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(
230 if (!VN_CACHED(vp)) {
237 * We need to get the I/O lock exclusively in order
238 * to safely invalidate pages and mappings.
241 XFS_IUNLOCK(mp, io, XFS_IOLOCK_SHARED);
242 XFS_ILOCK(mp, io, XFS_IOLOCK_EXCL);
245 /* Writing beyond EOF creates a hole that must be zeroed */
246 if (write && (offset > XFS_SIZE(mp, io))) {
249 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
250 isize = XFS_SIZE(mp, io);
251 if (offset > isize) {
252 xfs_zero_eof(vp, io, offset, isize, offset);
254 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
257 xfs_inval_cached_trace(io, offset, -1, ctooff(offtoct(offset)), -1);
258 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), -1, FI_REMAPF_LOCKED);
260 XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
264 ssize_t /* bytes read, or (-) error */
268 const struct iovec *iovp,
274 struct file *file = iocb->ki_filp;
283 ip = XFS_BHVTOI(bdp);
284 vp = BHV_TO_VNODE(bdp);
287 XFS_STATS_INC(xs_read_calls);
289 /* START copy & waste from filemap.c */
290 for (seg = 0; seg < segs; seg++) {
291 const struct iovec *iv = &iovp[seg];
294 * If any segment has a negative length, or the cumulative
295 * length ever wraps negative then return -EINVAL.
298 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
299 return XFS_ERROR(-EINVAL);
301 /* END copy & waste from filemap.c */
303 if (ioflags & IO_ISDIRECT) {
304 xfs_buftarg_t *target =
305 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
306 mp->m_rtdev_targp : mp->m_ddev_targp;
307 if ((*offset & target->pbr_smask) ||
308 (size & target->pbr_smask)) {
309 if (*offset == ip->i_d.di_size) {
312 return -XFS_ERROR(EINVAL);
316 n = XFS_MAXIOFFSET(mp) - *offset;
317 if ((n <= 0) || (size == 0))
323 if (XFS_FORCED_SHUTDOWN(mp)) {
327 /* OK so we are holding the I/O lock for the duration
328 * of the submission, then what happens if the I/O
329 * does not really happen here, but is scheduled
332 xfs_ilock(ip, XFS_IOLOCK_SHARED);
334 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
335 !(ioflags & IO_INVIS)) {
336 vrwlock_t locktype = VRWLOCK_READ;
338 ret = XFS_SEND_DATA(mp, DM_EVENT_READ,
339 BHV_TO_VNODE(bdp), *offset, size,
340 FILP_DELAY_FLAG(file), &locktype);
342 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
347 xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
348 iovp, segs, *offset, ioflags);
349 ret = __generic_file_aio_read(iocb, iovp, segs, offset);
350 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
353 XFS_STATS_ADD(xs_read_bytes, ret);
355 if (likely(!(ioflags & IO_INVIS)))
356 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
378 ip = XFS_BHVTOI(bdp);
379 vp = BHV_TO_VNODE(bdp);
382 XFS_STATS_INC(xs_read_calls);
384 n = XFS_MAXIOFFSET(mp) - *offset;
385 if ((n <= 0) || (count == 0))
391 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
394 xfs_ilock(ip, XFS_IOLOCK_SHARED);
396 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
397 (!(ioflags & IO_INVIS))) {
398 vrwlock_t locktype = VRWLOCK_READ;
401 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp), *offset, count,
402 FILP_DELAY_FLAG(filp), &locktype);
404 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
408 xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
409 target, count, *offset, ioflags);
410 ret = generic_file_sendfile(filp, offset, count, actor, target);
411 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
413 XFS_STATS_ADD(xs_read_bytes, ret);
414 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
419 * This routine is called to handle zeroing any space in the last
420 * block of the file that is beyond the EOF. We do this since the
421 * size is being increased without writing anything to that block
422 * and we don't want anyone to read the garbage on the disk.
424 STATIC int /* error (positive) */
430 xfs_fsize_t end_size)
432 xfs_fileoff_t last_fsb;
437 int isize_fsb_offset;
439 xfs_bmbt_irec_t imap;
443 ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
444 ASSERT(offset > isize);
448 isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
449 if (isize_fsb_offset == 0) {
451 * There are no extra bytes in the last block on disk to
457 last_fsb = XFS_B_TO_FSBT(mp, isize);
459 error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
466 * If the block underlying isize is just a hole, then there
467 * is nothing to zero.
469 if (imap.br_startblock == HOLESTARTBLOCK) {
473 * Zero the part of the last block beyond the EOF, and write it
474 * out sync. We need to drop the ilock while we do this so we
475 * don't deadlock when the buffer cache calls back to us.
477 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
478 loff = XFS_FSB_TO_B(mp, last_fsb);
479 lsize = XFS_FSB_TO_B(mp, 1);
481 zero_offset = isize_fsb_offset;
482 zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
484 error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
486 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
492 * Zero any on disk space between the current EOF and the new,
493 * larger EOF. This handles the normal case of zeroing the remainder
494 * of the last block in the file and the unusual case of zeroing blocks
495 * out beyond the size of the file. This second case only happens
496 * with fixed size extents and when the system crashes before the inode
497 * size was updated but after blocks were allocated. If fill is set,
498 * then any holes in the range are filled and zeroed. If not, the holes
499 * are left alone as holes.
502 int /* error (positive) */
506 xfs_off_t offset, /* starting I/O offset */
507 xfs_fsize_t isize, /* current inode size */
508 xfs_fsize_t end_size) /* terminal inode size */
510 struct inode *ip = LINVFS_GET_IP(vp);
511 xfs_fileoff_t start_zero_fsb;
512 xfs_fileoff_t end_zero_fsb;
513 xfs_fileoff_t prev_zero_fsb;
514 xfs_fileoff_t zero_count_fsb;
515 xfs_fileoff_t last_fsb;
516 xfs_extlen_t buf_len_fsb;
517 xfs_extlen_t prev_zero_count;
521 xfs_bmbt_irec_t imap;
525 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
526 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
531 * First handle zeroing the block on which isize resides.
532 * We only zero a part of that block so it is handled specially.
534 error = xfs_zero_last_block(ip, io, offset, isize, end_size);
536 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
537 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
542 * Calculate the range between the new size and the old
543 * where blocks needing to be zeroed may exist. To get the
544 * block where the last byte in the file currently resides,
545 * we need to subtract one from the size and truncate back
546 * to a block boundary. We subtract 1 in case the size is
547 * exactly on a block boundary.
549 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
550 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
551 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
552 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
553 if (last_fsb == end_zero_fsb) {
555 * The size was only incremented on its last block.
556 * We took care of that above, so just return.
561 ASSERT(start_zero_fsb <= end_zero_fsb);
562 prev_zero_fsb = NULLFILEOFF;
564 while (start_zero_fsb <= end_zero_fsb) {
566 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
567 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
568 0, NULL, 0, &imap, &nimaps, NULL);
570 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
571 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
576 if (imap.br_state == XFS_EXT_UNWRITTEN ||
577 imap.br_startblock == HOLESTARTBLOCK) {
579 * This loop handles initializing pages that were
580 * partially initialized by the code below this
581 * loop. It basically zeroes the part of the page
582 * that sits on a hole and sets the page as P_HOLE
583 * and calls remapf if it is a mapped file.
585 prev_zero_fsb = NULLFILEOFF;
587 start_zero_fsb = imap.br_startoff +
589 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
594 * There are blocks in the range requested.
595 * Zero them a single write at a time. We actually
596 * don't zero the entire range returned if it is
597 * too big and simply loop around to get the rest.
598 * That is not the most efficient thing to do, but it
599 * is simple and this path should not be exercised often.
601 buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
602 mp->m_writeio_blocks << 8);
604 * Drop the inode lock while we're doing the I/O.
605 * We'll still have the iolock to protect us.
607 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
609 loff = XFS_FSB_TO_B(mp, start_zero_fsb);
610 lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
612 error = xfs_iozero(ip, loff, lsize, end_size);
618 prev_zero_fsb = start_zero_fsb;
619 prev_zero_count = buf_len_fsb;
620 start_zero_fsb = imap.br_startoff + buf_len_fsb;
621 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
623 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
630 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
635 ssize_t /* bytes written, or (-) error */
639 const struct iovec *iovp,
645 struct file *file = iocb->ki_filp;
651 xfs_fsize_t isize, new_size;
652 xfs_fsize_t n, limit;
660 XFS_STATS_INC(xs_write_calls);
662 vp = BHV_TO_VNODE(bdp);
663 xip = XFS_BHVTOI(bdp);
665 /* START copy & waste from filemap.c */
666 for (seg = 0; seg < segs; seg++) {
667 const struct iovec *iv = &iovp[seg];
670 * If any segment has a negative length, or the cumulative
671 * length ever wraps negative then return -EINVAL.
674 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
675 return XFS_ERROR(-EINVAL);
677 /* END copy & waste from filemap.c */
685 if (XFS_FORCED_SHUTDOWN(mp)) {
689 if (ioflags & IO_ISDIRECT) {
690 xfs_buftarg_t *target =
691 (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
692 mp->m_rtdev_targp : mp->m_ddev_targp;
694 if ((*offset & target->pbr_smask) ||
695 (size & target->pbr_smask)) {
696 return XFS_ERROR(-EINVAL);
698 iolock = XFS_IOLOCK_SHARED;
699 locktype = VRWLOCK_WRITE_DIRECT;
701 iolock = XFS_IOLOCK_EXCL;
702 locktype = VRWLOCK_WRITE;
705 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
707 isize = xip->i_d.di_size;
708 limit = XFS_MAXIOFFSET(mp);
710 if (file->f_flags & O_APPEND)
716 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
723 new_size = *offset + size;
724 if (new_size > isize) {
725 io->io_new_size = new_size;
728 if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
729 !(ioflags & IO_INVIS) && !eventsent)) {
730 loff_t savedsize = *offset;
731 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
733 xfs_iunlock(xip, XFS_ILOCK_EXCL);
734 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
738 xfs_iunlock(xip, iolock);
741 xfs_ilock(xip, XFS_ILOCK_EXCL);
745 * The iolock was dropped and reaquired in XFS_SEND_DATA
746 * so we have to recheck the size when appending.
747 * We will only "goto start;" once, since having sent the
748 * event prevents another call to XFS_SEND_DATA, which is
749 * what allows the size to change in the first place.
751 if ((file->f_flags & O_APPEND) &&
752 savedsize != xip->i_d.di_size) {
753 *offset = isize = xip->i_d.di_size;
759 * On Linux, generic_file_write updates the times even if
760 * no data is copied in so long as the write had a size.
762 * We must update xfs' times since revalidate will overcopy xfs.
764 if (size && !(ioflags & IO_INVIS))
765 xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
768 * If the offset is beyond the size of the file, we have a couple
769 * of things to do. First, if there is already space allocated
770 * we need to either create holes or zero the disk or ...
772 * If there is a page where the previous size lands, we need
773 * to zero it out up to the new size.
776 if (!(ioflags & IO_ISDIRECT) && (*offset > isize && isize)) {
777 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, *offset,
778 isize, *offset + size);
780 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
784 xfs_iunlock(xip, XFS_ILOCK_EXCL);
787 * If we're writing the file then make sure to clear the
788 * setuid and setgid bits if the process is not being run
789 * by root. This keeps people from modifying setuid and
793 if (((xip->i_d.di_mode & S_ISUID) ||
794 ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
795 (S_ISGID | S_IXGRP))) &&
796 !capable(CAP_FSETID)) {
797 error = xfs_write_clear_setuid(xip);
799 xfs_iunlock(xip, iolock);
805 if (ioflags & IO_ISDIRECT) {
806 xfs_inval_cached_pages(vp, io, *offset, 1, 1);
807 xfs_rw_enter_trace(XFS_DIOWR_ENTER,
808 io, iovp, segs, *offset, ioflags);
810 xfs_rw_enter_trace(XFS_WRITE_ENTER,
811 io, iovp, segs, *offset, ioflags);
813 ret = generic_file_aio_write_nolock(iocb, iovp, segs, offset);
815 if ((ret == -ENOSPC) &&
816 DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
817 !(ioflags & IO_INVIS)) {
819 xfs_rwunlock(bdp, locktype);
820 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
821 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
822 0, 0, 0); /* Delay flag intentionally unused */
825 xfs_rwlock(bdp, locktype);
826 *offset = xip->i_d.di_size;
830 if (*offset > xip->i_d.di_size) {
831 xfs_ilock(xip, XFS_ILOCK_EXCL);
832 if (*offset > xip->i_d.di_size) {
833 struct inode *inode = LINVFS_GET_IP(vp);
835 xip->i_d.di_size = *offset;
836 i_size_write(inode, *offset);
837 xip->i_update_core = 1;
838 xip->i_update_size = 1;
840 xfs_iunlock(xip, XFS_ILOCK_EXCL);
844 xfs_rwunlock(bdp, locktype);
848 XFS_STATS_ADD(xs_write_bytes, ret);
850 /* Handle various SYNC-type writes */
851 if ((file->f_flags & O_SYNC) || IS_SYNC(file->f_dentry->d_inode)) {
854 * If we're treating this as O_DSYNC and we have not updated the
855 * size, force the log.
858 if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC)
859 && !(xip->i_update_size)) {
861 * If an allocation transaction occurred
862 * without extending the size, then we have to force
863 * the log up the proper point to ensure that the
864 * allocation is permanent. We can't count on
865 * the fact that buffered writes lock out direct I/O
866 * writes - the direct I/O write could have extended
867 * the size nontransactionally, then finished before
868 * we started. xfs_write_file will think that the file
869 * didn't grow but the update isn't safe unless the
870 * size change is logged.
872 * Force the log if we've committed a transaction
873 * against the inode or if someone else has and
874 * the commit record hasn't gone to disk (e.g.
875 * the inode is pinned). This guarantees that
876 * all changes affecting the inode are permanent
880 xfs_inode_log_item_t *iip;
884 if (iip && iip->ili_last_lsn) {
885 lsn = iip->ili_last_lsn;
886 xfs_log_force(mp, lsn,
887 XFS_LOG_FORCE | XFS_LOG_SYNC);
888 } else if (xfs_ipincount(xip) > 0) {
889 xfs_log_force(mp, (xfs_lsn_t)0,
890 XFS_LOG_FORCE | XFS_LOG_SYNC);
897 * O_SYNC or O_DSYNC _with_ a size update are handled
900 * If the write was synchronous then we need to make
901 * sure that the inode modification time is permanent.
902 * We'll have updated the timestamp above, so here
903 * we use a synchronous transaction to log the inode.
904 * It's not fast, but it's necessary.
906 * If this a dsync write and the size got changed
907 * non-transactionally, then we need to ensure that
908 * the size change gets logged in a synchronous
912 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
913 if ((error = xfs_trans_reserve(tp, 0,
914 XFS_SWRITE_LOG_RES(mp),
916 /* Transaction reserve failed */
917 xfs_trans_cancel(tp, 0);
919 /* Transaction reserve successful */
920 xfs_ilock(xip, XFS_ILOCK_EXCL);
921 xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
922 xfs_trans_ihold(tp, xip);
923 xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
924 xfs_trans_set_sync(tp);
925 error = xfs_trans_commit(tp, 0, (xfs_lsn_t)0);
926 xfs_iunlock(xip, XFS_ILOCK_EXCL);
929 } /* (ioflags & O_SYNC) */
931 xfs_rwunlock(bdp, locktype);
936 * All xfs metadata buffers except log state machine buffers
937 * get this attached as their b_bdstrat callback function.
938 * This is so that we can catch a buffer
939 * after prematurely unpinning it to forcibly shutdown the filesystem.
942 xfs_bdstrat_cb(struct xfs_buf *bp)
946 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
947 if (!XFS_FORCED_SHUTDOWN(mp)) {
948 pagebuf_iorequest(bp);
951 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
953 * Metadata write that didn't get logged but
954 * written delayed anyway. These aren't associated
955 * with a transaction, and can be ignored.
957 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
958 (XFS_BUF_ISREAD(bp)) == 0)
959 return (xfs_bioerror_relse(bp));
961 return (xfs_bioerror(bp));
967 xfs_bmap(bhv_desc_t *bdp,
974 xfs_inode_t *ip = XFS_BHVTOI(bdp);
975 xfs_iocore_t *io = &ip->i_iocore;
977 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
978 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
979 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
981 return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
985 * Wrapper around bdstrat so that we can stop data
986 * from going to disk in case we are shutting down the filesystem.
987 * Typically user data goes thru this path; one of the exceptions
992 struct xfs_mount *mp,
996 if (!XFS_FORCED_SHUTDOWN(mp)) {
997 /* Grio redirection would go here
998 * if (XFS_BUF_IS_GRIO(bp)) {
1001 pagebuf_iorequest(bp);
1005 xfs_buftrace("XFSBDSTRAT IOERROR", bp);
1006 return (xfs_bioerror_relse(bp));
1010 * If the underlying (data/log/rt) device is readonly, there are some
1011 * operations that cannot proceed.
1014 xfs_dev_is_read_only(
1018 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1019 xfs_readonly_buftarg(mp->m_logdev_targp) ||
1020 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1022 "XFS: %s required on read-only device.", message);
1024 "XFS: write access unavailable, cannot proceed.");