2 * Copyright (c) 1982, 1986, 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
35 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $
38 #include "opt_quota.h"
39 #include "opt_suiddir.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
51 #include <sys/namei.h>
52 #include <sys/mount.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <sys/malloc.h>
56 #include <sys/dirent.h>
57 #include <sys/lockf.h>
58 #include <sys/event.h>
61 #include <sys/file.h> /* XXX */
65 #include <vm/vm_extern.h>
67 #include <vfs/fifofs/fifo.h>
73 #include "ufs_extern.h"
74 #include "ffs_extern.h"
80 static int ufs_access (struct vop_access_args *);
81 static int ufs_advlock (struct vop_advlock_args *);
82 static int ufs_chmod (struct vnode *, int, struct ucred *);
83 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *);
84 static int ufs_close (struct vop_close_args *);
85 static int ufs_create (struct vop_old_create_args *);
86 static int ufs_getattr (struct vop_getattr_args *);
87 static int ufs_link (struct vop_old_link_args *);
88 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *);
89 static int ufs_markatime (struct vop_markatime_args *);
90 static int ufs_missingop (struct vop_generic_args *ap);
91 static int ufs_mkdir (struct vop_old_mkdir_args *);
92 static int ufs_mknod (struct vop_old_mknod_args *);
93 static int ufs_print (struct vop_print_args *);
94 static int ufs_readdir (struct vop_readdir_args *);
95 static int ufs_readlink (struct vop_readlink_args *);
96 static int ufs_remove (struct vop_old_remove_args *);
97 static int ufs_rename (struct vop_old_rename_args *);
98 static int ufs_rmdir (struct vop_old_rmdir_args *);
99 static int ufs_setattr (struct vop_setattr_args *);
100 static int ufs_strategy (struct vop_strategy_args *);
101 static int ufs_symlink (struct vop_old_symlink_args *);
102 static int ufs_whiteout (struct vop_old_whiteout_args *);
103 static int ufsfifo_close (struct vop_close_args *);
104 static int ufsfifo_kqfilter (struct vop_kqfilter_args *);
105 static int ufsfifo_read (struct vop_read_args *);
106 static int ufsfifo_write (struct vop_write_args *);
107 static int filt_ufsread (struct knote *kn, long hint);
108 static int filt_ufswrite (struct knote *kn, long hint);
109 static int filt_ufsvnode (struct knote *kn, long hint);
110 static void filt_ufsdetach (struct knote *kn);
111 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
113 #define VN_KNOTE(vp, b) \
114 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b))
116 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
119 * A virgin directory (no blushing please).
121 static struct dirtemplate mastertemplate = {
122 0, 12, DT_DIR, 1, ".",
123 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
125 static struct odirtemplate omastertemplate = {
127 0, DIRBLKSIZ - 12, 2, ".."
131 ufs_itimes(struct vnode *vp)
137 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
139 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
140 ip->i_flag |= IN_LAZYMOD;
142 ip->i_flag |= IN_MODIFIED;
144 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
146 if (ip->i_flag & IN_ACCESS) {
147 ip->i_atime = (uint32_t)ts.tv_sec;
148 ip->i_atime_ext = ts.tv_sec >> 32;
149 ip->i_atimensec = ts.tv_nsec;
151 if (ip->i_flag & IN_CHANGE) {
152 ip->i_ctime = (uint32_t)ts.tv_sec;
153 ip->i_ctime_ext = ts.tv_sec >> 32;
154 ip->i_ctimensec = ts.tv_nsec;
156 if (ip->i_flag & IN_UPDATE) {
157 if (ip->i_flag & IN_NOCOPYWRITE) {
158 if (vp->v_flag & VLASTWRITETS) {
159 ip->i_mtime = (uint32_t)
160 vp->v_lastwrite_ts.tv_sec;
162 vp->v_lastwrite_ts.tv_sec >> 32;
164 vp->v_lastwrite_ts.tv_nsec;
167 ip->i_mtime = (uint32_t)ts.tv_sec;
168 ip->i_mtime_ext = ts.tv_sec >> 32;
169 ip->i_mtimensec = ts.tv_nsec;
174 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
178 * Create a regular file
180 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
181 * struct componentname *a_cnp, struct vattr *a_vap)
185 ufs_create(struct vop_old_create_args *ap)
190 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
191 ap->a_dvp, ap->a_vpp, ap->a_cnp);
194 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
201 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
202 * struct componentname *a_cnp, struct vattr *a_vap)
207 ufs_mknod(struct vop_old_mknod_args *ap)
209 struct vattr *vap = ap->a_vap;
210 struct vnode **vpp = ap->a_vpp;
216 * UFS cannot represent the entire major/minor range supported by
219 if (vap->va_rmajor != VNOVAL &&
220 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
224 /* no special directory support */
225 if (vap->va_type == VDIR)
228 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
229 ap->a_dvp, vpp, ap->a_cnp);
232 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
234 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
235 if (vap->va_rmajor != VNOVAL) {
237 * Want to be able to use this to make badblock
238 * inodes, so don't truncate the dev number.
240 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
243 * Remove inode, then reload it through VFS_VGET so it is
244 * checked to see if it is an alias of an existing entry in
247 (*vpp)->v_type = VNON;
248 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
249 vgone_vxlocked(*vpp);
251 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp);
262 * Update the times on the inode.
264 * ufs_close(struct vnode *a_vp, int a_fflag)
269 ufs_close(struct vop_close_args *ap)
271 struct vnode *vp = ap->a_vp;
275 return (vop_stdclose(ap));
279 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
283 ufs_access(struct vop_access_args *ap)
285 struct vnode *vp = ap->a_vp;
286 struct inode *ip = VTOI(vp);
290 if (ap->a_mode & VWRITE) {
291 switch (vp->v_type) {
295 if ((error = ufs_getinoquota(ip)) != 0)
304 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0);
309 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
314 ufs_getattr(struct vop_getattr_args *ap)
316 struct vnode *vp = ap->a_vp;
317 struct inode *ip = VTOI(vp);
318 struct vattr *vap = ap->a_vap;
322 * Copy from inode table
324 vap->va_fsid = devid_from_dev(ip->i_dev);
325 vap->va_fileid = ip->i_number;
326 vap->va_mode = ip->i_mode & ~IFMT;
327 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
328 ip->i_effnlink : ip->i_nlink;
329 vap->va_uid = ip->i_uid;
330 vap->va_gid = ip->i_gid;
331 vap->va_rmajor = umajor(ip->i_rdev);
332 vap->va_rminor = uminor(ip->i_rdev);
333 vap->va_size = ip->i_din.di_size;
334 vap->va_atime.tv_sec =
335 (time_t)(ip->i_atime | ((uint64_t)ip->i_atime_ext << 32));
336 vap->va_atime.tv_nsec = ip->i_atimensec;
337 vap->va_mtime.tv_sec =
338 (time_t)(ip->i_mtime | ((uint64_t)ip->i_mtime_ext << 32));
339 vap->va_mtime.tv_nsec = ip->i_mtimensec;
340 vap->va_ctime.tv_sec =
341 (time_t)(ip->i_ctime | ((uint64_t)ip->i_ctime_ext << 32));
342 vap->va_ctime.tv_nsec = ip->i_ctimensec;
343 vap->va_flags = ip->i_flags;
344 vap->va_gen = ip->i_gen;
345 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
346 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
347 vap->va_type = IFTOVT(ip->i_mode);
348 vap->va_filerev = ip->i_modrev;
354 ufs_markatime(struct vop_markatime_args *ap)
356 struct vnode *vp = ap->a_vp;
357 struct inode *ip = VTOI(vp);
359 if (vp->v_mount->mnt_flag & MNT_RDONLY)
361 if (vp->v_mount->mnt_flag & MNT_NOATIME)
363 ip->i_flag |= IN_ACCESS;
364 VN_KNOTE(vp, NOTE_ATTRIB);
369 * Set attribute vnode op. called from several syscalls
371 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
372 * struct ucred *a_cred)
376 ufs_setattr(struct vop_setattr_args *ap)
378 struct vattr *vap = ap->a_vap;
379 struct vnode *vp = ap->a_vp;
380 struct inode *ip = VTOI(vp);
381 struct ucred *cred = ap->a_cred;
385 * Check for unsettable attributes.
387 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
388 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
389 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
390 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
393 if (vap->va_flags != VNOVAL) {
394 if (vp->v_mount->mnt_flag & MNT_RDONLY)
396 if (cred->cr_uid != ip->i_uid &&
397 (error = caps_priv_check(cred, SYSCAP_NOVFS_SETATTR)))
403 * Note that a root chflags becomes a user chflags when
404 * we are jailed, unless the jail vfs_chflags sysctl
407 if (cred->cr_uid == 0 &&
409 PRISON_CAP_ISSET(cred->cr_prison->pr_caps,
410 PRISON_CAP_VFS_CHFLAGS))) {
412 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
415 ip->i_flags = vap->va_flags;
418 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
419 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
421 ip->i_flags &= SF_SETTABLE;
422 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
424 ip->i_flag |= IN_CHANGE;
425 if (vap->va_flags & (IMMUTABLE | APPEND))
428 if (ip->i_flags & (IMMUTABLE | APPEND))
431 * Go through the fields and update iff not VNOVAL.
433 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
434 if (vp->v_mount->mnt_flag & MNT_RDONLY)
436 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
439 if (vap->va_size != VNOVAL) {
441 * Disallow write attempts on read-only filesystems;
442 * unless the file is a socket, fifo, or a block or
443 * character device resident on the filesystem.
445 switch (vp->v_type) {
450 if (vp->v_mount->mnt_flag & MNT_RDONLY)
456 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
460 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
461 if (vp->v_mount->mnt_flag & MNT_RDONLY)
463 if (cred->cr_uid != ip->i_uid &&
464 (error = caps_priv_check(cred, SYSCAP_NOVFS_SETATTR)) &&
465 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
466 (error = VOP_EACCESS(vp, VWRITE, cred))))
470 if (vap->va_atime.tv_sec != VNOVAL)
471 ip->i_flag |= IN_ACCESS;
472 if (vap->va_mtime.tv_sec != VNOVAL)
473 ip->i_flag |= IN_CHANGE | IN_UPDATE;
475 if (vap->va_atime.tv_sec != VNOVAL) {
476 ip->i_atime = (uint32_t)vap->va_atime.tv_sec;
477 ip->i_atime_ext = vap->va_atime.tv_sec >> 32;
478 ip->i_atimensec = vap->va_atime.tv_nsec;
480 if (vap->va_mtime.tv_sec != VNOVAL) {
481 ip->i_mtime = (uint32_t)vap->va_mtime.tv_sec;
482 ip->i_mtime_ext = vap->va_mtime.tv_sec >> 32;
483 ip->i_mtimensec = vap->va_mtime.tv_nsec;
484 vclrflags(vp, VLASTWRITETS);
486 error = ffs_update(vp, 0);
491 if (vap->va_mode != (mode_t)VNOVAL) {
492 if (vp->v_mount->mnt_flag & MNT_RDONLY)
494 error = ufs_chmod(vp, (int)vap->va_mode, cred);
496 VN_KNOTE(vp, NOTE_ATTRIB);
501 * Change the mode on a file.
502 * Inode must be locked before calling.
505 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
507 struct inode *ip = VTOI(vp);
509 mode_t cur_mode = ip->i_mode;
511 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid,
516 if (cred->cr_uid != ip->i_uid) {
517 error = caps_priv_check(cred, SYSCAP_NOVFS_CHMOD);
522 if (vp->v_type != VDIR && (mode & S_ISTXT))
524 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
528 ip->i_mode = cur_mode;
529 ip->i_flag |= IN_CHANGE;
534 * Perform chown operation on inode ip;
535 * inode must be locked prior to call.
538 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
540 struct inode *ip = VTOI(vp);
549 if (uid == (uid_t)VNOVAL)
551 if (gid == (gid_t)VNOVAL)
554 * If we don't own the file, are trying to change the owner
555 * of the file, or are not a member of the target group,
556 * the caller must be superuser or the call fails.
558 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
559 (gid != ip->i_gid && !(cred->cr_gid == gid ||
560 groupmember(gid, cred)))) &&
561 (error = caps_priv_check(cred, SYSCAP_NOVFS_CHOWN)))
568 if ((error = ufs_getinoquota(ip)) != 0)
571 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
572 ip->i_dquot[USRQUOTA] = NODQUOT;
575 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
576 ip->i_dquot[GRPQUOTA] = NODQUOT;
578 change = ip->i_blocks;
579 (void) ufs_chkdq(ip, -change, cred, CHOWN);
580 (void) ufs_chkiq(ip, -1, cred, CHOWN);
581 for (i = 0; i < MAXQUOTAS; i++) {
582 ufs_dqrele(vp, ip->i_dquot[i]);
583 ip->i_dquot[i] = NODQUOT;
589 if ((error = ufs_getinoquota(ip)) == 0) {
591 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
592 ip->i_dquot[USRQUOTA] = NODQUOT;
595 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
596 ip->i_dquot[GRPQUOTA] = NODQUOT;
598 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
599 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
602 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
604 for (i = 0; i < MAXQUOTAS; i++) {
605 ufs_dqrele(vp, ip->i_dquot[i]);
606 ip->i_dquot[i] = NODQUOT;
611 if (ufs_getinoquota(ip) == 0) {
613 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
614 ip->i_dquot[USRQUOTA] = NODQUOT;
617 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
618 ip->i_dquot[GRPQUOTA] = NODQUOT;
620 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
621 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
622 (void) ufs_getinoquota(ip);
626 if (ufs_getinoquota(ip))
627 panic("ufs_chown: lost quota");
629 ip->i_flag |= IN_CHANGE;
630 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
631 ip->i_mode &= ~(ISUID | ISGID);
636 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
637 * struct componentname *a_cnp)
641 ufs_remove(struct vop_old_remove_args *ap)
644 struct vnode *vp = ap->a_vp;
645 struct vnode *dvp = ap->a_dvp;
649 #if 0 /* handled by kernel now */
650 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
651 (VTOI(dvp)->i_flags & APPEND)) {
656 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
657 VN_KNOTE(vp, NOTE_DELETE);
658 VN_KNOTE(dvp, NOTE_WRITE);
668 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
669 * struct componentname *a_cnp)
673 ufs_link(struct vop_old_link_args *ap)
675 struct vnode *vp = ap->a_vp;
676 struct vnode *tdvp = ap->a_tdvp;
677 struct componentname *cnp = ap->a_cnp;
679 struct direct newdir;
682 if (tdvp->v_mount != vp->v_mount) {
687 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM);
692 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
696 #if 0 /* handled by kernel now, also DragonFly allows this */
697 if (ip->i_flags & (IMMUTABLE | APPEND)) {
704 ip->i_flag |= IN_CHANGE;
705 if (DOINGSOFTDEP(vp))
706 softdep_change_linkcnt(ip);
707 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
709 ufs_makedirentry(ip, cnp, &newdir);
710 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
716 ip->i_flag |= IN_CHANGE;
717 if (DOINGSOFTDEP(vp))
718 softdep_change_linkcnt(ip);
724 VN_KNOTE(vp, NOTE_LINK);
725 VN_KNOTE(tdvp, NOTE_WRITE);
730 * whiteout vnode call
732 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
736 ufs_whiteout(struct vop_old_whiteout_args *ap)
738 struct vnode *dvp = ap->a_dvp;
739 struct componentname *cnp = ap->a_cnp;
740 struct direct newdir;
743 switch (ap->a_flags) {
745 /* 4.4 format directories support whiteout operations */
746 if (dvp->v_mount->mnt_maxsymlinklen > 0)
751 /* create a new directory whiteout */
753 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
754 panic("ufs_whiteout: old format filesystem");
757 newdir.d_ino = UFS_WINO;
758 newdir.d_namlen = cnp->cn_namelen;
759 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
760 newdir.d_type = DT_WHT;
761 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
765 /* remove an existing directory whiteout */
767 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
768 panic("ufs_whiteout: old format filesystem");
771 cnp->cn_flags &= ~CNP_DOWHITEOUT;
772 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
775 panic("ufs_whiteout: unknown op");
781 * Rename system call.
782 * rename("foo", "bar");
785 * link("foo", "bar");
787 * but ``atomically''. Can't do full commit without saving state in the
788 * inode on disk which isn't feasible at this time. Best we can do is
789 * always guarantee the target exists.
791 * Basic algorithm is:
793 * 1) Bump link count on source while we're linking it to the
794 * target. This also ensure the inode won't be deleted out
795 * from underneath us while we work (it may be truncated by
796 * a concurrent `trunc' or `open' for creation).
797 * 2) Link source to destination. If destination already exists,
799 * 3) Unlink source reference to inode if still around. If a
800 * directory was moved and the parent of the destination
801 * is different from the source, patch the ".." entry in the
804 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
805 * struct componentname *a_fcnp, struct vnode *a_tdvp,
806 * struct vnode *a_tvp, struct componentname *a_tcnp)
810 ufs_rename(struct vop_old_rename_args *ap)
812 struct vnode *tvp = ap->a_tvp;
813 struct vnode *tdvp = ap->a_tdvp;
814 struct vnode *fvp = ap->a_fvp;
815 struct vnode *fdvp = ap->a_fdvp;
816 struct componentname *tcnp = ap->a_tcnp;
817 struct componentname *fcnp = ap->a_fcnp;
818 struct inode *ip, *xp, *dp;
819 struct direct newdir;
820 ino_t oldparent = 0, newparent = 0;
821 int doingdirectory = 0;
822 int error = 0, ioflag;
825 * Check for cross-device rename.
827 if ((fvp->v_mount != tdvp->v_mount) ||
828 (tvp && (fvp->v_mount != tvp->v_mount))) {
842 #if 0 /* handled by kernel now */
843 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
844 (VTOI(tdvp)->i_flags & APPEND))) {
851 * Renaming a file to itself has no effect. The upper layers should
852 * not call us in that case. Temporarily just warn if they do.
855 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
860 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
865 * Note: now that fvp is locked we have to be sure to unlock it before
866 * using the 'abortit' target.
870 if (ip->i_nlink >= LINK_MAX) {
875 #if 0 /* handled by kernel now */
876 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
877 || (dp->i_flags & APPEND)) {
883 if ((ip->i_mode & IFMT) == IFDIR) {
885 * Avoid ".", "..", and aliases of "." for obvious reasons.
887 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
888 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
889 (ip->i_flag & IN_RENAME)) {
894 ip->i_flag |= IN_RENAME;
895 oldparent = dp->i_number;
898 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
901 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
902 * Cleanup fvp requirements so we can unlock it.
904 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
905 * is setup we can use the 'bad' target if we unlock fvp. We cannot
906 * use the abortit target anymore because of IN_RENAME.
915 * 1) Bump link count while we're moving stuff
916 * around. If we crash somewhere before
917 * completing our work, the link count
918 * may be wrong, but correctable.
922 ip->i_flag |= IN_CHANGE;
923 if (DOINGSOFTDEP(fvp))
924 softdep_change_linkcnt(ip);
925 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
926 DOINGASYNC(fvp)))) != 0) {
932 * If ".." must be changed (ie the directory gets a new
933 * parent) then the source directory must not be in the
934 * directory heirarchy above the target, as this would
935 * orphan everything below the source directory. Also
936 * the user must have write permission in the source so
937 * as to be able to change "..". We must repeat the call
938 * to namei, as the parent directory is unlocked by the
939 * call to checkpath().
941 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred);
945 * We are now back to where we were in that fvp, fdvp are unlocked
946 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
947 * set. Only the bad target or, if we clean up tvp and tdvp, the
948 * out target, may be used.
950 if (oldparent != dp->i_number)
951 newparent = dp->i_number;
952 if (doingdirectory && newparent) {
953 if (error) /* write access check above */
957 * Once we start messing with tvp and tdvp we cannot use the
958 * 'bad' target, only finish cleaning tdvp and tvp up and
959 * use the 'out' target.
961 * This cleans up tvp.
969 * This is a real mess. ufs_checkpath vput's the target
970 * directory so retain an extra ref and note that tdvp will
971 * lose its lock on return. This leaves us with one good
972 * ref after ufs_checkpath returns.
975 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
976 tcnp->cn_flags |= CNP_PDIRUNLOCK;
983 * relookup no longer messes with tdvp's refs. tdvp must be
984 * unlocked on entry and will be locked on a successful
987 error = relookup(tdvp, &tvp, tcnp);
989 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
995 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1002 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1003 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1004 * be set if doingdirectory. This means we can use the 'bad' target
1009 * 2) If target doesn't exist, link the target
1010 * to the source and unlink the source.
1011 * Otherwise, rewrite the target directory
1012 * entry to reference the source inode and
1013 * expunge the original entry's existence.
1016 if (dp->i_dev != ip->i_dev)
1017 panic("ufs_rename: EXDEV");
1019 * Account for ".." in new directory.
1020 * When source and destination have the same
1021 * parent we don't fool with the link count.
1023 if (doingdirectory && newparent) {
1024 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1030 dp->i_flag |= IN_CHANGE;
1031 if (DOINGSOFTDEP(tdvp))
1032 softdep_change_linkcnt(dp);
1033 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1038 ufs_makedirentry(ip, tcnp, &newdir);
1039 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1041 if (doingdirectory && newparent) {
1044 dp->i_flag |= IN_CHANGE;
1045 if (DOINGSOFTDEP(tdvp))
1046 softdep_change_linkcnt(dp);
1047 (void)ffs_update(tdvp, 1);
1051 VN_KNOTE(tdvp, NOTE_WRITE);
1054 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1055 panic("ufs_rename: EXDEV");
1057 * Short circuit rename(foo, foo).
1059 if (xp->i_number == ip->i_number)
1060 panic("ufs_rename: same file");
1062 * If the parent directory is "sticky", then the user must
1063 * own the parent directory, or the destination of the rename,
1064 * otherwise the destination may not be changed (except by
1065 * root). This implements append-only directories.
1067 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1068 tcnp->cn_cred->cr_uid != dp->i_uid &&
1069 xp->i_uid != tcnp->cn_cred->cr_uid) {
1074 * Target must be empty if a directory and have no links
1075 * to it. Also, ensure source and target are compatible
1076 * (both directories, or both not directories).
1078 * Purge the file or directory being replaced from the
1081 if ((xp->i_mode&IFMT) == IFDIR) {
1082 if ((xp->i_effnlink > 2) ||
1083 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1087 if (!doingdirectory) {
1091 /* cache_purge removed - handled by VFS compat layer */
1092 } else if (doingdirectory == 0) {
1093 /* cache_purge removed - handled by VFS compat layer */
1099 * note: inode passed to ufs_dirrewrite() is 0 for a
1100 * non-directory file rename, 1 for a directory rename
1101 * in the same directory, and > 1 for an inode representing
1102 * the new directory.
1104 error = ufs_dirrewrite(dp, xp, ip->i_number,
1106 (doingdirectory && newparent) ?
1107 newparent : (ino_t)doingdirectory);
1110 if (doingdirectory) {
1113 if (DOINGSOFTDEP(tdvp))
1114 softdep_change_linkcnt(dp);
1117 if (DOINGSOFTDEP(tvp))
1118 softdep_change_linkcnt(xp);
1120 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1122 * Truncate inode. The only stuff left in the directory
1123 * is "." and "..". The "." reference is inconsequential
1124 * since we are quashing it. We have removed the "."
1125 * reference and the reference in the parent directory,
1126 * but there may be other hard links. The soft
1127 * dependency code will arrange to do these operations
1128 * after the parent directory entry has been deleted on
1129 * disk, so when running with that code we avoid doing
1134 dp->i_flag |= IN_CHANGE;
1137 xp->i_flag |= IN_CHANGE;
1138 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1139 error = ffs_truncate(tvp, (off_t)0, ioflag,
1144 VN_KNOTE(tdvp, NOTE_WRITE);
1146 VN_KNOTE(tvp, NOTE_DELETE);
1152 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1153 * unlocked) remain. We are about to overwrite fvp but we have to
1154 * keep 'ip' intact so we cannot release the old fvp, which is still
1155 * refd and accessible via ap->a_fvp.
1157 * This means we cannot use either 'bad' or 'out' to cleanup any
1162 * 3) Unlink the source.
1164 fcnp->cn_flags &= ~CNP_MODMASK;
1165 fcnp->cn_flags |= CNP_LOCKPARENT;
1166 error = relookup(fdvp, &fvp, fcnp);
1167 if (error || fvp == NULL) {
1169 * From name has disappeared. IN_RENAME will not be set if
1170 * we get past the panic so we don't have to clean it up.
1173 panic("ufs_rename: lost dir entry");
1175 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1181 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1184 * fdvp and fvp are locked.
1190 * Ensure that the directory entry still exists and has not
1191 * changed while the new name has been entered. If the source is
1192 * a file then the entry may have been unlinked or renamed. In
1193 * either case there is no further work to be done. If the source
1194 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1195 * flag ensures that it cannot be moved by another rename or removed
1196 * by a rmdir. Cleanup IN_RENAME.
1200 panic("ufs_rename: lost dir entry");
1203 * If the source is a directory with a
1204 * new parent, the link count of the old
1205 * parent directory must be decremented
1206 * and ".." set to point to the new parent.
1208 if (doingdirectory && newparent) {
1209 xp->i_offset = mastertemplate.dot_reclen;
1210 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1211 /* cache_purge removed - handled by VFS compat layer */
1213 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1214 xp->i_flag &= ~IN_RENAME;
1217 VN_KNOTE(fvp, NOTE_RENAME);
1229 ip->i_flag &= ~IN_RENAME;
1230 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) {
1233 ip->i_flag |= IN_CHANGE;
1234 ip->i_flag &= ~IN_RENAME;
1235 if (DOINGSOFTDEP(fvp))
1236 softdep_change_linkcnt(ip);
1247 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1248 * struct componentname *a_cnp, struct vattr *a_vap)
1252 ufs_mkdir(struct vop_old_mkdir_args *ap)
1254 struct vnode *dvp = ap->a_dvp;
1255 struct vattr *vap = ap->a_vap;
1256 struct componentname *cnp = ap->a_cnp;
1257 struct inode *ip, *dp;
1260 struct dirtemplate dirtemplate, *dtp;
1261 struct direct newdir;
1266 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1270 dmode = vap->va_mode & 0777;
1273 * Must simulate part of ufs_makeinode here to acquire the inode,
1274 * but not have it entered in the parent directory. The entry is
1275 * made later after writing "." and ".." entries.
1277 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1281 ip->i_gid = dp->i_gid;
1285 struct ucred ucred, *ucp;
1289 * If we are hacking owners here, (only do this where told to)
1290 * and we are not giving it TO root, (would subvert quotas)
1291 * then go ahead and give it to the other user.
1292 * The new directory also inherits the SUID bit.
1293 * If user's UID and dir UID are the same,
1294 * 'give it away' so that the SUID is still forced on.
1296 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1297 (dp->i_mode & ISUID) && dp->i_uid) {
1299 ip->i_uid = dp->i_uid;
1301 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1303 * Make sure the correct user gets charged
1305 * Make a dummy credential for the victim.
1306 * XXX This seems to never be accessed out of
1307 * our context so a stack variable is ok.
1310 ucred.cr_uid = ip->i_uid;
1311 ucred.cr_ngroups = 1;
1312 ucred.cr_groups[0] = dp->i_gid;
1317 ip->i_uid = cnp->cn_cred->cr_uid;
1319 if ((error = ufs_getinoquota(ip)) ||
1320 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1321 ffs_vfree(tvp, ip->i_number, dmode);
1327 #else /* !SUIDDIR */
1328 ip->i_uid = cnp->cn_cred->cr_uid;
1330 if ((error = ufs_getinoquota(ip)) ||
1331 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1332 ffs_vfree(tvp, ip->i_number, dmode);
1337 #endif /* !SUIDDIR */
1338 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1340 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1343 if (DOINGSOFTDEP(tvp))
1344 softdep_change_linkcnt(ip);
1345 if (cnp->cn_flags & CNP_ISWHITEOUT)
1346 ip->i_flags |= UF_OPAQUE;
1349 * Bump link count in parent directory to reflect work done below.
1350 * Should be done before reference is created so cleanup is
1351 * possible if we crash.
1355 dp->i_flag |= IN_CHANGE;
1356 if (DOINGSOFTDEP(dvp))
1357 softdep_change_linkcnt(dp);
1358 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1363 * The vnode must have a VM object in order to issue buffer cache
1366 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1);
1369 * Initialize directory with "." and ".." from static template.
1371 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1372 dtp = &mastertemplate;
1374 dtp = (struct dirtemplate *)&omastertemplate;
1376 dirtemplate.dot_ino = ip->i_number;
1377 dirtemplate.dotdot_ino = dp->i_number;
1378 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1);
1379 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1382 ip->i_size = DIRBLKSIZ;
1383 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1384 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1385 if (DOINGSOFTDEP(tvp)) {
1387 * Ensure that the entire newly allocated block is a
1388 * valid directory so that future growth within the
1389 * block does not have to ensure that the block is
1390 * written before the inode.
1393 while (blkoff < bp->b_bcount) {
1395 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1396 blkoff += DIRBLKSIZ;
1399 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1400 DOINGASYNC(tvp)))) != 0) {
1405 * Directory set up, now install its entry in the parent directory.
1407 * If we are not doing soft dependencies, then we must write out the
1408 * buffer containing the new directory body before entering the new
1409 * name in the parent. If we are doing soft dependencies, then the
1410 * buffer containing the new directory body will be passed to and
1411 * released in the soft dependency code after the code has attached
1412 * an appropriate ordering dependency to the buffer which ensures that
1413 * the buffer is written before the new name is written in the parent.
1415 if (DOINGASYNC(dvp))
1417 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1419 ufs_makedirentry(ip, cnp, &newdir);
1420 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1424 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1429 dp->i_flag |= IN_CHANGE;
1430 if (DOINGSOFTDEP(dvp))
1431 softdep_change_linkcnt(dp);
1433 * No need to do an explicit VOP_TRUNCATE here, vrele will
1434 * do this for us because we set the link count to 0.
1438 ip->i_flag |= IN_CHANGE;
1439 if (DOINGSOFTDEP(tvp))
1440 softdep_change_linkcnt(ip);
1448 * Rmdir system call.
1450 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1451 * struct componentname *a_cnp)
1455 ufs_rmdir(struct vop_old_rmdir_args *ap)
1457 struct vnode *vp = ap->a_vp;
1458 struct vnode *dvp = ap->a_dvp;
1459 struct componentname *cnp = ap->a_cnp;
1460 struct inode *ip, *dp;
1467 * Do not remove a directory that is in the process of being renamed.
1468 * Verify the directory is empty (and valid). Rmdir ".." will not be
1469 * valid since ".." will contain a reference to the current directory
1470 * and thus be non-empty. Do not allow the removal of mounted on
1471 * directories (this can happen when an NFS exported filesystem
1472 * tries to remove a locally mounted on directory).
1475 if (ip->i_flag & IN_RENAME) {
1479 if (ip->i_effnlink != 2 ||
1480 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1484 #if 0 /* handled by kernel now */
1485 if ((dp->i_flags & APPEND)
1486 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1492 * Delete reference to directory before purging
1493 * inode. If we crash in between, the directory
1494 * will be reattached to lost+found,
1498 if (DOINGSOFTDEP(vp)) {
1499 softdep_change_linkcnt(dp);
1500 softdep_change_linkcnt(ip);
1502 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1506 if (DOINGSOFTDEP(vp)) {
1507 softdep_change_linkcnt(dp);
1508 softdep_change_linkcnt(ip);
1512 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1514 * Truncate inode. The only stuff left in the directory is "." and
1515 * "..". The "." reference is inconsequential since we are quashing
1516 * it. The soft dependency code will arrange to do these operations
1517 * after the parent directory entry has been deleted on disk, so
1518 * when running with that code we avoid doing them now.
1520 if (!DOINGSOFTDEP(vp)) {
1522 dp->i_flag |= IN_CHANGE;
1524 ip->i_flag |= IN_CHANGE;
1525 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1526 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1528 /* cache_purge removed - handled by VFS compat layer */
1530 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1531 if (ip->i_dirhash != NULL)
1532 ufsdirhash_free(ip);
1535 VN_KNOTE(vp, NOTE_DELETE);
1540 * symlink -- make a symbolic link
1542 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1543 * struct componentname *a_cnp, struct vattr *a_vap,
1548 ufs_symlink(struct vop_old_symlink_args *ap)
1550 struct vnode *vp, **vpp = ap->a_vpp;
1554 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1558 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1560 len = strlen(ap->a_target);
1561 if (len < vp->v_mount->mnt_maxsymlinklen) {
1563 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1565 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1568 * Make sure we have a VM object in order to use
1571 if (vp->v_object == NULL)
1572 vinitvmio(vp, 0, PAGE_SIZE, -1);
1573 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1574 UIO_SYSSPACE, IO_NODELOCKED,
1575 ap->a_cnp->cn_cred, NULL);
1583 * Vnode op for reading directories.
1585 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1586 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1590 ufs_readdir(struct vop_readdir_args *ap)
1592 struct uio *uio = ap->a_uio;
1593 struct vnode *vp = ap->a_vp;
1598 int offset; /* offset into buffer cache buffer */
1599 int eoffset; /* end of buffer clipped to file EOF */
1600 int pickup; /* pickup point */
1605 if (uio->uio_offset < 0)
1608 * Guess the number of cookies needed. Make sure we compute at
1609 * least 1, and no more then a reasonable limit.
1611 if (ap->a_ncookies) {
1612 ncookies = uio->uio_resid / 16 + 1;
1613 if (ncookies > 1024)
1615 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1617 ncookies = -1; /* force conditionals below */
1622 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
1629 if (uio->uio_offset >= VTOI(vp)->i_size) {
1632 if (ap->a_ncookies) {
1633 *ap->a_ncookies = cookie_index;
1634 *ap->a_cookies = cookies;
1640 * Loop until we run out of cookies, we run out of user buffer,
1641 * or we hit the directory EOF.
1643 * Always start scans at the beginning of the buffer, don't trust
1644 * the offset supplied by userland.
1646 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1647 pickup = (int)(uio->uio_offset - bp->b_loffset);
1650 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1651 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1653 eoffset = bp->b_bcount;
1655 while (offset < eoffset) {
1656 dp = (struct direct *)(bp->b_data + offset);
1657 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1658 offset + dp->d_reclen > bp->b_bcount) {
1662 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1666 if (offset < pickup) {
1667 offset += dp->d_reclen;
1670 #if BYTE_ORDER == LITTLE_ENDIAN
1672 retval = vop_write_dirent(&error, uio,
1673 dp->d_ino, dp->d_namlen, dp->d_type,
1678 retval = vop_write_dirent(&error, uio,
1679 dp->d_ino, dp->d_type, dp->d_namlen,
1685 cookies[cookie_index] = bp->b_loffset + offset;
1687 offset += dp->d_reclen;
1688 if (cookie_index == ncookies)
1693 * This will align the next loop to the beginning of the
1694 * next block, and pickup will calculate to 0.
1696 uio->uio_offset = bp->b_loffset + offset;
1699 if (retval || error || cookie_index == ncookies ||
1700 uio->uio_offset >= VTOI(vp)->i_size) {
1705 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1708 * Report errors only if we didn't manage to read anything
1710 if (error && cookie_index == 0) {
1712 kfree(cookies, M_TEMP);
1713 *ap->a_ncookies = 0;
1714 *ap->a_cookies = NULL;
1719 *ap->a_ncookies = cookie_index;
1720 *ap->a_cookies = cookies;
1729 * Return target name of a symbolic link
1731 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1735 ufs_readlink(struct vop_readlink_args *ap)
1737 struct vnode *vp = ap->a_vp;
1738 struct inode *ip = VTOI(vp);
1742 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1743 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1744 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1749 * Perform the equivalent of an OPEN on vp so we can issue a
1752 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1756 * Calculate the logical to physical mapping if not done already,
1757 * then call the device strategy routine.
1759 * In order to be able to swap to a file, the VOP_BMAP operation may not
1760 * deadlock on memory. See ufs_bmap() for details.
1762 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1766 ufs_strategy(struct vop_strategy_args *ap)
1768 struct bio *bio = ap->a_bio;
1770 struct buf *bp = bio->bio_buf;
1771 struct vnode *vp = ap->a_vp;
1776 if (vp->v_type == VBLK || vp->v_type == VCHR)
1777 panic("ufs_strategy: spec");
1778 nbio = push_bio(bio);
1779 if (nbio->bio_offset == NOOFFSET) {
1780 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1781 NULL, NULL, bp->b_cmd);
1783 bp->b_error = error;
1784 bp->b_flags |= B_ERROR;
1785 /* I/O was never started on nbio, must biodone(bio) */
1789 if (nbio->bio_offset == NOOFFSET)
1792 if (nbio->bio_offset == NOOFFSET) {
1794 * We hit a hole in the file. The buffer has been zero-filled
1795 * so just biodone() it.
1799 vn_strategy(ip->i_devvp, nbio);
1805 * Print out the contents of an inode.
1807 * ufs_print(struct vnode *a_vp)
1811 ufs_print(struct vop_print_args *ap)
1813 struct vnode *vp = ap->a_vp;
1814 struct inode *ip = VTOI(vp);
1816 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1817 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1819 if (vp->v_type == VFIFO)
1821 lockmgr_printinfo(&vp->v_lock);
1827 * Read wrapper for fifos.
1829 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1830 * struct ucred *a_cred)
1834 ufsfifo_read(struct vop_read_args *ap)
1841 resid = uio->uio_resid;
1842 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1843 ip = VTOI(ap->a_vp);
1844 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1845 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1846 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1851 * Write wrapper for fifos.
1853 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1854 * struct ucred *a_cred)
1858 ufsfifo_write(struct vop_write_args *ap)
1865 resid = uio->uio_resid;
1866 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1867 ip = VTOI(ap->a_vp);
1868 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1869 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1874 * Close wrapper for fifos.
1876 * Update the times on the inode then do device close.
1878 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
1882 ufsfifo_close(struct vop_close_args *ap)
1884 struct vnode *vp = ap->a_vp;
1886 if (VREFCNT(vp) > 1)
1888 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1892 * Kqfilter wrapper for fifos.
1894 * Fall through to ufs kqfilter routines if needed
1898 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
1902 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1904 error = ufs_kqfilter(ap);
1909 * Advisory record locking support
1911 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
1916 ufs_advlock(struct vop_advlock_args *ap)
1918 struct inode *ip = VTOI(ap->a_vp);
1920 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
1924 * Initialize the vnode associated with a new inode, handle aliased
1927 * Make sure directories have their VM object now rather then later,
1928 * saving us from having to check on all the myrid directory VOPs
1929 * that might be executed without a VOP_OPEN being performed.
1932 ufs_vinit(struct mount *mntp, struct vnode **vpp)
1940 vp->v_type = IFTOVT(ip->i_mode);
1942 switch(vp->v_type) {
1945 vp->v_ops = &mntp->mnt_vn_spec_ops;
1946 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
1949 vp->v_ops = &mntp->mnt_vn_fifo_ops;
1953 vinitvmio(vp, ip->i_size,
1954 blkoffsize(ip->i_fs, ip, ip->i_size),
1955 blkoff(ip->i_fs, ip->i_size));
1958 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) {
1959 vinitvmio(vp, ip->i_size,
1960 blkoffsize(ip->i_fs, ip, ip->i_size),
1961 blkoff(ip->i_fs, ip->i_size));
1969 if (ip->i_number == UFS_ROOTINO)
1970 vsetflags(vp, VROOT);
1972 * Initialize modrev times
1974 ip->i_modrev = init_va_filerev();
1980 * Allocate a new inode.
1984 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1985 struct componentname *cnp)
1987 struct inode *ip, *pdir;
1988 struct direct newdir;
1994 if ((mode & IFMT) == 0)
1997 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2001 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
2002 ip->i_gid = pdir->i_gid;
2006 struct ucred ucred, *ucp;
2010 * If we are not the owner of the directory,
2011 * and we are hacking owners here, (only do this where told to)
2012 * and we are not giving it TO root, (would subvert quotas)
2013 * then go ahead and give it to the other user.
2014 * Note that this drops off the execute bits for security.
2016 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2017 (pdir->i_mode & ISUID) &&
2018 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2019 ip->i_uid = pdir->i_uid;
2023 * Make sure the correct user gets charged
2025 * Quickly knock up a dummy credential for the victim.
2026 * XXX This seems to never be accessed out of our
2027 * context so a stack variable is ok.
2030 ucred.cr_uid = ip->i_uid;
2031 ucred.cr_ngroups = 1;
2032 ucred.cr_groups[0] = pdir->i_gid;
2036 ip->i_uid = cnp->cn_cred->cr_uid;
2039 if ((error = ufs_getinoquota(ip)) ||
2040 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2041 ffs_vfree(tvp, ip->i_number, mode);
2047 #else /* !SUIDDIR */
2048 ip->i_uid = cnp->cn_cred->cr_uid;
2050 if ((error = ufs_getinoquota(ip)) ||
2051 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2052 ffs_vfree(tvp, ip->i_number, mode);
2057 #endif /* !SUIDDIR */
2058 ip->i_din.di_spare7E = 0;
2059 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2061 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2064 if (DOINGSOFTDEP(tvp))
2065 softdep_change_linkcnt(ip);
2066 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2067 caps_priv_check(cnp->cn_cred, SYSCAP_NOVFS_SETGID))
2069 ip->i_mode &= ~ISGID;
2072 if (cnp->cn_flags & CNP_ISWHITEOUT)
2073 ip->i_flags |= UF_OPAQUE;
2076 * Regular files and directories need VM objects. Softlinks do
2077 * not (not immediately anyway).
2079 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2080 vinitvmio(tvp, 0, PAGE_SIZE, -1);
2083 * Make sure inode goes to disk before directory entry.
2085 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2088 ufs_makedirentry(ip, cnp, &newdir);
2089 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2097 * Write error occurred trying to update the inode
2098 * or the directory so must deallocate the inode.
2102 ip->i_flag |= IN_CHANGE;
2103 if (DOINGSOFTDEP(tvp))
2104 softdep_change_linkcnt(ip);
2110 ufs_missingop(struct vop_generic_args *ap)
2112 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2113 return (EOPNOTSUPP);
2116 static struct filterops ufsread_filtops =
2117 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread };
2118 static struct filterops ufswrite_filtops =
2119 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite };
2120 static struct filterops ufsvnode_filtops =
2121 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode };
2124 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2127 ufs_kqfilter(struct vop_kqfilter_args *ap)
2129 struct vnode *vp = ap->a_vp;
2130 struct knote *kn = ap->a_kn;
2132 switch (kn->kn_filter) {
2134 kn->kn_fop = &ufsread_filtops;
2137 kn->kn_fop = &ufswrite_filtops;
2140 kn->kn_fop = &ufsvnode_filtops;
2143 return (EOPNOTSUPP);
2146 kn->kn_hook = (caddr_t)vp;
2148 /* XXX: kq token actually protects the list */
2149 lwkt_gettoken(&vp->v_token);
2150 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2151 lwkt_reltoken(&vp->v_token);
2157 filt_ufsdetach(struct knote *kn)
2159 struct vnode *vp = (struct vnode *)kn->kn_hook;
2161 lwkt_gettoken(&vp->v_token);
2162 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2163 lwkt_reltoken(&vp->v_token);
2168 filt_ufsread(struct knote *kn, long hint)
2170 struct vnode *vp = (struct vnode *)kn->kn_hook;
2171 struct inode *ip = VTOI(vp);
2175 * filesystem is gone, so set the EOF flag and schedule
2176 * the knote for deletion.
2178 if (hint == NOTE_REVOKE) {
2179 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2183 off = ip->i_size - kn->kn_fp->f_offset;
2184 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2185 if (kn->kn_sfflags & NOTE_OLDAPI)
2187 return (kn->kn_data != 0);
2192 filt_ufswrite(struct knote *kn, long hint)
2195 * filesystem is gone, so set the EOF flag and schedule
2196 * the knote for deletion.
2198 if (hint == NOTE_REVOKE)
2199 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2206 filt_ufsvnode(struct knote *kn, long hint)
2208 if (kn->kn_sfflags & hint)
2209 kn->kn_fflags |= hint;
2210 if (hint == NOTE_REVOKE) {
2211 kn->kn_flags |= (EV_EOF | EV_NODATA);
2214 return (kn->kn_fflags != 0);
2217 /* Global vfs data structures for ufs. */
2218 static struct vop_ops ufs_vnode_vops = {
2219 .vop_default = vop_defaultop,
2220 .vop_fsync = (void *)ufs_missingop,
2221 .vop_read = (void *)ufs_missingop,
2222 .vop_reallocblks = (void *)ufs_missingop,
2223 .vop_write = (void *)ufs_missingop,
2224 .vop_access = ufs_access,
2225 .vop_advlock = ufs_advlock,
2226 .vop_bmap = ufs_bmap,
2227 .vop_old_lookup = ufs_lookup,
2228 .vop_close = ufs_close,
2229 .vop_old_create = ufs_create,
2230 .vop_getattr = ufs_getattr,
2231 .vop_inactive = ufs_inactive,
2232 .vop_old_link = ufs_link,
2233 .vop_old_mkdir = ufs_mkdir,
2234 .vop_old_mknod = ufs_mknod,
2235 .vop_open = vop_stdopen,
2236 .vop_pathconf = vop_stdpathconf,
2237 .vop_kqfilter = ufs_kqfilter,
2238 .vop_print = ufs_print,
2239 .vop_readdir = ufs_readdir,
2240 .vop_readlink = ufs_readlink,
2241 .vop_reclaim = ufs_reclaim,
2242 .vop_old_remove = ufs_remove,
2243 .vop_old_rename = ufs_rename,
2244 .vop_old_rmdir = ufs_rmdir,
2245 .vop_setattr = ufs_setattr,
2246 .vop_markatime = ufs_markatime,
2247 .vop_strategy = ufs_strategy,
2248 .vop_old_symlink = ufs_symlink,
2249 .vop_old_whiteout = ufs_whiteout
2252 static struct vop_ops ufs_spec_vops = {
2253 .vop_default = vop_defaultop,
2254 .vop_fsync = (void *)ufs_missingop,
2255 .vop_access = ufs_access,
2256 .vop_close = ufs_close,
2257 .vop_getattr = ufs_getattr,
2258 .vop_inactive = ufs_inactive,
2259 .vop_print = ufs_print,
2260 .vop_read = vop_stdnoread,
2261 .vop_reclaim = ufs_reclaim,
2262 .vop_setattr = ufs_setattr,
2263 .vop_markatime = ufs_markatime,
2264 .vop_write = vop_stdnowrite
2267 static struct vop_ops ufs_fifo_vops = {
2268 .vop_default = fifo_vnoperate,
2269 .vop_fsync = (void *)ufs_missingop,
2270 .vop_access = ufs_access,
2271 .vop_close = ufsfifo_close,
2272 .vop_getattr = ufs_getattr,
2273 .vop_inactive = ufs_inactive,
2274 .vop_kqfilter = ufsfifo_kqfilter,
2275 .vop_print = ufs_print,
2276 .vop_read = ufsfifo_read,
2277 .vop_reclaim = ufs_reclaim,
2278 .vop_setattr = ufs_setattr,
2279 .vop_markatime = ufs_markatime,
2280 .vop_write = ufsfifo_write
2283 VNODEOP_SET(ufs_vnode_vops);
2284 VNODEOP_SET(ufs_spec_vops);
2285 VNODEOP_SET(ufs_fifo_vops);
2291 ufs_vnoperate(struct vop_generic_args *ap)
2293 return (VOCALL(&ufs_vnode_vops, ap));
2297 * ufs_vnoperatefifo()
2300 ufs_vnoperatefifo(struct vop_generic_args *ap)
2302 return (VOCALL(&ufs_fifo_vops, ap));
2306 * ufs_vnoperatespec()
2309 ufs_vnoperatespec(struct vop_generic_args *ap)
2311 return (VOCALL(&ufs_spec_vops, ap));