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>
79 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
83 #include <vm/vm_param.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vnode_pager.h>
90 #include <ufs/ufs/extattr.h>
91 #include <ufs/ufs/quota.h>
92 #include <ufs/ufs/inode.h>
93 #include <ufs/ufs/ufs_extern.h>
94 #include <ufs/ufs/ufsmount.h>
96 #include <ufs/ffs/fs.h>
97 #include <ufs/ffs/ffs_extern.h>
98 #include "opt_directio.h"
102 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
104 static vop_fsync_t ffs_fsync;
105 static vop_lock1_t ffs_lock;
106 static vop_getpages_t ffs_getpages;
107 static vop_read_t ffs_read;
108 static vop_write_t ffs_write;
109 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
110 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
112 static vop_strategy_t ffsext_strategy;
113 static vop_closeextattr_t ffs_closeextattr;
114 static vop_deleteextattr_t ffs_deleteextattr;
115 static vop_getextattr_t ffs_getextattr;
116 static vop_listextattr_t ffs_listextattr;
117 static vop_openextattr_t ffs_openextattr;
118 static vop_setextattr_t ffs_setextattr;
119 static vop_vptofh_t ffs_vptofh;
122 /* Global vfs data structures for ufs. */
123 struct vop_vector ffs_vnodeops1 = {
124 .vop_default = &ufs_vnodeops,
125 .vop_fsync = ffs_fsync,
126 .vop_getpages = ffs_getpages,
127 .vop_lock1 = ffs_lock,
128 .vop_read = ffs_read,
129 .vop_reallocblks = ffs_reallocblks,
130 .vop_write = ffs_write,
131 .vop_vptofh = ffs_vptofh,
134 struct vop_vector ffs_fifoops1 = {
135 .vop_default = &ufs_fifoops,
136 .vop_fsync = ffs_fsync,
137 .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */
138 .vop_vptofh = ffs_vptofh,
141 /* Global vfs data structures for ufs. */
142 struct vop_vector ffs_vnodeops2 = {
143 .vop_default = &ufs_vnodeops,
144 .vop_fsync = ffs_fsync,
145 .vop_getpages = ffs_getpages,
146 .vop_lock1 = ffs_lock,
147 .vop_read = ffs_read,
148 .vop_reallocblks = ffs_reallocblks,
149 .vop_write = ffs_write,
150 .vop_closeextattr = ffs_closeextattr,
151 .vop_deleteextattr = ffs_deleteextattr,
152 .vop_getextattr = ffs_getextattr,
153 .vop_listextattr = ffs_listextattr,
154 .vop_openextattr = ffs_openextattr,
155 .vop_setextattr = ffs_setextattr,
156 .vop_vptofh = ffs_vptofh,
159 struct vop_vector ffs_fifoops2 = {
160 .vop_default = &ufs_fifoops,
161 .vop_fsync = ffs_fsync,
162 .vop_lock1 = ffs_lock,
163 .vop_reallocblks = ffs_reallocblks,
164 .vop_strategy = ffsext_strategy,
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,
175 * Synch an open file.
179 ffs_fsync(struct vop_fsync_args *ap)
188 error = ffs_syncvnode(vp, ap->a_waitfor, 0);
191 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
192 error = softdep_fsync(vp);
197 * The softdep_fsync() function may drop vp lock,
198 * allowing for dirty buffers to reappear on the
199 * bo_dirty list. Recheck and resync as needed.
202 if (vp->v_type == VREG && (bo->bo_numoutput > 0 ||
203 bo->bo_dirty.bv_cnt > 0)) {
213 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
220 int error, wait, passes;
223 ip->i_flag &= ~IN_NEEDSYNC;
227 * When doing MNT_WAIT we must first flush all dependencies
230 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
231 (error = softdep_sync_metadata(vp)) != 0)
235 * Flush all dirty buffers associated with a vnode.
239 wait = 0; /* Always do an async pass first. */
240 lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
243 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
244 bp->b_vflags &= ~BV_SCANNED;
245 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
247 * Reasons to skip this buffer: it has already been considered
248 * on this pass, the buffer has dependencies that will cause
249 * it to be redirtied and it has not already been deferred,
250 * or it is already being written.
252 if ((bp->b_vflags & BV_SCANNED) != 0)
254 bp->b_vflags |= BV_SCANNED;
255 /* Flush indirects in order. */
256 if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
257 lbn_level(bp->b_lblkno) >= passes)
259 if (bp->b_lblkno > lbn)
260 panic("ffs_syncvnode: syncing truncated data.");
261 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
264 if ((bp->b_flags & B_DELWRI) == 0)
265 panic("ffs_fsync: not dirty");
267 * Check for dependencies and potentially complete them.
269 if (!LIST_EMPTY(&bp->b_dep) &&
270 (error = softdep_sync_buf(vp, bp,
271 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
273 if (error != EBUSY) {
277 /* If we deferred once, don't defer again. */
278 if ((bp->b_flags & B_DEFERRED) == 0) {
279 bp->b_flags |= B_DEFERRED;
286 if ((error = bwrite(bp)) != 0)
288 } else if ((bp->b_flags & B_CLUSTEROK)) {
289 (void) vfs_bio_awrite(bp);
296 * Since we may have slept during the I/O, we need
297 * to start from a known point.
300 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
302 if (waitfor != MNT_WAIT) {
304 if ((flags & NO_INO_UPDT) != 0)
307 return (ffs_update(vp, 0));
309 /* Drain IO to see if we're done. */
310 bufobj_wwait(bo, 0, 0);
312 * Block devices associated with filesystems may have new I/O
313 * requests posted for them even if the vnode is locked, so no
314 * amount of trying will get them clean. We make several passes
317 * Regular files may need multiple passes to flush all dependency
318 * work as it is possible that we must write once per indirect
319 * level, once for the leaf, and once for the inode and each of
320 * these will be done with one sync and one async pass.
322 if (bo->bo_dirty.bv_cnt > 0) {
323 /* Write the inode after sync passes to flush deps. */
324 if (wait && DOINGSOFTDEP(vp) && (flags & NO_INO_UPDT) == 0) {
329 /* switch between sync/async. */
331 if (wait == 1 || ++passes < NIADDR + 2)
334 if (!vn_isdisk(vp, NULL))
335 vprint("ffs_fsync: dirty", vp);
340 if ((flags & NO_INO_UPDT) == 0)
341 error = ffs_update(vp, 1);
343 softdep_journal_fsync(VTOI(vp));
349 struct vop_lock1_args /* {
357 #ifndef NO_FFS_SNAPSHOT
363 switch (ap->a_flags & LK_TYPE_MASK) {
370 #ifdef DEBUG_VFS_LOCKS
371 KASSERT(vp->v_holdcnt != 0,
372 ("ffs_lock %p: zero hold count", vp));
375 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
376 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
377 ap->a_file, ap->a_line);
378 if (lkp == vp->v_vnlock || result != 0)
381 * Apparent success, except that the vnode
382 * mutated between snapshot file vnode and
383 * regular file vnode while this process
384 * slept. The lock currently held is not the
385 * right lock. Release it, and try to get the
388 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
389 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
390 ap->a_file, ap->a_line);
391 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
392 (LK_INTERLOCK | LK_NOWAIT))
394 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
395 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
396 flags &= ~LK_INTERLOCK;
400 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
404 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
409 * Vnode op for reading.
413 struct vop_read_args /* {
417 struct ucred *a_cred;
425 ufs_lbn_t lbn, nextlbn;
427 long size, xfersize, blkoffset;
435 ioflag = ap->a_ioflag;
436 if (ap->a_ioflag & IO_EXT)
438 return (ffs_extread(vp, uio, ioflag));
440 panic("ffs_read+IO_EXT");
443 if ((ioflag & IO_DIRECT) != 0) {
446 error = ffs_rawread(vp, uio, &workdone);
447 if (error != 0 || workdone != 0)
452 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
456 if (uio->uio_rw != UIO_READ)
457 panic("ffs_read: mode");
459 if (vp->v_type == VLNK) {
460 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
461 panic("ffs_read: short symlink");
462 } else if (vp->v_type != VREG && vp->v_type != VDIR)
463 panic("ffs_read: type %d", vp->v_type);
465 orig_resid = uio->uio_resid;
466 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
469 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
471 if (uio->uio_offset < ip->i_size &&
472 uio->uio_offset >= fs->fs_maxfilesize)
475 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
476 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
478 lbn = lblkno(fs, uio->uio_offset);
482 * size of buffer. The buffer representing the
483 * end of the file is rounded up to the size of
484 * the block type ( fragment or full block,
487 size = blksize(fs, ip, lbn);
488 blkoffset = blkoff(fs, uio->uio_offset);
491 * The amount we want to transfer in this iteration is
492 * one FS block less the amount of the data before
493 * our startpoint (duh!)
495 xfersize = fs->fs_bsize - blkoffset;
498 * But if we actually want less than the block,
499 * or the file doesn't have a whole block more of data,
500 * then use the lesser number.
502 if (uio->uio_resid < xfersize)
503 xfersize = uio->uio_resid;
504 if (bytesinfile < xfersize)
505 xfersize = bytesinfile;
507 if (lblktosize(fs, nextlbn) >= ip->i_size) {
509 * Don't do readahead if this is the end of the file.
511 error = bread(vp, lbn, size, NOCRED, &bp);
512 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
514 * Otherwise if we are allowed to cluster,
515 * grab as much as we can.
517 * XXX This may not be a win if we are not
518 * doing sequential access.
520 error = cluster_read(vp, ip->i_size, lbn,
521 size, NOCRED, blkoffset + uio->uio_resid,
523 } else if (seqcount > 1) {
525 * If we are NOT allowed to cluster, then
526 * if we appear to be acting sequentially,
527 * fire off a request for a readahead
528 * as well as a read. Note that the 4th and 5th
529 * arguments point to arrays of the size specified in
532 int nextsize = blksize(fs, ip, nextlbn);
533 error = breadn(vp, lbn,
534 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
537 * Failing all of the above, just read what the
538 * user asked for. Interestingly, the same as
539 * the first option above.
541 error = bread(vp, lbn, size, NOCRED, &bp);
550 * If IO_DIRECT then set B_DIRECT for the buffer. This
551 * will cause us to attempt to release the buffer later on
552 * and will cause the buffer cache to attempt to free the
555 if (ioflag & IO_DIRECT)
556 bp->b_flags |= B_DIRECT;
559 * We should only get non-zero b_resid when an I/O error
560 * has occurred, which should cause us to break above.
561 * However, if the short read did not cause an error,
562 * then we want to ensure that we do not uiomove bad
563 * or uninitialized data.
566 if (size < xfersize) {
572 error = vn_io_fault_uiomove((char *)bp->b_data + blkoffset,
577 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
578 (LIST_EMPTY(&bp->b_dep))) {
580 * If there are no dependencies, and it's VMIO,
581 * then we don't need the buf, mark it available
582 * for freeing. For non-direct VMIO reads, the VM
585 bp->b_flags |= B_RELBUF;
589 * Otherwise let whoever
590 * made the request take care of
591 * freeing it. We just queue
592 * it onto another list.
599 * This can only happen in the case of an error
600 * because the loop above resets bp to NULL on each iteration
601 * and on normal completion has not set a new value into it.
602 * so it must have come from a 'break' statement
605 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
606 (LIST_EMPTY(&bp->b_dep))) {
607 bp->b_flags |= B_RELBUF;
614 if ((error == 0 || uio->uio_resid != orig_resid) &&
615 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
616 (ip->i_flag & IN_ACCESS) == 0) {
618 ip->i_flag |= IN_ACCESS;
625 * Vnode op for writing.
629 struct vop_write_args /* {
633 struct ucred *a_cred;
645 int blkoffset, error, flags, ioflag, size, xfersize;
649 ioflag = ap->a_ioflag;
650 if (ap->a_ioflag & IO_EXT)
652 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
654 panic("ffs_write+IO_EXT");
657 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
661 if (uio->uio_rw != UIO_WRITE)
662 panic("ffs_write: mode");
665 switch (vp->v_type) {
667 if (ioflag & IO_APPEND)
668 uio->uio_offset = ip->i_size;
669 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
675 panic("ffs_write: dir write");
678 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
679 (int)uio->uio_offset,
684 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
685 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
687 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
690 * Maybe this should be above the vnode op call, but so long as
691 * file servers have no limits, I don't think it matters.
693 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
696 resid = uio->uio_resid;
698 if (seqcount > BA_SEQMAX)
699 flags = BA_SEQMAX << BA_SEQSHIFT;
701 flags = seqcount << BA_SEQSHIFT;
702 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
705 for (error = 0; uio->uio_resid > 0;) {
706 lbn = lblkno(fs, uio->uio_offset);
707 blkoffset = blkoff(fs, uio->uio_offset);
708 xfersize = fs->fs_bsize - blkoffset;
709 if (uio->uio_resid < xfersize)
710 xfersize = uio->uio_resid;
711 if (uio->uio_offset + xfersize > ip->i_size)
712 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
715 * We must perform a read-before-write if the transfer size
716 * does not cover the entire buffer.
718 if (fs->fs_bsize > xfersize)
722 /* XXX is uio->uio_offset the right thing here? */
723 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
724 ap->a_cred, flags, &bp);
726 vnode_pager_setsize(vp, ip->i_size);
729 if (ioflag & IO_DIRECT)
730 bp->b_flags |= B_DIRECT;
731 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
732 bp->b_flags |= B_NOCACHE;
734 if (uio->uio_offset + xfersize > ip->i_size) {
735 ip->i_size = uio->uio_offset + xfersize;
736 DIP_SET(ip, i_size, ip->i_size);
739 size = blksize(fs, ip, lbn) - bp->b_resid;
743 error = vn_io_fault_uiomove((char *)bp->b_data + blkoffset,
746 * If the buffer is not already filled and we encounter an
747 * error while trying to fill it, we have to clear out any
748 * garbage data from the pages instantiated for the buffer.
749 * If we do not, a failed uiomove() during a write can leave
750 * the prior contents of the pages exposed to a userland mmap.
752 * Note that we need only clear buffers with a transfer size
753 * equal to the block size because buffers with a shorter
754 * transfer size were cleared above by the call to UFS_BALLOC()
755 * with the BA_CLRBUF flag set.
757 * If the source region for uiomove identically mmaps the
758 * buffer, uiomove() performed the NOP copy, and the buffer
759 * content remains valid because the page fault handler
760 * validated the pages.
762 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
763 fs->fs_bsize == xfersize)
765 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
766 (LIST_EMPTY(&bp->b_dep))) {
767 bp->b_flags |= B_RELBUF;
771 * If IO_SYNC each buffer is written synchronously. Otherwise
772 * if we have a severe page deficiency write the buffer
773 * asynchronously. Otherwise try to cluster, and if that
774 * doesn't do it then either do an async write (if O_DIRECT),
775 * or a delayed write (if not).
777 if (ioflag & IO_SYNC) {
779 } else if (vm_page_count_severe() ||
780 buf_dirty_count_severe() ||
781 (ioflag & IO_ASYNC)) {
782 bp->b_flags |= B_CLUSTEROK;
784 } else if (xfersize + blkoffset == fs->fs_bsize) {
785 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
786 bp->b_flags |= B_CLUSTEROK;
787 cluster_write(vp, bp, ip->i_size, seqcount);
791 } else if (ioflag & IO_DIRECT) {
792 bp->b_flags |= B_CLUSTEROK;
795 bp->b_flags |= B_CLUSTEROK;
798 if (error || xfersize == 0)
800 ip->i_flag |= IN_CHANGE | IN_UPDATE;
803 * If we successfully wrote any data, and we are not the superuser
804 * we clear the setuid and setgid bits as a precaution against
807 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
809 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
810 ip->i_mode &= ~(ISUID | ISGID);
811 DIP_SET(ip, i_mode, ip->i_mode);
815 if (ioflag & IO_UNIT) {
816 (void)ffs_truncate(vp, osize,
817 IO_NORMAL | (ioflag & IO_SYNC),
818 ap->a_cred, uio->uio_td);
819 uio->uio_offset -= resid - uio->uio_resid;
820 uio->uio_resid = resid;
822 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
823 error = ffs_update(vp, 1);
832 struct vop_getpages_args *ap;
838 pcount = round_page(ap->a_count) / PAGE_SIZE;
839 mreq = ap->a_m[ap->a_reqpage];
842 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
843 * then the entire page is valid. Since the page may be mapped,
844 * user programs might reference data beyond the actual end of file
845 * occuring within the page. We have to zero that data.
847 VM_OBJECT_LOCK(mreq->object);
849 if (mreq->valid != VM_PAGE_BITS_ALL)
850 vm_page_zero_invalid(mreq, TRUE);
851 for (i = 0; i < pcount; i++) {
852 if (i != ap->a_reqpage) {
853 vm_page_lock(ap->a_m[i]);
854 vm_page_free(ap->a_m[i]);
855 vm_page_unlock(ap->a_m[i]);
858 VM_OBJECT_UNLOCK(mreq->object);
861 VM_OBJECT_UNLOCK(mreq->object);
863 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
870 * Extended attribute area reading.
873 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
876 struct ufs2_dinode *dp;
879 ufs_lbn_t lbn, nextlbn;
881 long size, xfersize, blkoffset;
890 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
891 panic("ffs_extread: mode");
894 orig_resid = uio->uio_resid;
895 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
898 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
900 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
901 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
903 lbn = lblkno(fs, uio->uio_offset);
907 * size of buffer. The buffer representing the
908 * end of the file is rounded up to the size of
909 * the block type ( fragment or full block,
912 size = sblksize(fs, dp->di_extsize, lbn);
913 blkoffset = blkoff(fs, uio->uio_offset);
916 * The amount we want to transfer in this iteration is
917 * one FS block less the amount of the data before
918 * our startpoint (duh!)
920 xfersize = fs->fs_bsize - blkoffset;
923 * But if we actually want less than the block,
924 * or the file doesn't have a whole block more of data,
925 * then use the lesser number.
927 if (uio->uio_resid < xfersize)
928 xfersize = uio->uio_resid;
929 if (bytesinfile < xfersize)
930 xfersize = bytesinfile;
932 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
934 * Don't do readahead if this is the end of the info.
936 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
939 * If we have a second block, then
940 * fire off a request for a readahead
941 * as well as a read. Note that the 4th and 5th
942 * arguments point to arrays of the size specified in
945 int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
947 nextlbn = -1 - nextlbn;
948 error = breadn(vp, -1 - lbn,
949 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
958 * If IO_DIRECT then set B_DIRECT for the buffer. This
959 * will cause us to attempt to release the buffer later on
960 * and will cause the buffer cache to attempt to free the
963 if (ioflag & IO_DIRECT)
964 bp->b_flags |= B_DIRECT;
967 * We should only get non-zero b_resid when an I/O error
968 * has occurred, which should cause us to break above.
969 * However, if the short read did not cause an error,
970 * then we want to ensure that we do not uiomove bad
971 * or uninitialized data.
974 if (size < xfersize) {
980 error = uiomove((char *)bp->b_data + blkoffset,
985 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
986 (LIST_EMPTY(&bp->b_dep))) {
988 * If there are no dependencies, and it's VMIO,
989 * then we don't need the buf, mark it available
990 * for freeing. For non-direct VMIO reads, the VM
993 bp->b_flags |= B_RELBUF;
997 * Otherwise let whoever
998 * made the request take care of
999 * freeing it. We just queue
1000 * it onto another list.
1007 * This can only happen in the case of an error
1008 * because the loop above resets bp to NULL on each iteration
1009 * and on normal completion has not set a new value into it.
1010 * so it must have come from a 'break' statement
1013 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1014 (LIST_EMPTY(&bp->b_dep))) {
1015 bp->b_flags |= B_RELBUF;
1025 * Extended attribute area writing.
1028 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1031 struct ufs2_dinode *dp;
1037 int blkoffset, error, flags, size, xfersize;
1044 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1045 panic("ffs_extwrite: mode");
1048 if (ioflag & IO_APPEND)
1049 uio->uio_offset = dp->di_extsize;
1050 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1051 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1052 if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1055 resid = uio->uio_resid;
1056 osize = dp->di_extsize;
1058 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1061 for (error = 0; uio->uio_resid > 0;) {
1062 lbn = lblkno(fs, uio->uio_offset);
1063 blkoffset = blkoff(fs, uio->uio_offset);
1064 xfersize = fs->fs_bsize - blkoffset;
1065 if (uio->uio_resid < xfersize)
1066 xfersize = uio->uio_resid;
1069 * We must perform a read-before-write if the transfer size
1070 * does not cover the entire buffer.
1072 if (fs->fs_bsize > xfersize)
1075 flags &= ~BA_CLRBUF;
1076 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1081 * If the buffer is not valid we have to clear out any
1082 * garbage data from the pages instantiated for the buffer.
1083 * If we do not, a failed uiomove() during a write can leave
1084 * the prior contents of the pages exposed to a userland
1085 * mmap(). XXX deal with uiomove() errors a better way.
1087 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1089 if (ioflag & IO_DIRECT)
1090 bp->b_flags |= B_DIRECT;
1092 if (uio->uio_offset + xfersize > dp->di_extsize)
1093 dp->di_extsize = uio->uio_offset + xfersize;
1095 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1096 if (size < xfersize)
1100 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1101 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1102 (LIST_EMPTY(&bp->b_dep))) {
1103 bp->b_flags |= B_RELBUF;
1107 * If IO_SYNC each buffer is written synchronously. Otherwise
1108 * if we have a severe page deficiency write the buffer
1109 * asynchronously. Otherwise try to cluster, and if that
1110 * doesn't do it then either do an async write (if O_DIRECT),
1111 * or a delayed write (if not).
1113 if (ioflag & IO_SYNC) {
1115 } else if (vm_page_count_severe() ||
1116 buf_dirty_count_severe() ||
1117 xfersize + blkoffset == fs->fs_bsize ||
1118 (ioflag & (IO_ASYNC | IO_DIRECT)))
1122 if (error || xfersize == 0)
1124 ip->i_flag |= IN_CHANGE;
1127 * If we successfully wrote any data, and we are not the superuser
1128 * we clear the setuid and setgid bits as a precaution against
1131 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1132 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1133 ip->i_mode &= ~(ISUID | ISGID);
1134 dp->di_mode = ip->i_mode;
1138 if (ioflag & IO_UNIT) {
1139 (void)ffs_truncate(vp, osize,
1140 IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1141 uio->uio_offset -= resid - uio->uio_resid;
1142 uio->uio_resid = resid;
1144 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1145 error = ffs_update(vp, 1);
1151 * Vnode operating to retrieve a named extended attribute.
1153 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1154 * the length of the EA, and possibly the pointer to the entry and to the data.
1157 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1159 u_char *p, *pe, *pn, *p0;
1160 int eapad1, eapad2, ealength, ealen, nlen;
1164 nlen = strlen(name);
1166 for (p = ptr; p < pe; p = pn) {
1168 bcopy(p, &ul, sizeof(ul));
1170 /* make sure this entry is complete */
1173 p += sizeof(uint32_t);
1181 if (bcmp(p, name, nlen))
1183 ealength = sizeof(uint32_t) + 3 + nlen;
1184 eapad1 = 8 - (ealength % 8);
1188 ealen = ul - ealength - eapad2;
1200 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1203 struct ufs2_dinode *dp;
1206 struct iovec liovec;
1213 easize = dp->di_extsize;
1214 if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1217 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1219 liovec.iov_base = eae;
1220 liovec.iov_len = easize;
1221 luio.uio_iov = &liovec;
1222 luio.uio_iovcnt = 1;
1223 luio.uio_offset = 0;
1224 luio.uio_resid = easize;
1225 luio.uio_segflg = UIO_SYSSPACE;
1226 luio.uio_rw = UIO_READ;
1229 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1239 ffs_lock_ea(struct vnode *vp)
1245 while (ip->i_flag & IN_EA_LOCKED) {
1246 ip->i_flag |= IN_EA_LOCKWAIT;
1247 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1250 ip->i_flag |= IN_EA_LOCKED;
1255 ffs_unlock_ea(struct vnode *vp)
1261 if (ip->i_flag & IN_EA_LOCKWAIT)
1262 wakeup(&ip->i_ea_refs);
1263 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1268 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1271 struct ufs2_dinode *dp;
1277 if (ip->i_ea_area != NULL) {
1283 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1288 ip->i_ea_len = dp->di_extsize;
1296 * Vnode extattr transaction commit/abort
1299 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1303 struct iovec liovec;
1305 struct ufs2_dinode *dp;
1310 if (ip->i_ea_area == NULL) {
1315 error = ip->i_ea_error;
1316 if (commit && error == 0) {
1317 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1319 cred = vp->v_mount->mnt_cred;
1320 liovec.iov_base = ip->i_ea_area;
1321 liovec.iov_len = ip->i_ea_len;
1322 luio.uio_iov = &liovec;
1323 luio.uio_iovcnt = 1;
1324 luio.uio_offset = 0;
1325 luio.uio_resid = ip->i_ea_len;
1326 luio.uio_segflg = UIO_SYSSPACE;
1327 luio.uio_rw = UIO_WRITE;
1329 /* XXX: I'm not happy about truncating to zero size */
1330 if (ip->i_ea_len < dp->di_extsize)
1331 error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1332 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1334 if (--ip->i_ea_refs == 0) {
1335 free(ip->i_ea_area, M_TEMP);
1336 ip->i_ea_area = NULL;
1345 * Vnode extattr strategy routine for fifos.
1347 * We need to check for a read or write of the external attributes.
1348 * Otherwise we just fall through and do the usual thing.
1351 ffsext_strategy(struct vop_strategy_args *ap)
1353 struct vop_strategy_args {
1354 struct vnodeop_desc *a_desc;
1364 lbn = ap->a_bp->b_lblkno;
1365 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1366 lbn < 0 && lbn >= -NXADDR)
1367 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1368 if (vp->v_type == VFIFO)
1369 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1370 panic("spec nodes went here");
1374 * Vnode extattr transaction commit/abort
1377 ffs_openextattr(struct vop_openextattr_args *ap)
1379 struct vop_openextattr_args {
1380 struct vnodeop_desc *a_desc;
1382 IN struct ucred *a_cred;
1383 IN struct thread *a_td;
1390 ip = VTOI(ap->a_vp);
1393 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1394 return (EOPNOTSUPP);
1396 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1401 * Vnode extattr transaction commit/abort
1404 ffs_closeextattr(struct vop_closeextattr_args *ap)
1406 struct vop_closeextattr_args {
1407 struct vnodeop_desc *a_desc;
1410 IN struct ucred *a_cred;
1411 IN struct thread *a_td;
1418 ip = VTOI(ap->a_vp);
1421 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1422 return (EOPNOTSUPP);
1424 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1427 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1431 * Vnode operation to remove a named attribute.
1434 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1437 IN struct vnode *a_vp;
1438 IN int a_attrnamespace;
1439 IN const char *a_name;
1440 IN struct ucred *a_cred;
1441 IN struct thread *a_td;
1447 uint32_t ealength, ul;
1448 int ealen, olen, eapad1, eapad2, error, i, easize;
1451 ip = VTOI(ap->a_vp);
1454 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1455 return (EOPNOTSUPP);
1457 if (strlen(ap->a_name) == 0)
1460 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1463 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1464 ap->a_cred, ap->a_td, VWRITE);
1468 * ffs_lock_ea is not needed there, because the vnode
1469 * must be exclusively locked.
1471 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1472 ip->i_ea_error = error;
1476 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1480 ealength = eapad1 = ealen = eapad2 = 0;
1482 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1483 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1484 easize = ip->i_ea_len;
1486 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1489 /* delete but nonexistent */
1491 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1494 bcopy(p, &ul, sizeof ul);
1496 if (ul != ealength) {
1497 bcopy(p + ul, p + ealength, easize - i);
1498 easize += (ealength - ul);
1500 if (easize > NXADDR * fs->fs_bsize) {
1502 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1503 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1504 ip->i_ea_error = ENOSPC;
1508 ip->i_ea_area = eae;
1509 ip->i_ea_len = easize;
1511 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1516 * Vnode operation to retrieve a named extended attribute.
1519 ffs_getextattr(struct vop_getextattr_args *ap)
1522 IN struct vnode *a_vp;
1523 IN int a_attrnamespace;
1524 IN const char *a_name;
1525 INOUT struct uio *a_uio;
1527 IN struct ucred *a_cred;
1528 IN struct thread *a_td;
1538 ip = VTOI(ap->a_vp);
1541 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1542 return (EOPNOTSUPP);
1544 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1545 ap->a_cred, ap->a_td, VREAD);
1549 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1553 eae = ip->i_ea_area;
1554 easize = ip->i_ea_len;
1556 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1560 if (ap->a_size != NULL)
1561 *ap->a_size = ealen;
1562 else if (ap->a_uio != NULL)
1563 error = uiomove(p, ealen, ap->a_uio);
1567 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1572 * Vnode operation to retrieve extended attributes on a vnode.
1575 ffs_listextattr(struct vop_listextattr_args *ap)
1578 IN struct vnode *a_vp;
1579 IN int a_attrnamespace;
1580 INOUT struct uio *a_uio;
1582 IN struct ucred *a_cred;
1583 IN struct thread *a_td;
1589 u_char *eae, *p, *pe, *pn;
1594 ip = VTOI(ap->a_vp);
1597 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1598 return (EOPNOTSUPP);
1600 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1601 ap->a_cred, ap->a_td, VREAD);
1605 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1608 eae = ip->i_ea_area;
1609 easize = ip->i_ea_len;
1612 if (ap->a_size != NULL)
1615 for(p = eae; error == 0 && p < pe; p = pn) {
1616 bcopy(p, &ul, sizeof(ul));
1621 if (*p++ != ap->a_attrnamespace)
1625 if (ap->a_size != NULL) {
1626 *ap->a_size += ealen + 1;
1627 } else if (ap->a_uio != NULL) {
1628 error = uiomove(p, ealen + 1, ap->a_uio);
1631 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1636 * Vnode operation to set a named attribute.
1639 ffs_setextattr(struct vop_setextattr_args *ap)
1642 IN struct vnode *a_vp;
1643 IN int a_attrnamespace;
1644 IN const char *a_name;
1645 INOUT struct uio *a_uio;
1646 IN struct ucred *a_cred;
1647 IN struct thread *a_td;
1653 uint32_t ealength, ul;
1655 int olen, eapad1, eapad2, error, i, easize;
1658 ip = VTOI(ap->a_vp);
1661 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1662 return (EOPNOTSUPP);
1664 if (strlen(ap->a_name) == 0)
1667 /* XXX Now unsupported API to delete EAs using NULL uio. */
1668 if (ap->a_uio == NULL)
1669 return (EOPNOTSUPP);
1671 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1674 ealen = ap->a_uio->uio_resid;
1675 if (ealen < 0 || ealen > lblktosize(fs, NXADDR))
1678 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1679 ap->a_cred, ap->a_td, VWRITE);
1683 * ffs_lock_ea is not needed there, because the vnode
1684 * must be exclusively locked.
1686 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1687 ip->i_ea_error = error;
1691 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1695 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1696 eapad1 = 8 - (ealength % 8);
1699 eapad2 = 8 - (ealen % 8);
1702 ealength += eapad1 + ealen + eapad2;
1704 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1705 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1706 easize = ip->i_ea_len;
1708 olen = ffs_findextattr(eae, easize,
1709 ap->a_attrnamespace, ap->a_name, &p, NULL);
1711 /* new, append at end */
1715 bcopy(p, &ul, sizeof ul);
1717 if (ul != ealength) {
1718 bcopy(p + ul, p + ealength, easize - i);
1719 easize += (ealength - ul);
1722 if (easize > lblktosize(fs, NXADDR)) {
1724 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1725 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1726 ip->i_ea_error = ENOSPC;
1729 bcopy(&ealength, p, sizeof(ealength));
1730 p += sizeof(ealength);
1731 *p++ = ap->a_attrnamespace;
1733 *p++ = strlen(ap->a_name);
1734 strcpy(p, ap->a_name);
1735 p += strlen(ap->a_name);
1738 error = uiomove(p, ealen, ap->a_uio);
1741 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1742 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1743 ip->i_ea_error = error;
1750 ip->i_ea_area = eae;
1751 ip->i_ea_len = easize;
1753 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1758 * Vnode pointer to File handle
1761 ffs_vptofh(struct vop_vptofh_args *ap)
1764 IN struct vnode *a_vp;
1765 IN struct fid *a_fhp;
1772 ip = VTOI(ap->a_vp);
1773 ufhp = (struct ufid *)ap->a_fhp;
1774 ufhp->ufid_len = sizeof(struct ufid);
1775 ufhp->ufid_ino = ip->i_number;
1776 ufhp->ufid_gen = ip->i_gen;