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>
79 #if defined(XFS_RW_TRACE)
84 const struct iovec *iovp,
89 xfs_inode_t *ip = XFS_IO_INODE(io);
91 if (ip->i_rwtrace == NULL)
93 ktrace_enter(ip->i_rwtrace,
94 (void *)(unsigned long)tag,
96 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
97 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
98 (void *)(__psint_t)iovp,
99 (void *)((unsigned long)segs),
100 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
101 (void *)((unsigned long)(offset & 0xffffffff)),
102 (void *)((unsigned long)ioflags),
103 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
104 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
113 xfs_inval_cached_trace(
120 xfs_inode_t *ip = XFS_IO_INODE(io);
122 if (ip->i_rwtrace == NULL)
124 ktrace_enter(ip->i_rwtrace,
125 (void *)(__psint_t)XFS_INVAL_CACHED,
127 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
128 (void *)((unsigned long)(offset & 0xffffffff)),
129 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
130 (void *)((unsigned long)(len & 0xffffffff)),
131 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
132 (void *)((unsigned long)(first & 0xffffffff)),
133 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
134 (void *)((unsigned long)(last & 0xffffffff)),
147 * xfs_iozero clears the specified range of buffer supplied,
148 * and marks all the affected blocks as valid and modified. If
149 * an affected block is not allocated, it will be allocated. If
150 * an affected block is not completely overwritten, and is not
151 * valid before the operation, it will be read from disk before
152 * being partially zeroed.
156 struct inode *ip, /* inode */
157 loff_t pos, /* offset in file */
158 size_t count, /* size of data to zero */
159 loff_t end_size) /* max file size to set */
163 struct address_space *mapping;
167 mapping = ip->i_mapping;
169 unsigned long index, offset;
171 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
172 index = pos >> PAGE_CACHE_SHIFT;
173 bytes = PAGE_CACHE_SIZE - offset;
178 page = grab_cache_page(mapping, index);
183 status = mapping->a_ops->prepare_write(NULL, page, offset,
189 memset((void *) (kaddr + offset), 0, bytes);
190 flush_dcache_page(page);
191 status = mapping->a_ops->commit_write(NULL, page, offset,
196 if (pos > i_size_read(ip))
197 i_size_write(ip, pos < end_size ? pos : end_size);
203 page_cache_release(page);
212 * xfs_inval_cached_pages
214 * This routine is responsible for keeping direct I/O and buffered I/O
215 * somewhat coherent. From here we make sure that we're at least
216 * temporarily holding the inode I/O lock exclusively and then call
217 * the page cache to flush and invalidate any cached pages. If there
218 * are no cached pages this routine will be very quick.
221 xfs_inval_cached_pages(
232 if (!VN_CACHED(vp)) {
240 * We need to get the I/O lock exclusively in order
241 * to safely invalidate pages and mappings.
244 XFS_IUNLOCK(mp, io, XFS_IOLOCK_SHARED);
245 XFS_ILOCK(mp, io, XFS_IOLOCK_EXCL);
248 /* Writing beyond EOF creates a hole that must be zeroed */
249 if (write && (offset > XFS_SIZE(mp, io))) {
252 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
253 isize = XFS_SIZE(mp, io);
254 if (offset > isize) {
255 xfs_zero_eof(vp, io, offset, isize, offset);
257 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
261 xfs_inval_cached_trace(io, offset, -1, ctooff(offtoct(offset)), -1);
262 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), -1, FI_REMAPF_LOCKED);
265 XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
270 ssize_t /* bytes read, or (-) error */
274 const struct iovec *iovp,
280 struct file *file = iocb->ki_filp;
289 ip = XFS_BHVTOI(bdp);
290 vp = BHV_TO_VNODE(bdp);
293 XFS_STATS_INC(xs_read_calls);
295 /* START copy & waste from filemap.c */
296 for (seg = 0; seg < segs; seg++) {
297 const struct iovec *iv = &iovp[seg];
300 * If any segment has a negative length, or the cumulative
301 * length ever wraps negative then return -EINVAL.
304 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
305 return XFS_ERROR(-EINVAL);
307 /* END copy & waste from filemap.c */
309 if (ioflags & IO_ISDIRECT) {
310 xfs_buftarg_t *target =
311 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
312 mp->m_rtdev_targp : mp->m_ddev_targp;
313 if ((*offset & target->pbr_smask) ||
314 (size & target->pbr_smask)) {
315 if (*offset == ip->i_d.di_size) {
318 return -XFS_ERROR(EINVAL);
322 n = XFS_MAXIOFFSET(mp) - *offset;
323 if ((n <= 0) || (size == 0))
329 if (XFS_FORCED_SHUTDOWN(mp)) {
333 /* OK so we are holding the I/O lock for the duration
334 * of the submission, then what happens if the I/O
335 * does not really happen here, but is scheduled
338 xfs_ilock(ip, XFS_IOLOCK_SHARED);
340 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
341 !(ioflags & IO_INVIS)) {
342 vrwlock_t locktype = VRWLOCK_READ;
344 ret = XFS_SEND_DATA(mp, DM_EVENT_READ,
345 BHV_TO_VNODE(bdp), *offset, size,
346 FILP_DELAY_FLAG(file), &locktype);
348 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
353 xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
354 iovp, segs, *offset, ioflags);
355 ret = __generic_file_aio_read(iocb, iovp, segs, offset);
356 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
359 XFS_STATS_ADD(xs_read_bytes, ret);
361 if (likely(!(ioflags & IO_INVIS)))
362 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
384 ip = XFS_BHVTOI(bdp);
385 vp = BHV_TO_VNODE(bdp);
388 XFS_STATS_INC(xs_read_calls);
390 n = XFS_MAXIOFFSET(mp) - *offset;
391 if ((n <= 0) || (count == 0))
397 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
400 xfs_ilock(ip, XFS_IOLOCK_SHARED);
402 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
403 (!(ioflags & IO_INVIS))) {
404 vrwlock_t locktype = VRWLOCK_READ;
407 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp), *offset, count,
408 FILP_DELAY_FLAG(filp), &locktype);
410 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
414 xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
415 target, count, *offset, ioflags);
416 ret = generic_file_sendfile(filp, offset, count, actor, target);
417 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
419 XFS_STATS_ADD(xs_read_bytes, ret);
420 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
425 * This routine is called to handle zeroing any space in the last
426 * block of the file that is beyond the EOF. We do this since the
427 * size is being increased without writing anything to that block
428 * and we don't want anyone to read the garbage on the disk.
430 STATIC int /* error (positive) */
436 xfs_fsize_t end_size)
438 xfs_fileoff_t last_fsb;
443 int isize_fsb_offset;
445 xfs_bmbt_irec_t imap;
449 ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
450 ASSERT(offset > isize);
454 isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
455 if (isize_fsb_offset == 0) {
457 * There are no extra bytes in the last block on disk to
463 last_fsb = XFS_B_TO_FSBT(mp, isize);
465 error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
472 * If the block underlying isize is just a hole, then there
473 * is nothing to zero.
475 if (imap.br_startblock == HOLESTARTBLOCK) {
479 * Zero the part of the last block beyond the EOF, and write it
480 * out sync. We need to drop the ilock while we do this so we
481 * don't deadlock when the buffer cache calls back to us.
483 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
484 loff = XFS_FSB_TO_B(mp, last_fsb);
485 lsize = XFS_FSB_TO_B(mp, 1);
487 zero_offset = isize_fsb_offset;
488 zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
490 error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
492 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
498 * Zero any on disk space between the current EOF and the new,
499 * larger EOF. This handles the normal case of zeroing the remainder
500 * of the last block in the file and the unusual case of zeroing blocks
501 * out beyond the size of the file. This second case only happens
502 * with fixed size extents and when the system crashes before the inode
503 * size was updated but after blocks were allocated. If fill is set,
504 * then any holes in the range are filled and zeroed. If not, the holes
505 * are left alone as holes.
508 int /* error (positive) */
512 xfs_off_t offset, /* starting I/O offset */
513 xfs_fsize_t isize, /* current inode size */
514 xfs_fsize_t end_size) /* terminal inode size */
516 struct inode *ip = LINVFS_GET_IP(vp);
517 xfs_fileoff_t start_zero_fsb;
518 xfs_fileoff_t end_zero_fsb;
519 xfs_fileoff_t prev_zero_fsb;
520 xfs_fileoff_t zero_count_fsb;
521 xfs_fileoff_t last_fsb;
522 xfs_extlen_t buf_len_fsb;
523 xfs_extlen_t prev_zero_count;
527 xfs_bmbt_irec_t imap;
531 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
532 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
537 * First handle zeroing the block on which isize resides.
538 * We only zero a part of that block so it is handled specially.
540 error = xfs_zero_last_block(ip, io, offset, isize, end_size);
542 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
543 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
548 * Calculate the range between the new size and the old
549 * where blocks needing to be zeroed may exist. To get the
550 * block where the last byte in the file currently resides,
551 * we need to subtract one from the size and truncate back
552 * to a block boundary. We subtract 1 in case the size is
553 * exactly on a block boundary.
555 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
556 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
557 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
558 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
559 if (last_fsb == end_zero_fsb) {
561 * The size was only incremented on its last block.
562 * We took care of that above, so just return.
567 ASSERT(start_zero_fsb <= end_zero_fsb);
568 prev_zero_fsb = NULLFILEOFF;
570 while (start_zero_fsb <= end_zero_fsb) {
572 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
573 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
574 0, NULL, 0, &imap, &nimaps, NULL);
576 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
577 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
582 if (imap.br_state == XFS_EXT_UNWRITTEN ||
583 imap.br_startblock == HOLESTARTBLOCK) {
585 * This loop handles initializing pages that were
586 * partially initialized by the code below this
587 * loop. It basically zeroes the part of the page
588 * that sits on a hole and sets the page as P_HOLE
589 * and calls remapf if it is a mapped file.
591 prev_zero_fsb = NULLFILEOFF;
593 start_zero_fsb = imap.br_startoff +
595 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
600 * There are blocks in the range requested.
601 * Zero them a single write at a time. We actually
602 * don't zero the entire range returned if it is
603 * too big and simply loop around to get the rest.
604 * That is not the most efficient thing to do, but it
605 * is simple and this path should not be exercised often.
607 buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
608 mp->m_writeio_blocks << 8);
610 * Drop the inode lock while we're doing the I/O.
611 * We'll still have the iolock to protect us.
613 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
615 loff = XFS_FSB_TO_B(mp, start_zero_fsb);
616 lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
618 error = xfs_iozero(ip, loff, lsize, end_size);
624 prev_zero_fsb = start_zero_fsb;
625 prev_zero_count = buf_len_fsb;
626 start_zero_fsb = imap.br_startoff + buf_len_fsb;
627 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
629 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
636 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
641 ssize_t /* bytes written, or (-) error */
645 const struct iovec *iovp,
651 struct file *file = iocb->ki_filp;
657 xfs_fsize_t isize, new_size;
658 xfs_fsize_t n, limit;
666 XFS_STATS_INC(xs_write_calls);
668 vp = BHV_TO_VNODE(bdp);
669 xip = XFS_BHVTOI(bdp);
671 /* START copy & waste from filemap.c */
672 for (seg = 0; seg < segs; seg++) {
673 const struct iovec *iv = &iovp[seg];
676 * If any segment has a negative length, or the cumulative
677 * length ever wraps negative then return -EINVAL.
680 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
681 return XFS_ERROR(-EINVAL);
683 /* END copy & waste from filemap.c */
691 if (XFS_FORCED_SHUTDOWN(mp)) {
695 if (ioflags & IO_ISDIRECT) {
696 xfs_buftarg_t *target =
697 (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
698 mp->m_rtdev_targp : mp->m_ddev_targp;
700 if ((*offset & target->pbr_smask) ||
701 (size & target->pbr_smask)) {
702 return XFS_ERROR(-EINVAL);
705 iolock = XFS_IOLOCK_SHARED;
706 locktype = VRWLOCK_WRITE_DIRECT;
708 iolock = XFS_IOLOCK_EXCL;
709 locktype = VRWLOCK_WRITE;
713 iolock = XFS_IOLOCK_EXCL;
714 locktype = VRWLOCK_WRITE;
717 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
719 isize = xip->i_d.di_size;
720 limit = XFS_MAXIOFFSET(mp);
722 if (file->f_flags & O_APPEND)
728 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
735 new_size = *offset + size;
736 if (new_size > isize) {
737 io->io_new_size = new_size;
740 if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
741 !(ioflags & IO_INVIS) && !eventsent)) {
742 loff_t savedsize = *offset;
743 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
745 xfs_iunlock(xip, XFS_ILOCK_EXCL);
746 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
750 xfs_iunlock(xip, iolock);
753 xfs_ilock(xip, XFS_ILOCK_EXCL);
757 * The iolock was dropped and reaquired in XFS_SEND_DATA
758 * so we have to recheck the size when appending.
759 * We will only "goto start;" once, since having sent the
760 * event prevents another call to XFS_SEND_DATA, which is
761 * what allows the size to change in the first place.
763 if ((file->f_flags & O_APPEND) &&
764 savedsize != xip->i_d.di_size) {
765 *offset = isize = xip->i_d.di_size;
771 * On Linux, generic_file_write updates the times even if
772 * no data is copied in so long as the write had a size.
774 * We must update xfs' times since revalidate will overcopy xfs.
776 if (size && !(ioflags & IO_INVIS))
777 xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
780 * If the offset is beyond the size of the file, we have a couple
781 * of things to do. First, if there is already space allocated
782 * we need to either create holes or zero the disk or ...
784 * If there is a page where the previous size lands, we need
785 * to zero it out up to the new size.
788 if ((0 && !(ioflags & IO_ISDIRECT)) && (*offset > isize && isize)) {
789 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, *offset,
790 isize, *offset + size);
792 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
796 xfs_iunlock(xip, XFS_ILOCK_EXCL);
799 * If we're writing the file then make sure to clear the
800 * setuid and setgid bits if the process is not being run
801 * by root. This keeps people from modifying setuid and
805 if (((xip->i_d.di_mode & S_ISUID) ||
806 ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
807 (S_ISGID | S_IXGRP))) &&
808 !capable(CAP_FSETID)) {
809 error = xfs_write_clear_setuid(xip);
811 xfs_iunlock(xip, iolock);
817 if (ioflags & IO_ISDIRECT) {
818 xfs_inval_cached_pages(vp, io, *offset, 1, 1);
819 xfs_rw_enter_trace(XFS_DIOWR_ENTER,
820 io, iovp, segs, *offset, ioflags);
822 xfs_rw_enter_trace(XFS_WRITE_ENTER,
823 io, iovp, segs, *offset, ioflags);
825 ret = generic_file_aio_write_nolock(iocb, iovp, segs, offset);
827 if ((ret == -ENOSPC) &&
828 DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
829 !(ioflags & IO_INVIS)) {
831 xfs_rwunlock(bdp, locktype);
832 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
833 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
834 0, 0, 0); /* Delay flag intentionally unused */
837 xfs_rwlock(bdp, locktype);
838 *offset = xip->i_d.di_size;
842 if (*offset > xip->i_d.di_size) {
843 xfs_ilock(xip, XFS_ILOCK_EXCL);
844 if (*offset > xip->i_d.di_size) {
845 struct inode *inode = LINVFS_GET_IP(vp);
847 xip->i_d.di_size = *offset;
848 i_size_write(inode, *offset);
849 xip->i_update_core = 1;
850 xip->i_update_size = 1;
852 xfs_iunlock(xip, XFS_ILOCK_EXCL);
856 xfs_rwunlock(bdp, locktype);
860 XFS_STATS_ADD(xs_write_bytes, ret);
862 /* Handle various SYNC-type writes */
863 if ((file->f_flags & O_SYNC) || IS_SYNC(file->f_dentry->d_inode)) {
866 * If we're treating this as O_DSYNC and we have not updated the
867 * size, force the log.
870 if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC)
871 && !(xip->i_update_size)) {
873 * If an allocation transaction occurred
874 * without extending the size, then we have to force
875 * the log up the proper point to ensure that the
876 * allocation is permanent. We can't count on
877 * the fact that buffered writes lock out direct I/O
878 * writes - the direct I/O write could have extended
879 * the size nontransactionally, then finished before
880 * we started. xfs_write_file will think that the file
881 * didn't grow but the update isn't safe unless the
882 * size change is logged.
884 * Force the log if we've committed a transaction
885 * against the inode or if someone else has and
886 * the commit record hasn't gone to disk (e.g.
887 * the inode is pinned). This guarantees that
888 * all changes affecting the inode are permanent
892 xfs_inode_log_item_t *iip;
896 if (iip && iip->ili_last_lsn) {
897 lsn = iip->ili_last_lsn;
898 xfs_log_force(mp, lsn,
899 XFS_LOG_FORCE | XFS_LOG_SYNC);
900 } else if (xfs_ipincount(xip) > 0) {
901 xfs_log_force(mp, (xfs_lsn_t)0,
902 XFS_LOG_FORCE | XFS_LOG_SYNC);
909 * O_SYNC or O_DSYNC _with_ a size update are handled
912 * If the write was synchronous then we need to make
913 * sure that the inode modification time is permanent.
914 * We'll have updated the timestamp above, so here
915 * we use a synchronous transaction to log the inode.
916 * It's not fast, but it's necessary.
918 * If this a dsync write and the size got changed
919 * non-transactionally, then we need to ensure that
920 * the size change gets logged in a synchronous
924 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
925 if ((error = xfs_trans_reserve(tp, 0,
926 XFS_SWRITE_LOG_RES(mp),
928 /* Transaction reserve failed */
929 xfs_trans_cancel(tp, 0);
931 /* Transaction reserve successful */
932 xfs_ilock(xip, XFS_ILOCK_EXCL);
933 xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
934 xfs_trans_ihold(tp, xip);
935 xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
936 xfs_trans_set_sync(tp);
937 error = xfs_trans_commit(tp, 0, NULL);
938 xfs_iunlock(xip, XFS_ILOCK_EXCL);
941 } /* (ioflags & O_SYNC) */
943 xfs_rwunlock(bdp, locktype);
948 * All xfs metadata buffers except log state machine buffers
949 * get this attached as their b_bdstrat callback function.
950 * This is so that we can catch a buffer
951 * after prematurely unpinning it to forcibly shutdown the filesystem.
954 xfs_bdstrat_cb(struct xfs_buf *bp)
958 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
959 if (!XFS_FORCED_SHUTDOWN(mp)) {
960 pagebuf_iorequest(bp);
963 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
965 * Metadata write that didn't get logged but
966 * written delayed anyway. These aren't associated
967 * with a transaction, and can be ignored.
969 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
970 (XFS_BUF_ISREAD(bp)) == 0)
971 return (xfs_bioerror_relse(bp));
973 return (xfs_bioerror(bp));
979 xfs_bmap(bhv_desc_t *bdp,
986 xfs_inode_t *ip = XFS_BHVTOI(bdp);
987 xfs_iocore_t *io = &ip->i_iocore;
989 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
990 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
991 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
993 return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
997 * Wrapper around bdstrat so that we can stop data
998 * from going to disk in case we are shutting down the filesystem.
999 * Typically user data goes thru this path; one of the exceptions
1000 * is the superblock.
1004 struct xfs_mount *mp,
1008 if (!XFS_FORCED_SHUTDOWN(mp)) {
1009 /* Grio redirection would go here
1010 * if (XFS_BUF_IS_GRIO(bp)) {
1013 pagebuf_iorequest(bp);
1017 xfs_buftrace("XFSBDSTRAT IOERROR", bp);
1018 return (xfs_bioerror_relse(bp));
1022 * If the underlying (data/log/rt) device is readonly, there are some
1023 * operations that cannot proceed.
1026 xfs_dev_is_read_only(
1030 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1031 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1032 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1034 "XFS: %s required on read-only device.", message);
1036 "XFS: write access unavailable, cannot proceed.");