isize, end_size);
516 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
517 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
522 * Calculate the range between the new size and the old
523 * where blocks needing to be zeroed may exist. To get the
524 * block where the last byte in the file currently resides,
525 * we need to subtract one from the size and truncate back
526 * to a block boundary. We subtract 1 in case the size is
527 * exactly on a block boundary.
529 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
530 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
531 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
532 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
533 if (last_fsb == end_zero_fsb) {
535 * The size was only incremented on its last block.
536 * We took care of that above, so just return.
541 ASSERT(start_zero_fsb <= end_zero_fsb);
542 prev_zero_fsb = NULLFILEOFF;
544 while (start_zero_fsb <= end_zero_fsb) {
546 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
547 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
548 0, NULL, 0, &imap, &nimaps, NULL);
550 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
551 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
556 if (imap.br_state == XFS_EXT_UNWRITTEN ||
557 imap.br_startblock == HOLESTARTBLOCK) {
559 * This loop handles initializing pages that were
560 * partially initialized by the code below this
561 * loop. It basically zeroes the part of the page
562 * that sits on a hole and sets the page as P_HOLE
563 * and calls remapf if it is a mapped file.
565 prev_zero_fsb = NULLFILEOFF;
567 start_zero_fsb = imap.br_startoff +
569 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
574 * There are blocks in the range requested.
575 * Zero them a single write at a time. We actually
576 * don't zero the entire range returned if it is
577 * too big and simply loop around to get the rest.
578 * That is not the most efficient thing to do, but it
579 * is simple and this path should not be exercised often.
581 buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
582 mp->m_writeio_blocks << 8);
584 * Drop the inode lock while we're doing the I/O.
585 * We'll still have the iolock to protect us.
587 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
589 loff = XFS_FSB_TO_B(mp, start_zero_fsb);
590 lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
592 error = xfs_iozero(ip, loff, lsize, end_size);
598 prev_zero_fsb = start_zero_fsb;
599 prev_zero_count = buf_len_fsb;
600 start_zero_fsb = imap.br_startoff + buf_len_fsb;
601 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
603 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
610 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
615 ssize_t /* bytes written, or (-) error */
619 const struct iovec *iovp,
625 struct file *file = iocb->ki_filp;
626 struct address_space *mapping = file->f_mapping;
627 struct inode *inode = mapping->host;
628 unsigned long segs = nsegs;
631 ssize_t ret = 0, error = 0;
632 xfs_fsize_t isize, new_size;
639 size_t ocount = 0, count;
641 int need_isem = 1, need_flush = 0;
643 XFS_STATS_INC(xs_write_calls);
645 vp = BHV_TO_VNODE(bdp);
646 xip = XFS_BHVTOI(bdp);
648 for (seg = 0; seg < segs; seg++) {
649 const struct iovec *iv = &iovp[seg];
652 * If any segment has a negative length, or the cumulative
653 * length ever wraps negative then return -EINVAL.
655 ocount += iv->iov_len;
656 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
658 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
663 ocount -= iv->iov_len; /* This segment is no good */
676 if (XFS_FORCED_SHUTDOWN(mp))
679 fs_check_frozen(vp->v_vfsp, SB_FREEZE_WRITE);
681 if (ioflags & IO_ISDIRECT) {
682 xfs_buftarg_t *target =
683 (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
684 mp->m_rtdev_targp : mp->m_ddev_targp;
686 if (ioflags & IO_ISAIO)
687 return XFS_ERROR(-ENOSYS);
689 if ((pos & target->pbr_smask) || (count & target->pbr_smask))
690 return XFS_ERROR(-EINVAL);
692 if (!VN_CACHED(vp) && pos < i_size_read(inode))
701 iolock = XFS_IOLOCK_EXCL;
702 locktype = VRWLOCK_WRITE;
706 iolock = XFS_IOLOCK_SHARED;
707 locktype = VRWLOCK_WRITE_DIRECT;
710 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
712 isize = i_size_read(inode);
714 if (file->f_flags & O_APPEND)
718 error = -generic_write_checks(file, &pos, &count,
719 S_ISBLK(inode->i_mode));
721 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
722 goto out_unlock_isem;
725 new_size = pos + count;
726 if (new_size > isize)
727 io->io_new_size = new_size;
729 if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
730 !(ioflags & IO_INVIS) && !eventsent)) {
731 loff_t savedsize = pos;
732 int dmflags = FILP_DELAY_FLAG(file);
735 dmflags |= DM_FLAGS_ISEM;
737 xfs_iunlock(xip, XFS_ILOCK_EXCL);
738 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
742 xfs_iunlock(xip, iolock);
743 goto out_unlock_isem;
745 xfs_ilock(xip, XFS_ILOCK_EXCL);
749 * The iolock was dropped and reaquired in XFS_SEND_DATA
750 * so we have to recheck the size when appending.
751 * We will only "goto start;" once, since having sent the
752 * event prevents another call to XFS_SEND_DATA, which is
753 * what allows the size to change in the first place.
755 if ((file->f_flags & O_APPEND) && savedsize != isize) {
756 pos = isize = xip->i_d.di_size;
762 * On Linux, generic_file_write updates the times even if
763 * no data is copied in so long as the write had a size.
765 * We must update xfs' times since revalidate will overcopy xfs.
767 if (!(ioflags & IO_INVIS)) {
768 xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
769 inode_update_time(inode, 1);
773 * If the offset is beyond the size of the file, we have a couple
774 * of things to do. First, if there is already space allocated
775 * we need to either create holes or zero the disk or ...
777 * If there is a page where the previous size lands, we need
778 * to zero it out up to the new size.
782 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, pos,
785 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
786 goto out_unlock_isem;
789 xfs_iunlock(xip, XFS_ILOCK_EXCL);
792 * If we're writing the file then make sure to clear the
793 * setuid and setgid bits if the process is not being run
794 * by root. This keeps people from modifying setuid and
798 if (((xip->i_d.di_mode & S_ISUID) ||
799 ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
800 (S_ISGID | S_IXGRP))) &&
801 !capable(CAP_FSETID)) {
802 error = xfs_write_clear_setuid(xip);
804 error = -remove_suid(file->f_dentry);
805 if (unlikely(error)) {
806 xfs_iunlock(xip, iolock);
807 goto out_unlock_isem;
812 /* We can write back this queue in page reclaim */
813 current->backing_dev_info = mapping->backing_dev_info;
815 if ((ioflags & IO_ISDIRECT)) {
817 xfs_inval_cached_trace(io, pos, -1,
818 ctooff(offtoct(pos)), -1);
819 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(pos)),
820 -1, FI_REMAPF_LOCKED);
824 /* demote the lock now the cached pages are gone */
825 XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
828 iolock = XFS_IOLOCK_SHARED;
829 locktype = VRWLOCK_WRITE_DIRECT;
833 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
835 ret = generic_file_direct_write(iocb, iovp,
836 &segs, pos, offset, count, ocount);
839 * direct-io write to a hole: fall through to buffered I/O
840 * for completing the rest of the request.
842 if (ret >= 0 && ret != count) {
843 XFS_STATS_ADD(xs_write_bytes, ret);
849 ioflags &= ~IO_ISDIRECT;
850 xfs_iunlock(xip, iolock);
854 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
856 ret = generic_file_buffered_write(iocb, iovp, segs,
857 pos, offset, count, ret);
860 current->backing_dev_info = NULL;
862 if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
863 ret = wait_on_sync_kiocb(iocb);
865 if ((ret == -ENOSPC) &&
866 DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
867 !(ioflags & IO_INVIS)) {
869 xfs_rwunlock(bdp, locktype);
870 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
871 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
872 0, 0, 0); /* Delay flag intentionally unused */
874 goto out_unlock_isem;
875 xfs_rwlock(bdp, locktype);
876 pos = xip->i_d.di_size;
881 if (*offset > xip->i_d.di_size) {
882 xfs_ilock(xip, XFS_ILOCK_EXCL);
883 if (*offset > xip->i_d.di_size) {
884 xip->i_d.di_size = *offset;
885 i_size_write(inode, *offset);
886 xip->i_update_core = 1;
887 xip->i_update_size = 1;
889 xfs_iunlock(xip, XFS_ILOCK_EXCL);
894 goto out_unlock_internal;
896 XFS_STATS_ADD(xs_write_bytes, ret);
898 /* Handle various SYNC-type writes */
899 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
901 * If we're treating this as O_DSYNC and we have not updated the
902 * size, force the log.
904 if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC) &&
905 !(xip->i_update_size)) {
906 xfs_inode_log_item_t *iip = xip->i_itemp;
909 * If an allocation transaction occurred
910 * without extending the size, then we have to force
911 * the log up the proper point to ensure that the
912 * allocation is permanent. We can't count on
913 * the fact that buffered writes lock out direct I/O
914 * writes - the direct I/O write could have extended
915 * the size nontransactionally, then finished before
916 * we started. xfs_write_file will think that the file
917 * didn't grow but the update isn't safe unless the
918 * size change is logged.
920 * Force the log if we've committed a transaction
921 * against the inode or if someone else has and
922 * the commit record hasn't gone to disk (e.g.
923 * the inode is pinned). This guarantees that
924 * all changes affecting the inode are permanent
927 if (iip && iip->ili_last_lsn) {
928 xfs_log_force(mp, iip->ili_last_lsn,
929 XFS_LOG_FORCE | XFS_LOG_SYNC);
930 } else if (xfs_ipincount(xip) > 0) {
931 xfs_log_force(mp, (xfs_lsn_t)0,
932 XFS_LOG_FORCE | XFS_LOG_SYNC);
939 * O_SYNC or O_DSYNC _with_ a size update are handled
942 * If the write was synchronous then we need to make
943 * sure that the inode modification time is permanent.
944 * We'll have updated the timestamp above, so here
945 * we use a synchronous transaction to log the inode.
946 * It's not fast, but it's necessary.
948 * If this a dsync write and the size got changed
949 * non-transactionally, then we need to ensure that
950 * the size change gets logged in a synchronous
954 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
955 if ((error = xfs_trans_reserve(tp, 0,
956 XFS_SWRITE_LOG_RES(mp),
958 /* Transaction reserve failed */
959 xfs_trans_cancel(tp, 0);
961 /* Transaction reserve successful */
962 xfs_ilock(xip, XFS_ILOCK_EXCL);
963 xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
964 xfs_trans_ihold(tp, xip);
965 xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
966 xfs_trans_set_sync(tp);
967 error = xfs_trans_commit(tp, 0, NULL);
968 xfs_iunlock(xip, XFS_ILOCK_EXCL);
971 goto out_unlock_internal;
974 xfs_rwunlock(bdp, locktype);
978 error = sync_page_range(inode, mapping, pos, ret);
985 xfs_rwunlock(bdp, locktype);
993 * All xfs metadata buffers except log state machine buffers
994 * get this attached as their b_bdstrat callback function.
995 * This is so that we can catch a buffer
996 * after prematurely unpinning it to forcibly shutdown the filesystem.
999 xfs_bdstrat_cb(struct xfs_buf *bp)
1003 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
1004 if (!XFS_FORCED_SHUTDOWN(mp)) {
1005 pagebuf_iorequest(bp);
1008 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
1010 * Metadata write that didn't get logged but
1011 * written delayed anyway. These aren't associated
1012 * with a transaction, and can be ignored.
1014 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
1015 (XFS_BUF_ISREAD(bp)) == 0)
1016 return (xfs_bioerror_relse(bp));
1018 return (xfs_bioerror(bp));
1024 xfs_bmap(bhv_desc_t *bdp,
1028 xfs_iomap_t *iomapp,
1031 xfs_inode_t *ip = XFS_BHVTOI(bdp);
1032 xfs_iocore_t *io = &ip->i_iocore;
1034 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
1035 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
1036 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
1038 return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
1042 * Wrapper around bdstrat so that we can stop data
1043 * from going to disk in case we are shutting down the filesystem.
1044 * Typically user data goes thru this path; one of the exceptions
1045 * is the superblock.
1049 struct xfs_mount *mp,
1053 if (!XFS_FORCED_SHUTDOWN(mp)) {
1054 /* Grio redirection would go here
1055 * if (XFS_BUF_IS_GRIO(bp)) {
1058 pagebuf_iorequest(bp);
1062 xfs_buftrace("XFSBDSTRAT IOERROR", bp);
1063 return (xfs_bioerror_relse(bp));
1067 * If the underlying (data/log/rt) device is readonly, there are some
1068 * operations that cannot proceed.
1071 xfs_dev_is_read_only(
1075 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1076 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1077 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1079 "XFS: %s required on read-only device.", message);
1081 "XFS: write access unavailable, cannot proceed.");