2 * Copyright (c) 2000-2003,2005 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
23 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
47 #include "xfs_inode_item.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_utils.h"
50 #include "xfs_iomap.h"
51 #include "xfs_vnodeops.h"
53 #include <linux/capability.h>
54 #include <linux/writeback.h>
57 #if defined(XFS_RW_TRACE)
67 xfs_inode_t *ip = XFS_IO_INODE(io);
69 if (ip->i_rwtrace == NULL)
71 ktrace_enter(ip->i_rwtrace,
72 (void *)(unsigned long)tag,
74 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
75 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
77 (void *)((unsigned long)segs),
78 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
79 (void *)((unsigned long)(offset & 0xffffffff)),
80 (void *)((unsigned long)ioflags),
81 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
82 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
83 (void *)((unsigned long)current_pid()),
91 xfs_inval_cached_trace(
98 xfs_inode_t *ip = XFS_IO_INODE(io);
100 if (ip->i_rwtrace == NULL)
102 ktrace_enter(ip->i_rwtrace,
103 (void *)(__psint_t)XFS_INVAL_CACHED,
105 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
106 (void *)((unsigned long)(offset & 0xffffffff)),
107 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
108 (void *)((unsigned long)(len & 0xffffffff)),
109 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
110 (void *)((unsigned long)(first & 0xffffffff)),
111 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
112 (void *)((unsigned long)(last & 0xffffffff)),
113 (void *)((unsigned long)current_pid()),
125 * xfs_iozero clears the specified range of buffer supplied,
126 * and marks all the affected blocks as valid and modified. If
127 * an affected block is not allocated, it will be allocated. If
128 * an affected block is not completely overwritten, and is not
129 * valid before the operation, it will be read from disk before
130 * being partially zeroed.
134 struct inode *ip, /* inode */
135 loff_t pos, /* offset in file */
136 size_t count) /* size of data to zero */
140 struct address_space *mapping;
143 mapping = ip->i_mapping;
145 unsigned long index, offset;
147 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
148 index = pos >> PAGE_CACHE_SHIFT;
149 bytes = PAGE_CACHE_SIZE - offset;
154 page = grab_cache_page(mapping, index);
158 status = mapping->a_ops->prepare_write(NULL, page, offset,
163 zero_user_page(page, offset, bytes, KM_USER0);
165 status = mapping->a_ops->commit_write(NULL, page, offset,
174 page_cache_release(page);
182 ssize_t /* bytes read, or (-) error */
186 const struct iovec *iovp,
191 struct file *file = iocb->ki_filp;
192 struct inode *inode = file->f_mapping->host;
193 bhv_vnode_t *vp = XFS_ITOV(ip);
194 xfs_mount_t *mp = ip->i_mount;
201 XFS_STATS_INC(xs_read_calls);
203 /* START copy & waste from filemap.c */
204 for (seg = 0; seg < segs; seg++) {
205 const struct iovec *iv = &iovp[seg];
208 * If any segment has a negative length, or the cumulative
209 * length ever wraps negative then return -EINVAL.
212 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
213 return XFS_ERROR(-EINVAL);
215 /* END copy & waste from filemap.c */
217 if (unlikely(ioflags & IO_ISDIRECT)) {
218 xfs_buftarg_t *target =
219 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
220 mp->m_rtdev_targp : mp->m_ddev_targp;
221 if ((*offset & target->bt_smask) ||
222 (size & target->bt_smask)) {
223 if (*offset == ip->i_size) {
226 return -XFS_ERROR(EINVAL);
230 n = XFS_MAXIOFFSET(mp) - *offset;
231 if ((n <= 0) || (size == 0))
237 if (XFS_FORCED_SHUTDOWN(mp))
240 if (unlikely(ioflags & IO_ISDIRECT))
241 mutex_lock(&inode->i_mutex);
242 xfs_ilock(ip, XFS_IOLOCK_SHARED);
244 if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) {
245 bhv_vrwlock_t locktype = VRWLOCK_READ;
246 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
248 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ, vp, *offset, size,
251 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
252 if (unlikely(ioflags & IO_ISDIRECT))
253 mutex_unlock(&inode->i_mutex);
258 if (unlikely(ioflags & IO_ISDIRECT)) {
260 ret = bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
261 -1, FI_REMAPF_LOCKED);
262 mutex_unlock(&inode->i_mutex);
264 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
269 xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
270 (void *)iovp, segs, *offset, ioflags);
272 iocb->ki_pos = *offset;
273 ret = generic_file_aio_read(iocb, iovp, segs, *offset);
274 if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
275 ret = wait_on_sync_kiocb(iocb);
277 XFS_STATS_ADD(xs_read_bytes, ret);
279 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
288 struct pipe_inode_info *pipe,
293 bhv_vnode_t *vp = XFS_ITOV(ip);
294 xfs_mount_t *mp = ip->i_mount;
297 XFS_STATS_INC(xs_read_calls);
298 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
301 xfs_ilock(ip, XFS_IOLOCK_SHARED);
303 if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) {
304 bhv_vrwlock_t locktype = VRWLOCK_READ;
307 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, *ppos, count,
308 FILP_DELAY_FLAG(infilp), &locktype);
310 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
314 xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, &ip->i_iocore,
315 pipe, count, *ppos, ioflags);
316 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
318 XFS_STATS_ADD(xs_read_bytes, ret);
320 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
327 struct pipe_inode_info *pipe,
328 struct file *outfilp,
334 bhv_vnode_t *vp = XFS_ITOV(ip);
335 xfs_mount_t *mp = ip->i_mount;
336 xfs_iocore_t *io = &ip->i_iocore;
338 struct inode *inode = outfilp->f_mapping->host;
339 xfs_fsize_t isize, new_size;
341 XFS_STATS_INC(xs_write_calls);
342 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
345 xfs_ilock(ip, XFS_IOLOCK_EXCL);
347 if (DM_EVENT_ENABLED(ip, DM_EVENT_WRITE) && !(ioflags & IO_INVIS)) {
348 bhv_vrwlock_t locktype = VRWLOCK_WRITE;
351 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp, *ppos, count,
352 FILP_DELAY_FLAG(outfilp), &locktype);
354 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
359 new_size = *ppos + count;
361 xfs_ilock(ip, XFS_ILOCK_EXCL);
362 if (new_size > ip->i_size)
363 io->io_new_size = new_size;
364 xfs_iunlock(ip, XFS_ILOCK_EXCL);
366 xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, &ip->i_iocore,
367 pipe, count, *ppos, ioflags);
368 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
370 XFS_STATS_ADD(xs_write_bytes, ret);
372 isize = i_size_read(inode);
373 if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
376 if (*ppos > ip->i_size) {
377 xfs_ilock(ip, XFS_ILOCK_EXCL);
378 if (*ppos > ip->i_size)
380 xfs_iunlock(ip, XFS_ILOCK_EXCL);
383 if (io->io_new_size) {
384 xfs_ilock(ip, XFS_ILOCK_EXCL);
386 if (ip->i_d.di_size > ip->i_size)
387 ip->i_d.di_size = ip->i_size;
388 xfs_iunlock(ip, XFS_ILOCK_EXCL);
390 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
395 * This routine is called to handle zeroing any space in the last
396 * block of the file that is beyond the EOF. We do this since the
397 * size is being increased without writing anything to that block
398 * and we don't want anyone to read the garbage on the disk.
400 STATIC int /* error (positive) */
407 xfs_fileoff_t last_fsb;
408 xfs_mount_t *mp = io->io_mount;
413 xfs_bmbt_irec_t imap;
415 ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
417 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
418 if (zero_offset == 0) {
420 * There are no extra bytes in the last block on disk to
426 last_fsb = XFS_B_TO_FSBT(mp, isize);
428 error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
429 &nimaps, NULL, NULL);
435 * If the block underlying isize is just a hole, then there
436 * is nothing to zero.
438 if (imap.br_startblock == HOLESTARTBLOCK) {
442 * Zero the part of the last block beyond the EOF, and write it
443 * out sync. We need to drop the ilock while we do this so we
444 * don't deadlock when the buffer cache calls back to us.
446 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
448 zero_len = mp->m_sb.sb_blocksize - zero_offset;
449 if (isize + zero_len > offset)
450 zero_len = offset - isize;
451 error = xfs_iozero(ip, isize, zero_len);
453 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
459 * Zero any on disk space between the current EOF and the new,
460 * larger EOF. This handles the normal case of zeroing the remainder
461 * of the last block in the file and the unusual case of zeroing blocks
462 * out beyond the size of the file. This second case only happens
463 * with fixed size extents and when the system crashes before the inode
464 * size was updated but after blocks were allocated. If fill is set,
465 * then any holes in the range are filled and zeroed. If not, the holes
466 * are left alone as holes.
469 int /* error (positive) */
473 xfs_off_t offset, /* starting I/O offset */
474 xfs_fsize_t isize) /* current inode size */
476 struct inode *ip = vn_to_inode(vp);
477 xfs_fileoff_t start_zero_fsb;
478 xfs_fileoff_t end_zero_fsb;
479 xfs_fileoff_t zero_count_fsb;
480 xfs_fileoff_t last_fsb;
481 xfs_fileoff_t zero_off;
482 xfs_fsize_t zero_len;
483 xfs_mount_t *mp = io->io_mount;
486 xfs_bmbt_irec_t imap;
488 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
489 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
490 ASSERT(offset > isize);
493 * First handle zeroing the block on which isize resides.
494 * We only zero a part of that block so it is handled specially.
496 error = xfs_zero_last_block(ip, io, offset, isize);
498 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
499 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
504 * Calculate the range between the new size and the old
505 * where blocks needing to be zeroed may exist. To get the
506 * block where the last byte in the file currently resides,
507 * we need to subtract one from the size and truncate back
508 * to a block boundary. We subtract 1 in case the size is
509 * exactly on a block boundary.
511 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
512 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
513 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
514 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
515 if (last_fsb == end_zero_fsb) {
517 * The size was only incremented on its last block.
518 * We took care of that above, so just return.
523 ASSERT(start_zero_fsb <= end_zero_fsb);
524 while (start_zero_fsb <= end_zero_fsb) {
526 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
527 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
528 0, NULL, 0, &imap, &nimaps, NULL, NULL);
530 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
531 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
536 if (imap.br_state == XFS_EXT_UNWRITTEN ||
537 imap.br_startblock == HOLESTARTBLOCK) {
539 * This loop handles initializing pages that were
540 * partially initialized by the code below this
541 * loop. It basically zeroes the part of the page
542 * that sits on a hole and sets the page as P_HOLE
543 * and calls remapf if it is a mapped file.
545 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
546 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
551 * There are blocks we need to zero.
552 * Drop the inode lock while we're doing the I/O.
553 * We'll still have the iolock to protect us.
555 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
557 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
558 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
560 if ((zero_off + zero_len) > offset)
561 zero_len = offset - zero_off;
563 error = xfs_iozero(ip, zero_off, zero_len);
568 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
569 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
571 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
578 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
583 ssize_t /* bytes written, or (-) error */
585 struct xfs_inode *xip,
587 const struct iovec *iovp,
592 struct file *file = iocb->ki_filp;
593 struct address_space *mapping = file->f_mapping;
594 struct inode *inode = mapping->host;
595 bhv_vnode_t *vp = XFS_ITOV(xip);
596 unsigned long segs = nsegs;
598 ssize_t ret = 0, error = 0;
599 xfs_fsize_t isize, new_size;
603 bhv_vrwlock_t locktype;
604 size_t ocount = 0, count;
608 XFS_STATS_INC(xs_write_calls);
610 error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ);
623 vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_WRITE);
625 if (XFS_FORCED_SHUTDOWN(mp))
629 if (ioflags & IO_ISDIRECT) {
630 iolock = XFS_IOLOCK_SHARED;
631 locktype = VRWLOCK_WRITE_DIRECT;
634 iolock = XFS_IOLOCK_EXCL;
635 locktype = VRWLOCK_WRITE;
637 mutex_lock(&inode->i_mutex);
640 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
643 error = -generic_write_checks(file, &pos, &count,
644 S_ISBLK(inode->i_mode));
646 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
647 goto out_unlock_mutex;
650 if ((DM_EVENT_ENABLED(xip, DM_EVENT_WRITE) &&
651 !(ioflags & IO_INVIS) && !eventsent)) {
652 int dmflags = FILP_DELAY_FLAG(file);
655 dmflags |= DM_FLAGS_IMUX;
657 xfs_iunlock(xip, XFS_ILOCK_EXCL);
658 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
662 goto out_unlock_internal;
664 xfs_ilock(xip, XFS_ILOCK_EXCL);
668 * The iolock was dropped and reacquired in XFS_SEND_DATA
669 * so we have to recheck the size when appending.
670 * We will only "goto start;" once, since having sent the
671 * event prevents another call to XFS_SEND_DATA, which is
672 * what allows the size to change in the first place.
674 if ((file->f_flags & O_APPEND) && pos != xip->i_size)
678 if (ioflags & IO_ISDIRECT) {
679 xfs_buftarg_t *target =
680 (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
681 mp->m_rtdev_targp : mp->m_ddev_targp;
683 if ((pos & target->bt_smask) || (count & target->bt_smask)) {
684 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
685 return XFS_ERROR(-EINVAL);
688 if (!need_i_mutex && (VN_CACHED(vp) || pos > xip->i_size)) {
689 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
690 iolock = XFS_IOLOCK_EXCL;
691 locktype = VRWLOCK_WRITE;
693 mutex_lock(&inode->i_mutex);
694 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
699 new_size = pos + count;
700 if (new_size > xip->i_size)
701 io->io_new_size = new_size;
703 if (likely(!(ioflags & IO_INVIS))) {
704 file_update_time(file);
705 xfs_ichgtime_fast(xip, inode,
706 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
710 * If the offset is beyond the size of the file, we have a couple
711 * of things to do. First, if there is already space allocated
712 * we need to either create holes or zero the disk or ...
714 * If there is a page where the previous size lands, we need
715 * to zero it out up to the new size.
718 if (pos > xip->i_size) {
719 error = xfs_zero_eof(vp, io, pos, xip->i_size);
721 xfs_iunlock(xip, XFS_ILOCK_EXCL);
722 goto out_unlock_internal;
725 xfs_iunlock(xip, XFS_ILOCK_EXCL);
728 * If we're writing the file then make sure to clear the
729 * setuid and setgid bits if the process is not being run
730 * by root. This keeps people from modifying setuid and
734 if (((xip->i_d.di_mode & S_ISUID) ||
735 ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
736 (S_ISGID | S_IXGRP))) &&
737 !capable(CAP_FSETID)) {
738 error = xfs_write_clear_setuid(xip);
740 error = -remove_suid(file->f_path.dentry);
741 if (unlikely(error)) {
742 goto out_unlock_internal;
747 /* We can write back this queue in page reclaim */
748 current->backing_dev_info = mapping->backing_dev_info;
750 if ((ioflags & IO_ISDIRECT)) {
752 WARN_ON(need_i_mutex == 0);
753 xfs_inval_cached_trace(io, pos, -1,
754 ctooff(offtoct(pos)), -1);
755 error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
756 -1, FI_REMAPF_LOCKED);
758 goto out_unlock_internal;
762 /* demote the lock now the cached pages are gone */
763 XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
764 mutex_unlock(&inode->i_mutex);
766 iolock = XFS_IOLOCK_SHARED;
767 locktype = VRWLOCK_WRITE_DIRECT;
771 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
773 ret = generic_file_direct_write(iocb, iovp,
774 &segs, pos, offset, count, ocount);
777 * direct-io write to a hole: fall through to buffered I/O
778 * for completing the rest of the request.
780 if (ret >= 0 && ret != count) {
781 XFS_STATS_ADD(xs_write_bytes, ret);
786 ioflags &= ~IO_ISDIRECT;
787 xfs_iunlock(xip, iolock);
791 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
793 ret = generic_file_buffered_write(iocb, iovp, segs,
794 pos, offset, count, ret);
797 current->backing_dev_info = NULL;
799 if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
800 ret = wait_on_sync_kiocb(iocb);
802 if (ret == -ENOSPC &&
803 DM_EVENT_ENABLED(xip, DM_EVENT_NOSPACE) && !(ioflags & IO_INVIS)) {
804 xfs_rwunlock(xip, locktype);
806 mutex_unlock(&inode->i_mutex);
807 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
808 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
809 0, 0, 0); /* Delay flag intentionally unused */
811 mutex_lock(&inode->i_mutex);
812 xfs_rwlock(xip, locktype);
814 goto out_unlock_internal;
820 isize = i_size_read(inode);
821 if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
824 if (*offset > xip->i_size) {
825 xfs_ilock(xip, XFS_ILOCK_EXCL);
826 if (*offset > xip->i_size)
827 xip->i_size = *offset;
828 xfs_iunlock(xip, XFS_ILOCK_EXCL);
833 goto out_unlock_internal;
835 XFS_STATS_ADD(xs_write_bytes, ret);
837 /* Handle various SYNC-type writes */
838 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
839 error = xfs_write_sync_logforce(mp, xip);
841 goto out_unlock_internal;
843 xfs_rwunlock(xip, locktype);
845 mutex_unlock(&inode->i_mutex);
847 error = sync_page_range(inode, mapping, pos, ret);
851 mutex_lock(&inode->i_mutex);
852 xfs_rwlock(xip, locktype);
856 if (io->io_new_size) {
857 xfs_ilock(xip, XFS_ILOCK_EXCL);
860 * If this was a direct or synchronous I/O that failed (such
861 * as ENOSPC) then part of the I/O may have been written to
862 * disk before the error occured. In this case the on-disk
863 * file size may have been adjusted beyond the in-memory file
864 * size and now needs to be truncated back.
866 if (xip->i_d.di_size > xip->i_size)
867 xip->i_d.di_size = xip->i_size;
868 xfs_iunlock(xip, XFS_ILOCK_EXCL);
870 xfs_rwunlock(xip, locktype);
873 mutex_unlock(&inode->i_mutex);
878 * All xfs metadata buffers except log state machine buffers
879 * get this attached as their b_bdstrat callback function.
880 * This is so that we can catch a buffer
881 * after prematurely unpinning it to forcibly shutdown the filesystem.
884 xfs_bdstrat_cb(struct xfs_buf *bp)
888 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
889 if (!XFS_FORCED_SHUTDOWN(mp)) {
890 xfs_buf_iorequest(bp);
893 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
895 * Metadata write that didn't get logged but
896 * written delayed anyway. These aren't associated
897 * with a transaction, and can be ignored.
899 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
900 (XFS_BUF_ISREAD(bp)) == 0)
901 return (xfs_bioerror_relse(bp));
903 return (xfs_bioerror(bp));
917 xfs_iocore_t *io = &ip->i_iocore;
919 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
920 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
921 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
923 return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
927 * Wrapper around bdstrat so that we can stop data
928 * from going to disk in case we are shutting down the filesystem.
929 * Typically user data goes thru this path; one of the exceptions
934 struct xfs_mount *mp,
938 if (!XFS_FORCED_SHUTDOWN(mp)) {
939 /* Grio redirection would go here
940 * if (XFS_BUF_IS_GRIO(bp)) {
943 xfs_buf_iorequest(bp);
947 xfs_buftrace("XFSBDSTRAT IOERROR", bp);
948 return (xfs_bioerror_relse(bp));
952 * If the underlying (data/log/rt) device is readonly, there are some
953 * operations that cannot proceed.
956 xfs_dev_is_read_only(
960 if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
961 xfs_readonly_buftarg(mp->m_logdev_targp) ||
962 (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
964 "XFS: %s required on read-only device.", message);
966 "XFS: write access unavailable, cannot proceed.");