2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause)
4 * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
7 * This software was developed for the FreeBSD Project by Marshall
8 * Kirk McKusick and Network Associates Laboratories, the Security
9 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
10 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * Copyright (c) 1982, 1986, 1989, 1993
35 * The Regents of the University of California. All rights reserved.
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61 * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95
62 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
63 * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
69 #include <sys/param.h>
71 #include <sys/systm.h>
74 #include <sys/extattr.h>
75 #include <sys/kernel.h>
76 #include <sys/limits.h>
77 #include <sys/malloc.h>
78 #include <sys/mount.h>
80 #include <sys/rwlock.h>
82 #include <sys/sysctl.h>
83 #include <sys/vmmeter.h>
84 #include <sys/vnode.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vnode_pager.h>
94 #include <ufs/ufs/extattr.h>
95 #include <ufs/ufs/quota.h>
96 #include <ufs/ufs/inode.h>
97 #include <ufs/ufs/ufs_extern.h>
98 #include <ufs/ufs/ufsmount.h>
100 #include <ufs/ffs/fs.h>
101 #include <ufs/ffs/ffs_extern.h>
102 #include "opt_directio.h"
105 #define ALIGNED_TO(ptr, s) \
106 (((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
109 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
111 static vop_fdatasync_t ffs_fdatasync;
112 static vop_fsync_t ffs_fsync;
113 static vop_getpages_t ffs_getpages;
114 static vop_lock1_t ffs_lock;
115 static vop_read_t ffs_read;
116 static vop_write_t ffs_write;
117 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
118 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
120 static vop_strategy_t ffsext_strategy;
121 static vop_closeextattr_t ffs_closeextattr;
122 static vop_deleteextattr_t ffs_deleteextattr;
123 static vop_getextattr_t ffs_getextattr;
124 static vop_listextattr_t ffs_listextattr;
125 static vop_openextattr_t ffs_openextattr;
126 static vop_setextattr_t ffs_setextattr;
127 static vop_vptofh_t ffs_vptofh;
129 /* Global vfs data structures for ufs. */
130 struct vop_vector ffs_vnodeops1 = {
131 .vop_default = &ufs_vnodeops,
132 .vop_fsync = ffs_fsync,
133 .vop_fdatasync = ffs_fdatasync,
134 .vop_getpages = ffs_getpages,
135 .vop_getpages_async = vnode_pager_local_getpages_async,
136 .vop_lock1 = ffs_lock,
137 .vop_read = ffs_read,
138 .vop_reallocblks = ffs_reallocblks,
139 .vop_write = ffs_write,
140 .vop_vptofh = ffs_vptofh,
143 struct vop_vector ffs_fifoops1 = {
144 .vop_default = &ufs_fifoops,
145 .vop_fsync = ffs_fsync,
146 .vop_fdatasync = ffs_fdatasync,
147 .vop_lock1 = ffs_lock,
148 .vop_vptofh = ffs_vptofh,
151 /* Global vfs data structures for ufs. */
152 struct vop_vector ffs_vnodeops2 = {
153 .vop_default = &ufs_vnodeops,
154 .vop_fsync = ffs_fsync,
155 .vop_fdatasync = ffs_fdatasync,
156 .vop_getpages = ffs_getpages,
157 .vop_getpages_async = vnode_pager_local_getpages_async,
158 .vop_lock1 = ffs_lock,
159 .vop_read = ffs_read,
160 .vop_reallocblks = ffs_reallocblks,
161 .vop_write = ffs_write,
162 .vop_closeextattr = ffs_closeextattr,
163 .vop_deleteextattr = ffs_deleteextattr,
164 .vop_getextattr = ffs_getextattr,
165 .vop_listextattr = ffs_listextattr,
166 .vop_openextattr = ffs_openextattr,
167 .vop_setextattr = ffs_setextattr,
168 .vop_vptofh = ffs_vptofh,
171 struct vop_vector ffs_fifoops2 = {
172 .vop_default = &ufs_fifoops,
173 .vop_fsync = ffs_fsync,
174 .vop_fdatasync = ffs_fdatasync,
175 .vop_lock1 = ffs_lock,
176 .vop_reallocblks = ffs_reallocblks,
177 .vop_strategy = ffsext_strategy,
178 .vop_closeextattr = ffs_closeextattr,
179 .vop_deleteextattr = ffs_deleteextattr,
180 .vop_getextattr = ffs_getextattr,
181 .vop_listextattr = ffs_listextattr,
182 .vop_openextattr = ffs_openextattr,
183 .vop_setextattr = ffs_setextattr,
184 .vop_vptofh = ffs_vptofh,
188 * Synch an open file.
192 ffs_fsync(struct vop_fsync_args *ap)
201 error = ffs_syncvnode(vp, ap->a_waitfor, 0);
204 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
205 error = softdep_fsync(vp);
210 * The softdep_fsync() function may drop vp lock,
211 * allowing for dirty buffers to reappear on the
212 * bo_dirty list. Recheck and resync as needed.
215 if ((vp->v_type == VREG || vp->v_type == VDIR) &&
216 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
226 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
230 struct buf *bp, *nbp;
233 bool still_dirty, wait;
236 ip->i_flag &= ~IN_NEEDSYNC;
240 * When doing MNT_WAIT we must first flush all dependencies
243 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
244 (error = softdep_sync_metadata(vp)) != 0)
248 * Flush all dirty buffers associated with a vnode.
252 wait = false; /* Always do an async pass first. */
253 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
256 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
257 bp->b_vflags &= ~BV_SCANNED;
258 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
260 * Reasons to skip this buffer: it has already been considered
261 * on this pass, the buffer has dependencies that will cause
262 * it to be redirtied and it has not already been deferred,
263 * or it is already being written.
265 if ((bp->b_vflags & BV_SCANNED) != 0)
267 bp->b_vflags |= BV_SCANNED;
269 * Flush indirects in order, if requested.
271 * Note that if only datasync is requested, we can
272 * skip indirect blocks when softupdates are not
273 * active. Otherwise we must flush them with data,
274 * since dependencies prevent data block writes.
276 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
277 (lbn_level(bp->b_lblkno) >= passes ||
278 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
280 if (bp->b_lblkno > lbn)
281 panic("ffs_syncvnode: syncing truncated data.");
282 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
286 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
287 BO_LOCKPTR(bo)) != 0) {
288 bp->b_vflags &= ~BV_SCANNED;
293 if ((bp->b_flags & B_DELWRI) == 0)
294 panic("ffs_fsync: not dirty");
296 * Check for dependencies and potentially complete them.
298 if (!LIST_EMPTY(&bp->b_dep) &&
299 (error = softdep_sync_buf(vp, bp,
300 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
302 if (error != EBUSY) {
306 /* If we deferred once, don't defer again. */
307 if ((bp->b_flags & B_DEFERRED) == 0) {
308 bp->b_flags |= B_DEFERRED;
315 if ((error = bwrite(bp)) != 0)
317 } else if ((bp->b_flags & B_CLUSTEROK)) {
318 (void) vfs_bio_awrite(bp);
325 * Since we may have slept during the I/O, we need
326 * to start from a known point.
329 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
331 if (waitfor != MNT_WAIT) {
333 if ((flags & NO_INO_UPDT) != 0)
336 return (ffs_update(vp, 0));
338 /* Drain IO to see if we're done. */
339 bufobj_wwait(bo, 0, 0);
341 * Block devices associated with filesystems may have new I/O
342 * requests posted for them even if the vnode is locked, so no
343 * amount of trying will get them clean. We make several passes
346 * Regular files may need multiple passes to flush all dependency
347 * work as it is possible that we must write once per indirect
348 * level, once for the leaf, and once for the inode and each of
349 * these will be done with one sync and one async pass.
351 if (bo->bo_dirty.bv_cnt > 0) {
352 if ((flags & DATA_ONLY) == 0) {
356 * For data-only sync, dirty indirect buffers
360 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
361 if (bp->b_lblkno > -UFS_NDADDR) {
369 /* Write the inode after sync passes to flush deps. */
370 if (wait && DOINGSOFTDEP(vp) &&
371 (flags & NO_INO_UPDT) == 0) {
376 /* switch between sync/async. */
378 if (wait || ++passes < UFS_NIADDR + 2)
384 if ((flags & DATA_ONLY) == 0) {
385 if ((flags & NO_INO_UPDT) == 0)
386 error = ffs_update(vp, 1);
388 softdep_journal_fsync(VTOI(vp));
394 ffs_fdatasync(struct vop_fdatasync_args *ap)
397 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
402 struct vop_lock1_args /* {
410 #ifndef NO_FFS_SNAPSHOT
416 switch (ap->a_flags & LK_TYPE_MASK) {
423 #ifdef DEBUG_VFS_LOCKS
424 KASSERT(vp->v_holdcnt != 0,
425 ("ffs_lock %p: zero hold count", vp));
428 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
429 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
430 ap->a_file, ap->a_line);
431 if (lkp == vp->v_vnlock || result != 0)
434 * Apparent success, except that the vnode
435 * mutated between snapshot file vnode and
436 * regular file vnode while this process
437 * slept. The lock currently held is not the
438 * right lock. Release it, and try to get the
441 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
442 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
443 ap->a_file, ap->a_line);
444 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
445 (LK_INTERLOCK | LK_NOWAIT))
447 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
448 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
449 flags &= ~LK_INTERLOCK;
453 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
457 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
462 ffs_read_hole(struct uio *uio, long xfersize, long *size)
464 ssize_t saved_resid, tlen;
467 while (xfersize > 0) {
468 tlen = min(xfersize, ZERO_REGION_SIZE);
469 saved_resid = uio->uio_resid;
470 error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
474 tlen = saved_resid - uio->uio_resid;
482 * Vnode op for reading.
486 struct vop_read_args /* {
490 struct ucred *a_cred;
498 ufs_lbn_t lbn, nextlbn;
500 long size, xfersize, blkoffset;
502 int bflag, error, ioflag, seqcount;
506 ioflag = ap->a_ioflag;
507 if (ap->a_ioflag & IO_EXT)
509 return (ffs_extread(vp, uio, ioflag));
511 panic("ffs_read+IO_EXT");
514 if ((ioflag & IO_DIRECT) != 0) {
517 error = ffs_rawread(vp, uio, &workdone);
518 if (error != 0 || workdone != 0)
523 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
527 if (uio->uio_rw != UIO_READ)
528 panic("ffs_read: mode");
530 if (vp->v_type == VLNK) {
531 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
532 panic("ffs_read: short symlink");
533 } else if (vp->v_type != VREG && vp->v_type != VDIR)
534 panic("ffs_read: type %d", vp->v_type);
536 orig_resid = uio->uio_resid;
537 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
540 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
542 if (uio->uio_offset < ip->i_size &&
543 uio->uio_offset >= fs->fs_maxfilesize)
546 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
547 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
548 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
550 lbn = lblkno(fs, uio->uio_offset);
554 * size of buffer. The buffer representing the
555 * end of the file is rounded up to the size of
556 * the block type ( fragment or full block,
559 size = blksize(fs, ip, lbn);
560 blkoffset = blkoff(fs, uio->uio_offset);
563 * The amount we want to transfer in this iteration is
564 * one FS block less the amount of the data before
565 * our startpoint (duh!)
567 xfersize = fs->fs_bsize - blkoffset;
570 * But if we actually want less than the block,
571 * or the file doesn't have a whole block more of data,
572 * then use the lesser number.
574 if (uio->uio_resid < xfersize)
575 xfersize = uio->uio_resid;
576 if (bytesinfile < xfersize)
577 xfersize = bytesinfile;
579 if (lblktosize(fs, nextlbn) >= ip->i_size) {
581 * Don't do readahead if this is the end of the file.
583 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
584 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
586 * Otherwise if we are allowed to cluster,
587 * grab as much as we can.
589 * XXX This may not be a win if we are not
590 * doing sequential access.
592 error = cluster_read(vp, ip->i_size, lbn,
593 size, NOCRED, blkoffset + uio->uio_resid,
594 seqcount, bflag, &bp);
595 } else if (seqcount > 1) {
597 * If we are NOT allowed to cluster, then
598 * if we appear to be acting sequentially,
599 * fire off a request for a readahead
600 * as well as a read. Note that the 4th and 5th
601 * arguments point to arrays of the size specified in
604 u_int nextsize = blksize(fs, ip, nextlbn);
605 error = breadn_flags(vp, lbn, size, &nextlbn,
606 &nextsize, 1, NOCRED, bflag, NULL, &bp);
609 * Failing all of the above, just read what the
610 * user asked for. Interestingly, the same as
611 * the first option above.
613 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
615 if (error == EJUSTRETURN) {
616 error = ffs_read_hole(uio, xfersize, &size);
627 * We should only get non-zero b_resid when an I/O error
628 * has occurred, which should cause us to break above.
629 * However, if the short read did not cause an error,
630 * then we want to ensure that we do not uiomove bad
631 * or uninitialized data.
634 if (size < xfersize) {
640 if (buf_mapped(bp)) {
641 error = vn_io_fault_uiomove((char *)bp->b_data +
642 blkoffset, (int)xfersize, uio);
644 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
650 vfs_bio_brelse(bp, ioflag);
654 * This can only happen in the case of an error
655 * because the loop above resets bp to NULL on each iteration
656 * and on normal completion has not set a new value into it.
657 * so it must have come from a 'break' statement
660 vfs_bio_brelse(bp, ioflag);
662 if ((error == 0 || uio->uio_resid != orig_resid) &&
663 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 &&
664 (ip->i_flag & IN_ACCESS) == 0) {
666 ip->i_flag |= IN_ACCESS;
673 * Vnode op for writing.
677 struct vop_write_args /* {
681 struct ucred *a_cred;
693 int blkoffset, error, flags, ioflag, size, xfersize;
697 ioflag = ap->a_ioflag;
698 if (ap->a_ioflag & IO_EXT)
700 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
702 panic("ffs_write+IO_EXT");
705 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
709 if (uio->uio_rw != UIO_WRITE)
710 panic("ffs_write: mode");
713 switch (vp->v_type) {
715 if (ioflag & IO_APPEND)
716 uio->uio_offset = ip->i_size;
717 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
723 panic("ffs_write: dir write");
726 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
727 (int)uio->uio_offset,
732 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
733 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
735 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
738 * Maybe this should be above the vnode op call, but so long as
739 * file servers have no limits, I don't think it matters.
741 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
744 resid = uio->uio_resid;
746 if (seqcount > BA_SEQMAX)
747 flags = BA_SEQMAX << BA_SEQSHIFT;
749 flags = seqcount << BA_SEQSHIFT;
750 if (ioflag & IO_SYNC)
752 flags |= BA_UNMAPPED;
754 for (error = 0; uio->uio_resid > 0;) {
755 lbn = lblkno(fs, uio->uio_offset);
756 blkoffset = blkoff(fs, uio->uio_offset);
757 xfersize = fs->fs_bsize - blkoffset;
758 if (uio->uio_resid < xfersize)
759 xfersize = uio->uio_resid;
760 if (uio->uio_offset + xfersize > ip->i_size)
761 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
764 * We must perform a read-before-write if the transfer size
765 * does not cover the entire buffer.
767 if (fs->fs_bsize > xfersize)
771 /* XXX is uio->uio_offset the right thing here? */
772 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
773 ap->a_cred, flags, &bp);
775 vnode_pager_setsize(vp, ip->i_size);
778 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
779 bp->b_flags |= B_NOCACHE;
781 if (uio->uio_offset + xfersize > ip->i_size) {
782 ip->i_size = uio->uio_offset + xfersize;
783 DIP_SET(ip, i_size, ip->i_size);
786 size = blksize(fs, ip, lbn) - bp->b_resid;
790 if (buf_mapped(bp)) {
791 error = vn_io_fault_uiomove((char *)bp->b_data +
792 blkoffset, (int)xfersize, uio);
794 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
798 * If the buffer is not already filled and we encounter an
799 * error while trying to fill it, we have to clear out any
800 * garbage data from the pages instantiated for the buffer.
801 * If we do not, a failed uiomove() during a write can leave
802 * the prior contents of the pages exposed to a userland mmap.
804 * Note that we need only clear buffers with a transfer size
805 * equal to the block size because buffers with a shorter
806 * transfer size were cleared above by the call to UFS_BALLOC()
807 * with the BA_CLRBUF flag set.
809 * If the source region for uiomove identically mmaps the
810 * buffer, uiomove() performed the NOP copy, and the buffer
811 * content remains valid because the page fault handler
812 * validated the pages.
814 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
815 fs->fs_bsize == xfersize)
818 vfs_bio_set_flags(bp, ioflag);
821 * If IO_SYNC each buffer is written synchronously. Otherwise
822 * if we have a severe page deficiency write the buffer
823 * asynchronously. Otherwise try to cluster, and if that
824 * doesn't do it then either do an async write (if O_DIRECT),
825 * or a delayed write (if not).
827 if (ioflag & IO_SYNC) {
829 } else if (vm_page_count_severe() ||
830 buf_dirty_count_severe() ||
831 (ioflag & IO_ASYNC)) {
832 bp->b_flags |= B_CLUSTEROK;
834 } else if (xfersize + blkoffset == fs->fs_bsize) {
835 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
836 bp->b_flags |= B_CLUSTEROK;
837 cluster_write(vp, bp, ip->i_size, seqcount,
842 } else if (ioflag & IO_DIRECT) {
843 bp->b_flags |= B_CLUSTEROK;
846 bp->b_flags |= B_CLUSTEROK;
849 if (error || xfersize == 0)
851 ip->i_flag |= IN_CHANGE | IN_UPDATE;
854 * If we successfully wrote any data, and we are not the superuser
855 * we clear the setuid and setgid bits as a precaution against
858 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
860 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
861 ip->i_mode &= ~(ISUID | ISGID);
862 DIP_SET(ip, i_mode, ip->i_mode);
866 if (ioflag & IO_UNIT) {
867 (void)ffs_truncate(vp, osize,
868 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
869 uio->uio_offset -= resid - uio->uio_resid;
870 uio->uio_resid = resid;
872 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
873 error = ffs_update(vp, 1);
878 * Extended attribute area reading.
881 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
884 struct ufs2_dinode *dp;
887 ufs_lbn_t lbn, nextlbn;
889 long size, xfersize, blkoffset;
898 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
899 panic("ffs_extread: mode");
902 orig_resid = uio->uio_resid;
903 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
906 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
908 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
909 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
911 lbn = lblkno(fs, uio->uio_offset);
915 * size of buffer. The buffer representing the
916 * end of the file is rounded up to the size of
917 * the block type ( fragment or full block,
920 size = sblksize(fs, dp->di_extsize, lbn);
921 blkoffset = blkoff(fs, uio->uio_offset);
924 * The amount we want to transfer in this iteration is
925 * one FS block less the amount of the data before
926 * our startpoint (duh!)
928 xfersize = fs->fs_bsize - blkoffset;
931 * But if we actually want less than the block,
932 * or the file doesn't have a whole block more of data,
933 * then use the lesser number.
935 if (uio->uio_resid < xfersize)
936 xfersize = uio->uio_resid;
937 if (bytesinfile < xfersize)
938 xfersize = bytesinfile;
940 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
942 * Don't do readahead if this is the end of the info.
944 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
947 * If we have a second block, then
948 * fire off a request for a readahead
949 * as well as a read. Note that the 4th and 5th
950 * arguments point to arrays of the size specified in
953 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
955 nextlbn = -1 - nextlbn;
956 error = breadn(vp, -1 - lbn,
957 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
966 * We should only get non-zero b_resid when an I/O error
967 * has occurred, which should cause us to break above.
968 * However, if the short read did not cause an error,
969 * then we want to ensure that we do not uiomove bad
970 * or uninitialized data.
973 if (size < xfersize) {
979 error = uiomove((char *)bp->b_data + blkoffset,
983 vfs_bio_brelse(bp, ioflag);
987 * This can only happen in the case of an error
988 * because the loop above resets bp to NULL on each iteration
989 * and on normal completion has not set a new value into it.
990 * so it must have come from a 'break' statement
993 vfs_bio_brelse(bp, ioflag);
998 * Extended attribute area writing.
1001 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1004 struct ufs2_dinode *dp;
1010 int blkoffset, error, flags, size, xfersize;
1017 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1018 panic("ffs_extwrite: mode");
1021 if (ioflag & IO_APPEND)
1022 uio->uio_offset = dp->di_extsize;
1023 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1024 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1025 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1026 UFS_NXADDR * fs->fs_bsize)
1029 resid = uio->uio_resid;
1030 osize = dp->di_extsize;
1032 if (ioflag & IO_SYNC)
1035 for (error = 0; uio->uio_resid > 0;) {
1036 lbn = lblkno(fs, uio->uio_offset);
1037 blkoffset = blkoff(fs, uio->uio_offset);
1038 xfersize = fs->fs_bsize - blkoffset;
1039 if (uio->uio_resid < xfersize)
1040 xfersize = uio->uio_resid;
1043 * We must perform a read-before-write if the transfer size
1044 * does not cover the entire buffer.
1046 if (fs->fs_bsize > xfersize)
1049 flags &= ~BA_CLRBUF;
1050 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1055 * If the buffer is not valid we have to clear out any
1056 * garbage data from the pages instantiated for the buffer.
1057 * If we do not, a failed uiomove() during a write can leave
1058 * the prior contents of the pages exposed to a userland
1059 * mmap(). XXX deal with uiomove() errors a better way.
1061 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1064 if (uio->uio_offset + xfersize > dp->di_extsize)
1065 dp->di_extsize = uio->uio_offset + xfersize;
1067 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1068 if (size < xfersize)
1072 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1074 vfs_bio_set_flags(bp, ioflag);
1077 * If IO_SYNC each buffer is written synchronously. Otherwise
1078 * if we have a severe page deficiency write the buffer
1079 * asynchronously. Otherwise try to cluster, and if that
1080 * doesn't do it then either do an async write (if O_DIRECT),
1081 * or a delayed write (if not).
1083 if (ioflag & IO_SYNC) {
1085 } else if (vm_page_count_severe() ||
1086 buf_dirty_count_severe() ||
1087 xfersize + blkoffset == fs->fs_bsize ||
1088 (ioflag & (IO_ASYNC | IO_DIRECT)))
1092 if (error || xfersize == 0)
1094 ip->i_flag |= IN_CHANGE;
1097 * If we successfully wrote any data, and we are not the superuser
1098 * we clear the setuid and setgid bits as a precaution against
1101 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1102 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1103 ip->i_mode &= ~(ISUID | ISGID);
1104 dp->di_mode = ip->i_mode;
1108 if (ioflag & IO_UNIT) {
1109 (void)ffs_truncate(vp, osize,
1110 IO_EXT | (ioflag&IO_SYNC), ucred);
1111 uio->uio_offset -= resid - uio->uio_resid;
1112 uio->uio_resid = resid;
1114 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1115 error = ffs_update(vp, 1);
1121 * Vnode operating to retrieve a named extended attribute.
1123 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1124 * the length of the EA, and possibly the pointer to the entry and to the data.
1127 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1128 struct extattr **eapp, u_char **eac)
1130 struct extattr *eap, *eaend;
1133 nlen = strlen(name);
1134 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1135 eap = (struct extattr *)ptr;
1136 eaend = (struct extattr *)(ptr + length);
1137 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1138 /* make sure this entry is complete */
1139 if (EXTATTR_NEXT(eap) > eaend)
1141 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1142 || memcmp(eap->ea_name, name, nlen) != 0)
1147 *eac = EXTATTR_CONTENT(eap);
1148 return (EXTATTR_CONTENT_SIZE(eap));
1154 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1157 struct ufs2_dinode *dp;
1160 struct iovec liovec;
1168 easize = dp->di_extsize;
1169 if ((uoff_t)easize + extra > UFS_NXADDR * fs->fs_bsize)
1172 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1174 liovec.iov_base = eae;
1175 liovec.iov_len = easize;
1176 luio.uio_iov = &liovec;
1177 luio.uio_iovcnt = 1;
1178 luio.uio_offset = 0;
1179 luio.uio_resid = easize;
1180 luio.uio_segflg = UIO_SYSSPACE;
1181 luio.uio_rw = UIO_READ;
1184 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1194 ffs_lock_ea(struct vnode *vp)
1200 while (ip->i_flag & IN_EA_LOCKED) {
1201 ip->i_flag |= IN_EA_LOCKWAIT;
1202 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1205 ip->i_flag |= IN_EA_LOCKED;
1210 ffs_unlock_ea(struct vnode *vp)
1216 if (ip->i_flag & IN_EA_LOCKWAIT)
1217 wakeup(&ip->i_ea_refs);
1218 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1223 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1226 struct ufs2_dinode *dp;
1232 if (ip->i_ea_area != NULL) {
1238 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1243 ip->i_ea_len = dp->di_extsize;
1251 * Vnode extattr transaction commit/abort
1254 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1258 struct iovec liovec;
1260 struct ufs2_dinode *dp;
1265 if (ip->i_ea_area == NULL) {
1270 error = ip->i_ea_error;
1271 if (commit && error == 0) {
1272 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1274 cred = vp->v_mount->mnt_cred;
1275 liovec.iov_base = ip->i_ea_area;
1276 liovec.iov_len = ip->i_ea_len;
1277 luio.uio_iov = &liovec;
1278 luio.uio_iovcnt = 1;
1279 luio.uio_offset = 0;
1280 luio.uio_resid = ip->i_ea_len;
1281 luio.uio_segflg = UIO_SYSSPACE;
1282 luio.uio_rw = UIO_WRITE;
1284 /* XXX: I'm not happy about truncating to zero size */
1285 if (ip->i_ea_len < dp->di_extsize)
1286 error = ffs_truncate(vp, 0, IO_EXT, cred);
1287 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1289 if (--ip->i_ea_refs == 0) {
1290 free(ip->i_ea_area, M_TEMP);
1291 ip->i_ea_area = NULL;
1300 * Vnode extattr strategy routine for fifos.
1302 * We need to check for a read or write of the external attributes.
1303 * Otherwise we just fall through and do the usual thing.
1306 ffsext_strategy(struct vop_strategy_args *ap)
1308 struct vop_strategy_args {
1309 struct vnodeop_desc *a_desc;
1319 lbn = ap->a_bp->b_lblkno;
1320 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1321 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1322 if (vp->v_type == VFIFO)
1323 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1324 panic("spec nodes went here");
1328 * Vnode extattr transaction commit/abort
1331 ffs_openextattr(struct vop_openextattr_args *ap)
1333 struct vop_openextattr_args {
1334 struct vnodeop_desc *a_desc;
1336 IN struct ucred *a_cred;
1337 IN struct thread *a_td;
1342 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1343 return (EOPNOTSUPP);
1345 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1350 * Vnode extattr transaction commit/abort
1353 ffs_closeextattr(struct vop_closeextattr_args *ap)
1355 struct vop_closeextattr_args {
1356 struct vnodeop_desc *a_desc;
1359 IN struct ucred *a_cred;
1360 IN struct thread *a_td;
1365 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1366 return (EOPNOTSUPP);
1368 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1371 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1375 * Vnode operation to remove a named attribute.
1378 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1381 IN struct vnode *a_vp;
1382 IN int a_attrnamespace;
1383 IN const char *a_name;
1384 IN struct ucred *a_cred;
1385 IN struct thread *a_td;
1390 struct extattr *eap;
1392 int olen, error, i, easize;
1396 ip = VTOI(ap->a_vp);
1398 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1399 return (EOPNOTSUPP);
1401 if (strlen(ap->a_name) == 0)
1404 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1407 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1408 ap->a_cred, ap->a_td, VWRITE);
1412 * ffs_lock_ea is not needed there, because the vnode
1413 * must be exclusively locked.
1415 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1416 ip->i_ea_error = error;
1420 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1424 /* CEM: delete could be done in-place instead */
1425 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1426 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1427 easize = ip->i_ea_len;
1429 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1432 /* delete but nonexistent */
1434 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1437 ul = eap->ea_length;
1438 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1439 bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1442 tmp = ip->i_ea_area;
1443 ip->i_ea_area = eae;
1444 ip->i_ea_len = easize;
1446 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1451 * Vnode operation to retrieve a named extended attribute.
1454 ffs_getextattr(struct vop_getextattr_args *ap)
1457 IN struct vnode *a_vp;
1458 IN int a_attrnamespace;
1459 IN const char *a_name;
1460 INOUT struct uio *a_uio;
1462 IN struct ucred *a_cred;
1463 IN struct thread *a_td;
1472 ip = VTOI(ap->a_vp);
1474 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1475 return (EOPNOTSUPP);
1477 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1478 ap->a_cred, ap->a_td, VREAD);
1482 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1486 eae = ip->i_ea_area;
1487 easize = ip->i_ea_len;
1489 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1493 if (ap->a_size != NULL)
1494 *ap->a_size = ealen;
1495 else if (ap->a_uio != NULL)
1496 error = uiomove(p, ealen, ap->a_uio);
1500 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1505 * Vnode operation to retrieve extended attributes on a vnode.
1508 ffs_listextattr(struct vop_listextattr_args *ap)
1511 IN struct vnode *a_vp;
1512 IN int a_attrnamespace;
1513 INOUT struct uio *a_uio;
1515 IN struct ucred *a_cred;
1516 IN struct thread *a_td;
1521 struct extattr *eap, *eaend;
1524 ip = VTOI(ap->a_vp);
1526 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1527 return (EOPNOTSUPP);
1529 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1530 ap->a_cred, ap->a_td, VREAD);
1534 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1539 if (ap->a_size != NULL)
1542 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1543 eap = (struct extattr *)ip->i_ea_area;
1544 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1545 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1546 /* make sure this entry is complete */
1547 if (EXTATTR_NEXT(eap) > eaend)
1549 if (eap->ea_namespace != ap->a_attrnamespace)
1552 ealen = eap->ea_namelength;
1553 if (ap->a_size != NULL)
1554 *ap->a_size += ealen + 1;
1555 else if (ap->a_uio != NULL)
1556 error = uiomove(&eap->ea_namelength, ealen + 1,
1560 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1565 * Vnode operation to set a named attribute.
1568 ffs_setextattr(struct vop_setextattr_args *ap)
1571 IN struct vnode *a_vp;
1572 IN int a_attrnamespace;
1573 IN const char *a_name;
1574 INOUT struct uio *a_uio;
1575 IN struct ucred *a_cred;
1576 IN struct thread *a_td;
1582 struct extattr *eap;
1583 uint32_t ealength, ul;
1585 int olen, eapad1, eapad2, error, i, easize;
1589 ip = VTOI(ap->a_vp);
1592 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1593 return (EOPNOTSUPP);
1595 if (strlen(ap->a_name) == 0)
1598 /* XXX Now unsupported API to delete EAs using NULL uio. */
1599 if (ap->a_uio == NULL)
1600 return (EOPNOTSUPP);
1602 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1605 ealen = ap->a_uio->uio_resid;
1606 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1609 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1610 ap->a_cred, ap->a_td, VWRITE);
1614 * ffs_lock_ea is not needed there, because the vnode
1615 * must be exclusively locked.
1617 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1618 ip->i_ea_error = error;
1622 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1626 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1627 eapad1 = roundup2(ealength, 8) - ealength;
1628 eapad2 = roundup2(ealen, 8) - ealen;
1629 ealength += eapad1 + ealen + eapad2;
1632 * CEM: rewrites of the same size or smaller could be done in-place
1633 * instead. (We don't acquire any fine-grained locks in here either,
1634 * so we could also do bigger writes in-place.)
1636 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1637 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1638 easize = ip->i_ea_len;
1640 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1643 /* new, append at end */
1644 KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1646 eap = (struct extattr *)(eae + easize);
1649 ul = eap->ea_length;
1650 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1651 if (ul != ealength) {
1652 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1654 easize += (ealength - ul);
1657 if (easize > lblktosize(fs, UFS_NXADDR)) {
1659 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1660 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1661 ip->i_ea_error = ENOSPC;
1664 eap->ea_length = ealength;
1665 eap->ea_namespace = ap->a_attrnamespace;
1666 eap->ea_contentpadlen = eapad2;
1667 eap->ea_namelength = strlen(ap->a_name);
1668 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1669 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1670 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1673 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1674 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1675 ip->i_ea_error = error;
1678 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1680 tmp = ip->i_ea_area;
1681 ip->i_ea_area = eae;
1682 ip->i_ea_len = easize;
1684 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1689 * Vnode pointer to File handle
1692 ffs_vptofh(struct vop_vptofh_args *ap)
1695 IN struct vnode *a_vp;
1696 IN struct fid *a_fhp;
1703 ip = VTOI(ap->a_vp);
1704 ufhp = (struct ufid *)ap->a_fhp;
1705 ufhp->ufid_len = sizeof(struct ufid);
1706 ufhp->ufid_ino = ip->i_number;
1707 ufhp->ufid_gen = ip->i_gen;
1711 SYSCTL_DECL(_vfs_ffs);
1712 static int use_buf_pager = 1;
1713 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1714 "Always use buffer pager instead of bmap");
1717 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1720 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1724 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1727 return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn));
1731 ffs_getpages(struct vop_getpages_args *ap)
1734 struct ufsmount *um;
1737 um = VFSTOUFS(vp->v_mount);
1739 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1740 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1741 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1742 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1743 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));