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_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vnode_pager.h>
89 #include <ufs/ufs/extattr.h>
90 #include <ufs/ufs/quota.h>
91 #include <ufs/ufs/inode.h>
92 #include <ufs/ufs/ufs_extern.h>
93 #include <ufs/ufs/ufsmount.h>
95 #include <ufs/ffs/fs.h>
96 #include <ufs/ffs/ffs_extern.h>
97 #include "opt_directio.h"
101 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
103 static vop_fsync_t ffs_fsync;
104 static vop_lock1_t ffs_lock;
105 static vop_getpages_t ffs_getpages;
106 static vop_read_t ffs_read;
107 static vop_write_t ffs_write;
108 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
109 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
111 static vop_strategy_t ffsext_strategy;
112 static vop_closeextattr_t ffs_closeextattr;
113 static vop_deleteextattr_t ffs_deleteextattr;
114 static vop_getextattr_t ffs_getextattr;
115 static vop_listextattr_t ffs_listextattr;
116 static vop_openextattr_t ffs_openextattr;
117 static vop_setextattr_t ffs_setextattr;
118 static vop_vptofh_t ffs_vptofh;
121 /* Global vfs data structures for ufs. */
122 struct vop_vector ffs_vnodeops1 = {
123 .vop_default = &ufs_vnodeops,
124 .vop_fsync = ffs_fsync,
125 .vop_getpages = ffs_getpages,
126 .vop_lock1 = ffs_lock,
127 .vop_read = ffs_read,
128 .vop_reallocblks = ffs_reallocblks,
129 .vop_write = ffs_write,
130 .vop_vptofh = ffs_vptofh,
133 struct vop_vector ffs_fifoops1 = {
134 .vop_default = &ufs_fifoops,
135 .vop_fsync = ffs_fsync,
136 .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */
137 .vop_vptofh = ffs_vptofh,
140 /* Global vfs data structures for ufs. */
141 struct vop_vector ffs_vnodeops2 = {
142 .vop_default = &ufs_vnodeops,
143 .vop_fsync = ffs_fsync,
144 .vop_getpages = ffs_getpages,
145 .vop_lock1 = ffs_lock,
146 .vop_read = ffs_read,
147 .vop_reallocblks = ffs_reallocblks,
148 .vop_write = ffs_write,
149 .vop_closeextattr = ffs_closeextattr,
150 .vop_deleteextattr = ffs_deleteextattr,
151 .vop_getextattr = ffs_getextattr,
152 .vop_listextattr = ffs_listextattr,
153 .vop_openextattr = ffs_openextattr,
154 .vop_setextattr = ffs_setextattr,
155 .vop_vptofh = ffs_vptofh,
158 struct vop_vector ffs_fifoops2 = {
159 .vop_default = &ufs_fifoops,
160 .vop_fsync = ffs_fsync,
161 .vop_lock1 = ffs_lock,
162 .vop_reallocblks = ffs_reallocblks,
163 .vop_strategy = ffsext_strategy,
164 .vop_closeextattr = ffs_closeextattr,
165 .vop_deleteextattr = ffs_deleteextattr,
166 .vop_getextattr = ffs_getextattr,
167 .vop_listextattr = ffs_listextattr,
168 .vop_openextattr = ffs_openextattr,
169 .vop_setextattr = ffs_setextattr,
170 .vop_vptofh = ffs_vptofh,
174 * Synch an open file.
178 ffs_fsync(struct vop_fsync_args *ap)
187 error = ffs_syncvnode(vp, ap->a_waitfor);
190 if (ap->a_waitfor == MNT_WAIT &&
191 (vp->v_mount->mnt_flag & MNT_SOFTDEP)) {
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)
215 struct inode *ip = VTOI(vp);
219 int s, error, wait, passes, skipmeta;
222 wait = (waitfor == MNT_WAIT);
223 lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
227 * Flush all dirty buffers associated with a vnode.
236 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
237 bp->b_vflags &= ~BV_SCANNED;
238 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
240 * Reasons to skip this buffer: it has already been considered
241 * on this pass, this pass is the first time through on a
242 * synchronous flush request and the buffer being considered
243 * is metadata, the buffer has dependencies that will cause
244 * it to be redirtied and it has not already been deferred,
245 * or it is already being written.
247 if ((bp->b_vflags & BV_SCANNED) != 0)
249 bp->b_vflags |= BV_SCANNED;
250 if ((skipmeta == 1 && bp->b_lblkno < 0))
252 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
255 if (!wait && !LIST_EMPTY(&bp->b_dep) &&
256 (bp->b_flags & B_DEFERRED) == 0 &&
257 buf_countdeps(bp, 0)) {
258 bp->b_flags |= B_DEFERRED;
263 if ((bp->b_flags & B_DELWRI) == 0)
264 panic("ffs_fsync: not dirty");
266 * If this is a synchronous flush request, or it is not a
267 * file or device, start the write on this buffer immediately.
269 if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
272 * On our final pass through, do all I/O synchronously
273 * so that we can find out if our flush is failing
274 * because of write errors.
276 if (passes > 0 || !wait) {
277 if ((bp->b_flags & B_CLUSTEROK) && !wait) {
278 (void) vfs_bio_awrite(bp);
288 if ((error = bwrite(bp)) != 0)
292 } else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
294 * If the buffer is for data that has been truncated
295 * off the file, then throw it away.
298 bp->b_flags |= B_INVAL | B_NOCACHE;
306 * Since we may have slept during the I/O, we need
307 * to start from a known point.
310 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
313 * If we were asked to do this synchronously, then go back for
314 * another pass, this time doing the metadata.
322 bufobj_wwait(bo, 0, 0);
326 * Ensure that any filesystem metatdata associated
327 * with the vnode has been written.
330 if ((error = softdep_sync_metadata(vp)) != 0)
335 if (bo->bo_dirty.bv_cnt > 0) {
337 * Block devices associated with filesystems may
338 * have new I/O requests posted for them even if
339 * the vnode is locked, so no amount of trying will
340 * get them clean. Thus we give block devices a
341 * good effort, then just give up. For all other file
342 * types, go around and try again until it is clean.
349 if (!vn_isdisk(vp, NULL))
350 vprint("ffs_fsync: dirty", vp);
356 return (ffs_update(vp, wait));
361 struct vop_lock1_args /* {
369 #ifndef NO_FFS_SNAPSHOT
375 switch (ap->a_flags & LK_TYPE_MASK) {
382 #ifdef DEBUG_VFS_LOCKS
383 KASSERT(vp->v_holdcnt != 0,
384 ("ffs_lock %p: zero hold count", vp));
387 result = _lockmgr_args(lkp, flags, VI_MTX(vp),
388 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
389 ap->a_file, ap->a_line);
390 if (lkp == vp->v_vnlock || result != 0)
393 * Apparent success, except that the vnode
394 * mutated between snapshot file vnode and
395 * regular file vnode while this process
396 * slept. The lock currently held is not the
397 * right lock. Release it, and try to get the
400 (void) _lockmgr_args(lkp, LK_RELEASE, NULL,
401 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
402 ap->a_file, ap->a_line);
403 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
404 (LK_INTERLOCK | LK_NOWAIT))
406 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
407 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
408 flags &= ~LK_INTERLOCK;
412 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
416 return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
421 * Vnode op for reading.
425 struct vop_read_args /* {
429 struct ucred *a_cred;
437 ufs_lbn_t lbn, nextlbn;
439 long size, xfersize, blkoffset;
440 int error, orig_resid;
446 ioflag = ap->a_ioflag;
447 if (ap->a_ioflag & IO_EXT)
449 return (ffs_extread(vp, uio, ioflag));
451 panic("ffs_read+IO_EXT");
454 if ((ioflag & IO_DIRECT) != 0) {
457 error = ffs_rawread(vp, uio, &workdone);
458 if (error != 0 || workdone != 0)
463 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
467 if (uio->uio_rw != UIO_READ)
468 panic("ffs_read: mode");
470 if (vp->v_type == VLNK) {
471 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
472 panic("ffs_read: short symlink");
473 } else if (vp->v_type != VREG && vp->v_type != VDIR)
474 panic("ffs_read: type %d", vp->v_type);
476 orig_resid = uio->uio_resid;
477 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
480 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
482 if (uio->uio_offset < ip->i_size &&
483 uio->uio_offset >= fs->fs_maxfilesize)
486 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
487 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
489 lbn = lblkno(fs, uio->uio_offset);
493 * size of buffer. The buffer representing the
494 * end of the file is rounded up to the size of
495 * the block type ( fragment or full block,
498 size = blksize(fs, ip, lbn);
499 blkoffset = blkoff(fs, uio->uio_offset);
502 * The amount we want to transfer in this iteration is
503 * one FS block less the amount of the data before
504 * our startpoint (duh!)
506 xfersize = fs->fs_bsize - blkoffset;
509 * But if we actually want less than the block,
510 * or the file doesn't have a whole block more of data,
511 * then use the lesser number.
513 if (uio->uio_resid < xfersize)
514 xfersize = uio->uio_resid;
515 if (bytesinfile < xfersize)
516 xfersize = bytesinfile;
518 if (lblktosize(fs, nextlbn) >= ip->i_size) {
520 * Don't do readahead if this is the end of the file.
522 error = bread(vp, lbn, size, NOCRED, &bp);
523 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
525 * Otherwise if we are allowed to cluster,
526 * grab as much as we can.
528 * XXX This may not be a win if we are not
529 * doing sequential access.
531 error = cluster_read(vp, ip->i_size, lbn,
532 size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
533 } else if (seqcount > 1) {
535 * If we are NOT allowed to cluster, then
536 * if we appear to be acting sequentially,
537 * fire off a request for a readahead
538 * as well as a read. Note that the 4th and 5th
539 * arguments point to arrays of the size specified in
542 u_int nextsize = blksize(fs, ip, nextlbn);
543 error = breadn(vp, lbn,
544 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
547 * Failing all of the above, just read what the
548 * user asked for. Interestingly, the same as
549 * the first option above.
551 error = bread(vp, lbn, size, NOCRED, &bp);
560 * If IO_DIRECT then set B_DIRECT for the buffer. This
561 * will cause us to attempt to release the buffer later on
562 * and will cause the buffer cache to attempt to free the
565 if (ioflag & IO_DIRECT)
566 bp->b_flags |= B_DIRECT;
569 * We should only get non-zero b_resid when an I/O error
570 * has occurred, which should cause us to break above.
571 * However, if the short read did not cause an error,
572 * then we want to ensure that we do not uiomove bad
573 * or uninitialized data.
576 if (size < xfersize) {
582 error = uiomove((char *)bp->b_data + blkoffset,
587 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
588 (LIST_EMPTY(&bp->b_dep))) {
590 * If there are no dependencies, and it's VMIO,
591 * then we don't need the buf, mark it available
592 * for freeing. The VM has the data.
594 bp->b_flags |= B_RELBUF;
598 * Otherwise let whoever
599 * made the request take care of
600 * freeing it. We just queue
601 * it onto another list.
608 * This can only happen in the case of an error
609 * because the loop above resets bp to NULL on each iteration
610 * and on normal completion has not set a new value into it.
611 * so it must have come from a 'break' statement
614 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
615 (LIST_EMPTY(&bp->b_dep))) {
616 bp->b_flags |= B_RELBUF;
623 if ((error == 0 || uio->uio_resid != orig_resid) &&
624 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
625 (ip->i_flag & IN_ACCESS) == 0) {
627 ip->i_flag |= IN_ACCESS;
634 * Vnode op for writing.
638 struct vop_write_args /* {
642 struct ucred *a_cred;
653 int blkoffset, error, flags, ioflag, resid, 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))
713 for (error = 0; uio->uio_resid > 0;) {
714 lbn = lblkno(fs, uio->uio_offset);
715 blkoffset = blkoff(fs, uio->uio_offset);
716 xfersize = fs->fs_bsize - blkoffset;
717 if (uio->uio_resid < xfersize)
718 xfersize = uio->uio_resid;
719 if (uio->uio_offset + xfersize > ip->i_size)
720 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
723 * We must perform a read-before-write if the transfer size
724 * does not cover the entire buffer.
726 if (fs->fs_bsize > xfersize)
730 /* XXX is uio->uio_offset the right thing here? */
731 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
732 ap->a_cred, flags, &bp);
734 vnode_pager_setsize(vp, ip->i_size);
738 * If the buffer is not valid we have to clear out any
739 * garbage data from the pages instantiated for the buffer.
740 * If we do not, a failed uiomove() during a write can leave
741 * the prior contents of the pages exposed to a userland
742 * mmap(). XXX deal with uiomove() errors a better way.
744 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
746 if (ioflag & IO_DIRECT)
747 bp->b_flags |= B_DIRECT;
748 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
749 bp->b_flags |= B_NOCACHE;
751 if (uio->uio_offset + xfersize > ip->i_size) {
752 ip->i_size = uio->uio_offset + xfersize;
753 DIP_SET(ip, i_size, ip->i_size);
756 size = blksize(fs, ip, lbn) - bp->b_resid;
761 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
762 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
763 (LIST_EMPTY(&bp->b_dep))) {
764 bp->b_flags |= B_RELBUF;
768 * If IO_SYNC each buffer is written synchronously. Otherwise
769 * if we have a severe page deficiency write the buffer
770 * asynchronously. Otherwise try to cluster, and if that
771 * doesn't do it then either do an async write (if O_DIRECT),
772 * or a delayed write (if not).
774 if (ioflag & IO_SYNC) {
776 } else if (vm_page_count_severe() ||
777 buf_dirty_count_severe() ||
778 (ioflag & IO_ASYNC)) {
779 bp->b_flags |= B_CLUSTEROK;
781 } else if (xfersize + blkoffset == fs->fs_bsize) {
782 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
783 bp->b_flags |= B_CLUSTEROK;
784 cluster_write(vp, bp, ip->i_size, seqcount);
788 } else if (ioflag & IO_DIRECT) {
789 bp->b_flags |= B_CLUSTEROK;
792 bp->b_flags |= B_CLUSTEROK;
795 if (error || xfersize == 0)
797 ip->i_flag |= IN_CHANGE | IN_UPDATE;
800 * If we successfully wrote any data, and we are not the superuser
801 * we clear the setuid and setgid bits as a precaution against
804 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
806 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
807 ip->i_mode &= ~(ISUID | ISGID);
808 DIP_SET(ip, i_mode, ip->i_mode);
812 if (ioflag & IO_UNIT) {
813 (void)ffs_truncate(vp, osize,
814 IO_NORMAL | (ioflag & IO_SYNC),
815 ap->a_cred, uio->uio_td);
816 uio->uio_offset -= resid - uio->uio_resid;
817 uio->uio_resid = resid;
819 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
820 error = ffs_update(vp, 1);
829 struct vop_getpages_args *ap;
835 pcount = round_page(ap->a_count) / PAGE_SIZE;
836 mreq = ap->a_m[ap->a_reqpage];
839 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
840 * then the entire page is valid. Since the page may be mapped,
841 * user programs might reference data beyond the actual end of file
842 * occuring within the page. We have to zero that data.
844 VM_OBJECT_LOCK(mreq->object);
846 if (mreq->valid != VM_PAGE_BITS_ALL)
847 vm_page_zero_invalid(mreq, TRUE);
848 vm_page_lock_queues();
849 for (i = 0; i < pcount; i++) {
850 if (i != ap->a_reqpage) {
851 vm_page_free(ap->a_m[i]);
854 vm_page_unlock_queues();
855 VM_OBJECT_UNLOCK(mreq->object);
858 VM_OBJECT_UNLOCK(mreq->object);
860 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
867 * Extended attribute area reading.
870 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
873 struct ufs2_dinode *dp;
876 ufs_lbn_t lbn, nextlbn;
878 long size, xfersize, blkoffset;
879 int error, orig_resid;
886 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
887 panic("ffs_extread: mode");
890 orig_resid = uio->uio_resid;
891 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
894 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
896 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
897 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
899 lbn = lblkno(fs, uio->uio_offset);
903 * size of buffer. The buffer representing the
904 * end of the file is rounded up to the size of
905 * the block type ( fragment or full block,
908 size = sblksize(fs, dp->di_extsize, lbn);
909 blkoffset = blkoff(fs, uio->uio_offset);
912 * The amount we want to transfer in this iteration is
913 * one FS block less the amount of the data before
914 * our startpoint (duh!)
916 xfersize = fs->fs_bsize - blkoffset;
919 * But if we actually want less than the block,
920 * or the file doesn't have a whole block more of data,
921 * then use the lesser number.
923 if (uio->uio_resid < xfersize)
924 xfersize = uio->uio_resid;
925 if (bytesinfile < xfersize)
926 xfersize = bytesinfile;
928 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
930 * Don't do readahead if this is the end of the info.
932 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
935 * If we have a second block, then
936 * fire off a request for a readahead
937 * as well as a read. Note that the 4th and 5th
938 * arguments point to arrays of the size specified in
941 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
943 nextlbn = -1 - nextlbn;
944 error = breadn(vp, -1 - lbn,
945 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
954 * If IO_DIRECT then set B_DIRECT for the buffer. This
955 * will cause us to attempt to release the buffer later on
956 * and will cause the buffer cache to attempt to free the
959 if (ioflag & IO_DIRECT)
960 bp->b_flags |= B_DIRECT;
963 * We should only get non-zero b_resid when an I/O error
964 * has occurred, which should cause us to break above.
965 * However, if the short read did not cause an error,
966 * then we want to ensure that we do not uiomove bad
967 * or uninitialized data.
970 if (size < xfersize) {
976 error = uiomove((char *)bp->b_data + blkoffset,
981 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
982 (LIST_EMPTY(&bp->b_dep))) {
984 * If there are no dependencies, and it's VMIO,
985 * then we don't need the buf, mark it available
986 * for freeing. The VM has the data.
988 bp->b_flags |= B_RELBUF;
992 * Otherwise let whoever
993 * made the request take care of
994 * freeing it. We just queue
995 * it onto another list.
1002 * This can only happen in the case of an error
1003 * because the loop above resets bp to NULL on each iteration
1004 * and on normal completion has not set a new value into it.
1005 * so it must have come from a 'break' statement
1008 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1009 (LIST_EMPTY(&bp->b_dep))) {
1010 bp->b_flags |= B_RELBUF;
1020 * Extended attribute area writing.
1023 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1026 struct ufs2_dinode *dp;
1031 int blkoffset, error, flags, resid, size, xfersize;
1037 KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1041 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1042 panic("ffs_extwrite: mode");
1045 if (ioflag & IO_APPEND)
1046 uio->uio_offset = dp->di_extsize;
1047 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1048 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1049 if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1052 resid = uio->uio_resid;
1053 osize = dp->di_extsize;
1055 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1058 for (error = 0; uio->uio_resid > 0;) {
1059 lbn = lblkno(fs, uio->uio_offset);
1060 blkoffset = blkoff(fs, uio->uio_offset);
1061 xfersize = fs->fs_bsize - blkoffset;
1062 if (uio->uio_resid < xfersize)
1063 xfersize = uio->uio_resid;
1066 * We must perform a read-before-write if the transfer size
1067 * does not cover the entire buffer.
1069 if (fs->fs_bsize > xfersize)
1072 flags &= ~BA_CLRBUF;
1073 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1078 * If the buffer is not valid we have to clear out any
1079 * garbage data from the pages instantiated for the buffer.
1080 * If we do not, a failed uiomove() during a write can leave
1081 * the prior contents of the pages exposed to a userland
1082 * mmap(). XXX deal with uiomove() errors a better way.
1084 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1086 if (ioflag & IO_DIRECT)
1087 bp->b_flags |= B_DIRECT;
1089 if (uio->uio_offset + xfersize > dp->di_extsize)
1090 dp->di_extsize = uio->uio_offset + xfersize;
1092 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1093 if (size < xfersize)
1097 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1098 if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1099 (LIST_EMPTY(&bp->b_dep))) {
1100 bp->b_flags |= B_RELBUF;
1104 * If IO_SYNC each buffer is written synchronously. Otherwise
1105 * if we have a severe page deficiency write the buffer
1106 * asynchronously. Otherwise try to cluster, and if that
1107 * doesn't do it then either do an async write (if O_DIRECT),
1108 * or a delayed write (if not).
1110 if (ioflag & IO_SYNC) {
1112 } else if (vm_page_count_severe() ||
1113 buf_dirty_count_severe() ||
1114 xfersize + blkoffset == fs->fs_bsize ||
1115 (ioflag & (IO_ASYNC | IO_DIRECT)))
1119 if (error || xfersize == 0)
1121 ip->i_flag |= IN_CHANGE;
1124 * If we successfully wrote any data, and we are not the superuser
1125 * we clear the setuid and setgid bits as a precaution against
1128 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1129 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1130 ip->i_mode &= ~(ISUID | ISGID);
1131 dp->di_mode = ip->i_mode;
1135 if (ioflag & IO_UNIT) {
1136 (void)ffs_truncate(vp, osize,
1137 IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1138 uio->uio_offset -= resid - uio->uio_resid;
1139 uio->uio_resid = resid;
1141 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1142 error = ffs_update(vp, 1);
1148 * Vnode operating to retrieve a named extended attribute.
1150 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1151 * the length of the EA, and possibly the pointer to the entry and to the data.
1154 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1156 u_char *p, *pe, *pn, *p0;
1157 int eapad1, eapad2, ealength, ealen, nlen;
1161 nlen = strlen(name);
1163 for (p = ptr; p < pe; p = pn) {
1165 bcopy(p, &ul, sizeof(ul));
1167 /* make sure this entry is complete */
1170 p += sizeof(uint32_t);
1178 if (bcmp(p, name, nlen))
1180 ealength = sizeof(uint32_t) + 3 + nlen;
1181 eapad1 = 8 - (ealength % 8);
1185 ealen = ul - ealength - eapad2;
1197 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1200 struct ufs2_dinode *dp;
1203 struct iovec liovec;
1211 easize = dp->di_extsize;
1212 if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1215 eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1217 liovec.iov_base = eae;
1218 liovec.iov_len = easize;
1219 luio.uio_iov = &liovec;
1220 luio.uio_iovcnt = 1;
1221 luio.uio_offset = 0;
1222 luio.uio_resid = easize;
1223 luio.uio_segflg = UIO_SYSSPACE;
1224 luio.uio_rw = UIO_READ;
1227 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1237 ffs_lock_ea(struct vnode *vp)
1243 while (ip->i_flag & IN_EA_LOCKED) {
1244 ip->i_flag |= IN_EA_LOCKWAIT;
1245 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1248 ip->i_flag |= IN_EA_LOCKED;
1253 ffs_unlock_ea(struct vnode *vp)
1259 if (ip->i_flag & IN_EA_LOCKWAIT)
1260 wakeup(&ip->i_ea_refs);
1261 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1266 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1269 struct ufs2_dinode *dp;
1275 if (ip->i_ea_area != NULL) {
1281 error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1286 ip->i_ea_len = dp->di_extsize;
1294 * Vnode extattr transaction commit/abort
1297 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1301 struct iovec liovec;
1303 struct ufs2_dinode *dp;
1308 if (ip->i_ea_area == NULL) {
1313 error = ip->i_ea_error;
1314 if (commit && error == 0) {
1315 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1317 cred = vp->v_mount->mnt_cred;
1318 liovec.iov_base = ip->i_ea_area;
1319 liovec.iov_len = ip->i_ea_len;
1320 luio.uio_iov = &liovec;
1321 luio.uio_iovcnt = 1;
1322 luio.uio_offset = 0;
1323 luio.uio_resid = ip->i_ea_len;
1324 luio.uio_segflg = UIO_SYSSPACE;
1325 luio.uio_rw = UIO_WRITE;
1327 /* XXX: I'm not happy about truncating to zero size */
1328 if (ip->i_ea_len < dp->di_extsize)
1329 error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1330 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1332 if (--ip->i_ea_refs == 0) {
1333 free(ip->i_ea_area, M_TEMP);
1334 ip->i_ea_area = NULL;
1343 * Vnode extattr strategy routine for fifos.
1345 * We need to check for a read or write of the external attributes.
1346 * Otherwise we just fall through and do the usual thing.
1349 ffsext_strategy(struct vop_strategy_args *ap)
1351 struct vop_strategy_args {
1352 struct vnodeop_desc *a_desc;
1362 lbn = ap->a_bp->b_lblkno;
1363 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1364 lbn < 0 && lbn >= -NXADDR)
1365 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1366 if (vp->v_type == VFIFO)
1367 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1368 panic("spec nodes went here");
1372 * Vnode extattr transaction commit/abort
1375 ffs_openextattr(struct vop_openextattr_args *ap)
1377 struct vop_openextattr_args {
1378 struct vnodeop_desc *a_desc;
1380 IN struct ucred *a_cred;
1381 IN struct thread *a_td;
1388 ip = VTOI(ap->a_vp);
1391 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1392 return (EOPNOTSUPP);
1394 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1399 * Vnode extattr transaction commit/abort
1402 ffs_closeextattr(struct vop_closeextattr_args *ap)
1404 struct vop_closeextattr_args {
1405 struct vnodeop_desc *a_desc;
1408 IN struct ucred *a_cred;
1409 IN struct thread *a_td;
1416 ip = VTOI(ap->a_vp);
1419 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1420 return (EOPNOTSUPP);
1422 if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1425 return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1429 * Vnode operation to remove a named attribute.
1432 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1435 IN struct vnode *a_vp;
1436 IN int a_attrnamespace;
1437 IN const char *a_name;
1438 IN struct ucred *a_cred;
1439 IN struct thread *a_td;
1445 uint32_t ealength, ul;
1446 int ealen, olen, eapad1, eapad2, error, i, easize;
1449 ip = VTOI(ap->a_vp);
1452 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1453 return (EOPNOTSUPP);
1455 if (strlen(ap->a_name) == 0)
1458 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1461 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1462 ap->a_cred, ap->a_td, VWRITE);
1466 * ffs_lock_ea is not needed there, because the vnode
1467 * must be exclusively locked.
1469 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1470 ip->i_ea_error = error;
1474 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1478 ealength = eapad1 = ealen = eapad2 = 0;
1480 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1481 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1482 easize = ip->i_ea_len;
1484 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1487 /* delete but nonexistent */
1489 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1492 bcopy(p, &ul, sizeof ul);
1494 if (ul != ealength) {
1495 bcopy(p + ul, p + ealength, easize - i);
1496 easize += (ealength - ul);
1498 if (easize > NXADDR * fs->fs_bsize) {
1500 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1501 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1502 ip->i_ea_error = ENOSPC;
1506 ip->i_ea_area = eae;
1507 ip->i_ea_len = easize;
1509 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1514 * Vnode operation to retrieve a named extended attribute.
1517 ffs_getextattr(struct vop_getextattr_args *ap)
1520 IN struct vnode *a_vp;
1521 IN int a_attrnamespace;
1522 IN const char *a_name;
1523 INOUT struct uio *a_uio;
1525 IN struct ucred *a_cred;
1526 IN struct thread *a_td;
1536 ip = VTOI(ap->a_vp);
1539 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1540 return (EOPNOTSUPP);
1542 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1543 ap->a_cred, ap->a_td, VREAD);
1547 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1551 eae = ip->i_ea_area;
1552 easize = ip->i_ea_len;
1554 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1558 if (ap->a_size != NULL)
1559 *ap->a_size = ealen;
1560 else if (ap->a_uio != NULL)
1561 error = uiomove(p, ealen, ap->a_uio);
1565 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1570 * Vnode operation to retrieve extended attributes on a vnode.
1573 ffs_listextattr(struct vop_listextattr_args *ap)
1576 IN struct vnode *a_vp;
1577 IN int a_attrnamespace;
1578 INOUT struct uio *a_uio;
1580 IN struct ucred *a_cred;
1581 IN struct thread *a_td;
1587 u_char *eae, *p, *pe, *pn;
1592 ip = VTOI(ap->a_vp);
1595 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1596 return (EOPNOTSUPP);
1598 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1599 ap->a_cred, ap->a_td, VREAD);
1603 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1606 eae = ip->i_ea_area;
1607 easize = ip->i_ea_len;
1610 if (ap->a_size != NULL)
1613 for(p = eae; error == 0 && p < pe; p = pn) {
1614 bcopy(p, &ul, sizeof(ul));
1619 if (*p++ != ap->a_attrnamespace)
1623 if (ap->a_size != NULL) {
1624 *ap->a_size += ealen + 1;
1625 } else if (ap->a_uio != NULL) {
1626 error = uiomove(p, ealen + 1, ap->a_uio);
1629 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1634 * Vnode operation to set a named attribute.
1637 ffs_setextattr(struct vop_setextattr_args *ap)
1640 IN struct vnode *a_vp;
1641 IN int a_attrnamespace;
1642 IN const char *a_name;
1643 INOUT struct uio *a_uio;
1644 IN struct ucred *a_cred;
1645 IN struct thread *a_td;
1651 uint32_t ealength, ul;
1653 int olen, eapad1, eapad2, error, i, easize;
1656 ip = VTOI(ap->a_vp);
1659 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1660 return (EOPNOTSUPP);
1662 if (strlen(ap->a_name) == 0)
1665 /* XXX Now unsupported API to delete EAs using NULL uio. */
1666 if (ap->a_uio == NULL)
1667 return (EOPNOTSUPP);
1669 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1672 ealen = ap->a_uio->uio_resid;
1673 if (ealen < 0 || ealen > lblktosize(fs, NXADDR))
1676 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1677 ap->a_cred, ap->a_td, VWRITE);
1681 * ffs_lock_ea is not needed there, because the vnode
1682 * must be exclusively locked.
1684 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1685 ip->i_ea_error = error;
1689 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1693 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1694 eapad1 = 8 - (ealength % 8);
1697 eapad2 = 8 - (ealen % 8);
1700 ealength += eapad1 + ealen + eapad2;
1702 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1703 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1704 easize = ip->i_ea_len;
1706 olen = ffs_findextattr(eae, easize,
1707 ap->a_attrnamespace, ap->a_name, &p, NULL);
1709 /* new, append at end */
1713 bcopy(p, &ul, sizeof ul);
1715 if (ul != ealength) {
1716 bcopy(p + ul, p + ealength, easize - i);
1717 easize += (ealength - ul);
1720 if (easize > lblktosize(fs, NXADDR)) {
1722 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1723 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1724 ip->i_ea_error = ENOSPC;
1727 bcopy(&ealength, p, sizeof(ealength));
1728 p += sizeof(ealength);
1729 *p++ = ap->a_attrnamespace;
1731 *p++ = strlen(ap->a_name);
1732 strcpy(p, ap->a_name);
1733 p += strlen(ap->a_name);
1736 error = uiomove(p, ealen, ap->a_uio);
1739 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1740 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1741 ip->i_ea_error = error;
1748 ip->i_ea_area = eae;
1749 ip->i_ea_len = easize;
1751 error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1756 * Vnode pointer to File handle
1759 ffs_vptofh(struct vop_vptofh_args *ap)
1762 IN struct vnode *a_vp;
1763 IN struct fid *a_fhp;
1770 ip = VTOI(ap->a_vp);
1771 ufhp = (struct ufid *)ap->a_fhp;
1772 ufhp->ufid_len = sizeof(struct ufid);
1773 ufhp->ufid_ino = ip->i_number;
1774 ufhp->ufid_gen = ip->i_gen;