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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
39 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $
40 * $DragonFly: src/sys/vfs/ufs/ufs_vnops.c,v 1.67 2008/09/28 05:04:22 dillon Exp $
43 #include "opt_quota.h"
44 #include "opt_suiddir.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
55 #include <sys/namei.h>
56 #include <sys/mount.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/malloc.h>
60 #include <sys/dirent.h>
61 #include <sys/lockf.h>
62 #include <sys/event.h>
65 #include <sys/file.h> /* XXX */
69 #include <vm/vm_extern.h>
71 #include <vfs/fifofs/fifo.h>
77 #include "ufs_extern.h"
78 #include "ffs_extern.h"
83 static int ufs_access (struct vop_access_args *);
84 static int ufs_advlock (struct vop_advlock_args *);
85 static int ufs_chmod (struct vnode *, int, struct ucred *);
86 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *);
87 static int ufs_close (struct vop_close_args *);
88 static int ufs_create (struct vop_old_create_args *);
89 static int ufs_getattr (struct vop_getattr_args *);
90 static int ufs_link (struct vop_old_link_args *);
91 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *);
92 static int ufs_markatime (struct vop_markatime_args *);
93 static int ufs_missingop (struct vop_generic_args *ap);
94 static int ufs_mkdir (struct vop_old_mkdir_args *);
95 static int ufs_mknod (struct vop_old_mknod_args *);
96 static int ufs_mmap (struct vop_mmap_args *);
97 static int ufs_print (struct vop_print_args *);
98 static int ufs_readdir (struct vop_readdir_args *);
99 static int ufs_readlink (struct vop_readlink_args *);
100 static int ufs_remove (struct vop_old_remove_args *);
101 static int ufs_rename (struct vop_old_rename_args *);
102 static int ufs_rmdir (struct vop_old_rmdir_args *);
103 static int ufs_setattr (struct vop_setattr_args *);
104 static int ufs_strategy (struct vop_strategy_args *);
105 static int ufs_symlink (struct vop_old_symlink_args *);
106 static int ufs_whiteout (struct vop_old_whiteout_args *);
107 static int ufsfifo_close (struct vop_close_args *);
108 static int ufsfifo_kqfilter (struct vop_kqfilter_args *);
109 static int ufsfifo_read (struct vop_read_args *);
110 static int ufsfifo_write (struct vop_write_args *);
111 static int ufsspec_close (struct vop_close_args *);
112 static int ufsspec_read (struct vop_read_args *);
113 static int ufsspec_write (struct vop_write_args *);
114 static int ufsspec_getattr (struct vop_getattr_args *);
115 static int filt_ufsread (struct knote *kn, long hint);
116 static int filt_ufswrite (struct knote *kn, long hint);
117 static int filt_ufsvnode (struct knote *kn, long hint);
118 static void filt_ufsdetach (struct knote *kn);
119 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
125 #define SETHIGH(q, h) { \
128 tmp.val[_QUAD_HIGHWORD] = (h); \
131 #define SETLOW(q, l) { \
134 tmp.val[_QUAD_LOWWORD] = (l); \
137 #define VN_KNOTE(vp, b) \
138 KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, (b))
140 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
143 * A virgin directory (no blushing please).
145 static struct dirtemplate mastertemplate = {
146 0, 12, DT_DIR, 1, ".",
147 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
149 static struct odirtemplate omastertemplate = {
151 0, DIRBLKSIZ - 12, 2, ".."
155 ufs_itimes(struct vnode *vp)
161 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
163 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
164 ip->i_flag |= IN_LAZYMOD;
166 ip->i_flag |= IN_MODIFIED;
167 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
169 if (ip->i_flag & IN_ACCESS) {
170 ip->i_atime = ts.tv_sec;
171 ip->i_atimensec = ts.tv_nsec;
173 if (ip->i_flag & IN_UPDATE) {
174 ip->i_mtime = ts.tv_sec;
175 ip->i_mtimensec = ts.tv_nsec;
178 if (ip->i_flag & IN_CHANGE) {
179 ip->i_ctime = ts.tv_sec;
180 ip->i_ctimensec = ts.tv_nsec;
183 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
187 * Create a regular file
189 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
190 * struct componentname *a_cnp, struct vattr *a_vap)
194 ufs_create(struct vop_old_create_args *ap)
199 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
200 ap->a_dvp, ap->a_vpp, ap->a_cnp);
203 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
210 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
211 * struct componentname *a_cnp, struct vattr *a_vap)
216 ufs_mknod(struct vop_old_mknod_args *ap)
218 struct vattr *vap = ap->a_vap;
219 struct vnode **vpp = ap->a_vpp;
225 * UFS cannot represent the entire major/minor range supported by
228 if (vap->va_rmajor != VNOVAL &&
229 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
233 /* no special directory support */
234 if (vap->va_type == VDIR)
237 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
238 ap->a_dvp, vpp, ap->a_cnp);
241 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
243 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
244 if (vap->va_rmajor != VNOVAL) {
246 * Want to be able to use this to make badblock
247 * inodes, so don't truncate the dev number.
249 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
252 * Remove inode, then reload it through VFS_VGET so it is
253 * checked to see if it is an alias of an existing entry in
256 (*vpp)->v_type = VNON;
257 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
258 vgone_vxlocked(*vpp);
260 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp);
271 * Update the times on the inode.
273 * ufs_close(struct vnode *a_vp, int a_fflag)
278 ufs_close(struct vop_close_args *ap)
280 struct vnode *vp = ap->a_vp;
282 if (vp->v_sysref.refcnt > 1)
284 return (vop_stdclose(ap));
288 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
292 ufs_access(struct vop_access_args *ap)
294 struct vnode *vp = ap->a_vp;
295 struct inode *ip = VTOI(vp);
296 struct ucred *cred = ap->a_cred;
297 mode_t mask, mode = ap->a_mode;
305 * Disallow write attempts on read-only filesystems;
306 * unless the file is a socket, fifo, or a block or
307 * character device resident on the filesystem.
310 switch (vp->v_type) {
314 if (vp->v_mount->mnt_flag & MNT_RDONLY)
317 if ((error = ufs_getinoquota(ip)) != 0)
326 #if 0 /* handled by kernel now */
327 /* If immutable bit set, nobody gets to write it. */
328 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
332 /* Otherwise, user id 0 always gets access. */
333 if (cred->cr_uid == 0)
338 /* Otherwise, check the owner. */
339 if (cred->cr_uid == ip->i_uid) {
346 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
349 /* Otherwise, check the groups. */
350 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
351 if (ip->i_gid == *gp) {
358 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
361 /* Otherwise, check everyone else. */
368 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
372 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
377 ufs_getattr(struct vop_getattr_args *ap)
379 struct vnode *vp = ap->a_vp;
380 struct inode *ip = VTOI(vp);
381 struct vattr *vap = ap->a_vap;
384 * This may cause i_fsmid to be updated even if no change (0)
385 * is returned, but we should only write out the inode if non-zero
386 * is returned and if the mount is read-write.
388 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
389 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
391 ip->i_flag |= IN_LAZYMOD;
396 * Copy from inode table
398 vap->va_fsid = dev2udev(ip->i_dev);
399 vap->va_fileid = ip->i_number;
400 vap->va_mode = ip->i_mode & ~IFMT;
401 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
402 ip->i_effnlink : ip->i_nlink;
403 vap->va_uid = ip->i_uid;
404 vap->va_gid = ip->i_gid;
405 vap->va_rmajor = umajor(ip->i_rdev);
406 vap->va_rminor = uminor(ip->i_rdev);
407 vap->va_size = ip->i_din.di_size;
408 vap->va_atime.tv_sec = ip->i_atime;
409 vap->va_atime.tv_nsec = ip->i_atimensec;
410 vap->va_mtime.tv_sec = ip->i_mtime;
411 vap->va_mtime.tv_nsec = ip->i_mtimensec;
412 vap->va_ctime.tv_sec = ip->i_ctime;
413 vap->va_ctime.tv_nsec = ip->i_ctimensec;
414 vap->va_flags = ip->i_flags;
415 vap->va_gen = ip->i_gen;
416 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
417 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
418 vap->va_type = IFTOVT(ip->i_mode);
419 vap->va_filerev = ip->i_modrev;
420 vap->va_fsmid = ip->i_fsmid;
426 ufs_markatime(struct vop_markatime_args *ap)
428 struct vnode *vp = ap->a_vp;
429 struct inode *ip = VTOI(vp);
431 if (vp->v_mount->mnt_flag & MNT_RDONLY)
433 if (vp->v_mount->mnt_flag & MNT_NOATIME)
435 ip->i_flag |= IN_ACCESS;
436 VN_KNOTE(vp, NOTE_ATTRIB);
441 * Set attribute vnode op. called from several syscalls
443 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
444 * struct ucred *a_cred)
448 ufs_setattr(struct vop_setattr_args *ap)
450 struct vattr *vap = ap->a_vap;
451 struct vnode *vp = ap->a_vp;
452 struct inode *ip = VTOI(vp);
453 struct ucred *cred = ap->a_cred;
457 * Check for unsettable attributes.
459 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
460 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
461 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
462 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
465 if (vap->va_flags != VNOVAL) {
466 if (vp->v_mount->mnt_flag & MNT_RDONLY)
468 if (cred->cr_uid != ip->i_uid &&
469 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
472 * Note that a root chflags becomes a user chflags when
473 * we are jailed, unless the jail.chflags_allowed sysctl
476 if (cred->cr_uid == 0 &&
477 (!jailed(cred) || jail_chflags_allowed)) {
479 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
482 ip->i_flags = vap->va_flags;
485 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
486 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
488 ip->i_flags &= SF_SETTABLE;
489 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
491 ip->i_flag |= IN_CHANGE;
492 if (vap->va_flags & (IMMUTABLE | APPEND))
495 if (ip->i_flags & (IMMUTABLE | APPEND))
498 * Go through the fields and update iff not VNOVAL.
500 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
501 if (vp->v_mount->mnt_flag & MNT_RDONLY)
503 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
506 if (vap->va_size != VNOVAL) {
508 * Disallow write attempts on read-only filesystems;
509 * unless the file is a socket, fifo, or a block or
510 * character device resident on the filesystem.
512 switch (vp->v_type) {
517 if (vp->v_mount->mnt_flag & MNT_RDONLY)
523 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
527 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
528 if (vp->v_mount->mnt_flag & MNT_RDONLY)
530 if (cred->cr_uid != ip->i_uid &&
531 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) &&
532 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
533 (error = VOP_ACCESS(vp, VWRITE, cred))))
535 if (vap->va_atime.tv_sec != VNOVAL)
536 ip->i_flag |= IN_ACCESS;
537 if (vap->va_mtime.tv_sec != VNOVAL)
538 ip->i_flag |= IN_CHANGE | IN_UPDATE;
540 if (vap->va_atime.tv_sec != VNOVAL) {
541 ip->i_atime = vap->va_atime.tv_sec;
542 ip->i_atimensec = vap->va_atime.tv_nsec;
544 if (vap->va_mtime.tv_sec != VNOVAL) {
545 ip->i_mtime = vap->va_mtime.tv_sec;
546 ip->i_mtimensec = vap->va_mtime.tv_nsec;
548 error = ffs_update(vp, 0);
553 if (vap->va_mode != (mode_t)VNOVAL) {
554 if (vp->v_mount->mnt_flag & MNT_RDONLY)
556 error = ufs_chmod(vp, (int)vap->va_mode, cred);
558 VN_KNOTE(vp, NOTE_ATTRIB);
563 * Change the mode on a file.
564 * Inode must be locked before calling.
567 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
569 struct inode *ip = VTOI(vp);
571 mode_t cur_mode = ip->i_mode;
573 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid,
578 if (cred->cr_uid != ip->i_uid) {
579 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
584 if (vp->v_type != VDIR && (mode & S_ISTXT))
586 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
590 ip->i_mode = cur_mode;
591 ip->i_flag |= IN_CHANGE;
596 * Perform chown operation on inode ip;
597 * inode must be locked prior to call.
600 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
602 struct inode *ip = VTOI(vp);
611 if (uid == (uid_t)VNOVAL)
613 if (gid == (gid_t)VNOVAL)
616 * If we don't own the file, are trying to change the owner
617 * of the file, or are not a member of the target group,
618 * the caller must be superuser or the call fails.
620 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
621 (gid != ip->i_gid && !(cred->cr_gid == gid ||
622 groupmember((gid_t)gid, cred)))) &&
623 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
628 if ((error = ufs_getinoquota(ip)) != 0)
631 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
632 ip->i_dquot[USRQUOTA] = NODQUOT;
635 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
636 ip->i_dquot[GRPQUOTA] = NODQUOT;
638 change = ip->i_blocks;
639 (void) ufs_chkdq(ip, -change, cred, CHOWN);
640 (void) ufs_chkiq(ip, -1, cred, CHOWN);
641 for (i = 0; i < MAXQUOTAS; i++) {
642 ufs_dqrele(vp, ip->i_dquot[i]);
643 ip->i_dquot[i] = NODQUOT;
649 if ((error = ufs_getinoquota(ip)) == 0) {
651 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
652 ip->i_dquot[USRQUOTA] = NODQUOT;
655 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
656 ip->i_dquot[GRPQUOTA] = NODQUOT;
658 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
659 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
662 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
664 for (i = 0; i < MAXQUOTAS; i++) {
665 ufs_dqrele(vp, ip->i_dquot[i]);
666 ip->i_dquot[i] = NODQUOT;
671 if (ufs_getinoquota(ip) == 0) {
673 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
674 ip->i_dquot[USRQUOTA] = NODQUOT;
677 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
678 ip->i_dquot[GRPQUOTA] = NODQUOT;
680 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
681 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
682 (void) ufs_getinoquota(ip);
686 if (ufs_getinoquota(ip))
687 panic("ufs_chown: lost quota");
689 ip->i_flag |= IN_CHANGE;
690 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
691 ip->i_mode &= ~(ISUID | ISGID);
698 * NB Currently unsupported.
700 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
705 ufs_mmap(struct vop_mmap_args *ap)
711 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
712 * struct componentname *a_cnp)
716 ufs_remove(struct vop_old_remove_args *ap)
719 struct vnode *vp = ap->a_vp;
720 struct vnode *dvp = ap->a_dvp;
724 #if 0 /* handled by kernel now */
725 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
726 (VTOI(dvp)->i_flags & APPEND)) {
731 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
732 VN_KNOTE(vp, NOTE_DELETE);
733 VN_KNOTE(dvp, NOTE_WRITE);
743 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
744 * struct componentname *a_cnp)
748 ufs_link(struct vop_old_link_args *ap)
750 struct vnode *vp = ap->a_vp;
751 struct vnode *tdvp = ap->a_tdvp;
752 struct componentname *cnp = ap->a_cnp;
754 struct direct newdir;
757 if (tdvp->v_mount != vp->v_mount) {
761 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
765 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
769 #if 0 /* handled by kernel now, also DragonFly allows this */
770 if (ip->i_flags & (IMMUTABLE | APPEND)) {
777 ip->i_flag |= IN_CHANGE;
778 if (DOINGSOFTDEP(vp))
779 softdep_change_linkcnt(ip);
780 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
782 ufs_makedirentry(ip, cnp, &newdir);
783 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
789 ip->i_flag |= IN_CHANGE;
790 if (DOINGSOFTDEP(vp))
791 softdep_change_linkcnt(ip);
797 VN_KNOTE(vp, NOTE_LINK);
798 VN_KNOTE(tdvp, NOTE_WRITE);
803 * whiteout vnode call
805 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
809 ufs_whiteout(struct vop_old_whiteout_args *ap)
811 struct vnode *dvp = ap->a_dvp;
812 struct componentname *cnp = ap->a_cnp;
813 struct direct newdir;
816 switch (ap->a_flags) {
818 /* 4.4 format directories support whiteout operations */
819 if (dvp->v_mount->mnt_maxsymlinklen > 0)
824 /* create a new directory whiteout */
826 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
827 panic("ufs_whiteout: old format filesystem");
831 newdir.d_namlen = cnp->cn_namelen;
832 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
833 newdir.d_type = DT_WHT;
834 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
838 /* remove an existing directory whiteout */
840 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
841 panic("ufs_whiteout: old format filesystem");
844 cnp->cn_flags &= ~CNP_DOWHITEOUT;
845 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
848 panic("ufs_whiteout: unknown op");
854 * Rename system call.
855 * rename("foo", "bar");
858 * link("foo", "bar");
860 * but ``atomically''. Can't do full commit without saving state in the
861 * inode on disk which isn't feasible at this time. Best we can do is
862 * always guarantee the target exists.
864 * Basic algorithm is:
866 * 1) Bump link count on source while we're linking it to the
867 * target. This also ensure the inode won't be deleted out
868 * from underneath us while we work (it may be truncated by
869 * a concurrent `trunc' or `open' for creation).
870 * 2) Link source to destination. If destination already exists,
872 * 3) Unlink source reference to inode if still around. If a
873 * directory was moved and the parent of the destination
874 * is different from the source, patch the ".." entry in the
877 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
878 * struct componentname *a_fcnp, struct vnode *a_tdvp,
879 * struct vnode *a_tvp, struct componentname *a_tcnp)
883 ufs_rename(struct vop_old_rename_args *ap)
885 struct vnode *tvp = ap->a_tvp;
886 struct vnode *tdvp = ap->a_tdvp;
887 struct vnode *fvp = ap->a_fvp;
888 struct vnode *fdvp = ap->a_fdvp;
889 struct componentname *tcnp = ap->a_tcnp;
890 struct componentname *fcnp = ap->a_fcnp;
891 struct inode *ip, *xp, *dp;
892 struct direct newdir;
893 ino_t oldparent = 0, newparent = 0;
894 int doingdirectory = 0;
895 int error = 0, ioflag;
898 * Check for cross-device rename.
900 if ((fvp->v_mount != tdvp->v_mount) ||
901 (tvp && (fvp->v_mount != tvp->v_mount))) {
915 #if 0 /* handled by kernel now */
916 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
917 (VTOI(tdvp)->i_flags & APPEND))) {
924 * Renaming a file to itself has no effect. The upper layers should
925 * not call us in that case. Temporarily just warn if they do.
928 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
933 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
937 * Note: now that fvp is locked we have to be sure to unlock it before
938 * using the 'abortit' target.
942 if (ip->i_nlink >= LINK_MAX) {
947 #if 0 /* handled by kernel now */
948 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
949 || (dp->i_flags & APPEND)) {
955 if ((ip->i_mode & IFMT) == IFDIR) {
957 * Avoid ".", "..", and aliases of "." for obvious reasons.
959 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
960 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
961 (ip->i_flag & IN_RENAME)) {
966 ip->i_flag |= IN_RENAME;
967 oldparent = dp->i_number;
970 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
973 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
974 * Cleanup fvp requirements so we can unlock it.
976 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
977 * is setup we can use the 'bad' target if we unlock fvp. We cannot
978 * use the abortit target anymore because of IN_RENAME.
987 * 1) Bump link count while we're moving stuff
988 * around. If we crash somewhere before
989 * completing our work, the link count
990 * may be wrong, but correctable.
994 ip->i_flag |= IN_CHANGE;
995 if (DOINGSOFTDEP(fvp))
996 softdep_change_linkcnt(ip);
997 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
998 DOINGASYNC(fvp)))) != 0) {
1004 * If ".." must be changed (ie the directory gets a new
1005 * parent) then the source directory must not be in the
1006 * directory heirarchy above the target, as this would
1007 * orphan everything below the source directory. Also
1008 * the user must have write permission in the source so
1009 * as to be able to change "..". We must repeat the call
1010 * to namei, as the parent directory is unlocked by the
1011 * call to checkpath().
1013 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
1017 * We are now back to where we were in that fvp, fdvp are unlocked
1018 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
1019 * set. Only the bad target or, if we clean up tvp and tdvp, the
1020 * out target, may be used.
1022 if (oldparent != dp->i_number)
1023 newparent = dp->i_number;
1024 if (doingdirectory && newparent) {
1025 if (error) /* write access check above */
1029 * Once we start messing with tvp and tdvp we cannot use the
1030 * 'bad' target, only finish cleaning tdvp and tvp up and
1031 * use the 'out' target.
1033 * This cleans up tvp.
1041 * This is a real mess. ufs_checkpath vput's the target
1042 * directory so retain an extra ref and note that tdvp will
1043 * lose its lock on return. This leaves us with one good
1044 * ref after ufs_checkpath returns.
1047 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
1048 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1055 * relookup no longer messes with tdvp's refs. tdvp must be
1056 * unlocked on entry and will be locked on a successful
1059 error = relookup(tdvp, &tvp, tcnp);
1061 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1067 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1074 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1075 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1076 * be set if doingdirectory. This means we can use the 'bad' target
1081 * 2) If target doesn't exist, link the target
1082 * to the source and unlink the source.
1083 * Otherwise, rewrite the target directory
1084 * entry to reference the source inode and
1085 * expunge the original entry's existence.
1088 if (dp->i_dev != ip->i_dev)
1089 panic("ufs_rename: EXDEV");
1091 * Account for ".." in new directory.
1092 * When source and destination have the same
1093 * parent we don't fool with the link count.
1095 if (doingdirectory && newparent) {
1096 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1102 dp->i_flag |= IN_CHANGE;
1103 if (DOINGSOFTDEP(tdvp))
1104 softdep_change_linkcnt(dp);
1105 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1110 ufs_makedirentry(ip, tcnp, &newdir);
1111 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1113 if (doingdirectory && newparent) {
1116 dp->i_flag |= IN_CHANGE;
1117 if (DOINGSOFTDEP(tdvp))
1118 softdep_change_linkcnt(dp);
1119 (void)ffs_update(tdvp, 1);
1123 VN_KNOTE(tdvp, NOTE_WRITE);
1126 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1127 panic("ufs_rename: EXDEV");
1129 * Short circuit rename(foo, foo).
1131 if (xp->i_number == ip->i_number)
1132 panic("ufs_rename: same file");
1134 * If the parent directory is "sticky", then the user must
1135 * own the parent directory, or the destination of the rename,
1136 * otherwise the destination may not be changed (except by
1137 * root). This implements append-only directories.
1139 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1140 tcnp->cn_cred->cr_uid != dp->i_uid &&
1141 xp->i_uid != tcnp->cn_cred->cr_uid) {
1146 * Target must be empty if a directory and have no links
1147 * to it. Also, ensure source and target are compatible
1148 * (both directories, or both not directories).
1150 * Purge the file or directory being replaced from the
1153 if ((xp->i_mode&IFMT) == IFDIR) {
1154 if ((xp->i_effnlink > 2) ||
1155 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1159 if (!doingdirectory) {
1163 /* cache_purge removed - handled by VFS compat layer */
1164 } else if (doingdirectory == 0) {
1165 /* cache_purge removed - handled by VFS compat layer */
1171 * note: inode passed to ufs_dirrewrite() is 0 for a
1172 * non-directory file rename, 1 for a directory rename
1173 * in the same directory, and > 1 for an inode representing
1174 * the new directory.
1176 error = ufs_dirrewrite(dp, xp, ip->i_number,
1178 (doingdirectory && newparent) ?
1179 newparent : (ino_t)doingdirectory);
1182 if (doingdirectory) {
1185 if (DOINGSOFTDEP(tdvp))
1186 softdep_change_linkcnt(dp);
1189 if (DOINGSOFTDEP(tvp))
1190 softdep_change_linkcnt(xp);
1192 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1194 * Truncate inode. The only stuff left in the directory
1195 * is "." and "..". The "." reference is inconsequential
1196 * since we are quashing it. We have removed the "."
1197 * reference and the reference in the parent directory,
1198 * but there may be other hard links. The soft
1199 * dependency code will arrange to do these operations
1200 * after the parent directory entry has been deleted on
1201 * disk, so when running with that code we avoid doing
1206 dp->i_flag |= IN_CHANGE;
1209 xp->i_flag |= IN_CHANGE;
1210 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1211 error = ffs_truncate(tvp, (off_t)0, ioflag,
1216 VN_KNOTE(tdvp, NOTE_WRITE);
1218 VN_KNOTE(tvp, NOTE_DELETE);
1224 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1225 * unlocked) remain. We are about to overwrite fvp but we have to
1226 * keep 'ip' intact so we cannot release the old fvp, which is still
1227 * refd and accessible via ap->a_fvp.
1229 * This means we cannot use either 'bad' or 'out' to cleanup any
1234 * 3) Unlink the source.
1236 fcnp->cn_flags &= ~CNP_MODMASK;
1237 fcnp->cn_flags |= CNP_LOCKPARENT;
1238 error = relookup(fdvp, &fvp, fcnp);
1239 if (error || fvp == NULL) {
1241 * From name has disappeared. IN_RENAME will not be set if
1242 * we get past the panic so we don't have to clean it up.
1245 panic("ufs_rename: lost dir entry");
1247 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1253 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1256 * fdvp and fvp are locked.
1262 * Ensure that the directory entry still exists and has not
1263 * changed while the new name has been entered. If the source is
1264 * a file then the entry may have been unlinked or renamed. In
1265 * either case there is no further work to be done. If the source
1266 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1267 * flag ensures that it cannot be moved by another rename or removed
1268 * by a rmdir. Cleanup IN_RENAME.
1272 panic("ufs_rename: lost dir entry");
1275 * If the source is a directory with a
1276 * new parent, the link count of the old
1277 * parent directory must be decremented
1278 * and ".." set to point to the new parent.
1280 if (doingdirectory && newparent) {
1281 xp->i_offset = mastertemplate.dot_reclen;
1282 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1283 /* cache_purge removed - handled by VFS compat layer */
1285 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1286 xp->i_flag &= ~IN_RENAME;
1289 VN_KNOTE(fvp, NOTE_RENAME);
1301 ip->i_flag &= ~IN_RENAME;
1302 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1305 ip->i_flag |= IN_CHANGE;
1306 ip->i_flag &= ~IN_RENAME;
1307 if (DOINGSOFTDEP(fvp))
1308 softdep_change_linkcnt(ip);
1319 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1320 * struct componentname *a_cnp, struct vattr *a_vap)
1324 ufs_mkdir(struct vop_old_mkdir_args *ap)
1326 struct vnode *dvp = ap->a_dvp;
1327 struct vattr *vap = ap->a_vap;
1328 struct componentname *cnp = ap->a_cnp;
1329 struct inode *ip, *dp;
1332 struct dirtemplate dirtemplate, *dtp;
1333 struct direct newdir;
1338 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1342 dmode = vap->va_mode & 0777;
1345 * Must simulate part of ufs_makeinode here to acquire the inode,
1346 * but not have it entered in the parent directory. The entry is
1347 * made later after writing "." and ".." entries.
1349 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1353 ip->i_gid = dp->i_gid;
1357 struct ucred ucred, *ucp;
1361 * If we are hacking owners here, (only do this where told to)
1362 * and we are not giving it TO root, (would subvert quotas)
1363 * then go ahead and give it to the other user.
1364 * The new directory also inherits the SUID bit.
1365 * If user's UID and dir UID are the same,
1366 * 'give it away' so that the SUID is still forced on.
1368 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1369 (dp->i_mode & ISUID) && dp->i_uid) {
1371 ip->i_uid = dp->i_uid;
1373 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1375 * Make sure the correct user gets charged
1377 * Make a dummy credential for the victim.
1378 * XXX This seems to never be accessed out of
1379 * our context so a stack variable is ok.
1382 ucred.cr_uid = ip->i_uid;
1383 ucred.cr_ngroups = 1;
1384 ucred.cr_groups[0] = dp->i_gid;
1389 ip->i_uid = cnp->cn_cred->cr_uid;
1391 if ((error = ufs_getinoquota(ip)) ||
1392 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1393 ffs_vfree(tvp, ip->i_number, dmode);
1399 #else /* !SUIDDIR */
1400 ip->i_uid = cnp->cn_cred->cr_uid;
1402 if ((error = ufs_getinoquota(ip)) ||
1403 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1404 ffs_vfree(tvp, ip->i_number, dmode);
1409 #endif /* !SUIDDIR */
1410 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1412 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1415 if (DOINGSOFTDEP(tvp))
1416 softdep_change_linkcnt(ip);
1417 if (cnp->cn_flags & CNP_ISWHITEOUT)
1418 ip->i_flags |= UF_OPAQUE;
1421 * Bump link count in parent directory to reflect work done below.
1422 * Should be done before reference is created so cleanup is
1423 * possible if we crash.
1427 dp->i_flag |= IN_CHANGE;
1428 if (DOINGSOFTDEP(dvp))
1429 softdep_change_linkcnt(dp);
1430 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1435 * The vnode must have a VM object in order to issue buffer cache
1438 vinitvmio(tvp, DIRBLKSIZ);
1441 * Initialize directory with "." and ".." from static template.
1443 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1444 dtp = &mastertemplate;
1446 dtp = (struct dirtemplate *)&omastertemplate;
1448 dirtemplate.dot_ino = ip->i_number;
1449 dirtemplate.dotdot_ino = dp->i_number;
1450 vnode_pager_setsize(tvp, DIRBLKSIZ);
1451 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1454 ip->i_size = DIRBLKSIZ;
1455 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1456 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1457 if (DOINGSOFTDEP(tvp)) {
1459 * Ensure that the entire newly allocated block is a
1460 * valid directory so that future growth within the
1461 * block does not have to ensure that the block is
1462 * written before the inode.
1465 while (blkoff < bp->b_bcount) {
1467 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1468 blkoff += DIRBLKSIZ;
1471 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1472 DOINGASYNC(tvp)))) != 0) {
1477 * Directory set up, now install its entry in the parent directory.
1479 * If we are not doing soft dependencies, then we must write out the
1480 * buffer containing the new directory body before entering the new
1481 * name in the parent. If we are doing soft dependencies, then the
1482 * buffer containing the new directory body will be passed to and
1483 * released in the soft dependency code after the code has attached
1484 * an appropriate ordering dependency to the buffer which ensures that
1485 * the buffer is written before the new name is written in the parent.
1487 if (DOINGASYNC(dvp))
1489 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1491 ufs_makedirentry(ip, cnp, &newdir);
1492 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1496 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1501 dp->i_flag |= IN_CHANGE;
1502 if (DOINGSOFTDEP(dvp))
1503 softdep_change_linkcnt(dp);
1505 * No need to do an explicit VOP_TRUNCATE here, vrele will
1506 * do this for us because we set the link count to 0.
1510 ip->i_flag |= IN_CHANGE;
1511 if (DOINGSOFTDEP(tvp))
1512 softdep_change_linkcnt(ip);
1520 * Rmdir system call.
1522 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1523 * struct componentname *a_cnp)
1527 ufs_rmdir(struct vop_old_rmdir_args *ap)
1529 struct vnode *vp = ap->a_vp;
1530 struct vnode *dvp = ap->a_dvp;
1531 struct componentname *cnp = ap->a_cnp;
1532 struct inode *ip, *dp;
1539 * Do not remove a directory that is in the process of being renamed.
1540 * Verify the directory is empty (and valid). Rmdir ".." will not be
1541 * valid since ".." will contain a reference to the current directory
1542 * and thus be non-empty. Do not allow the removal of mounted on
1543 * directories (this can happen when an NFS exported filesystem
1544 * tries to remove a locally mounted on directory).
1547 if (ip->i_flag & IN_RENAME) {
1551 if (ip->i_effnlink != 2 ||
1552 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1556 #if 0 /* handled by kernel now */
1557 if ((dp->i_flags & APPEND)
1558 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1564 * Delete reference to directory before purging
1565 * inode. If we crash in between, the directory
1566 * will be reattached to lost+found,
1570 if (DOINGSOFTDEP(vp)) {
1571 softdep_change_linkcnt(dp);
1572 softdep_change_linkcnt(ip);
1574 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1578 if (DOINGSOFTDEP(vp)) {
1579 softdep_change_linkcnt(dp);
1580 softdep_change_linkcnt(ip);
1584 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1586 * Truncate inode. The only stuff left in the directory is "." and
1587 * "..". The "." reference is inconsequential since we are quashing
1588 * it. The soft dependency code will arrange to do these operations
1589 * after the parent directory entry has been deleted on disk, so
1590 * when running with that code we avoid doing them now.
1592 if (!DOINGSOFTDEP(vp)) {
1594 dp->i_flag |= IN_CHANGE;
1596 ip->i_flag |= IN_CHANGE;
1597 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1598 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1600 /* cache_purge removed - handled by VFS compat layer */
1602 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1603 if (ip->i_dirhash != NULL)
1604 ufsdirhash_free(ip);
1607 VN_KNOTE(vp, NOTE_DELETE);
1612 * symlink -- make a symbolic link
1614 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1615 * struct componentname *a_cnp, struct vattr *a_vap,
1620 ufs_symlink(struct vop_old_symlink_args *ap)
1622 struct vnode *vp, **vpp = ap->a_vpp;
1626 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1630 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1632 len = strlen(ap->a_target);
1633 if (len < vp->v_mount->mnt_maxsymlinklen) {
1635 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1637 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1640 * Make sure we have a VM object in order to use
1643 if (vp->v_object == NULL)
1645 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1646 UIO_SYSSPACE, IO_NODELOCKED,
1647 ap->a_cnp->cn_cred, NULL);
1655 * Vnode op for reading directories.
1657 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1658 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1662 ufs_readdir(struct vop_readdir_args *ap)
1664 struct uio *uio = ap->a_uio;
1665 struct vnode *vp = ap->a_vp;
1670 int offset; /* offset into buffer cache buffer */
1671 int eoffset; /* end of buffer clipped to file EOF */
1672 int pickup; /* pickup point */
1677 if (uio->uio_offset < 0)
1680 * Guess the number of cookies needed. Make sure we compute at
1681 * least 1, and no more then a reasonable limit.
1683 if (ap->a_ncookies) {
1684 ncookies = uio->uio_resid / 16 + 1;
1685 if (ncookies > 1024)
1687 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1689 ncookies = -1; /* force conditionals below */
1694 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1700 if (uio->uio_offset >= VTOI(vp)->i_size) {
1703 if (ap->a_ncookies) {
1704 *ap->a_ncookies = cookie_index;
1705 *ap->a_cookies = cookies;
1711 * Loop until we run out of cookies, we run out of user buffer,
1712 * or we hit the directory EOF.
1714 * Always start scans at the beginning of the buffer, don't trust
1715 * the offset supplied by userland.
1717 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1718 pickup = (int)(uio->uio_offset - bp->b_loffset);
1721 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1722 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1724 eoffset = bp->b_bcount;
1726 while (offset < eoffset) {
1727 dp = (struct direct *)(bp->b_data + offset);
1728 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1729 offset + dp->d_reclen > bp->b_bcount) {
1733 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1737 if (offset < pickup) {
1738 offset += dp->d_reclen;
1741 #if BYTE_ORDER == LITTLE_ENDIAN
1743 retval = vop_write_dirent(&error, uio,
1744 dp->d_ino, dp->d_namlen, dp->d_type,
1749 retval = vop_write_dirent(&error, uio,
1750 dp->d_ino, dp->d_type, dp->d_namlen,
1756 cookies[cookie_index] = bp->b_loffset + offset;
1758 offset += dp->d_reclen;
1759 if (cookie_index == ncookies)
1764 * This will align the next loop to the beginning of the
1765 * next block, and pickup will calculate to 0.
1767 uio->uio_offset = bp->b_loffset + offset;
1770 if (retval || error || cookie_index == ncookies ||
1771 uio->uio_offset >= VTOI(vp)->i_size) {
1776 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1779 * Report errors only if we didn't manage to read anything
1781 if (error && cookie_index == 0) {
1783 kfree(cookies, M_TEMP);
1784 *ap->a_ncookies = 0;
1785 *ap->a_cookies = NULL;
1790 *ap->a_ncookies = cookie_index;
1791 *ap->a_cookies = cookies;
1800 * Return target name of a symbolic link
1802 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1806 ufs_readlink(struct vop_readlink_args *ap)
1808 struct vnode *vp = ap->a_vp;
1809 struct inode *ip = VTOI(vp);
1813 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1814 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1815 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1820 * Perform the equivalent of an OPEN on vp so we can issue a
1823 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1827 * Calculate the logical to physical mapping if not done already,
1828 * then call the device strategy routine.
1830 * In order to be able to swap to a file, the VOP_BMAP operation may not
1831 * deadlock on memory. See ufs_bmap() for details.
1833 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1837 ufs_strategy(struct vop_strategy_args *ap)
1839 struct bio *bio = ap->a_bio;
1841 struct buf *bp = bio->bio_buf;
1842 struct vnode *vp = ap->a_vp;
1847 if (vp->v_type == VBLK || vp->v_type == VCHR)
1848 panic("ufs_strategy: spec");
1849 nbio = push_bio(bio);
1850 if (nbio->bio_offset == NOOFFSET) {
1851 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1852 NULL, NULL, bp->b_cmd);
1854 bp->b_error = error;
1855 bp->b_flags |= B_ERROR;
1856 /* I/O was never started on nbio, must biodone(bio) */
1860 if (nbio->bio_offset == NOOFFSET)
1863 if (nbio->bio_offset == NOOFFSET) {
1865 * We hit a hole in the file. The buffer has been zero-filled
1866 * so just biodone() it.
1870 vn_strategy(ip->i_devvp, nbio);
1876 * Print out the contents of an inode.
1878 * ufs_print(struct vnode *a_vp)
1882 ufs_print(struct vop_print_args *ap)
1884 struct vnode *vp = ap->a_vp;
1885 struct inode *ip = VTOI(vp);
1887 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1888 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1890 if (vp->v_type == VFIFO)
1892 lockmgr_printinfo(&vp->v_lock);
1898 * Read wrapper for special devices.
1900 * ufsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1901 * struct ucred *a_cred)
1905 ufsspec_read(struct vop_read_args *ap)
1912 resid = uio->uio_resid;
1913 error = VOCALL(&spec_vnode_vops, &ap->a_head);
1915 * The inode may have been revoked during the call, so it must not
1916 * be accessed blindly here or in the other wrapper functions.
1918 ip = VTOI(ap->a_vp);
1919 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1920 ip->i_flag |= IN_ACCESS;
1925 * Write wrapper for special devices.
1927 * ufsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1928 * struct ucred *a_cred)
1932 ufsspec_write(struct vop_write_args *ap)
1939 resid = uio->uio_resid;
1940 error = VOCALL(&spec_vnode_vops, &ap->a_head);
1941 ip = VTOI(ap->a_vp);
1942 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1943 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1948 * SPECFS's getattr will override fields as necessary, but does not fill
1949 * stuff in from scratch.
1953 ufsspec_getattr (struct vop_getattr_args *ap)
1957 error = ufs_getattr(ap);
1959 VOCALL(&spec_vnode_vops, &ap->a_head);
1964 * Close wrapper for special devices.
1966 * Update the times on the inode then do device close.
1968 * ufsspec_close(struct vnode *a_vp, int a_fflag)
1972 ufsspec_close(struct vop_close_args *ap)
1974 struct vnode *vp = ap->a_vp;
1976 if (vp->v_sysref.refcnt > 1)
1978 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1982 * Read wrapper for fifos.
1984 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1985 * struct ucred *a_cred)
1989 ufsfifo_read(struct vop_read_args *ap)
1996 resid = uio->uio_resid;
1997 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1998 ip = VTOI(ap->a_vp);
1999 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
2000 (uio->uio_resid != resid || (error == 0 && resid != 0)))
2001 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
2006 * Write wrapper for fifos.
2008 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
2009 * struct ucred *a_cred)
2013 ufsfifo_write(struct vop_write_args *ap)
2020 resid = uio->uio_resid;
2021 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2022 ip = VTOI(ap->a_vp);
2023 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
2024 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
2029 * Close wrapper for fifos.
2031 * Update the times on the inode then do device close.
2033 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
2037 ufsfifo_close(struct vop_close_args *ap)
2039 struct vnode *vp = ap->a_vp;
2041 if (vp->v_sysref.refcnt > 1)
2043 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2047 * Kqfilter wrapper for fifos.
2049 * Fall through to ufs kqfilter routines if needed
2053 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
2057 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2059 error = ufs_kqfilter(ap);
2064 * Advisory record locking support
2066 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
2071 ufs_advlock(struct vop_advlock_args *ap)
2073 struct inode *ip = VTOI(ap->a_vp);
2075 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
2079 * Initialize the vnode associated with a new inode, handle aliased
2082 * Make sure directories have their VM object now rather then later,
2083 * saving us from having to check on all the myrid directory VOPs
2084 * that might be executed without a VOP_OPEN being performed.
2087 ufs_vinit(struct mount *mntp, struct vnode **vpp)
2096 vp->v_type = IFTOVT(ip->i_mode);
2098 switch(vp->v_type) {
2101 vp->v_ops = &mntp->mnt_vn_spec_ops;
2102 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
2105 vp->v_ops = &mntp->mnt_vn_fifo_ops;
2109 vinitvmio(vp, ip->i_size);
2112 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
2113 vinitvmio(vp, ip->i_size);
2120 if (ip->i_number == ROOTINO)
2121 vp->v_flag |= VROOT;
2123 * Initialize modrev times
2125 getmicrouptime(&tv);
2126 SETHIGH(ip->i_modrev, tv.tv_sec);
2127 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
2133 * Allocate a new inode.
2137 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2138 struct componentname *cnp)
2140 struct inode *ip, *pdir;
2141 struct direct newdir;
2147 if ((mode & IFMT) == 0)
2150 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2154 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
2155 ip->i_gid = pdir->i_gid;
2159 struct ucred ucred, *ucp;
2163 * If we are not the owner of the directory,
2164 * and we are hacking owners here, (only do this where told to)
2165 * and we are not giving it TO root, (would subvert quotas)
2166 * then go ahead and give it to the other user.
2167 * Note that this drops off the execute bits for security.
2169 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2170 (pdir->i_mode & ISUID) &&
2171 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2172 ip->i_uid = pdir->i_uid;
2176 * Make sure the correct user gets charged
2178 * Quickly knock up a dummy credential for the victim.
2179 * XXX This seems to never be accessed out of our
2180 * context so a stack variable is ok.
2183 ucred.cr_uid = ip->i_uid;
2184 ucred.cr_ngroups = 1;
2185 ucred.cr_groups[0] = pdir->i_gid;
2189 ip->i_uid = cnp->cn_cred->cr_uid;
2192 if ((error = ufs_getinoquota(ip)) ||
2193 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2194 ffs_vfree(tvp, ip->i_number, mode);
2200 #else /* !SUIDDIR */
2201 ip->i_uid = cnp->cn_cred->cr_uid;
2203 if ((error = ufs_getinoquota(ip)) ||
2204 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2205 ffs_vfree(tvp, ip->i_number, mode);
2210 #endif /* !SUIDDIR */
2211 ip->i_fsmid = cache_getnewfsmid();
2212 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2214 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2217 if (DOINGSOFTDEP(tvp))
2218 softdep_change_linkcnt(ip);
2219 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2220 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2221 ip->i_mode &= ~ISGID;
2224 if (cnp->cn_flags & CNP_ISWHITEOUT)
2225 ip->i_flags |= UF_OPAQUE;
2228 * Regular files and directories need VM objects. Softlinks do
2229 * not (not immediately anyway).
2231 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2235 * Make sure inode goes to disk before directory entry.
2237 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2240 ufs_makedirentry(ip, cnp, &newdir);
2241 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2249 * Write error occurred trying to update the inode
2250 * or the directory so must deallocate the inode.
2254 ip->i_flag |= IN_CHANGE;
2255 if (DOINGSOFTDEP(tvp))
2256 softdep_change_linkcnt(ip);
2262 ufs_missingop(struct vop_generic_args *ap)
2264 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2265 return (EOPNOTSUPP);
2268 static struct filterops ufsread_filtops =
2269 { 1, NULL, filt_ufsdetach, filt_ufsread };
2270 static struct filterops ufswrite_filtops =
2271 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2272 static struct filterops ufsvnode_filtops =
2273 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2276 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2279 ufs_kqfilter(struct vop_kqfilter_args *ap)
2281 struct vnode *vp = ap->a_vp;
2282 struct knote *kn = ap->a_kn;
2285 switch (kn->kn_filter) {
2287 kn->kn_fop = &ufsread_filtops;
2290 kn->kn_fop = &ufswrite_filtops;
2293 kn->kn_fop = &ufsvnode_filtops;
2299 kn->kn_hook = (caddr_t)vp;
2301 lwkt_gettoken(&vlock, &vp->v_token);
2302 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2303 lwkt_reltoken(&vlock);
2309 filt_ufsdetach(struct knote *kn)
2311 struct vnode *vp = (struct vnode *)kn->kn_hook;
2314 lwkt_gettoken(&vlock, &vp->v_token);
2315 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2316 kn, knote, kn_selnext);
2317 lwkt_reltoken(&vlock);
2322 filt_ufsread(struct knote *kn, long hint)
2324 struct vnode *vp = (struct vnode *)kn->kn_hook;
2325 struct inode *ip = VTOI(vp);
2328 * filesystem is gone, so set the EOF flag and schedule
2329 * the knote for deletion.
2331 if (hint == NOTE_REVOKE) {
2332 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2336 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2337 return (kn->kn_data != 0);
2342 filt_ufswrite(struct knote *kn, long hint)
2345 * filesystem is gone, so set the EOF flag and schedule
2346 * the knote for deletion.
2348 if (hint == NOTE_REVOKE)
2349 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2356 filt_ufsvnode(struct knote *kn, long hint)
2358 if (kn->kn_sfflags & hint)
2359 kn->kn_fflags |= hint;
2360 if (hint == NOTE_REVOKE) {
2361 kn->kn_flags |= EV_EOF;
2364 return (kn->kn_fflags != 0);
2367 /* Global vfs data structures for ufs. */
2368 static struct vop_ops ufs_vnode_vops = {
2369 .vop_default = vop_defaultop,
2370 .vop_fsync = (void *)ufs_missingop,
2371 .vop_read = (void *)ufs_missingop,
2372 .vop_reallocblks = (void *)ufs_missingop,
2373 .vop_write = (void *)ufs_missingop,
2374 .vop_access = ufs_access,
2375 .vop_advlock = ufs_advlock,
2376 .vop_bmap = ufs_bmap,
2377 .vop_old_lookup = ufs_lookup,
2378 .vop_close = ufs_close,
2379 .vop_old_create = ufs_create,
2380 .vop_getattr = ufs_getattr,
2381 .vop_inactive = ufs_inactive,
2382 .vop_old_link = ufs_link,
2383 .vop_old_mkdir = ufs_mkdir,
2384 .vop_old_mknod = ufs_mknod,
2385 .vop_mmap = ufs_mmap,
2386 .vop_open = vop_stdopen,
2387 .vop_pathconf = vop_stdpathconf,
2388 .vop_poll = vop_stdpoll,
2389 .vop_kqfilter = ufs_kqfilter,
2390 .vop_print = ufs_print,
2391 .vop_readdir = ufs_readdir,
2392 .vop_readlink = ufs_readlink,
2393 .vop_reclaim = ufs_reclaim,
2394 .vop_old_remove = ufs_remove,
2395 .vop_old_rename = ufs_rename,
2396 .vop_old_rmdir = ufs_rmdir,
2397 .vop_setattr = ufs_setattr,
2398 .vop_markatime = ufs_markatime,
2399 .vop_strategy = ufs_strategy,
2400 .vop_old_symlink = ufs_symlink,
2401 .vop_old_whiteout = ufs_whiteout
2404 static struct vop_ops ufs_spec_vops = {
2405 .vop_default = spec_vnoperate,
2406 .vop_fsync = (void *)ufs_missingop,
2407 .vop_access = ufs_access,
2408 .vop_close = ufsspec_close,
2409 .vop_getattr = ufsspec_getattr,
2410 .vop_inactive = ufs_inactive,
2411 .vop_print = ufs_print,
2412 .vop_read = ufsspec_read,
2413 .vop_reclaim = ufs_reclaim,
2414 .vop_setattr = ufs_setattr,
2415 .vop_markatime = ufs_markatime,
2416 .vop_write = ufsspec_write
2419 static struct vop_ops ufs_fifo_vops = {
2420 .vop_default = fifo_vnoperate,
2421 .vop_fsync = (void *)ufs_missingop,
2422 .vop_access = ufs_access,
2423 .vop_close = ufsfifo_close,
2424 .vop_getattr = ufs_getattr,
2425 .vop_inactive = ufs_inactive,
2426 .vop_kqfilter = ufsfifo_kqfilter,
2427 .vop_print = ufs_print,
2428 .vop_read = ufsfifo_read,
2429 .vop_reclaim = ufs_reclaim,
2430 .vop_setattr = ufs_setattr,
2431 .vop_markatime = ufs_markatime,
2432 .vop_write = ufsfifo_write
2435 VNODEOP_SET(ufs_vnode_vops);
2436 VNODEOP_SET(ufs_spec_vops);
2437 VNODEOP_SET(ufs_fifo_vops);
2443 ufs_vnoperate(struct vop_generic_args *ap)
2445 return (VOCALL(&ufs_vnode_vops, ap));
2449 * ufs_vnoperatefifo()
2452 ufs_vnoperatefifo(struct vop_generic_args *ap)
2454 return (VOCALL(&ufs_fifo_vops, ap));
2458 * ufs_vnoperatespec()
2461 ufs_vnoperatespec(struct vop_generic_args *ap)
2463 return (VOCALL(&ufs_spec_vops, ap));