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_getpages_async_t ffs_getpages_async;
115 static vop_lock1_t ffs_lock;
116 static vop_read_t ffs_read;
117 static vop_write_t ffs_write;
118 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
119 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
121 static vop_strategy_t ffsext_strategy;
122 static vop_closeextattr_t ffs_closeextattr;
123 static vop_deleteextattr_t ffs_deleteextattr;
124 static vop_getextattr_t ffs_getextattr;
125 static vop_listextattr_t ffs_listextattr;
126 static vop_openextattr_t ffs_openextattr;
127 static vop_setextattr_t ffs_setextattr;
128 static vop_vptofh_t ffs_vptofh;
130 /* Global vfs data structures for ufs. */
131 struct vop_vector ffs_vnodeops1 = {
132 .vop_default = &ufs_vnodeops,
133 .vop_fsync = ffs_fsync,
134 .vop_fdatasync = ffs_fdatasync,
135 .vop_getpages = ffs_getpages,
136 .vop_getpages_async = ffs_getpages_async,
137 .vop_lock1 = ffs_lock,
138 .vop_read = ffs_read,
139 .vop_reallocblks = ffs_reallocblks,
140 .vop_write = ffs_write,
141 .vop_vptofh = ffs_vptofh,
143 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1);
145 struct vop_vector ffs_fifoops1 = {
146 .vop_default = &ufs_fifoops,
147 .vop_fsync = ffs_fsync,
148 .vop_fdatasync = ffs_fdatasync,
149 .vop_lock1 = ffs_lock,
150 .vop_vptofh = ffs_vptofh,
152 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1);
154 /* Global vfs data structures for ufs. */
155 struct vop_vector ffs_vnodeops2 = {
156 .vop_default = &ufs_vnodeops,
157 .vop_fsync = ffs_fsync,
158 .vop_fdatasync = ffs_fdatasync,
159 .vop_getpages = ffs_getpages,
160 .vop_getpages_async = ffs_getpages_async,
161 .vop_lock1 = ffs_lock,
162 .vop_read = ffs_read,
163 .vop_reallocblks = ffs_reallocblks,
164 .vop_write = ffs_write,
165 .vop_closeextattr = ffs_closeextattr,
166 .vop_deleteextattr = ffs_deleteextattr,
167 .vop_getextattr = ffs_getextattr,
168 .vop_listextattr = ffs_listextattr,
169 .vop_openextattr = ffs_openextattr,
170 .vop_setextattr = ffs_setextattr,
171 .vop_vptofh = ffs_vptofh,
173 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2);
175 struct vop_vector ffs_fifoops2 = {
176 .vop_default = &ufs_fifoops,
177 .vop_fsync = ffs_fsync,
178 .vop_fdatasync = ffs_fdatasync,
179 .vop_lock1 = ffs_lock,
180 .vop_reallocblks = ffs_reallocblks,
181 .vop_strategy = ffsext_strategy,
182 .vop_closeextattr = ffs_closeextattr,
183 .vop_deleteextattr = ffs_deleteextattr,
184 .vop_getextattr = ffs_getextattr,
185 .vop_listextattr = ffs_listextattr,
186 .vop_openextattr = ffs_openextattr,
187 .vop_setextattr = ffs_setextattr,
188 .vop_vptofh = ffs_vptofh,
190 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2);
193 * Synch an open file.
197 ffs_fsync(struct vop_fsync_args *ap)
206 error = ffs_syncvnode(vp, ap->a_waitfor, 0);
209 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
210 error = softdep_fsync(vp);
215 * The softdep_fsync() function may drop vp lock,
216 * allowing for dirty buffers to reappear on the
217 * bo_dirty list. Recheck and resync as needed.
220 if ((vp->v_type == VREG || vp->v_type == VDIR) &&
221 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
231 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
235 struct buf *bp, *nbp;
238 bool still_dirty, wait;
241 ip->i_flag &= ~IN_NEEDSYNC;
245 * When doing MNT_WAIT we must first flush all dependencies
248 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
249 (error = softdep_sync_metadata(vp)) != 0)
253 * Flush all dirty buffers associated with a vnode.
257 wait = false; /* Always do an async pass first. */
258 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
261 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
262 bp->b_vflags &= ~BV_SCANNED;
263 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
265 * Reasons to skip this buffer: it has already been considered
266 * on this pass, the buffer has dependencies that will cause
267 * it to be redirtied and it has not already been deferred,
268 * or it is already being written.
270 if ((bp->b_vflags & BV_SCANNED) != 0)
272 bp->b_vflags |= BV_SCANNED;
274 * Flush indirects in order, if requested.
276 * Note that if only datasync is requested, we can
277 * skip indirect blocks when softupdates are not
278 * active. Otherwise we must flush them with data,
279 * since dependencies prevent data block writes.
281 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
282 (lbn_level(bp->b_lblkno) >= passes ||
283 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
285 if (bp->b_lblkno > lbn)
286 panic("ffs_syncvnode: syncing truncated data.");
287 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
291 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
292 BO_LOCKPTR(bo)) != 0) {
293 bp->b_vflags &= ~BV_SCANNED;
298 if ((bp->b_flags & B_DELWRI) == 0)
299 panic("ffs_fsync: not dirty");
301 * Check for dependencies and potentially complete them.
303 if (!LIST_EMPTY(&bp->b_dep) &&
304 (error = softdep_sync_buf(vp, bp,
305 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
307 if (error != EBUSY) {
311 /* If we deferred once, don't defer again. */
312 if ((bp->b_flags & B_DEFERRED) == 0) {
313 bp->b_flags |= B_DEFERRED;
320 if ((error = bwrite(bp)) != 0)
322 } else if ((bp->b_flags & B_CLUSTEROK)) {
323 (void) vfs_bio_awrite(bp);
330 * Since we may have slept during the I/O, we need
331 * to start from a known point.
334 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
336 if (waitfor != MNT_WAIT) {
338 if ((flags & NO_INO_UPDT) != 0)
341 return (ffs_update(vp, 0));
343 /* Drain IO to see if we're done. */
344 bufobj_wwait(bo, 0, 0);
346 * Block devices associated with filesystems may have new I/O
347 * requests posted for them even if the vnode is locked, so no
348 * amount of trying will get them clean. We make several passes
351 * Regular files may need multiple passes to flush all dependency
352 * work as it is possible that we must write once per indirect
353 * level, once for the leaf, and once for the inode and each of
354 * these will be done with one sync and one async pass.
356 if (bo->bo_dirty.bv_cnt > 0) {
357 if ((flags & DATA_ONLY) == 0) {
361 * For data-only sync, dirty indirect buffers
365 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
366 if (bp->b_lblkno > -UFS_NDADDR) {
374 /* Write the inode after sync passes to flush deps. */
375 if (wait && DOINGSOFTDEP(vp) &&
376 (flags & NO_INO_UPDT) == 0) {
381 /* switch between sync/async. */
383 if (wait || ++passes < UFS_NIADDR + 2)
389 if ((flags & DATA_ONLY) == 0) {
390 if ((flags & NO_INO_UPDT) == 0)
391 error = ffs_update(vp, 1);
393 softdep_journal_fsync(VTOI(vp));
399 ffs_fdatasync(struct vop_fdatasync_args *ap)
402 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
407 struct vop_lock1_args /* {
415 #ifndef NO_FFS_SNAPSHOT
421 switch (ap->a_flags & LK_TYPE_MASK) {
428 #ifdef DEBUG_VFS_LOCKS
429 KASSERT(vp->v_holdcnt != 0,
430 ("ffs_lock %p: zero hold count", vp));
433 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
434 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
435 ap->a_file, ap->a_line);
436 if (lkp == vp->v_vnlock || result != 0)
439 * Apparent success, except that the vnode
440 * mutated between snapshot file vnode and
441 * regular file vnode while this process
442 * slept. The lock currently held is not the
443 * right lock. Release it, and try to get the
446 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
447 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
448 ap->a_file, ap->a_line);
449 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
450 (LK_INTERLOCK | LK_NOWAIT))
452 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
453 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
454 flags &= ~LK_INTERLOCK;
458 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
462 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
467 ffs_read_hole(struct uio *uio, long xfersize, long *size)
469 ssize_t saved_resid, tlen;
472 while (xfersize > 0) {
473 tlen = min(xfersize, ZERO_REGION_SIZE);
474 saved_resid = uio->uio_resid;
475 error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
479 tlen = saved_resid - uio->uio_resid;
487 * Vnode op for reading.
491 struct vop_read_args /* {
495 struct ucred *a_cred;
503 ufs_lbn_t lbn, nextlbn;
505 long size, xfersize, blkoffset;
507 int bflag, error, ioflag, seqcount;
511 ioflag = ap->a_ioflag;
512 if (ap->a_ioflag & IO_EXT)
514 return (ffs_extread(vp, uio, ioflag));
516 panic("ffs_read+IO_EXT");
519 if ((ioflag & IO_DIRECT) != 0) {
522 error = ffs_rawread(vp, uio, &workdone);
523 if (error != 0 || workdone != 0)
528 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
532 if (uio->uio_rw != UIO_READ)
533 panic("ffs_read: mode");
535 if (vp->v_type == VLNK) {
536 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
537 panic("ffs_read: short symlink");
538 } else if (vp->v_type != VREG && vp->v_type != VDIR)
539 panic("ffs_read: type %d", vp->v_type);
541 orig_resid = uio->uio_resid;
542 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
545 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
547 if (uio->uio_offset < ip->i_size &&
548 uio->uio_offset >= fs->fs_maxfilesize)
551 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
552 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
553 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
555 lbn = lblkno(fs, uio->uio_offset);
559 * size of buffer. The buffer representing the
560 * end of the file is rounded up to the size of
561 * the block type ( fragment or full block,
564 size = blksize(fs, ip, lbn);
565 blkoffset = blkoff(fs, uio->uio_offset);
568 * The amount we want to transfer in this iteration is
569 * one FS block less the amount of the data before
570 * our startpoint (duh!)
572 xfersize = fs->fs_bsize - blkoffset;
575 * But if we actually want less than the block,
576 * or the file doesn't have a whole block more of data,
577 * then use the lesser number.
579 if (uio->uio_resid < xfersize)
580 xfersize = uio->uio_resid;
581 if (bytesinfile < xfersize)
582 xfersize = bytesinfile;
584 if (lblktosize(fs, nextlbn) >= ip->i_size) {
586 * Don't do readahead if this is the end of the file.
588 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
589 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
591 * Otherwise if we are allowed to cluster,
592 * grab as much as we can.
594 * XXX This may not be a win if we are not
595 * doing sequential access.
597 error = cluster_read(vp, ip->i_size, lbn,
598 size, NOCRED, blkoffset + uio->uio_resid,
599 seqcount, bflag, &bp);
600 } else if (seqcount > 1) {
602 * If we are NOT allowed to cluster, then
603 * if we appear to be acting sequentially,
604 * fire off a request for a readahead
605 * as well as a read. Note that the 4th and 5th
606 * arguments point to arrays of the size specified in
609 u_int nextsize = blksize(fs, ip, nextlbn);
610 error = breadn_flags(vp, lbn, lbn, size, &nextlbn,
611 &nextsize, 1, NOCRED, bflag, NULL, &bp);
614 * Failing all of the above, just read what the
615 * user asked for. Interestingly, the same as
616 * the first option above.
618 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
620 if (error == EJUSTRETURN) {
621 error = ffs_read_hole(uio, xfersize, &size);
632 * We should only get non-zero b_resid when an I/O error
633 * has occurred, which should cause us to break above.
634 * However, if the short read did not cause an error,
635 * then we want to ensure that we do not uiomove bad
636 * or uninitialized data.
639 if (size < xfersize) {
645 if (buf_mapped(bp)) {
646 error = vn_io_fault_uiomove((char *)bp->b_data +
647 blkoffset, (int)xfersize, uio);
649 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
655 vfs_bio_brelse(bp, ioflag);
659 * This can only happen in the case of an error
660 * because the loop above resets bp to NULL on each iteration
661 * and on normal completion has not set a new value into it.
662 * so it must have come from a 'break' statement
665 vfs_bio_brelse(bp, ioflag);
667 if ((error == 0 || uio->uio_resid != orig_resid) &&
668 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 &&
669 (ip->i_flag & IN_ACCESS) == 0) {
671 ip->i_flag |= IN_ACCESS;
678 * Vnode op for writing.
682 struct vop_write_args /* {
686 struct ucred *a_cred;
698 int blkoffset, error, flags, ioflag, size, xfersize;
702 ioflag = ap->a_ioflag;
703 if (ap->a_ioflag & IO_EXT)
705 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
707 panic("ffs_write+IO_EXT");
710 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
714 if (uio->uio_rw != UIO_WRITE)
715 panic("ffs_write: mode");
718 switch (vp->v_type) {
720 if (ioflag & IO_APPEND)
721 uio->uio_offset = ip->i_size;
722 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
728 panic("ffs_write: dir write");
731 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
732 (int)uio->uio_offset,
737 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
738 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
740 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
743 * Maybe this should be above the vnode op call, but so long as
744 * file servers have no limits, I don't think it matters.
746 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
749 resid = uio->uio_resid;
751 if (seqcount > BA_SEQMAX)
752 flags = BA_SEQMAX << BA_SEQSHIFT;
754 flags = seqcount << BA_SEQSHIFT;
755 if (ioflag & IO_SYNC)
757 flags |= BA_UNMAPPED;
759 for (error = 0; uio->uio_resid > 0;) {
760 lbn = lblkno(fs, uio->uio_offset);
761 blkoffset = blkoff(fs, uio->uio_offset);
762 xfersize = fs->fs_bsize - blkoffset;
763 if (uio->uio_resid < xfersize)
764 xfersize = uio->uio_resid;
765 if (uio->uio_offset + xfersize > ip->i_size)
766 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
769 * We must perform a read-before-write if the transfer size
770 * does not cover the entire buffer.
772 if (fs->fs_bsize > xfersize)
776 /* XXX is uio->uio_offset the right thing here? */
777 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
778 ap->a_cred, flags, &bp);
780 vnode_pager_setsize(vp, ip->i_size);
783 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
784 bp->b_flags |= B_NOCACHE;
786 if (uio->uio_offset + xfersize > ip->i_size) {
787 ip->i_size = uio->uio_offset + xfersize;
788 DIP_SET(ip, i_size, ip->i_size);
791 size = blksize(fs, ip, lbn) - bp->b_resid;
795 if (buf_mapped(bp)) {
796 error = vn_io_fault_uiomove((char *)bp->b_data +
797 blkoffset, (int)xfersize, uio);
799 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
803 * If the buffer is not already filled and we encounter an
804 * error while trying to fill it, we have to clear out any
805 * garbage data from the pages instantiated for the buffer.
806 * If we do not, a failed uiomove() during a write can leave
807 * the prior contents of the pages exposed to a userland mmap.
809 * Note that we need only clear buffers with a transfer size
810 * equal to the block size because buffers with a shorter
811 * transfer size were cleared above by the call to UFS_BALLOC()
812 * with the BA_CLRBUF flag set.
814 * If the source region for uiomove identically mmaps the
815 * buffer, uiomove() performed the NOP copy, and the buffer
816 * content remains valid because the page fault handler
817 * validated the pages.
819 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
820 fs->fs_bsize == xfersize)
823 vfs_bio_set_flags(bp, ioflag);
826 * If IO_SYNC each buffer is written synchronously. Otherwise
827 * if we have a severe page deficiency write the buffer
828 * asynchronously. Otherwise try to cluster, and if that
829 * doesn't do it then either do an async write (if O_DIRECT),
830 * or a delayed write (if not).
832 if (ioflag & IO_SYNC) {
834 } else if (vm_page_count_severe() ||
835 buf_dirty_count_severe() ||
836 (ioflag & IO_ASYNC)) {
837 bp->b_flags |= B_CLUSTEROK;
839 } else if (xfersize + blkoffset == fs->fs_bsize) {
840 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
841 bp->b_flags |= B_CLUSTEROK;
842 cluster_write(vp, bp, ip->i_size, seqcount,
847 } else if (ioflag & IO_DIRECT) {
848 bp->b_flags |= B_CLUSTEROK;
851 bp->b_flags |= B_CLUSTEROK;
854 if (error || xfersize == 0)
856 ip->i_flag |= IN_CHANGE | IN_UPDATE;
859 * If we successfully wrote any data, and we are not the superuser
860 * we clear the setuid and setgid bits as a precaution against
863 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
865 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
866 ip->i_mode &= ~(ISUID | ISGID);
867 DIP_SET(ip, i_mode, ip->i_mode);
871 if (ioflag & IO_UNIT) {
872 (void)ffs_truncate(vp, osize,
873 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
874 uio->uio_offset -= resid - uio->uio_resid;
875 uio->uio_resid = resid;
877 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
878 error = ffs_update(vp, 1);
883 * Extended attribute area reading.
886 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
889 struct ufs2_dinode *dp;
892 ufs_lbn_t lbn, nextlbn;
894 long size, xfersize, blkoffset;
903 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
904 panic("ffs_extread: mode");
907 orig_resid = uio->uio_resid;
908 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
911 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
913 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
914 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
916 lbn = lblkno(fs, uio->uio_offset);
920 * size of buffer. The buffer representing the
921 * end of the file is rounded up to the size of
922 * the block type ( fragment or full block,
925 size = sblksize(fs, dp->di_extsize, lbn);
926 blkoffset = blkoff(fs, uio->uio_offset);
929 * The amount we want to transfer in this iteration is
930 * one FS block less the amount of the data before
931 * our startpoint (duh!)
933 xfersize = fs->fs_bsize - blkoffset;
936 * But if we actually want less than the block,
937 * or the file doesn't have a whole block more of data,
938 * then use the lesser number.
940 if (uio->uio_resid < xfersize)
941 xfersize = uio->uio_resid;
942 if (bytesinfile < xfersize)
943 xfersize = bytesinfile;
945 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
947 * Don't do readahead if this is the end of the info.
949 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
952 * If we have a second block, then
953 * fire off a request for a readahead
954 * as well as a read. Note that the 4th and 5th
955 * arguments point to arrays of the size specified in
958 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
960 nextlbn = -1 - nextlbn;
961 error = breadn(vp, -1 - lbn,
962 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
971 * We should only get non-zero b_resid when an I/O error
972 * has occurred, which should cause us to break above.
973 * However, if the short read did not cause an error,
974 * then we want to ensure that we do not uiomove bad
975 * or uninitialized data.
978 if (size < xfersize) {
984 error = uiomove((char *)bp->b_data + blkoffset,
988 vfs_bio_brelse(bp, ioflag);
992 * This can only happen in the case of an error
993 * because the loop above resets bp to NULL on each iteration
994 * and on normal completion has not set a new value into it.
995 * so it must have come from a 'break' statement
998 vfs_bio_brelse(bp, ioflag);
1003 * Extended attribute area writing.
1006 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1009 struct ufs2_dinode *dp;
1015 int blkoffset, error, flags, size, xfersize;
1022 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1023 panic("ffs_extwrite: mode");
1026 if (ioflag & IO_APPEND)
1027 uio->uio_offset = dp->di_extsize;
1028 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1029 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1030 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1031 UFS_NXADDR * fs->fs_bsize)
1034 resid = uio->uio_resid;
1035 osize = dp->di_extsize;
1037 if (ioflag & IO_SYNC)
1040 for (error = 0; uio->uio_resid > 0;) {
1041 lbn = lblkno(fs, uio->uio_offset);
1042 blkoffset = blkoff(fs, uio->uio_offset);
1043 xfersize = fs->fs_bsize - blkoffset;
1044 if (uio->uio_resid < xfersize)
1045 xfersize = uio->uio_resid;
1048 * We must perform a read-before-write if the transfer size
1049 * does not cover the entire buffer.
1051 if (fs->fs_bsize > xfersize)
1054 flags &= ~BA_CLRBUF;
1055 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1060 * If the buffer is not valid we have to clear out any
1061 * garbage data from the pages instantiated for the buffer.
1062 * If we do not, a failed uiomove() during a write can leave
1063 * the prior contents of the pages exposed to a userland
1064 * mmap(). XXX deal with uiomove() errors a better way.
1066 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1069 if (uio->uio_offset + xfersize > dp->di_extsize)
1070 dp->di_extsize = uio->uio_offset + xfersize;
1072 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1073 if (size < xfersize)
1077 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1079 vfs_bio_set_flags(bp, ioflag);
1082 * If IO_SYNC each buffer is written synchronously. Otherwise
1083 * if we have a severe page deficiency write the buffer
1084 * asynchronously. Otherwise try to cluster, and if that
1085 * doesn't do it then either do an async write (if O_DIRECT),
1086 * or a delayed write (if not).
1088 if (ioflag & IO_SYNC) {
1090 } else if (vm_page_count_severe() ||
1091 buf_dirty_count_severe() ||
1092 xfersize + blkoffset == fs->fs_bsize ||
1093 (ioflag & (IO_ASYNC | IO_DIRECT)))
1097 if (error || xfersize == 0)
1099 ip->i_flag |= IN_CHANGE;
1102 * If we successfully wrote any data, and we are not the superuser
1103 * we clear the setuid and setgid bits as a precaution against
1106 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1107 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1108 ip->i_mode &= ~(ISUID | ISGID);
1109 dp->di_mode = ip->i_mode;
1113 if (ioflag & IO_UNIT) {
1114 (void)ffs_truncate(vp, osize,
1115 IO_EXT | (ioflag&IO_SYNC), ucred);
1116 uio->uio_offset -= resid - uio->uio_resid;
1117 uio->uio_resid = resid;
1119 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1120 error = ffs_update(vp, 1);
1126 * Vnode operating to retrieve a named extended attribute.
1128 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1129 * the length of the EA, and possibly the pointer to the entry and to the data.
1132 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1133 struct extattr **eapp, u_char **eac)
1135 struct extattr *eap, *eaend;
1138 nlen = strlen(name);
1139 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1140 eap = (struct extattr *)ptr;
1141 eaend = (struct extattr *)(ptr + length);
1142 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1143 /* make sure this entry is complete */
1144 if (EXTATTR_NEXT(eap) > eaend)
1146 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1147 || memcmp(eap->ea_name, name, nlen) != 0)
1152 *eac = EXTATTR_CONTENT(eap);
1153 return (EXTATTR_CONTENT_SIZE(eap));
1159 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1162 struct ufs2_dinode *dp;
1165 struct iovec liovec;
1173 easize = dp->di_extsize;
1174 if ((uoff_t)easize + extra > UFS_NXADDR * fs->fs_bsize)
1177 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1179 liovec.iov_base = eae;
1180 liovec.iov_len = easize;
1181 luio.uio_iov = &liovec;
1182 luio.uio_iovcnt = 1;
1183 luio.uio_offset = 0;
1184 luio.uio_resid = easize;
1185 luio.uio_segflg = UIO_SYSSPACE;
1186 luio.uio_rw = UIO_READ;
1189 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1199 ffs_lock_ea(struct vnode *vp)
1205 while (ip->i_flag & IN_EA_LOCKED) {
1206 ip->i_flag |= IN_EA_LOCKWAIT;
1207 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1210 ip->i_flag |= IN_EA_LOCKED;
1215 ffs_unlock_ea(struct vnode *vp)
1221 if (ip->i_flag & IN_EA_LOCKWAIT)
1222 wakeup(&ip->i_ea_refs);
1223 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1228 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1231 struct ufs2_dinode *dp;
1237 if (ip->i_ea_area != NULL) {
1243 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1248 ip->i_ea_len = dp->di_extsize;
1256 * Vnode extattr transaction commit/abort
1259 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1263 struct iovec liovec;
1265 struct ufs2_dinode *dp;
1270 if (ip->i_ea_area == NULL) {
1275 error = ip->i_ea_error;
1276 if (commit && error == 0) {
1277 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1279 cred = vp->v_mount->mnt_cred;
1280 liovec.iov_base = ip->i_ea_area;
1281 liovec.iov_len = ip->i_ea_len;
1282 luio.uio_iov = &liovec;
1283 luio.uio_iovcnt = 1;
1284 luio.uio_offset = 0;
1285 luio.uio_resid = ip->i_ea_len;
1286 luio.uio_segflg = UIO_SYSSPACE;
1287 luio.uio_rw = UIO_WRITE;
1289 /* XXX: I'm not happy about truncating to zero size */
1290 if (ip->i_ea_len < dp->di_extsize)
1291 error = ffs_truncate(vp, 0, IO_EXT, cred);
1292 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1294 if (--ip->i_ea_refs == 0) {
1295 free(ip->i_ea_area, M_TEMP);
1296 ip->i_ea_area = NULL;
1305 * Vnode extattr strategy routine for fifos.
1307 * We need to check for a read or write of the external attributes.
1308 * Otherwise we just fall through and do the usual thing.
1311 ffsext_strategy(struct vop_strategy_args *ap)
1313 struct vop_strategy_args {
1314 struct vnodeop_desc *a_desc;
1324 lbn = ap->a_bp->b_lblkno;
1325 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1326 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1327 if (vp->v_type == VFIFO)
1328 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1329 panic("spec nodes went here");
1333 * Vnode extattr transaction commit/abort
1336 ffs_openextattr(struct vop_openextattr_args *ap)
1338 struct vop_openextattr_args {
1339 struct vnodeop_desc *a_desc;
1341 IN struct ucred *a_cred;
1342 IN struct thread *a_td;
1347 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1348 return (EOPNOTSUPP);
1350 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1355 * Vnode extattr transaction commit/abort
1358 ffs_closeextattr(struct vop_closeextattr_args *ap)
1360 struct vop_closeextattr_args {
1361 struct vnodeop_desc *a_desc;
1364 IN struct ucred *a_cred;
1365 IN struct thread *a_td;
1370 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1371 return (EOPNOTSUPP);
1373 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1376 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1380 * Vnode operation to remove a named attribute.
1383 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1386 IN struct vnode *a_vp;
1387 IN int a_attrnamespace;
1388 IN const char *a_name;
1389 IN struct ucred *a_cred;
1390 IN struct thread *a_td;
1395 struct extattr *eap;
1397 int olen, error, i, easize;
1401 ip = VTOI(ap->a_vp);
1403 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1404 return (EOPNOTSUPP);
1406 if (strlen(ap->a_name) == 0)
1409 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1412 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1413 ap->a_cred, ap->a_td, VWRITE);
1417 * ffs_lock_ea is not needed there, because the vnode
1418 * must be exclusively locked.
1420 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1421 ip->i_ea_error = error;
1425 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1429 /* CEM: delete could be done in-place instead */
1430 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1431 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1432 easize = ip->i_ea_len;
1434 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1437 /* delete but nonexistent */
1439 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1442 ul = eap->ea_length;
1443 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1444 bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1447 tmp = ip->i_ea_area;
1448 ip->i_ea_area = eae;
1449 ip->i_ea_len = easize;
1451 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1456 * Vnode operation to retrieve a named extended attribute.
1459 ffs_getextattr(struct vop_getextattr_args *ap)
1462 IN struct vnode *a_vp;
1463 IN int a_attrnamespace;
1464 IN const char *a_name;
1465 INOUT struct uio *a_uio;
1467 IN struct ucred *a_cred;
1468 IN struct thread *a_td;
1477 ip = VTOI(ap->a_vp);
1479 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1480 return (EOPNOTSUPP);
1482 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1483 ap->a_cred, ap->a_td, VREAD);
1487 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1491 eae = ip->i_ea_area;
1492 easize = ip->i_ea_len;
1494 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1498 if (ap->a_size != NULL)
1499 *ap->a_size = ealen;
1500 else if (ap->a_uio != NULL)
1501 error = uiomove(p, ealen, ap->a_uio);
1505 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1510 * Vnode operation to retrieve extended attributes on a vnode.
1513 ffs_listextattr(struct vop_listextattr_args *ap)
1516 IN struct vnode *a_vp;
1517 IN int a_attrnamespace;
1518 INOUT struct uio *a_uio;
1520 IN struct ucred *a_cred;
1521 IN struct thread *a_td;
1526 struct extattr *eap, *eaend;
1529 ip = VTOI(ap->a_vp);
1531 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1532 return (EOPNOTSUPP);
1534 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1535 ap->a_cred, ap->a_td, VREAD);
1539 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1544 if (ap->a_size != NULL)
1547 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1548 eap = (struct extattr *)ip->i_ea_area;
1549 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1550 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1551 /* make sure this entry is complete */
1552 if (EXTATTR_NEXT(eap) > eaend)
1554 if (eap->ea_namespace != ap->a_attrnamespace)
1557 ealen = eap->ea_namelength;
1558 if (ap->a_size != NULL)
1559 *ap->a_size += ealen + 1;
1560 else if (ap->a_uio != NULL)
1561 error = uiomove(&eap->ea_namelength, ealen + 1,
1565 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1570 * Vnode operation to set a named attribute.
1573 ffs_setextattr(struct vop_setextattr_args *ap)
1576 IN struct vnode *a_vp;
1577 IN int a_attrnamespace;
1578 IN const char *a_name;
1579 INOUT struct uio *a_uio;
1580 IN struct ucred *a_cred;
1581 IN struct thread *a_td;
1587 struct extattr *eap;
1588 uint32_t ealength, ul;
1590 int olen, eapad1, eapad2, error, i, easize;
1594 ip = VTOI(ap->a_vp);
1597 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1598 return (EOPNOTSUPP);
1600 if (strlen(ap->a_name) == 0)
1603 /* XXX Now unsupported API to delete EAs using NULL uio. */
1604 if (ap->a_uio == NULL)
1605 return (EOPNOTSUPP);
1607 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1610 ealen = ap->a_uio->uio_resid;
1611 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1614 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1615 ap->a_cred, ap->a_td, VWRITE);
1619 * ffs_lock_ea is not needed there, because the vnode
1620 * must be exclusively locked.
1622 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1623 ip->i_ea_error = error;
1627 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1631 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1632 eapad1 = roundup2(ealength, 8) - ealength;
1633 eapad2 = roundup2(ealen, 8) - ealen;
1634 ealength += eapad1 + ealen + eapad2;
1637 * CEM: rewrites of the same size or smaller could be done in-place
1638 * instead. (We don't acquire any fine-grained locks in here either,
1639 * so we could also do bigger writes in-place.)
1641 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1642 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1643 easize = ip->i_ea_len;
1645 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1648 /* new, append at end */
1649 KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1651 eap = (struct extattr *)(eae + easize);
1654 ul = eap->ea_length;
1655 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1656 if (ul != ealength) {
1657 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1659 easize += (ealength - ul);
1662 if (easize > lblktosize(fs, UFS_NXADDR)) {
1664 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1665 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1666 ip->i_ea_error = ENOSPC;
1669 eap->ea_length = ealength;
1670 eap->ea_namespace = ap->a_attrnamespace;
1671 eap->ea_contentpadlen = eapad2;
1672 eap->ea_namelength = strlen(ap->a_name);
1673 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1674 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1675 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1678 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1679 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1680 ip->i_ea_error = error;
1683 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1685 tmp = ip->i_ea_area;
1686 ip->i_ea_area = eae;
1687 ip->i_ea_len = easize;
1689 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1694 * Vnode pointer to File handle
1697 ffs_vptofh(struct vop_vptofh_args *ap)
1700 IN struct vnode *a_vp;
1701 IN struct fid *a_fhp;
1708 ip = VTOI(ap->a_vp);
1709 ufhp = (struct ufid *)ap->a_fhp;
1710 ufhp->ufid_len = sizeof(struct ufid);
1711 ufhp->ufid_ino = ip->i_number;
1712 ufhp->ufid_gen = ip->i_gen;
1716 SYSCTL_DECL(_vfs_ffs);
1717 static int use_buf_pager = 1;
1718 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1719 "Always use buffer pager instead of bmap");
1722 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1725 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1729 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1732 return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn));
1736 ffs_getpages(struct vop_getpages_args *ap)
1739 struct ufsmount *um;
1742 um = VFSTOUFS(vp->v_mount);
1744 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1745 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1746 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1747 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1748 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1752 ffs_getpages_async(struct vop_getpages_async_args *ap)
1755 struct ufsmount *um;
1759 um = VFSTOUFS(vp->v_mount);
1761 if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1762 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1763 ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg));
1765 error = vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1766 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz);
1767 ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);