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/resourcevar.h>
80 #include <sys/signalvar.h>
82 #include <sys/vmmeter.h>
83 #include <sys/vnode.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vnode_pager.h>
92 #include <ufs/ufs/extattr.h>
93 #include <ufs/ufs/quota.h>
94 #include <ufs/ufs/inode.h>
95 #include <ufs/ufs/ufs_extern.h>
96 #include <ufs/ufs/ufsmount.h>
98 #include <ufs/ffs/fs.h>
99 #include <ufs/ffs/ffs_extern.h>
100 #include "opt_directio.h"
104 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
106 static vop_fsync_t ffs_fsync;
107 static vop_lock1_t ffs_lock;
108 static vop_getpages_t ffs_getpages;
109 static vop_read_t ffs_read;
110 static vop_write_t ffs_write;
111 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
112 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
114 static vop_strategy_t ffsext_strategy;
115 static vop_closeextattr_t ffs_closeextattr;
116 static vop_deleteextattr_t ffs_deleteextattr;
117 static vop_getextattr_t ffs_getextattr;
118 static vop_listextattr_t ffs_listextattr;
119 static vop_openextattr_t ffs_openextattr;
120 static vop_setextattr_t ffs_setextattr;
121 static vop_vptofh_t ffs_vptofh;
124 /* Global vfs data structures for ufs. */
125 struct vop_vector ffs_vnodeops1 = {
126 .vop_default = &ufs_vnodeops,
127 .vop_fsync = ffs_fsync,
128 .vop_getpages = ffs_getpages,
129 .vop_lock1 = ffs_lock,
130 .vop_read = ffs_read,
131 .vop_reallocblks = ffs_reallocblks,
132 .vop_write = ffs_write,
133 .vop_vptofh = ffs_vptofh,
136 struct vop_vector ffs_fifoops1 = {
137 .vop_default = &ufs_fifoops,
138 .vop_fsync = ffs_fsync,
139 .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */
140 .vop_vptofh = ffs_vptofh,
143 /* Global vfs data structures for ufs. */
144 struct vop_vector ffs_vnodeops2 = {
145 .vop_default = &ufs_vnodeops,
146 .vop_fsync = ffs_fsync,
147 .vop_getpages = ffs_getpages,
148 .vop_lock1 = ffs_lock,
149 .vop_read = ffs_read,
150 .vop_reallocblks = ffs_reallocblks,
151 .vop_write = ffs_write,
152 .vop_closeextattr = ffs_closeextattr,
153 .vop_deleteextattr = ffs_deleteextattr,
154 .vop_getextattr = ffs_getextattr,
155 .vop_listextattr = ffs_listextattr,
156 .vop_openextattr = ffs_openextattr,
157 .vop_setextattr = ffs_setextattr,
158 .vop_vptofh = ffs_vptofh,
161 struct vop_vector ffs_fifoops2 = {
162 .vop_default = &ufs_fifoops,
163 .vop_fsync = ffs_fsync,
164 .vop_lock1 = ffs_lock,
165 .vop_reallocblks = ffs_reallocblks,
166 .vop_strategy = ffsext_strategy,
167 .vop_closeextattr = ffs_closeextattr,
168 .vop_deleteextattr = ffs_deleteextattr,
169 .vop_getextattr = ffs_getextattr,
170 .vop_listextattr = ffs_listextattr,
171 .vop_openextattr = ffs_openextattr,
172 .vop_setextattr = ffs_setextattr,
173 .vop_vptofh = ffs_vptofh,
177 * Synch an open file.
181 ffs_fsync(struct vop_fsync_args *ap)
190 error = ffs_syncvnode(vp, ap->a_waitfor);
193 if (ap->a_waitfor == MNT_WAIT &&
194 (vp->v_mount->mnt_flag & MNT_SOFTDEP)) {
195 error = softdep_fsync(vp);
200 * The softdep_fsync() function may drop vp lock,
201 * allowing for dirty buffers to reappear on the
202 * bo_dirty list. Recheck and resync as needed.
205 if (vp->v_type == VREG && (bo->bo_numoutput > 0 ||
206 bo->bo_dirty.bv_cnt > 0)) {
216 ffs_syncvnode(struct vnode *vp, int waitfor)
218 struct inode *ip = VTOI(vp);
222 int s, error, wait, passes, skipmeta;
225 wait = (waitfor == MNT_WAIT);
226 lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
230 * Flush all dirty buffers associated with a vnode.
239 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
240 bp->b_vflags &= ~BV_SCANNED;
241 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
243 * Reasons to skip this buffer: it has already been considered
244 * on this pass, this pass is the first time through on a
245 * synchronous flush request and the buffer being considered
246 * is metadata, the buffer has dependencies that will cause
247 * it to be redirtied and it has not already been deferred,
248 * or it is already being written.
250 if ((bp->b_vflags & BV_SCANNED) != 0)
252 bp->b_vflags |= BV_SCANNED;
253 if ((skipmeta == 1 && bp->b_lblkno < 0))
255 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
258 if (!wait && !LIST_EMPTY(&bp->b_dep) &&
259 (bp->b_flags & B_DEFERRED) == 0 &&
260 buf_countdeps(bp, 0)) {
261 bp->b_flags |= B_DEFERRED;
266 if ((bp->b_flags & B_DELWRI) == 0)
267 panic("ffs_fsync: not dirty");
269 * If this is a synchronous flush request, or it is not a
270 * file or device, start the write on this buffer immediately.
272 if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
275 * On our final pass through, do all I/O synchronously
276 * so that we can find out if our flush is failing
277 * because of write errors.
279 if (passes > 0 || !wait) {
280 if ((bp->b_flags & B_CLUSTEROK) && !wait) {
281 (void) vfs_bio_awrite(bp);
291 if ((error = bwrite(bp)) != 0)
295 } else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
297 * If the buffer is for data that has been truncated
298 * off the file, then throw it away.
301 bp->b_flags |= B_INVAL | B_NOCACHE;
309 * Since we may have slept during the I/O, we need
310 * to start from a known point.
313 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
316 * If we were asked to do this synchronously, then go back for
317 * another pass, this time doing the metadata.
325 bufobj_wwait(bo, 3, 0);
329 * Ensure that any filesystem metatdata associated
330 * with the vnode has been written.
333 if ((error = softdep_sync_metadata(vp)) != 0)
338 if (bo->bo_dirty.bv_cnt > 0) {
340 * Block devices associated with filesystems may
341 * have new I/O requests posted for them even if
342 * the vnode is locked, so no amount of trying will
343 * get them clean. Thus we give block devices a
344 * good effort, then just give up. For all other file
345 * types, go around and try again until it is clean.
352 if (!vn_isdisk(vp, NULL))
353 vprint("ffs_fsync: dirty", vp);
359 return (ffs_update(vp, wait));
364 struct vop_lock1_args /* {
372 #ifndef NO_FFS_SNAPSHOT
378 switch (ap->a_flags & LK_TYPE_MASK) {
385 #ifdef DEBUG_VFS_LOCKS
386 KASSERT(vp->v_holdcnt != 0,
387 ("ffs_lock %p: zero hold count", vp));
390 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
391 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
392 ap->a_file, ap->a_line);
393 if (lkp == vp->v_vnlock || result != 0)
396 * Apparent success, except that the vnode
397 * mutated between snapshot file vnode and
398 * regular file vnode while this process
399 * slept. The lock currently held is not the
400 * right lock. Release it, and try to get the
403 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
404 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
405 ap->a_file, ap->a_line);
406 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
407 (LK_INTERLOCK | LK_NOWAIT))
409 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
410 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
411 flags &= ~LK_INTERLOCK;
415 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
419 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
424 * Vnode op for reading.
429 struct vop_read_args /* {
433 struct ucred *a_cred;
441 ufs_lbn_t lbn, nextlbn;
443 long size, xfersize, blkoffset;
444 int error, orig_resid;
450 ioflag = ap->a_ioflag;
451 if (ap->a_ioflag & IO_EXT)
453 return (ffs_extread(vp, uio, ioflag));
455 panic("ffs_read+IO_EXT");
458 if ((ioflag & IO_DIRECT) != 0) {
461 error = ffs_rawread(vp, uio, &workdone);
462 if (error != 0 || workdone != 0)
467 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
471 if (uio->uio_rw != UIO_READ)
472 panic("ffs_read: mode");
474 if (vp->v_type == VLNK) {
475 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
476 panic("ffs_read: short symlink");
477 } else if (vp->v_type != VREG && vp->v_type != VDIR)
478 panic("ffs_read: type %d", vp->v_type);
480 orig_resid = uio->uio_resid;
481 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
484 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
486 if (uio->uio_offset < ip->i_size &&
487 uio->uio_offset >= fs->fs_maxfilesize)
490 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
491 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
493 lbn = lblkno(fs, uio->uio_offset);
497 * size of buffer. The buffer representing the
498 * end of the file is rounded up to the size of
499 * the block type ( fragment or full block,
502 size = blksize(fs, ip, lbn);
503 blkoffset = blkoff(fs, uio->uio_offset);
506 * The amount we want to transfer in this iteration is
507 * one FS block less the amount of the data before
508 * our startpoint (duh!)
510 xfersize = fs->fs_bsize - blkoffset;
513 * But if we actually want less than the block,
514 * or the file doesn't have a whole block more of data,
515 * then use the lesser number.
517 if (uio->uio_resid < xfersize)
518 xfersize = uio->uio_resid;
519 if (bytesinfile < xfersize)
520 xfersize = bytesinfile;
522 if (lblktosize(fs, nextlbn) >= ip->i_size) {
524 * Don't do readahead if this is the end of the file.
526 error = bread(vp, lbn, size, NOCRED, &bp);
527 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
529 * Otherwise if we are allowed to cluster,
530 * grab as much as we can.
532 * XXX This may not be a win if we are not
533 * doing sequential access.
535 error = cluster_read(vp, ip->i_size, lbn,
536 size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
537 } else if (seqcount > 1) {
539 * If we are NOT allowed to cluster, then
540 * if we appear to be acting sequentially,
541 * fire off a request for a readahead
542 * as well as a read. Note that the 4th and 5th
543 * arguments point to arrays of the size specified in
546 int nextsize = blksize(fs, ip, nextlbn);
547 error = breadn(vp, lbn,
548 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
551 * Failing all of the above, just read what the
552 * user asked for. Interestingly, the same as
553 * the first option above.
555 error = bread(vp, lbn, size, NOCRED, &bp);
564 * If IO_DIRECT then set B_DIRECT for the buffer. This
565 * will cause us to attempt to release the buffer later on
566 * and will cause the buffer cache to attempt to free the
569 if (ioflag & IO_DIRECT)
570 bp->b_flags |= B_DIRECT;
573 * We should only get non-zero b_resid when an I/O error
574 * has occurred, which should cause us to break above.
575 * However, if the short read did not cause an error,
576 * then we want to ensure that we do not uiomove bad
577 * or uninitialized data.
580 if (size < xfersize) {
586 error = uiomove((char *)bp->b_data + blkoffset,
591 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
592 (LIST_EMPTY(&bp->b_dep))) {
594 * If there are no dependencies, and it's VMIO,
595 * then we don't need the buf, mark it available
596 * for freeing. The VM has the data.
598 bp->b_flags |= B_RELBUF;
602 * Otherwise let whoever
603 * made the request take care of
604 * freeing it. We just queue
605 * it onto another list.
612 * This can only happen in the case of an error
613 * because the loop above resets bp to NULL on each iteration
614 * and on normal completion has not set a new value into it.
615 * so it must have come from a 'break' statement
618 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
619 (LIST_EMPTY(&bp->b_dep))) {
620 bp->b_flags |= B_RELBUF;
627 if ((error == 0 || uio->uio_resid != orig_resid) &&
628 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
629 (ip->i_flag & IN_ACCESS) == 0) {
631 ip->i_flag |= IN_ACCESS;
638 * Vnode op for writing.
642 struct vop_write_args /* {
646 struct ucred *a_cred;
658 int blkoffset, error, flags, ioflag, resid, size, xfersize;
662 ioflag = ap->a_ioflag;
663 if (ap->a_ioflag & IO_EXT)
665 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
667 panic("ffs_write+IO_EXT");
670 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
674 if (uio->uio_rw != UIO_WRITE)
675 panic("ffs_write: mode");
678 switch (vp->v_type) {
680 if (ioflag & IO_APPEND)
681 uio->uio_offset = ip->i_size;
682 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
688 panic("ffs_write: dir write");
691 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
692 (int)uio->uio_offset,
697 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
698 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
700 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
703 * Maybe this should be above the vnode op call, but so long as
704 * file servers have no limits, I don't think it matters.
707 if (vp->v_type == VREG && td != NULL) {
708 PROC_LOCK(td->td_proc);
709 if (uio->uio_offset + uio->uio_resid >
710 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
711 psignal(td->td_proc, SIGXFSZ);
712 PROC_UNLOCK(td->td_proc);
715 PROC_UNLOCK(td->td_proc);
718 resid = uio->uio_resid;
720 if (seqcount > BA_SEQMAX)
721 flags = BA_SEQMAX << BA_SEQSHIFT;
723 flags = seqcount << BA_SEQSHIFT;
724 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
727 for (error = 0; uio->uio_resid > 0;) {
728 lbn = lblkno(fs, uio->uio_offset);
729 blkoffset = blkoff(fs, uio->uio_offset);
730 xfersize = fs->fs_bsize - blkoffset;
731 if (uio->uio_resid < xfersize)
732 xfersize = uio->uio_resid;
733 if (uio->uio_offset + xfersize > ip->i_size)
734 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
737 * We must perform a read-before-write if the transfer size
738 * does not cover the entire buffer.
740 if (fs->fs_bsize > xfersize)
744 /* XXX is uio->uio_offset the right thing here? */
745 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
746 ap->a_cred, flags, &bp);
748 vnode_pager_setsize(vp, ip->i_size);
752 * If the buffer is not valid we have to clear out any
753 * garbage data from the pages instantiated for the buffer.
754 * If we do not, a failed uiomove() during a write can leave
755 * the prior contents of the pages exposed to a userland
756 * mmap(). XXX deal with uiomove() errors a better way.
758 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
760 if (ioflag & IO_DIRECT)
761 bp->b_flags |= B_DIRECT;
762 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
763 bp->b_flags |= B_NOCACHE;
765 if (uio->uio_offset + xfersize > ip->i_size) {
766 ip->i_size = uio->uio_offset + xfersize;
767 DIP_SET(ip, i_size, ip->i_size);
770 size = blksize(fs, ip, lbn) - bp->b_resid;
775 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
776 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
777 (LIST_EMPTY(&bp->b_dep))) {
778 bp->b_flags |= B_RELBUF;
782 * If IO_SYNC each buffer is written synchronously. Otherwise
783 * if we have a severe page deficiency write the buffer
784 * asynchronously. Otherwise try to cluster, and if that
785 * doesn't do it then either do an async write (if O_DIRECT),
786 * or a delayed write (if not).
788 if (ioflag & IO_SYNC) {
790 } else if (vm_page_count_severe() ||
791 buf_dirty_count_severe() ||
792 (ioflag & IO_ASYNC)) {
793 bp->b_flags |= B_CLUSTEROK;
795 } else if (xfersize + blkoffset == fs->fs_bsize) {
796 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
797 bp->b_flags |= B_CLUSTEROK;
798 cluster_write(vp, bp, ip->i_size, seqcount);
802 } else if (ioflag & IO_DIRECT) {
803 bp->b_flags |= B_CLUSTEROK;
806 bp->b_flags |= B_CLUSTEROK;
809 if (error || xfersize == 0)
811 ip->i_flag |= IN_CHANGE | IN_UPDATE;
814 * If we successfully wrote any data, and we are not the superuser
815 * we clear the setuid and setgid bits as a precaution against
818 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
820 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
821 ip->i_mode &= ~(ISUID | ISGID);
822 DIP_SET(ip, i_mode, ip->i_mode);
826 if (ioflag & IO_UNIT) {
827 (void)ffs_truncate(vp, osize,
828 IO_NORMAL | (ioflag & IO_SYNC),
829 ap->a_cred, uio->uio_td);
830 uio->uio_offset -= resid - uio->uio_resid;
831 uio->uio_resid = resid;
833 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
834 error = ffs_update(vp, 1);
843 struct vop_getpages_args *ap;
849 pcount = round_page(ap->a_count) / PAGE_SIZE;
850 mreq = ap->a_m[ap->a_reqpage];
853 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
854 * then the entire page is valid. Since the page may be mapped,
855 * user programs might reference data beyond the actual end of file
856 * occuring within the page. We have to zero that data.
858 VM_OBJECT_LOCK(mreq->object);
860 if (mreq->valid != VM_PAGE_BITS_ALL)
861 vm_page_zero_invalid(mreq, TRUE);
862 vm_page_lock_queues();
863 for (i = 0; i < pcount; i++) {
864 if (i != ap->a_reqpage) {
865 vm_page_free(ap->a_m[i]);
868 vm_page_unlock_queues();
869 VM_OBJECT_UNLOCK(mreq->object);
872 VM_OBJECT_UNLOCK(mreq->object);
874 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
881 * Extended attribute area reading.
884 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
887 struct ufs2_dinode *dp;
890 ufs_lbn_t lbn, nextlbn;
892 long size, xfersize, blkoffset;
893 int error, orig_resid;
900 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
901 panic("ffs_extread: mode");
904 orig_resid = uio->uio_resid;
905 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
908 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
910 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
911 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
913 lbn = lblkno(fs, uio->uio_offset);
917 * size of buffer. The buffer representing the
918 * end of the file is rounded up to the size of
919 * the block type ( fragment or full block,
922 size = sblksize(fs, dp->di_extsize, lbn);
923 blkoffset = blkoff(fs, uio->uio_offset);
926 * The amount we want to transfer in this iteration is
927 * one FS block less the amount of the data before
928 * our startpoint (duh!)
930 xfersize = fs->fs_bsize - blkoffset;
933 * But if we actually want less than the block,
934 * or the file doesn't have a whole block more of data,
935 * then use the lesser number.
937 if (uio->uio_resid < xfersize)
938 xfersize = uio->uio_resid;
939 if (bytesinfile < xfersize)
940 xfersize = bytesinfile;
942 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
944 * Don't do readahead if this is the end of the info.
946 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
949 * If we have a second block, then
950 * fire off a request for a readahead
951 * as well as a read. Note that the 4th and 5th
952 * arguments point to arrays of the size specified in
955 int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
957 nextlbn = -1 - nextlbn;
958 error = breadn(vp, -1 - lbn,
959 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
968 * If IO_DIRECT then set B_DIRECT for the buffer. This
969 * will cause us to attempt to release the buffer later on
970 * and will cause the buffer cache to attempt to free the
973 if (ioflag & IO_DIRECT)
974 bp->b_flags |= B_DIRECT;
977 * We should only get non-zero b_resid when an I/O error
978 * has occurred, which should cause us to break above.
979 * However, if the short read did not cause an error,
980 * then we want to ensure that we do not uiomove bad
981 * or uninitialized data.
984 if (size < xfersize) {
990 error = uiomove((char *)bp->b_data + blkoffset,
995 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
996 (LIST_EMPTY(&bp->b_dep))) {
998 * If there are no dependencies, and it's VMIO,
999 * then we don't need the buf, mark it available
1000 * for freeing. The VM has the data.
1002 bp->b_flags |= B_RELBUF;
1006 * Otherwise let whoever
1007 * made the request take care of
1008 * freeing it. We just queue
1009 * it onto another list.
1016 * This can only happen in the case of an error
1017 * because the loop above resets bp to NULL on each iteration
1018 * and on normal completion has not set a new value into it.
1019 * so it must have come from a 'break' statement
1022 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1023 (LIST_EMPTY(&bp->b_dep))) {
1024 bp->b_flags |= B_RELBUF;
1034 * Extended attribute area writing.
1037 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1040 struct ufs2_dinode *dp;
1045 int blkoffset, error, flags, resid, size, xfersize;
1051 KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1055 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1056 panic("ffs_extwrite: mode");
1059 if (ioflag & IO_APPEND)
1060 uio->uio_offset = dp->di_extsize;
1061 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1062 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1063 if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1066 resid = uio->uio_resid;
1067 osize = dp->di_extsize;
1069 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1072 for (error = 0; uio->uio_resid > 0;) {
1073 lbn = lblkno(fs, uio->uio_offset);
1074 blkoffset = blkoff(fs, uio->uio_offset);
1075 xfersize = fs->fs_bsize - blkoffset;
1076 if (uio->uio_resid < xfersize)
1077 xfersize = uio->uio_resid;
1080 * We must perform a read-before-write if the transfer size
1081 * does not cover the entire buffer.
1083 if (fs->fs_bsize > xfersize)
1086 flags &= ~BA_CLRBUF;
1087 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1092 * If the buffer is not valid we have to clear out any
1093 * garbage data from the pages instantiated for the buffer.
1094 * If we do not, a failed uiomove() during a write can leave
1095 * the prior contents of the pages exposed to a userland
1096 * mmap(). XXX deal with uiomove() errors a better way.
1098 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1100 if (ioflag & IO_DIRECT)
1101 bp->b_flags |= B_DIRECT;
1103 if (uio->uio_offset + xfersize > dp->di_extsize)
1104 dp->di_extsize = uio->uio_offset + xfersize;
1106 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1107 if (size < xfersize)
1111 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1112 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1113 (LIST_EMPTY(&bp->b_dep))) {
1114 bp->b_flags |= B_RELBUF;
1118 * If IO_SYNC each buffer is written synchronously. Otherwise
1119 * if we have a severe page deficiency write the buffer
1120 * asynchronously. Otherwise try to cluster, and if that
1121 * doesn't do it then either do an async write (if O_DIRECT),
1122 * or a delayed write (if not).
1124 if (ioflag & IO_SYNC) {
1126 } else if (vm_page_count_severe() ||
1127 buf_dirty_count_severe() ||
1128 xfersize + blkoffset == fs->fs_bsize ||
1129 (ioflag & (IO_ASYNC | IO_DIRECT)))
1133 if (error || xfersize == 0)
1135 ip->i_flag |= IN_CHANGE;
1138 * If we successfully wrote any data, and we are not the superuser
1139 * we clear the setuid and setgid bits as a precaution against
1142 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1143 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1144 ip->i_mode &= ~(ISUID | ISGID);
1145 dp->di_mode = ip->i_mode;
1149 if (ioflag & IO_UNIT) {
1150 (void)ffs_truncate(vp, osize,
1151 IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1152 uio->uio_offset -= resid - uio->uio_resid;
1153 uio->uio_resid = resid;
1155 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1156 error = ffs_update(vp, 1);
1162 * Vnode operating to retrieve a named extended attribute.
1164 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1165 * the length of the EA, and possibly the pointer to the entry and to the data.
1168 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1170 u_char *p, *pe, *pn, *p0;
1171 int eapad1, eapad2, ealength, ealen, nlen;
1175 nlen = strlen(name);
1177 for (p = ptr; p < pe; p = pn) {
1179 bcopy(p, &ul, sizeof(ul));
1181 /* make sure this entry is complete */
1184 p += sizeof(uint32_t);
1192 if (bcmp(p, name, nlen))
1194 ealength = sizeof(uint32_t) + 3 + nlen;
1195 eapad1 = 8 - (ealength % 8);
1199 ealen = ul - ealength - eapad2;
1211 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1214 struct ufs2_dinode *dp;
1217 struct iovec liovec;
1224 easize = dp->di_extsize;
1225 if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1228 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1230 liovec.iov_base = eae;
1231 liovec.iov_len = easize;
1232 luio.uio_iov = &liovec;
1233 luio.uio_iovcnt = 1;
1234 luio.uio_offset = 0;
1235 luio.uio_resid = easize;
1236 luio.uio_segflg = UIO_SYSSPACE;
1237 luio.uio_rw = UIO_READ;
1240 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1250 ffs_lock_ea(struct vnode *vp)
1256 while (ip->i_flag & IN_EA_LOCKED) {
1257 ip->i_flag |= IN_EA_LOCKWAIT;
1258 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1261 ip->i_flag |= IN_EA_LOCKED;
1266 ffs_unlock_ea(struct vnode *vp)
1272 if (ip->i_flag & IN_EA_LOCKWAIT)
1273 wakeup(&ip->i_ea_refs);
1274 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1279 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1282 struct ufs2_dinode *dp;
1288 if (ip->i_ea_area != NULL) {
1294 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1299 ip->i_ea_len = dp->di_extsize;
1307 * Vnode extattr transaction commit/abort
1310 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1314 struct iovec liovec;
1316 struct ufs2_dinode *dp;
1321 if (ip->i_ea_area == NULL) {
1326 error = ip->i_ea_error;
1327 if (commit && error == 0) {
1328 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1330 cred = vp->v_mount->mnt_cred;
1331 liovec.iov_base = ip->i_ea_area;
1332 liovec.iov_len = ip->i_ea_len;
1333 luio.uio_iov = &liovec;
1334 luio.uio_iovcnt = 1;
1335 luio.uio_offset = 0;
1336 luio.uio_resid = ip->i_ea_len;
1337 luio.uio_segflg = UIO_SYSSPACE;
1338 luio.uio_rw = UIO_WRITE;
1340 /* XXX: I'm not happy about truncating to zero size */
1341 if (ip->i_ea_len < dp->di_extsize)
1342 error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1343 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1345 if (--ip->i_ea_refs == 0) {
1346 free(ip->i_ea_area, M_TEMP);
1347 ip->i_ea_area = NULL;
1356 * Vnode extattr strategy routine for fifos.
1358 * We need to check for a read or write of the external attributes.
1359 * Otherwise we just fall through and do the usual thing.
1362 ffsext_strategy(struct vop_strategy_args *ap)
1364 struct vop_strategy_args {
1365 struct vnodeop_desc *a_desc;
1375 lbn = ap->a_bp->b_lblkno;
1376 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1377 lbn < 0 && lbn >= -NXADDR)
1378 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1379 if (vp->v_type == VFIFO)
1380 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1381 panic("spec nodes went here");
1385 * Vnode extattr transaction commit/abort
1388 ffs_openextattr(struct vop_openextattr_args *ap)
1390 struct vop_openextattr_args {
1391 struct vnodeop_desc *a_desc;
1393 IN struct ucred *a_cred;
1394 IN struct thread *a_td;
1401 ip = VTOI(ap->a_vp);
1404 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1405 return (EOPNOTSUPP);
1407 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1412 * Vnode extattr transaction commit/abort
1415 ffs_closeextattr(struct vop_closeextattr_args *ap)
1417 struct vop_closeextattr_args {
1418 struct vnodeop_desc *a_desc;
1421 IN struct ucred *a_cred;
1422 IN struct thread *a_td;
1429 ip = VTOI(ap->a_vp);
1432 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1433 return (EOPNOTSUPP);
1435 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1438 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1442 * Vnode operation to remove a named attribute.
1445 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1448 IN struct vnode *a_vp;
1449 IN int a_attrnamespace;
1450 IN const char *a_name;
1451 IN struct ucred *a_cred;
1452 IN struct thread *a_td;
1458 uint32_t ealength, ul;
1459 int ealen, olen, eapad1, eapad2, error, i, easize;
1462 ip = VTOI(ap->a_vp);
1465 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1466 return (EOPNOTSUPP);
1468 if (strlen(ap->a_name) == 0)
1471 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1474 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1475 ap->a_cred, ap->a_td, VWRITE);
1479 * ffs_lock_ea is not needed there, because the vnode
1480 * must be exclusively locked.
1482 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1483 ip->i_ea_error = error;
1487 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1491 ealength = eapad1 = ealen = eapad2 = 0;
1493 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1494 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1495 easize = ip->i_ea_len;
1497 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1500 /* delete but nonexistent */
1502 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1505 bcopy(p, &ul, sizeof ul);
1507 if (ul != ealength) {
1508 bcopy(p + ul, p + ealength, easize - i);
1509 easize += (ealength - ul);
1511 if (easize > NXADDR * fs->fs_bsize) {
1513 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1514 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1515 ip->i_ea_error = ENOSPC;
1519 ip->i_ea_area = eae;
1520 ip->i_ea_len = easize;
1522 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1527 * Vnode operation to retrieve a named extended attribute.
1530 ffs_getextattr(struct vop_getextattr_args *ap)
1533 IN struct vnode *a_vp;
1534 IN int a_attrnamespace;
1535 IN const char *a_name;
1536 INOUT struct uio *a_uio;
1538 IN struct ucred *a_cred;
1539 IN struct thread *a_td;
1549 ip = VTOI(ap->a_vp);
1552 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1553 return (EOPNOTSUPP);
1555 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1556 ap->a_cred, ap->a_td, VREAD);
1560 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1564 eae = ip->i_ea_area;
1565 easize = ip->i_ea_len;
1567 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1571 if (ap->a_size != NULL)
1572 *ap->a_size = ealen;
1573 else if (ap->a_uio != NULL)
1574 error = uiomove(p, ealen, ap->a_uio);
1578 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1583 * Vnode operation to retrieve extended attributes on a vnode.
1586 ffs_listextattr(struct vop_listextattr_args *ap)
1589 IN struct vnode *a_vp;
1590 IN int a_attrnamespace;
1591 INOUT struct uio *a_uio;
1593 IN struct ucred *a_cred;
1594 IN struct thread *a_td;
1600 u_char *eae, *p, *pe, *pn;
1605 ip = VTOI(ap->a_vp);
1608 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1609 return (EOPNOTSUPP);
1611 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1612 ap->a_cred, ap->a_td, VREAD);
1616 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1619 eae = ip->i_ea_area;
1620 easize = ip->i_ea_len;
1623 if (ap->a_size != NULL)
1626 for(p = eae; error == 0 && p < pe; p = pn) {
1627 bcopy(p, &ul, sizeof(ul));
1632 if (*p++ != ap->a_attrnamespace)
1636 if (ap->a_size != NULL) {
1637 *ap->a_size += ealen + 1;
1638 } else if (ap->a_uio != NULL) {
1639 error = uiomove(p, ealen + 1, ap->a_uio);
1642 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1647 * Vnode operation to set a named attribute.
1650 ffs_setextattr(struct vop_setextattr_args *ap)
1653 IN struct vnode *a_vp;
1654 IN int a_attrnamespace;
1655 IN const char *a_name;
1656 INOUT struct uio *a_uio;
1657 IN struct ucred *a_cred;
1658 IN struct thread *a_td;
1664 uint32_t ealength, ul;
1665 int ealen, olen, eapad1, eapad2, error, i, easize;
1668 ip = VTOI(ap->a_vp);
1671 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1672 return (EOPNOTSUPP);
1674 if (strlen(ap->a_name) == 0)
1677 /* XXX Now unsupported API to delete EAs using NULL uio. */
1678 if (ap->a_uio == NULL)
1679 return (EOPNOTSUPP);
1681 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1684 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1685 ap->a_cred, ap->a_td, VWRITE);
1689 * ffs_lock_ea is not needed there, because the vnode
1690 * must be exclusively locked.
1692 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1693 ip->i_ea_error = error;
1697 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1701 ealen = ap->a_uio->uio_resid;
1702 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1703 eapad1 = 8 - (ealength % 8);
1706 eapad2 = 8 - (ealen % 8);
1709 ealength += eapad1 + ealen + eapad2;
1711 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1712 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1713 easize = ip->i_ea_len;
1715 olen = ffs_findextattr(eae, easize,
1716 ap->a_attrnamespace, ap->a_name, &p, NULL);
1718 /* new, append at end */
1722 bcopy(p, &ul, sizeof ul);
1724 if (ul != ealength) {
1725 bcopy(p + ul, p + ealength, easize - i);
1726 easize += (ealength - ul);
1729 if (easize > NXADDR * fs->fs_bsize) {
1731 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1732 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1733 ip->i_ea_error = ENOSPC;
1736 bcopy(&ealength, p, sizeof(ealength));
1737 p += sizeof(ealength);
1738 *p++ = ap->a_attrnamespace;
1740 *p++ = strlen(ap->a_name);
1741 strcpy(p, ap->a_name);
1742 p += strlen(ap->a_name);
1745 error = uiomove(p, ealen, ap->a_uio);
1748 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1749 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1750 ip->i_ea_error = error;
1757 ip->i_ea_area = eae;
1758 ip->i_ea_len = easize;
1760 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1765 * Vnode pointer to File handle
1768 ffs_vptofh(struct vop_vptofh_args *ap)
1771 IN struct vnode *a_vp;
1772 IN struct fid *a_fhp;
1779 ip = VTOI(ap->a_vp);
1780 ufhp = (struct ufid *)ap->a_fhp;
1781 ufhp->ufid_len = sizeof(struct ufid);
1782 ufhp->ufid_ino = ip->i_number;
1783 ufhp->ufid_gen = ip->i_gen;