2 * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
5 * This software was developed for the FreeBSD Project by Marshall
6 * Kirk McKusick and Network Associates Laboratories, the Security
7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * Copyright (c) 1982, 1986, 1989, 1993
33 * The Regents of the University of California. All rights reserved.
35 * Redistribution and use in source and binary forms, with or without
36 * modification, are permitted provided that the following conditions
38 * 1. Redistributions of source code must retain the above copyright
39 * notice, this list of conditions and the following disclaimer.
40 * 2. Redistributions in binary form must reproduce the above copyright
41 * notice, this list of conditions and the following disclaimer in the
42 * documentation and/or other materials provided with the distribution.
43 * 4. Neither the name of the University nor the names of its contributors
44 * may be used to endorse or promote products derived from this software
45 * without specific prior written permission.
47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95
60 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61 * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
67 #include <sys/param.h>
69 #include <sys/systm.h>
72 #include <sys/extattr.h>
73 #include <sys/kernel.h>
74 #include <sys/limits.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
78 #include <sys/rwlock.h>
80 #include <sys/vmmeter.h>
81 #include <sys/vnode.h>
84 #include <vm/vm_param.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vnode_pager.h>
91 #include <ufs/ufs/extattr.h>
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufs_extern.h>
95 #include <ufs/ufs/ufsmount.h>
97 #include <ufs/ffs/fs.h>
98 #include <ufs/ffs/ffs_extern.h>
99 #include "opt_directio.h"
103 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
105 static vop_fsync_t ffs_fsync;
106 static vop_lock1_t ffs_lock;
107 static vop_getpages_t ffs_getpages;
108 static vop_read_t ffs_read;
109 static vop_write_t ffs_write;
110 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
111 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
113 static vop_strategy_t ffsext_strategy;
114 static vop_closeextattr_t ffs_closeextattr;
115 static vop_deleteextattr_t ffs_deleteextattr;
116 static vop_getextattr_t ffs_getextattr;
117 static vop_listextattr_t ffs_listextattr;
118 static vop_openextattr_t ffs_openextattr;
119 static vop_setextattr_t ffs_setextattr;
120 static vop_vptofh_t ffs_vptofh;
123 /* Global vfs data structures for ufs. */
124 struct vop_vector ffs_vnodeops1 = {
125 .vop_default = &ufs_vnodeops,
126 .vop_fsync = ffs_fsync,
127 .vop_getpages = ffs_getpages,
128 .vop_lock1 = ffs_lock,
129 .vop_read = ffs_read,
130 .vop_reallocblks = ffs_reallocblks,
131 .vop_write = ffs_write,
132 .vop_vptofh = ffs_vptofh,
135 struct vop_vector ffs_fifoops1 = {
136 .vop_default = &ufs_fifoops,
137 .vop_fsync = ffs_fsync,
138 .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */
139 .vop_vptofh = ffs_vptofh,
142 /* Global vfs data structures for ufs. */
143 struct vop_vector ffs_vnodeops2 = {
144 .vop_default = &ufs_vnodeops,
145 .vop_fsync = ffs_fsync,
146 .vop_getpages = ffs_getpages,
147 .vop_lock1 = ffs_lock,
148 .vop_read = ffs_read,
149 .vop_reallocblks = ffs_reallocblks,
150 .vop_write = ffs_write,
151 .vop_closeextattr = ffs_closeextattr,
152 .vop_deleteextattr = ffs_deleteextattr,
153 .vop_getextattr = ffs_getextattr,
154 .vop_listextattr = ffs_listextattr,
155 .vop_openextattr = ffs_openextattr,
156 .vop_setextattr = ffs_setextattr,
157 .vop_vptofh = ffs_vptofh,
160 struct vop_vector ffs_fifoops2 = {
161 .vop_default = &ufs_fifoops,
162 .vop_fsync = ffs_fsync,
163 .vop_lock1 = ffs_lock,
164 .vop_reallocblks = ffs_reallocblks,
165 .vop_strategy = ffsext_strategy,
166 .vop_closeextattr = ffs_closeextattr,
167 .vop_deleteextattr = ffs_deleteextattr,
168 .vop_getextattr = ffs_getextattr,
169 .vop_listextattr = ffs_listextattr,
170 .vop_openextattr = ffs_openextattr,
171 .vop_setextattr = ffs_setextattr,
172 .vop_vptofh = ffs_vptofh,
176 * Synch an open file.
180 ffs_fsync(struct vop_fsync_args *ap)
189 error = ffs_syncvnode(vp, ap->a_waitfor, 0);
192 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
193 error = softdep_fsync(vp);
198 * The softdep_fsync() function may drop vp lock,
199 * allowing for dirty buffers to reappear on the
200 * bo_dirty list. Recheck and resync as needed.
203 if (vp->v_type == VREG && (bo->bo_numoutput > 0 ||
204 bo->bo_dirty.bv_cnt > 0)) {
214 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
221 int error, wait, passes;
224 ip->i_flag &= ~IN_NEEDSYNC;
228 * When doing MNT_WAIT we must first flush all dependencies
231 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
232 (error = softdep_sync_metadata(vp)) != 0)
236 * Flush all dirty buffers associated with a vnode.
240 wait = 0; /* Always do an async pass first. */
241 lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
244 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
245 bp->b_vflags &= ~BV_SCANNED;
246 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
248 * Reasons to skip this buffer: it has already been considered
249 * on this pass, the buffer has dependencies that will cause
250 * it to be redirtied and it has not already been deferred,
251 * or it is already being written.
253 if ((bp->b_vflags & BV_SCANNED) != 0)
255 bp->b_vflags |= BV_SCANNED;
256 /* Flush indirects in order. */
257 if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
258 lbn_level(bp->b_lblkno) >= passes)
260 if (bp->b_lblkno > lbn)
261 panic("ffs_syncvnode: syncing truncated data.");
262 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
265 if ((bp->b_flags & B_DELWRI) == 0)
266 panic("ffs_fsync: not dirty");
268 * Check for dependencies and potentially complete them.
270 if (!LIST_EMPTY(&bp->b_dep) &&
271 (error = softdep_sync_buf(vp, bp,
272 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
274 if (error != EBUSY) {
278 /* If we deferred once, don't defer again. */
279 if ((bp->b_flags & B_DEFERRED) == 0) {
280 bp->b_flags |= B_DEFERRED;
287 if ((error = bwrite(bp)) != 0)
289 } else if ((bp->b_flags & B_CLUSTEROK)) {
290 (void) vfs_bio_awrite(bp);
297 * Since we may have slept during the I/O, we need
298 * to start from a known point.
301 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
303 if (waitfor != MNT_WAIT) {
305 if ((flags & NO_INO_UPDT) != 0)
308 return (ffs_update(vp, 0));
310 /* Drain IO to see if we're done. */
311 bufobj_wwait(bo, 0, 0);
313 * Block devices associated with filesystems may have new I/O
314 * requests posted for them even if the vnode is locked, so no
315 * amount of trying will get them clean. We make several passes
318 * Regular files may need multiple passes to flush all dependency
319 * work as it is possible that we must write once per indirect
320 * level, once for the leaf, and once for the inode and each of
321 * these will be done with one sync and one async pass.
323 if (bo->bo_dirty.bv_cnt > 0) {
324 /* Write the inode after sync passes to flush deps. */
325 if (wait && DOINGSOFTDEP(vp) && (flags & NO_INO_UPDT) == 0) {
330 /* switch between sync/async. */
332 if (wait == 1 || ++passes < NIADDR + 2)
335 if (!vn_isdisk(vp, NULL))
336 vprint("ffs_fsync: dirty", vp);
341 if ((flags & NO_INO_UPDT) == 0)
342 error = ffs_update(vp, 1);
344 softdep_journal_fsync(VTOI(vp));
350 struct vop_lock1_args /* {
358 #ifndef NO_FFS_SNAPSHOT
364 switch (ap->a_flags & LK_TYPE_MASK) {
371 #ifdef DEBUG_VFS_LOCKS
372 KASSERT(vp->v_holdcnt != 0,
373 ("ffs_lock %p: zero hold count", vp));
376 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
377 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
378 ap->a_file, ap->a_line);
379 if (lkp == vp->v_vnlock || result != 0)
382 * Apparent success, except that the vnode
383 * mutated between snapshot file vnode and
384 * regular file vnode while this process
385 * slept. The lock currently held is not the
386 * right lock. Release it, and try to get the
389 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
390 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
391 ap->a_file, ap->a_line);
392 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
393 (LK_INTERLOCK | LK_NOWAIT))
395 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
396 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
397 flags &= ~LK_INTERLOCK;
401 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
405 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
410 * Vnode op for reading.
414 struct vop_read_args /* {
418 struct ucred *a_cred;
426 ufs_lbn_t lbn, nextlbn;
428 long size, xfersize, blkoffset;
436 ioflag = ap->a_ioflag;
437 if (ap->a_ioflag & IO_EXT)
439 return (ffs_extread(vp, uio, ioflag));
441 panic("ffs_read+IO_EXT");
444 if ((ioflag & IO_DIRECT) != 0) {
447 error = ffs_rawread(vp, uio, &workdone);
448 if (error != 0 || workdone != 0)
453 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
457 if (uio->uio_rw != UIO_READ)
458 panic("ffs_read: mode");
460 if (vp->v_type == VLNK) {
461 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
462 panic("ffs_read: short symlink");
463 } else if (vp->v_type != VREG && vp->v_type != VDIR)
464 panic("ffs_read: type %d", vp->v_type);
466 orig_resid = uio->uio_resid;
467 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
470 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
472 if (uio->uio_offset < ip->i_size &&
473 uio->uio_offset >= fs->fs_maxfilesize)
476 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
477 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
479 lbn = lblkno(fs, uio->uio_offset);
483 * size of buffer. The buffer representing the
484 * end of the file is rounded up to the size of
485 * the block type ( fragment or full block,
488 size = blksize(fs, ip, lbn);
489 blkoffset = blkoff(fs, uio->uio_offset);
492 * The amount we want to transfer in this iteration is
493 * one FS block less the amount of the data before
494 * our startpoint (duh!)
496 xfersize = fs->fs_bsize - blkoffset;
499 * But if we actually want less than the block,
500 * or the file doesn't have a whole block more of data,
501 * then use the lesser number.
503 if (uio->uio_resid < xfersize)
504 xfersize = uio->uio_resid;
505 if (bytesinfile < xfersize)
506 xfersize = bytesinfile;
508 if (lblktosize(fs, nextlbn) >= ip->i_size) {
510 * Don't do readahead if this is the end of the file.
512 error = bread_gb(vp, lbn, size, NOCRED,
514 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
516 * Otherwise if we are allowed to cluster,
517 * grab as much as we can.
519 * XXX This may not be a win if we are not
520 * doing sequential access.
522 error = cluster_read(vp, ip->i_size, lbn,
523 size, NOCRED, blkoffset + uio->uio_resid,
524 seqcount, GB_UNMAPPED, &bp);
525 } else if (seqcount > 1) {
527 * If we are NOT allowed to cluster, then
528 * if we appear to be acting sequentially,
529 * fire off a request for a readahead
530 * as well as a read. Note that the 4th and 5th
531 * arguments point to arrays of the size specified in
534 int nextsize = blksize(fs, ip, nextlbn);
535 error = breadn_flags(vp, lbn, size, &nextlbn,
536 &nextsize, 1, NOCRED, GB_UNMAPPED, &bp);
539 * Failing all of the above, just read what the
540 * user asked for. Interestingly, the same as
541 * the first option above.
543 error = bread_gb(vp, lbn, size, NOCRED,
553 * If IO_DIRECT then set B_DIRECT for the buffer. This
554 * will cause us to attempt to release the buffer later on
555 * and will cause the buffer cache to attempt to free the
558 if (ioflag & IO_DIRECT)
559 bp->b_flags |= B_DIRECT;
562 * We should only get non-zero b_resid when an I/O error
563 * has occurred, which should cause us to break above.
564 * However, if the short read did not cause an error,
565 * then we want to ensure that we do not uiomove bad
566 * or uninitialized data.
569 if (size < xfersize) {
575 if ((bp->b_flags & B_UNMAPPED) == 0) {
576 error = vn_io_fault_uiomove((char *)bp->b_data +
577 blkoffset, (int)xfersize, uio);
579 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
585 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
586 (LIST_EMPTY(&bp->b_dep))) {
588 * If there are no dependencies, and it's VMIO,
589 * then we don't need the buf, mark it available
590 * for freeing. For non-direct VMIO reads, the VM
593 bp->b_flags |= B_RELBUF;
597 * Otherwise let whoever
598 * made the request take care of
599 * freeing it. We just queue
600 * it onto another list.
607 * This can only happen in the case of an error
608 * because the loop above resets bp to NULL on each iteration
609 * and on normal completion has not set a new value into it.
610 * so it must have come from a 'break' statement
613 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
614 (LIST_EMPTY(&bp->b_dep))) {
615 bp->b_flags |= B_RELBUF;
622 if ((error == 0 || uio->uio_resid != orig_resid) &&
623 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
624 (ip->i_flag & IN_ACCESS) == 0) {
626 ip->i_flag |= IN_ACCESS;
633 * Vnode op for writing.
637 struct vop_write_args /* {
641 struct ucred *a_cred;
653 int blkoffset, error, flags, ioflag, size, xfersize;
657 ioflag = ap->a_ioflag;
658 if (ap->a_ioflag & IO_EXT)
660 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
662 panic("ffs_write+IO_EXT");
665 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
669 if (uio->uio_rw != UIO_WRITE)
670 panic("ffs_write: mode");
673 switch (vp->v_type) {
675 if (ioflag & IO_APPEND)
676 uio->uio_offset = ip->i_size;
677 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
683 panic("ffs_write: dir write");
686 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
687 (int)uio->uio_offset,
692 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
693 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
695 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
698 * Maybe this should be above the vnode op call, but so long as
699 * file servers have no limits, I don't think it matters.
701 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
704 resid = uio->uio_resid;
706 if (seqcount > BA_SEQMAX)
707 flags = BA_SEQMAX << BA_SEQSHIFT;
709 flags = seqcount << BA_SEQSHIFT;
710 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
712 flags |= BA_UNMAPPED;
714 for (error = 0; uio->uio_resid > 0;) {
715 lbn = lblkno(fs, uio->uio_offset);
716 blkoffset = blkoff(fs, uio->uio_offset);
717 xfersize = fs->fs_bsize - blkoffset;
718 if (uio->uio_resid < xfersize)
719 xfersize = uio->uio_resid;
720 if (uio->uio_offset + xfersize > ip->i_size)
721 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
724 * We must perform a read-before-write if the transfer size
725 * does not cover the entire buffer.
727 if (fs->fs_bsize > xfersize)
731 /* XXX is uio->uio_offset the right thing here? */
732 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
733 ap->a_cred, flags, &bp);
735 vnode_pager_setsize(vp, ip->i_size);
738 if (ioflag & IO_DIRECT)
739 bp->b_flags |= B_DIRECT;
740 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
741 bp->b_flags |= B_NOCACHE;
743 if (uio->uio_offset + xfersize > ip->i_size) {
744 ip->i_size = uio->uio_offset + xfersize;
745 DIP_SET(ip, i_size, ip->i_size);
748 size = blksize(fs, ip, lbn) - bp->b_resid;
752 if ((bp->b_flags & B_UNMAPPED) == 0) {
753 error = vn_io_fault_uiomove((char *)bp->b_data +
754 blkoffset, (int)xfersize, uio);
756 error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
760 * If the buffer is not already filled and we encounter an
761 * error while trying to fill it, we have to clear out any
762 * garbage data from the pages instantiated for the buffer.
763 * If we do not, a failed uiomove() during a write can leave
764 * the prior contents of the pages exposed to a userland mmap.
766 * Note that we need only clear buffers with a transfer size
767 * equal to the block size because buffers with a shorter
768 * transfer size were cleared above by the call to UFS_BALLOC()
769 * with the BA_CLRBUF flag set.
771 * If the source region for uiomove identically mmaps the
772 * buffer, uiomove() performed the NOP copy, and the buffer
773 * content remains valid because the page fault handler
774 * validated the pages.
776 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
777 fs->fs_bsize == xfersize)
779 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
780 (LIST_EMPTY(&bp->b_dep))) {
781 bp->b_flags |= B_RELBUF;
785 * If IO_SYNC each buffer is written synchronously. Otherwise
786 * if we have a severe page deficiency write the buffer
787 * asynchronously. Otherwise try to cluster, and if that
788 * doesn't do it then either do an async write (if O_DIRECT),
789 * or a delayed write (if not).
791 if (ioflag & IO_SYNC) {
793 } else if (vm_page_count_severe() ||
794 buf_dirty_count_severe() ||
795 (ioflag & IO_ASYNC)) {
796 bp->b_flags |= B_CLUSTEROK;
798 } else if (xfersize + blkoffset == fs->fs_bsize) {
799 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
800 bp->b_flags |= B_CLUSTEROK;
801 cluster_write(vp, bp, ip->i_size, seqcount,
806 } else if (ioflag & IO_DIRECT) {
807 bp->b_flags |= B_CLUSTEROK;
810 bp->b_flags |= B_CLUSTEROK;
813 if (error || xfersize == 0)
815 ip->i_flag |= IN_CHANGE | IN_UPDATE;
818 * If we successfully wrote any data, and we are not the superuser
819 * we clear the setuid and setgid bits as a precaution against
822 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
824 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
825 ip->i_mode &= ~(ISUID | ISGID);
826 DIP_SET(ip, i_mode, ip->i_mode);
830 if (ioflag & IO_UNIT) {
831 (void)ffs_truncate(vp, osize,
832 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
833 uio->uio_offset -= resid - uio->uio_resid;
834 uio->uio_resid = resid;
836 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
837 error = ffs_update(vp, 1);
846 struct vop_getpages_args *ap;
852 pcount = round_page(ap->a_count) / PAGE_SIZE;
853 mreq = ap->a_m[ap->a_reqpage];
856 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
857 * then the entire page is valid. Since the page may be mapped,
858 * user programs might reference data beyond the actual end of file
859 * occuring within the page. We have to zero that data.
861 VM_OBJECT_WLOCK(mreq->object);
863 if (mreq->valid != VM_PAGE_BITS_ALL)
864 vm_page_zero_invalid(mreq, TRUE);
865 for (i = 0; i < pcount; i++) {
866 if (i != ap->a_reqpage) {
867 vm_page_lock(ap->a_m[i]);
868 vm_page_free(ap->a_m[i]);
869 vm_page_unlock(ap->a_m[i]);
872 VM_OBJECT_WUNLOCK(mreq->object);
875 VM_OBJECT_WUNLOCK(mreq->object);
877 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
884 * Extended attribute area reading.
887 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
890 struct ufs2_dinode *dp;
893 ufs_lbn_t lbn, nextlbn;
895 long size, xfersize, blkoffset;
904 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
905 panic("ffs_extread: mode");
908 orig_resid = uio->uio_resid;
909 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
912 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
914 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
915 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
917 lbn = lblkno(fs, uio->uio_offset);
921 * size of buffer. The buffer representing the
922 * end of the file is rounded up to the size of
923 * the block type ( fragment or full block,
926 size = sblksize(fs, dp->di_extsize, lbn);
927 blkoffset = blkoff(fs, uio->uio_offset);
930 * The amount we want to transfer in this iteration is
931 * one FS block less the amount of the data before
932 * our startpoint (duh!)
934 xfersize = fs->fs_bsize - blkoffset;
937 * But if we actually want less than the block,
938 * or the file doesn't have a whole block more of data,
939 * then use the lesser number.
941 if (uio->uio_resid < xfersize)
942 xfersize = uio->uio_resid;
943 if (bytesinfile < xfersize)
944 xfersize = bytesinfile;
946 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
948 * Don't do readahead if this is the end of the info.
950 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
953 * If we have a second block, then
954 * fire off a request for a readahead
955 * as well as a read. Note that the 4th and 5th
956 * arguments point to arrays of the size specified in
959 int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
961 nextlbn = -1 - nextlbn;
962 error = breadn(vp, -1 - lbn,
963 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
972 * If IO_DIRECT then set B_DIRECT for the buffer. This
973 * will cause us to attempt to release the buffer later on
974 * and will cause the buffer cache to attempt to free the
977 if (ioflag & IO_DIRECT)
978 bp->b_flags |= B_DIRECT;
981 * We should only get non-zero b_resid when an I/O error
982 * has occurred, which should cause us to break above.
983 * However, if the short read did not cause an error,
984 * then we want to ensure that we do not uiomove bad
985 * or uninitialized data.
988 if (size < xfersize) {
994 error = uiomove((char *)bp->b_data + blkoffset,
999 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1000 (LIST_EMPTY(&bp->b_dep))) {
1002 * If there are no dependencies, and it's VMIO,
1003 * then we don't need the buf, mark it available
1004 * for freeing. For non-direct VMIO reads, the VM
1007 bp->b_flags |= B_RELBUF;
1011 * Otherwise let whoever
1012 * made the request take care of
1013 * freeing it. We just queue
1014 * it onto another list.
1021 * This can only happen in the case of an error
1022 * because the loop above resets bp to NULL on each iteration
1023 * and on normal completion has not set a new value into it.
1024 * so it must have come from a 'break' statement
1027 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1028 (LIST_EMPTY(&bp->b_dep))) {
1029 bp->b_flags |= B_RELBUF;
1039 * Extended attribute area writing.
1042 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1045 struct ufs2_dinode *dp;
1051 int blkoffset, error, flags, size, xfersize;
1058 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1059 panic("ffs_extwrite: mode");
1062 if (ioflag & IO_APPEND)
1063 uio->uio_offset = dp->di_extsize;
1064 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1065 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1066 if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1069 resid = uio->uio_resid;
1070 osize = dp->di_extsize;
1072 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1075 for (error = 0; uio->uio_resid > 0;) {
1076 lbn = lblkno(fs, uio->uio_offset);
1077 blkoffset = blkoff(fs, uio->uio_offset);
1078 xfersize = fs->fs_bsize - blkoffset;
1079 if (uio->uio_resid < xfersize)
1080 xfersize = uio->uio_resid;
1083 * We must perform a read-before-write if the transfer size
1084 * does not cover the entire buffer.
1086 if (fs->fs_bsize > xfersize)
1089 flags &= ~BA_CLRBUF;
1090 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1095 * If the buffer is not valid we have to clear out any
1096 * garbage data from the pages instantiated for the buffer.
1097 * If we do not, a failed uiomove() during a write can leave
1098 * the prior contents of the pages exposed to a userland
1099 * mmap(). XXX deal with uiomove() errors a better way.
1101 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1103 if (ioflag & IO_DIRECT)
1104 bp->b_flags |= B_DIRECT;
1106 if (uio->uio_offset + xfersize > dp->di_extsize)
1107 dp->di_extsize = uio->uio_offset + xfersize;
1109 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1110 if (size < xfersize)
1114 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1115 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1116 (LIST_EMPTY(&bp->b_dep))) {
1117 bp->b_flags |= B_RELBUF;
1121 * If IO_SYNC each buffer is written synchronously. Otherwise
1122 * if we have a severe page deficiency write the buffer
1123 * asynchronously. Otherwise try to cluster, and if that
1124 * doesn't do it then either do an async write (if O_DIRECT),
1125 * or a delayed write (if not).
1127 if (ioflag & IO_SYNC) {
1129 } else if (vm_page_count_severe() ||
1130 buf_dirty_count_severe() ||
1131 xfersize + blkoffset == fs->fs_bsize ||
1132 (ioflag & (IO_ASYNC | IO_DIRECT)))
1136 if (error || xfersize == 0)
1138 ip->i_flag |= IN_CHANGE;
1141 * If we successfully wrote any data, and we are not the superuser
1142 * we clear the setuid and setgid bits as a precaution against
1145 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1146 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1147 ip->i_mode &= ~(ISUID | ISGID);
1148 dp->di_mode = ip->i_mode;
1152 if (ioflag & IO_UNIT) {
1153 (void)ffs_truncate(vp, osize,
1154 IO_EXT | (ioflag&IO_SYNC), ucred);
1155 uio->uio_offset -= resid - uio->uio_resid;
1156 uio->uio_resid = resid;
1158 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1159 error = ffs_update(vp, 1);
1165 * Vnode operating to retrieve a named extended attribute.
1167 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1168 * the length of the EA, and possibly the pointer to the entry and to the data.
1171 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1173 u_char *p, *pe, *pn, *p0;
1174 int eapad1, eapad2, ealength, ealen, nlen;
1178 nlen = strlen(name);
1180 for (p = ptr; p < pe; p = pn) {
1182 bcopy(p, &ul, sizeof(ul));
1184 /* make sure this entry is complete */
1187 p += sizeof(uint32_t);
1195 if (bcmp(p, name, nlen))
1197 ealength = sizeof(uint32_t) + 3 + nlen;
1198 eapad1 = 8 - (ealength % 8);
1202 ealen = ul - ealength - eapad2;
1214 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1217 struct ufs2_dinode *dp;
1220 struct iovec liovec;
1227 easize = dp->di_extsize;
1228 if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1231 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1233 liovec.iov_base = eae;
1234 liovec.iov_len = easize;
1235 luio.uio_iov = &liovec;
1236 luio.uio_iovcnt = 1;
1237 luio.uio_offset = 0;
1238 luio.uio_resid = easize;
1239 luio.uio_segflg = UIO_SYSSPACE;
1240 luio.uio_rw = UIO_READ;
1243 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1253 ffs_lock_ea(struct vnode *vp)
1259 while (ip->i_flag & IN_EA_LOCKED) {
1260 ip->i_flag |= IN_EA_LOCKWAIT;
1261 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1264 ip->i_flag |= IN_EA_LOCKED;
1269 ffs_unlock_ea(struct vnode *vp)
1275 if (ip->i_flag & IN_EA_LOCKWAIT)
1276 wakeup(&ip->i_ea_refs);
1277 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1282 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1285 struct ufs2_dinode *dp;
1291 if (ip->i_ea_area != NULL) {
1297 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1302 ip->i_ea_len = dp->di_extsize;
1310 * Vnode extattr transaction commit/abort
1313 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1317 struct iovec liovec;
1319 struct ufs2_dinode *dp;
1324 if (ip->i_ea_area == NULL) {
1329 error = ip->i_ea_error;
1330 if (commit && error == 0) {
1331 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1333 cred = vp->v_mount->mnt_cred;
1334 liovec.iov_base = ip->i_ea_area;
1335 liovec.iov_len = ip->i_ea_len;
1336 luio.uio_iov = &liovec;
1337 luio.uio_iovcnt = 1;
1338 luio.uio_offset = 0;
1339 luio.uio_resid = ip->i_ea_len;
1340 luio.uio_segflg = UIO_SYSSPACE;
1341 luio.uio_rw = UIO_WRITE;
1343 /* XXX: I'm not happy about truncating to zero size */
1344 if (ip->i_ea_len < dp->di_extsize)
1345 error = ffs_truncate(vp, 0, IO_EXT, cred);
1346 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1348 if (--ip->i_ea_refs == 0) {
1349 free(ip->i_ea_area, M_TEMP);
1350 ip->i_ea_area = NULL;
1359 * Vnode extattr strategy routine for fifos.
1361 * We need to check for a read or write of the external attributes.
1362 * Otherwise we just fall through and do the usual thing.
1365 ffsext_strategy(struct vop_strategy_args *ap)
1367 struct vop_strategy_args {
1368 struct vnodeop_desc *a_desc;
1378 lbn = ap->a_bp->b_lblkno;
1379 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1380 lbn < 0 && lbn >= -NXADDR)
1381 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1382 if (vp->v_type == VFIFO)
1383 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1384 panic("spec nodes went here");
1388 * Vnode extattr transaction commit/abort
1391 ffs_openextattr(struct vop_openextattr_args *ap)
1393 struct vop_openextattr_args {
1394 struct vnodeop_desc *a_desc;
1396 IN struct ucred *a_cred;
1397 IN struct thread *a_td;
1404 ip = VTOI(ap->a_vp);
1407 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1408 return (EOPNOTSUPP);
1410 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1415 * Vnode extattr transaction commit/abort
1418 ffs_closeextattr(struct vop_closeextattr_args *ap)
1420 struct vop_closeextattr_args {
1421 struct vnodeop_desc *a_desc;
1424 IN struct ucred *a_cred;
1425 IN struct thread *a_td;
1432 ip = VTOI(ap->a_vp);
1435 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1436 return (EOPNOTSUPP);
1438 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1441 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1445 * Vnode operation to remove a named attribute.
1448 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1451 IN struct vnode *a_vp;
1452 IN int a_attrnamespace;
1453 IN const char *a_name;
1454 IN struct ucred *a_cred;
1455 IN struct thread *a_td;
1461 uint32_t ealength, ul;
1462 int ealen, olen, eapad1, eapad2, error, i, easize;
1465 ip = VTOI(ap->a_vp);
1468 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1469 return (EOPNOTSUPP);
1471 if (strlen(ap->a_name) == 0)
1474 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1477 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1478 ap->a_cred, ap->a_td, VWRITE);
1482 * ffs_lock_ea is not needed there, because the vnode
1483 * must be exclusively locked.
1485 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1486 ip->i_ea_error = error;
1490 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1494 ealength = eapad1 = ealen = eapad2 = 0;
1496 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1497 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1498 easize = ip->i_ea_len;
1500 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1503 /* delete but nonexistent */
1505 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1508 bcopy(p, &ul, sizeof ul);
1510 if (ul != ealength) {
1511 bcopy(p + ul, p + ealength, easize - i);
1512 easize += (ealength - ul);
1514 if (easize > NXADDR * fs->fs_bsize) {
1516 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1517 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1518 ip->i_ea_error = ENOSPC;
1522 ip->i_ea_area = eae;
1523 ip->i_ea_len = easize;
1525 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1530 * Vnode operation to retrieve a named extended attribute.
1533 ffs_getextattr(struct vop_getextattr_args *ap)
1536 IN struct vnode *a_vp;
1537 IN int a_attrnamespace;
1538 IN const char *a_name;
1539 INOUT struct uio *a_uio;
1541 IN struct ucred *a_cred;
1542 IN struct thread *a_td;
1552 ip = VTOI(ap->a_vp);
1555 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1556 return (EOPNOTSUPP);
1558 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1559 ap->a_cred, ap->a_td, VREAD);
1563 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1567 eae = ip->i_ea_area;
1568 easize = ip->i_ea_len;
1570 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1574 if (ap->a_size != NULL)
1575 *ap->a_size = ealen;
1576 else if (ap->a_uio != NULL)
1577 error = uiomove(p, ealen, ap->a_uio);
1581 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1586 * Vnode operation to retrieve extended attributes on a vnode.
1589 ffs_listextattr(struct vop_listextattr_args *ap)
1592 IN struct vnode *a_vp;
1593 IN int a_attrnamespace;
1594 INOUT struct uio *a_uio;
1596 IN struct ucred *a_cred;
1597 IN struct thread *a_td;
1603 u_char *eae, *p, *pe, *pn;
1608 ip = VTOI(ap->a_vp);
1611 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1612 return (EOPNOTSUPP);
1614 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1615 ap->a_cred, ap->a_td, VREAD);
1619 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1622 eae = ip->i_ea_area;
1623 easize = ip->i_ea_len;
1626 if (ap->a_size != NULL)
1629 for(p = eae; error == 0 && p < pe; p = pn) {
1630 bcopy(p, &ul, sizeof(ul));
1635 if (*p++ != ap->a_attrnamespace)
1639 if (ap->a_size != NULL) {
1640 *ap->a_size += ealen + 1;
1641 } else if (ap->a_uio != NULL) {
1642 error = uiomove(p, ealen + 1, ap->a_uio);
1645 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1650 * Vnode operation to set a named attribute.
1653 ffs_setextattr(struct vop_setextattr_args *ap)
1656 IN struct vnode *a_vp;
1657 IN int a_attrnamespace;
1658 IN const char *a_name;
1659 INOUT struct uio *a_uio;
1660 IN struct ucred *a_cred;
1661 IN struct thread *a_td;
1667 uint32_t ealength, ul;
1669 int olen, eapad1, eapad2, error, i, easize;
1672 ip = VTOI(ap->a_vp);
1675 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1676 return (EOPNOTSUPP);
1678 if (strlen(ap->a_name) == 0)
1681 /* XXX Now unsupported API to delete EAs using NULL uio. */
1682 if (ap->a_uio == NULL)
1683 return (EOPNOTSUPP);
1685 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1688 ealen = ap->a_uio->uio_resid;
1689 if (ealen < 0 || ealen > lblktosize(fs, NXADDR))
1692 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1693 ap->a_cred, ap->a_td, VWRITE);
1697 * ffs_lock_ea is not needed there, because the vnode
1698 * must be exclusively locked.
1700 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1701 ip->i_ea_error = error;
1705 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1709 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1710 eapad1 = 8 - (ealength % 8);
1713 eapad2 = 8 - (ealen % 8);
1716 ealength += eapad1 + ealen + eapad2;
1718 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1719 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1720 easize = ip->i_ea_len;
1722 olen = ffs_findextattr(eae, easize,
1723 ap->a_attrnamespace, ap->a_name, &p, NULL);
1725 /* new, append at end */
1729 bcopy(p, &ul, sizeof ul);
1731 if (ul != ealength) {
1732 bcopy(p + ul, p + ealength, easize - i);
1733 easize += (ealength - ul);
1736 if (easize > lblktosize(fs, NXADDR)) {
1738 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1739 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1740 ip->i_ea_error = ENOSPC;
1743 bcopy(&ealength, p, sizeof(ealength));
1744 p += sizeof(ealength);
1745 *p++ = ap->a_attrnamespace;
1747 *p++ = strlen(ap->a_name);
1748 strcpy(p, ap->a_name);
1749 p += strlen(ap->a_name);
1752 error = uiomove(p, ealen, ap->a_uio);
1755 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1756 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1757 ip->i_ea_error = error;
1764 ip->i_ea_area = eae;
1765 ip->i_ea_len = easize;
1767 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1772 * Vnode pointer to File handle
1775 ffs_vptofh(struct vop_vptofh_args *ap)
1778 IN struct vnode *a_vp;
1779 IN struct fid *a_fhp;
1786 ip = VTOI(ap->a_vp);
1787 ufhp = (struct ufid *)ap->a_fhp;
1788 ufhp->ufid_len = sizeof(struct ufid);
1789 ufhp->ufid_ino = ip->i_number;
1790 ufhp->ufid_gen = ip->i_gen;