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 filt_ufsread (struct knote *kn, long hint);
112 static int filt_ufswrite (struct knote *kn, long hint);
113 static int filt_ufsvnode (struct knote *kn, long hint);
114 static void filt_ufsdetach (struct knote *kn);
115 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
121 #define SETHIGH(q, h) { \
124 tmp.val[_QUAD_HIGHWORD] = (h); \
127 #define SETLOW(q, l) { \
130 tmp.val[_QUAD_LOWWORD] = (l); \
133 #define VN_KNOTE(vp, b) \
134 KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, (b))
136 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
139 * A virgin directory (no blushing please).
141 static struct dirtemplate mastertemplate = {
142 0, 12, DT_DIR, 1, ".",
143 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
145 static struct odirtemplate omastertemplate = {
147 0, DIRBLKSIZ - 12, 2, ".."
151 ufs_itimes(struct vnode *vp)
157 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
159 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
160 ip->i_flag |= IN_LAZYMOD;
162 ip->i_flag |= IN_MODIFIED;
163 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
165 if (ip->i_flag & IN_ACCESS) {
166 ip->i_atime = ts.tv_sec;
167 ip->i_atimensec = ts.tv_nsec;
169 if (ip->i_flag & IN_UPDATE) {
170 ip->i_mtime = ts.tv_sec;
171 ip->i_mtimensec = ts.tv_nsec;
174 if (ip->i_flag & IN_CHANGE) {
175 ip->i_ctime = ts.tv_sec;
176 ip->i_ctimensec = ts.tv_nsec;
179 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
183 * Create a regular file
185 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
186 * struct componentname *a_cnp, struct vattr *a_vap)
190 ufs_create(struct vop_old_create_args *ap)
195 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
196 ap->a_dvp, ap->a_vpp, ap->a_cnp);
199 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
206 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
207 * struct componentname *a_cnp, struct vattr *a_vap)
212 ufs_mknod(struct vop_old_mknod_args *ap)
214 struct vattr *vap = ap->a_vap;
215 struct vnode **vpp = ap->a_vpp;
221 * UFS cannot represent the entire major/minor range supported by
224 if (vap->va_rmajor != VNOVAL &&
225 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
229 /* no special directory support */
230 if (vap->va_type == VDIR)
233 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
234 ap->a_dvp, vpp, ap->a_cnp);
237 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
239 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
240 if (vap->va_rmajor != VNOVAL) {
242 * Want to be able to use this to make badblock
243 * inodes, so don't truncate the dev number.
245 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
248 * Remove inode, then reload it through VFS_VGET so it is
249 * checked to see if it is an alias of an existing entry in
252 (*vpp)->v_type = VNON;
253 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
254 vgone_vxlocked(*vpp);
256 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp);
267 * Update the times on the inode.
269 * ufs_close(struct vnode *a_vp, int a_fflag)
274 ufs_close(struct vop_close_args *ap)
276 struct vnode *vp = ap->a_vp;
278 if (vp->v_sysref.refcnt > 1)
280 return (vop_stdclose(ap));
284 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
288 ufs_access(struct vop_access_args *ap)
290 struct vnode *vp = ap->a_vp;
291 struct inode *ip = VTOI(vp);
292 struct ucred *cred = ap->a_cred;
293 mode_t mask, mode = ap->a_mode;
301 * Disallow write attempts on read-only filesystems;
302 * unless the file is a socket, fifo, or a block or
303 * character device resident on the filesystem.
306 switch (vp->v_type) {
310 if (vp->v_mount->mnt_flag & MNT_RDONLY)
313 if ((error = ufs_getinoquota(ip)) != 0)
322 #if 0 /* handled by kernel now */
323 /* If immutable bit set, nobody gets to write it. */
324 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
328 /* Otherwise, user id 0 always gets access. */
329 if (cred->cr_uid == 0)
334 /* Otherwise, check the owner. */
335 if (cred->cr_uid == ip->i_uid) {
342 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
345 /* Otherwise, check the groups. */
346 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
347 if (ip->i_gid == *gp) {
354 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
357 /* Otherwise, check everyone else. */
364 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
368 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
373 ufs_getattr(struct vop_getattr_args *ap)
375 struct vnode *vp = ap->a_vp;
376 struct inode *ip = VTOI(vp);
377 struct vattr *vap = ap->a_vap;
380 * This may cause i_fsmid to be updated even if no change (0)
381 * is returned, but we should only write out the inode if non-zero
382 * is returned and if the mount is read-write.
384 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
385 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
387 ip->i_flag |= IN_LAZYMOD;
392 * Copy from inode table
394 vap->va_fsid = dev2udev(ip->i_dev);
395 vap->va_fileid = ip->i_number;
396 vap->va_mode = ip->i_mode & ~IFMT;
397 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
398 ip->i_effnlink : ip->i_nlink;
399 vap->va_uid = ip->i_uid;
400 vap->va_gid = ip->i_gid;
401 vap->va_rmajor = umajor(ip->i_rdev);
402 vap->va_rminor = uminor(ip->i_rdev);
403 vap->va_size = ip->i_din.di_size;
404 vap->va_atime.tv_sec = ip->i_atime;
405 vap->va_atime.tv_nsec = ip->i_atimensec;
406 vap->va_mtime.tv_sec = ip->i_mtime;
407 vap->va_mtime.tv_nsec = ip->i_mtimensec;
408 vap->va_ctime.tv_sec = ip->i_ctime;
409 vap->va_ctime.tv_nsec = ip->i_ctimensec;
410 vap->va_flags = ip->i_flags;
411 vap->va_gen = ip->i_gen;
412 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
413 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
414 vap->va_type = IFTOVT(ip->i_mode);
415 vap->va_filerev = ip->i_modrev;
416 vap->va_fsmid = ip->i_fsmid;
422 ufs_markatime(struct vop_markatime_args *ap)
424 struct vnode *vp = ap->a_vp;
425 struct inode *ip = VTOI(vp);
427 if (vp->v_mount->mnt_flag & MNT_RDONLY)
429 if (vp->v_mount->mnt_flag & MNT_NOATIME)
431 ip->i_flag |= IN_ACCESS;
432 VN_KNOTE(vp, NOTE_ATTRIB);
437 * Set attribute vnode op. called from several syscalls
439 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
440 * struct ucred *a_cred)
444 ufs_setattr(struct vop_setattr_args *ap)
446 struct vattr *vap = ap->a_vap;
447 struct vnode *vp = ap->a_vp;
448 struct inode *ip = VTOI(vp);
449 struct ucred *cred = ap->a_cred;
453 * Check for unsettable attributes.
455 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
456 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
457 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
458 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
461 if (vap->va_flags != VNOVAL) {
462 if (vp->v_mount->mnt_flag & MNT_RDONLY)
464 if (cred->cr_uid != ip->i_uid &&
465 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
468 * Note that a root chflags becomes a user chflags when
469 * we are jailed, unless the jail.chflags_allowed sysctl
472 if (cred->cr_uid == 0 &&
473 (!jailed(cred) || jail_chflags_allowed)) {
475 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
478 ip->i_flags = vap->va_flags;
481 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
482 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
484 ip->i_flags &= SF_SETTABLE;
485 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
487 ip->i_flag |= IN_CHANGE;
488 if (vap->va_flags & (IMMUTABLE | APPEND))
491 if (ip->i_flags & (IMMUTABLE | APPEND))
494 * Go through the fields and update iff not VNOVAL.
496 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
497 if (vp->v_mount->mnt_flag & MNT_RDONLY)
499 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
502 if (vap->va_size != VNOVAL) {
504 * Disallow write attempts on read-only filesystems;
505 * unless the file is a socket, fifo, or a block or
506 * character device resident on the filesystem.
508 switch (vp->v_type) {
513 if (vp->v_mount->mnt_flag & MNT_RDONLY)
519 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
523 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
524 if (vp->v_mount->mnt_flag & MNT_RDONLY)
526 if (cred->cr_uid != ip->i_uid &&
527 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) &&
528 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
529 (error = VOP_ACCESS(vp, VWRITE, cred))))
531 if (vap->va_atime.tv_sec != VNOVAL)
532 ip->i_flag |= IN_ACCESS;
533 if (vap->va_mtime.tv_sec != VNOVAL)
534 ip->i_flag |= IN_CHANGE | IN_UPDATE;
536 if (vap->va_atime.tv_sec != VNOVAL) {
537 ip->i_atime = vap->va_atime.tv_sec;
538 ip->i_atimensec = vap->va_atime.tv_nsec;
540 if (vap->va_mtime.tv_sec != VNOVAL) {
541 ip->i_mtime = vap->va_mtime.tv_sec;
542 ip->i_mtimensec = vap->va_mtime.tv_nsec;
544 error = ffs_update(vp, 0);
549 if (vap->va_mode != (mode_t)VNOVAL) {
550 if (vp->v_mount->mnt_flag & MNT_RDONLY)
552 error = ufs_chmod(vp, (int)vap->va_mode, cred);
554 VN_KNOTE(vp, NOTE_ATTRIB);
559 * Change the mode on a file.
560 * Inode must be locked before calling.
563 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
565 struct inode *ip = VTOI(vp);
567 mode_t cur_mode = ip->i_mode;
569 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid,
574 if (cred->cr_uid != ip->i_uid) {
575 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
580 if (vp->v_type != VDIR && (mode & S_ISTXT))
582 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
586 ip->i_mode = cur_mode;
587 ip->i_flag |= IN_CHANGE;
592 * Perform chown operation on inode ip;
593 * inode must be locked prior to call.
596 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
598 struct inode *ip = VTOI(vp);
607 if (uid == (uid_t)VNOVAL)
609 if (gid == (gid_t)VNOVAL)
612 * If we don't own the file, are trying to change the owner
613 * of the file, or are not a member of the target group,
614 * the caller must be superuser or the call fails.
616 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
617 (gid != ip->i_gid && !(cred->cr_gid == gid ||
618 groupmember((gid_t)gid, cred)))) &&
619 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
624 if ((error = ufs_getinoquota(ip)) != 0)
627 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
628 ip->i_dquot[USRQUOTA] = NODQUOT;
631 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
632 ip->i_dquot[GRPQUOTA] = NODQUOT;
634 change = ip->i_blocks;
635 (void) ufs_chkdq(ip, -change, cred, CHOWN);
636 (void) ufs_chkiq(ip, -1, cred, CHOWN);
637 for (i = 0; i < MAXQUOTAS; i++) {
638 ufs_dqrele(vp, ip->i_dquot[i]);
639 ip->i_dquot[i] = NODQUOT;
645 if ((error = ufs_getinoquota(ip)) == 0) {
647 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
648 ip->i_dquot[USRQUOTA] = NODQUOT;
651 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
652 ip->i_dquot[GRPQUOTA] = NODQUOT;
654 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
655 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
658 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
660 for (i = 0; i < MAXQUOTAS; i++) {
661 ufs_dqrele(vp, ip->i_dquot[i]);
662 ip->i_dquot[i] = NODQUOT;
667 if (ufs_getinoquota(ip) == 0) {
669 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
670 ip->i_dquot[USRQUOTA] = NODQUOT;
673 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
674 ip->i_dquot[GRPQUOTA] = NODQUOT;
676 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
677 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
678 (void) ufs_getinoquota(ip);
682 if (ufs_getinoquota(ip))
683 panic("ufs_chown: lost quota");
685 ip->i_flag |= IN_CHANGE;
686 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
687 ip->i_mode &= ~(ISUID | ISGID);
694 * NB Currently unsupported.
696 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
701 ufs_mmap(struct vop_mmap_args *ap)
707 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
708 * struct componentname *a_cnp)
712 ufs_remove(struct vop_old_remove_args *ap)
715 struct vnode *vp = ap->a_vp;
716 struct vnode *dvp = ap->a_dvp;
720 #if 0 /* handled by kernel now */
721 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
722 (VTOI(dvp)->i_flags & APPEND)) {
727 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
728 VN_KNOTE(vp, NOTE_DELETE);
729 VN_KNOTE(dvp, NOTE_WRITE);
739 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
740 * struct componentname *a_cnp)
744 ufs_link(struct vop_old_link_args *ap)
746 struct vnode *vp = ap->a_vp;
747 struct vnode *tdvp = ap->a_tdvp;
748 struct componentname *cnp = ap->a_cnp;
750 struct direct newdir;
753 if (tdvp->v_mount != vp->v_mount) {
757 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
761 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
765 #if 0 /* handled by kernel now, also DragonFly allows this */
766 if (ip->i_flags & (IMMUTABLE | APPEND)) {
773 ip->i_flag |= IN_CHANGE;
774 if (DOINGSOFTDEP(vp))
775 softdep_change_linkcnt(ip);
776 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
778 ufs_makedirentry(ip, cnp, &newdir);
779 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
785 ip->i_flag |= IN_CHANGE;
786 if (DOINGSOFTDEP(vp))
787 softdep_change_linkcnt(ip);
793 VN_KNOTE(vp, NOTE_LINK);
794 VN_KNOTE(tdvp, NOTE_WRITE);
799 * whiteout vnode call
801 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
805 ufs_whiteout(struct vop_old_whiteout_args *ap)
807 struct vnode *dvp = ap->a_dvp;
808 struct componentname *cnp = ap->a_cnp;
809 struct direct newdir;
812 switch (ap->a_flags) {
814 /* 4.4 format directories support whiteout operations */
815 if (dvp->v_mount->mnt_maxsymlinklen > 0)
820 /* create a new directory whiteout */
822 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
823 panic("ufs_whiteout: old format filesystem");
827 newdir.d_namlen = cnp->cn_namelen;
828 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
829 newdir.d_type = DT_WHT;
830 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
834 /* remove an existing directory whiteout */
836 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
837 panic("ufs_whiteout: old format filesystem");
840 cnp->cn_flags &= ~CNP_DOWHITEOUT;
841 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
844 panic("ufs_whiteout: unknown op");
850 * Rename system call.
851 * rename("foo", "bar");
854 * link("foo", "bar");
856 * but ``atomically''. Can't do full commit without saving state in the
857 * inode on disk which isn't feasible at this time. Best we can do is
858 * always guarantee the target exists.
860 * Basic algorithm is:
862 * 1) Bump link count on source while we're linking it to the
863 * target. This also ensure the inode won't be deleted out
864 * from underneath us while we work (it may be truncated by
865 * a concurrent `trunc' or `open' for creation).
866 * 2) Link source to destination. If destination already exists,
868 * 3) Unlink source reference to inode if still around. If a
869 * directory was moved and the parent of the destination
870 * is different from the source, patch the ".." entry in the
873 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
874 * struct componentname *a_fcnp, struct vnode *a_tdvp,
875 * struct vnode *a_tvp, struct componentname *a_tcnp)
879 ufs_rename(struct vop_old_rename_args *ap)
881 struct vnode *tvp = ap->a_tvp;
882 struct vnode *tdvp = ap->a_tdvp;
883 struct vnode *fvp = ap->a_fvp;
884 struct vnode *fdvp = ap->a_fdvp;
885 struct componentname *tcnp = ap->a_tcnp;
886 struct componentname *fcnp = ap->a_fcnp;
887 struct inode *ip, *xp, *dp;
888 struct direct newdir;
889 ino_t oldparent = 0, newparent = 0;
890 int doingdirectory = 0;
891 int error = 0, ioflag;
894 * Check for cross-device rename.
896 if ((fvp->v_mount != tdvp->v_mount) ||
897 (tvp && (fvp->v_mount != tvp->v_mount))) {
911 #if 0 /* handled by kernel now */
912 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
913 (VTOI(tdvp)->i_flags & APPEND))) {
920 * Renaming a file to itself has no effect. The upper layers should
921 * not call us in that case. Temporarily just warn if they do.
924 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
929 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
933 * Note: now that fvp is locked we have to be sure to unlock it before
934 * using the 'abortit' target.
938 if (ip->i_nlink >= LINK_MAX) {
943 #if 0 /* handled by kernel now */
944 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
945 || (dp->i_flags & APPEND)) {
951 if ((ip->i_mode & IFMT) == IFDIR) {
953 * Avoid ".", "..", and aliases of "." for obvious reasons.
955 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
956 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
957 (ip->i_flag & IN_RENAME)) {
962 ip->i_flag |= IN_RENAME;
963 oldparent = dp->i_number;
966 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
969 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
970 * Cleanup fvp requirements so we can unlock it.
972 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
973 * is setup we can use the 'bad' target if we unlock fvp. We cannot
974 * use the abortit target anymore because of IN_RENAME.
983 * 1) Bump link count while we're moving stuff
984 * around. If we crash somewhere before
985 * completing our work, the link count
986 * may be wrong, but correctable.
990 ip->i_flag |= IN_CHANGE;
991 if (DOINGSOFTDEP(fvp))
992 softdep_change_linkcnt(ip);
993 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
994 DOINGASYNC(fvp)))) != 0) {
1000 * If ".." must be changed (ie the directory gets a new
1001 * parent) then the source directory must not be in the
1002 * directory heirarchy above the target, as this would
1003 * orphan everything below the source directory. Also
1004 * the user must have write permission in the source so
1005 * as to be able to change "..". We must repeat the call
1006 * to namei, as the parent directory is unlocked by the
1007 * call to checkpath().
1009 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
1013 * We are now back to where we were in that fvp, fdvp are unlocked
1014 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
1015 * set. Only the bad target or, if we clean up tvp and tdvp, the
1016 * out target, may be used.
1018 if (oldparent != dp->i_number)
1019 newparent = dp->i_number;
1020 if (doingdirectory && newparent) {
1021 if (error) /* write access check above */
1025 * Once we start messing with tvp and tdvp we cannot use the
1026 * 'bad' target, only finish cleaning tdvp and tvp up and
1027 * use the 'out' target.
1029 * This cleans up tvp.
1037 * This is a real mess. ufs_checkpath vput's the target
1038 * directory so retain an extra ref and note that tdvp will
1039 * lose its lock on return. This leaves us with one good
1040 * ref after ufs_checkpath returns.
1043 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
1044 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1051 * relookup no longer messes with tdvp's refs. tdvp must be
1052 * unlocked on entry and will be locked on a successful
1055 error = relookup(tdvp, &tvp, tcnp);
1057 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1063 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1070 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1071 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1072 * be set if doingdirectory. This means we can use the 'bad' target
1077 * 2) If target doesn't exist, link the target
1078 * to the source and unlink the source.
1079 * Otherwise, rewrite the target directory
1080 * entry to reference the source inode and
1081 * expunge the original entry's existence.
1084 if (dp->i_dev != ip->i_dev)
1085 panic("ufs_rename: EXDEV");
1087 * Account for ".." in new directory.
1088 * When source and destination have the same
1089 * parent we don't fool with the link count.
1091 if (doingdirectory && newparent) {
1092 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1098 dp->i_flag |= IN_CHANGE;
1099 if (DOINGSOFTDEP(tdvp))
1100 softdep_change_linkcnt(dp);
1101 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1106 ufs_makedirentry(ip, tcnp, &newdir);
1107 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1109 if (doingdirectory && newparent) {
1112 dp->i_flag |= IN_CHANGE;
1113 if (DOINGSOFTDEP(tdvp))
1114 softdep_change_linkcnt(dp);
1115 (void)ffs_update(tdvp, 1);
1119 VN_KNOTE(tdvp, NOTE_WRITE);
1122 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1123 panic("ufs_rename: EXDEV");
1125 * Short circuit rename(foo, foo).
1127 if (xp->i_number == ip->i_number)
1128 panic("ufs_rename: same file");
1130 * If the parent directory is "sticky", then the user must
1131 * own the parent directory, or the destination of the rename,
1132 * otherwise the destination may not be changed (except by
1133 * root). This implements append-only directories.
1135 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1136 tcnp->cn_cred->cr_uid != dp->i_uid &&
1137 xp->i_uid != tcnp->cn_cred->cr_uid) {
1142 * Target must be empty if a directory and have no links
1143 * to it. Also, ensure source and target are compatible
1144 * (both directories, or both not directories).
1146 * Purge the file or directory being replaced from the
1149 if ((xp->i_mode&IFMT) == IFDIR) {
1150 if ((xp->i_effnlink > 2) ||
1151 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1155 if (!doingdirectory) {
1159 /* cache_purge removed - handled by VFS compat layer */
1160 } else if (doingdirectory == 0) {
1161 /* cache_purge removed - handled by VFS compat layer */
1167 * note: inode passed to ufs_dirrewrite() is 0 for a
1168 * non-directory file rename, 1 for a directory rename
1169 * in the same directory, and > 1 for an inode representing
1170 * the new directory.
1172 error = ufs_dirrewrite(dp, xp, ip->i_number,
1174 (doingdirectory && newparent) ?
1175 newparent : (ino_t)doingdirectory);
1178 if (doingdirectory) {
1181 if (DOINGSOFTDEP(tdvp))
1182 softdep_change_linkcnt(dp);
1185 if (DOINGSOFTDEP(tvp))
1186 softdep_change_linkcnt(xp);
1188 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1190 * Truncate inode. The only stuff left in the directory
1191 * is "." and "..". The "." reference is inconsequential
1192 * since we are quashing it. We have removed the "."
1193 * reference and the reference in the parent directory,
1194 * but there may be other hard links. The soft
1195 * dependency code will arrange to do these operations
1196 * after the parent directory entry has been deleted on
1197 * disk, so when running with that code we avoid doing
1202 dp->i_flag |= IN_CHANGE;
1205 xp->i_flag |= IN_CHANGE;
1206 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1207 error = ffs_truncate(tvp, (off_t)0, ioflag,
1212 VN_KNOTE(tdvp, NOTE_WRITE);
1214 VN_KNOTE(tvp, NOTE_DELETE);
1220 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1221 * unlocked) remain. We are about to overwrite fvp but we have to
1222 * keep 'ip' intact so we cannot release the old fvp, which is still
1223 * refd and accessible via ap->a_fvp.
1225 * This means we cannot use either 'bad' or 'out' to cleanup any
1230 * 3) Unlink the source.
1232 fcnp->cn_flags &= ~CNP_MODMASK;
1233 fcnp->cn_flags |= CNP_LOCKPARENT;
1234 error = relookup(fdvp, &fvp, fcnp);
1235 if (error || fvp == NULL) {
1237 * From name has disappeared. IN_RENAME will not be set if
1238 * we get past the panic so we don't have to clean it up.
1241 panic("ufs_rename: lost dir entry");
1243 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1249 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1252 * fdvp and fvp are locked.
1258 * Ensure that the directory entry still exists and has not
1259 * changed while the new name has been entered. If the source is
1260 * a file then the entry may have been unlinked or renamed. In
1261 * either case there is no further work to be done. If the source
1262 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1263 * flag ensures that it cannot be moved by another rename or removed
1264 * by a rmdir. Cleanup IN_RENAME.
1268 panic("ufs_rename: lost dir entry");
1271 * If the source is a directory with a
1272 * new parent, the link count of the old
1273 * parent directory must be decremented
1274 * and ".." set to point to the new parent.
1276 if (doingdirectory && newparent) {
1277 xp->i_offset = mastertemplate.dot_reclen;
1278 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1279 /* cache_purge removed - handled by VFS compat layer */
1281 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1282 xp->i_flag &= ~IN_RENAME;
1285 VN_KNOTE(fvp, NOTE_RENAME);
1297 ip->i_flag &= ~IN_RENAME;
1298 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1301 ip->i_flag |= IN_CHANGE;
1302 ip->i_flag &= ~IN_RENAME;
1303 if (DOINGSOFTDEP(fvp))
1304 softdep_change_linkcnt(ip);
1315 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1316 * struct componentname *a_cnp, struct vattr *a_vap)
1320 ufs_mkdir(struct vop_old_mkdir_args *ap)
1322 struct vnode *dvp = ap->a_dvp;
1323 struct vattr *vap = ap->a_vap;
1324 struct componentname *cnp = ap->a_cnp;
1325 struct inode *ip, *dp;
1328 struct dirtemplate dirtemplate, *dtp;
1329 struct direct newdir;
1334 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1338 dmode = vap->va_mode & 0777;
1341 * Must simulate part of ufs_makeinode here to acquire the inode,
1342 * but not have it entered in the parent directory. The entry is
1343 * made later after writing "." and ".." entries.
1345 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1349 ip->i_gid = dp->i_gid;
1353 struct ucred ucred, *ucp;
1357 * If we are hacking owners here, (only do this where told to)
1358 * and we are not giving it TO root, (would subvert quotas)
1359 * then go ahead and give it to the other user.
1360 * The new directory also inherits the SUID bit.
1361 * If user's UID and dir UID are the same,
1362 * 'give it away' so that the SUID is still forced on.
1364 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1365 (dp->i_mode & ISUID) && dp->i_uid) {
1367 ip->i_uid = dp->i_uid;
1369 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1371 * Make sure the correct user gets charged
1373 * Make a dummy credential for the victim.
1374 * XXX This seems to never be accessed out of
1375 * our context so a stack variable is ok.
1378 ucred.cr_uid = ip->i_uid;
1379 ucred.cr_ngroups = 1;
1380 ucred.cr_groups[0] = dp->i_gid;
1385 ip->i_uid = cnp->cn_cred->cr_uid;
1387 if ((error = ufs_getinoquota(ip)) ||
1388 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1389 ffs_vfree(tvp, ip->i_number, dmode);
1395 #else /* !SUIDDIR */
1396 ip->i_uid = cnp->cn_cred->cr_uid;
1398 if ((error = ufs_getinoquota(ip)) ||
1399 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1400 ffs_vfree(tvp, ip->i_number, dmode);
1405 #endif /* !SUIDDIR */
1406 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1408 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1411 if (DOINGSOFTDEP(tvp))
1412 softdep_change_linkcnt(ip);
1413 if (cnp->cn_flags & CNP_ISWHITEOUT)
1414 ip->i_flags |= UF_OPAQUE;
1417 * Bump link count in parent directory to reflect work done below.
1418 * Should be done before reference is created so cleanup is
1419 * possible if we crash.
1423 dp->i_flag |= IN_CHANGE;
1424 if (DOINGSOFTDEP(dvp))
1425 softdep_change_linkcnt(dp);
1426 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1431 * The vnode must have a VM object in order to issue buffer cache
1434 vinitvmio(tvp, DIRBLKSIZ);
1437 * Initialize directory with "." and ".." from static template.
1439 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1440 dtp = &mastertemplate;
1442 dtp = (struct dirtemplate *)&omastertemplate;
1444 dirtemplate.dot_ino = ip->i_number;
1445 dirtemplate.dotdot_ino = dp->i_number;
1446 vnode_pager_setsize(tvp, DIRBLKSIZ);
1447 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1450 ip->i_size = DIRBLKSIZ;
1451 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1452 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1453 if (DOINGSOFTDEP(tvp)) {
1455 * Ensure that the entire newly allocated block is a
1456 * valid directory so that future growth within the
1457 * block does not have to ensure that the block is
1458 * written before the inode.
1461 while (blkoff < bp->b_bcount) {
1463 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1464 blkoff += DIRBLKSIZ;
1467 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1468 DOINGASYNC(tvp)))) != 0) {
1473 * Directory set up, now install its entry in the parent directory.
1475 * If we are not doing soft dependencies, then we must write out the
1476 * buffer containing the new directory body before entering the new
1477 * name in the parent. If we are doing soft dependencies, then the
1478 * buffer containing the new directory body will be passed to and
1479 * released in the soft dependency code after the code has attached
1480 * an appropriate ordering dependency to the buffer which ensures that
1481 * the buffer is written before the new name is written in the parent.
1483 if (DOINGASYNC(dvp))
1485 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1487 ufs_makedirentry(ip, cnp, &newdir);
1488 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1492 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1497 dp->i_flag |= IN_CHANGE;
1498 if (DOINGSOFTDEP(dvp))
1499 softdep_change_linkcnt(dp);
1501 * No need to do an explicit VOP_TRUNCATE here, vrele will
1502 * do this for us because we set the link count to 0.
1506 ip->i_flag |= IN_CHANGE;
1507 if (DOINGSOFTDEP(tvp))
1508 softdep_change_linkcnt(ip);
1516 * Rmdir system call.
1518 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1519 * struct componentname *a_cnp)
1523 ufs_rmdir(struct vop_old_rmdir_args *ap)
1525 struct vnode *vp = ap->a_vp;
1526 struct vnode *dvp = ap->a_dvp;
1527 struct componentname *cnp = ap->a_cnp;
1528 struct inode *ip, *dp;
1535 * Do not remove a directory that is in the process of being renamed.
1536 * Verify the directory is empty (and valid). Rmdir ".." will not be
1537 * valid since ".." will contain a reference to the current directory
1538 * and thus be non-empty. Do not allow the removal of mounted on
1539 * directories (this can happen when an NFS exported filesystem
1540 * tries to remove a locally mounted on directory).
1543 if (ip->i_flag & IN_RENAME) {
1547 if (ip->i_effnlink != 2 ||
1548 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1552 #if 0 /* handled by kernel now */
1553 if ((dp->i_flags & APPEND)
1554 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1560 * Delete reference to directory before purging
1561 * inode. If we crash in between, the directory
1562 * will be reattached to lost+found,
1566 if (DOINGSOFTDEP(vp)) {
1567 softdep_change_linkcnt(dp);
1568 softdep_change_linkcnt(ip);
1570 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1574 if (DOINGSOFTDEP(vp)) {
1575 softdep_change_linkcnt(dp);
1576 softdep_change_linkcnt(ip);
1580 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1582 * Truncate inode. The only stuff left in the directory is "." and
1583 * "..". The "." reference is inconsequential since we are quashing
1584 * it. The soft dependency code will arrange to do these operations
1585 * after the parent directory entry has been deleted on disk, so
1586 * when running with that code we avoid doing them now.
1588 if (!DOINGSOFTDEP(vp)) {
1590 dp->i_flag |= IN_CHANGE;
1592 ip->i_flag |= IN_CHANGE;
1593 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1594 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1596 /* cache_purge removed - handled by VFS compat layer */
1598 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1599 if (ip->i_dirhash != NULL)
1600 ufsdirhash_free(ip);
1603 VN_KNOTE(vp, NOTE_DELETE);
1608 * symlink -- make a symbolic link
1610 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1611 * struct componentname *a_cnp, struct vattr *a_vap,
1616 ufs_symlink(struct vop_old_symlink_args *ap)
1618 struct vnode *vp, **vpp = ap->a_vpp;
1622 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1626 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1628 len = strlen(ap->a_target);
1629 if (len < vp->v_mount->mnt_maxsymlinklen) {
1631 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1633 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1636 * Make sure we have a VM object in order to use
1639 if (vp->v_object == NULL)
1641 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1642 UIO_SYSSPACE, IO_NODELOCKED,
1643 ap->a_cnp->cn_cred, NULL);
1651 * Vnode op for reading directories.
1653 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1654 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1658 ufs_readdir(struct vop_readdir_args *ap)
1660 struct uio *uio = ap->a_uio;
1661 struct vnode *vp = ap->a_vp;
1666 int offset; /* offset into buffer cache buffer */
1667 int eoffset; /* end of buffer clipped to file EOF */
1668 int pickup; /* pickup point */
1673 if (uio->uio_offset < 0)
1676 * Guess the number of cookies needed. Make sure we compute at
1677 * least 1, and no more then a reasonable limit.
1679 if (ap->a_ncookies) {
1680 ncookies = uio->uio_resid / 16 + 1;
1681 if (ncookies > 1024)
1683 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1685 ncookies = -1; /* force conditionals below */
1690 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1696 if (uio->uio_offset >= VTOI(vp)->i_size) {
1699 if (ap->a_ncookies) {
1700 *ap->a_ncookies = cookie_index;
1701 *ap->a_cookies = cookies;
1707 * Loop until we run out of cookies, we run out of user buffer,
1708 * or we hit the directory EOF.
1710 * Always start scans at the beginning of the buffer, don't trust
1711 * the offset supplied by userland.
1713 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1714 pickup = (int)(uio->uio_offset - bp->b_loffset);
1717 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1718 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1720 eoffset = bp->b_bcount;
1722 while (offset < eoffset) {
1723 dp = (struct direct *)(bp->b_data + offset);
1724 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1725 offset + dp->d_reclen > bp->b_bcount) {
1729 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1733 if (offset < pickup) {
1734 offset += dp->d_reclen;
1737 #if BYTE_ORDER == LITTLE_ENDIAN
1739 retval = vop_write_dirent(&error, uio,
1740 dp->d_ino, dp->d_namlen, dp->d_type,
1745 retval = vop_write_dirent(&error, uio,
1746 dp->d_ino, dp->d_type, dp->d_namlen,
1752 cookies[cookie_index] = bp->b_loffset + offset;
1754 offset += dp->d_reclen;
1755 if (cookie_index == ncookies)
1760 * This will align the next loop to the beginning of the
1761 * next block, and pickup will calculate to 0.
1763 uio->uio_offset = bp->b_loffset + offset;
1766 if (retval || error || cookie_index == ncookies ||
1767 uio->uio_offset >= VTOI(vp)->i_size) {
1772 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1775 * Report errors only if we didn't manage to read anything
1777 if (error && cookie_index == 0) {
1779 kfree(cookies, M_TEMP);
1780 *ap->a_ncookies = 0;
1781 *ap->a_cookies = NULL;
1786 *ap->a_ncookies = cookie_index;
1787 *ap->a_cookies = cookies;
1796 * Return target name of a symbolic link
1798 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1802 ufs_readlink(struct vop_readlink_args *ap)
1804 struct vnode *vp = ap->a_vp;
1805 struct inode *ip = VTOI(vp);
1809 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1810 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1811 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1816 * Perform the equivalent of an OPEN on vp so we can issue a
1819 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1823 * Calculate the logical to physical mapping if not done already,
1824 * then call the device strategy routine.
1826 * In order to be able to swap to a file, the VOP_BMAP operation may not
1827 * deadlock on memory. See ufs_bmap() for details.
1829 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1833 ufs_strategy(struct vop_strategy_args *ap)
1835 struct bio *bio = ap->a_bio;
1837 struct buf *bp = bio->bio_buf;
1838 struct vnode *vp = ap->a_vp;
1843 if (vp->v_type == VBLK || vp->v_type == VCHR)
1844 panic("ufs_strategy: spec");
1845 nbio = push_bio(bio);
1846 if (nbio->bio_offset == NOOFFSET) {
1847 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1848 NULL, NULL, bp->b_cmd);
1850 bp->b_error = error;
1851 bp->b_flags |= B_ERROR;
1852 /* I/O was never started on nbio, must biodone(bio) */
1856 if (nbio->bio_offset == NOOFFSET)
1859 if (nbio->bio_offset == NOOFFSET) {
1861 * We hit a hole in the file. The buffer has been zero-filled
1862 * so just biodone() it.
1866 vn_strategy(ip->i_devvp, nbio);
1872 * Print out the contents of an inode.
1874 * ufs_print(struct vnode *a_vp)
1878 ufs_print(struct vop_print_args *ap)
1880 struct vnode *vp = ap->a_vp;
1881 struct inode *ip = VTOI(vp);
1883 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1884 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1886 if (vp->v_type == VFIFO)
1888 lockmgr_printinfo(&vp->v_lock);
1894 * Read wrapper for fifos.
1896 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1897 * struct ucred *a_cred)
1901 ufsfifo_read(struct vop_read_args *ap)
1908 resid = uio->uio_resid;
1909 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1910 ip = VTOI(ap->a_vp);
1911 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1912 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1913 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1918 * Write wrapper for fifos.
1920 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1921 * struct ucred *a_cred)
1925 ufsfifo_write(struct vop_write_args *ap)
1932 resid = uio->uio_resid;
1933 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1934 ip = VTOI(ap->a_vp);
1935 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1936 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1941 * Close wrapper for fifos.
1943 * Update the times on the inode then do device close.
1945 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
1949 ufsfifo_close(struct vop_close_args *ap)
1951 struct vnode *vp = ap->a_vp;
1953 if (vp->v_sysref.refcnt > 1)
1955 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1959 * Kqfilter wrapper for fifos.
1961 * Fall through to ufs kqfilter routines if needed
1965 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
1969 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1971 error = ufs_kqfilter(ap);
1976 * Advisory record locking support
1978 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
1983 ufs_advlock(struct vop_advlock_args *ap)
1985 struct inode *ip = VTOI(ap->a_vp);
1987 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
1991 * Initialize the vnode associated with a new inode, handle aliased
1994 * Make sure directories have their VM object now rather then later,
1995 * saving us from having to check on all the myrid directory VOPs
1996 * that might be executed without a VOP_OPEN being performed.
1999 ufs_vinit(struct mount *mntp, struct vnode **vpp)
2008 vp->v_type = IFTOVT(ip->i_mode);
2010 switch(vp->v_type) {
2013 vp->v_ops = &mntp->mnt_vn_spec_ops;
2014 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
2017 vp->v_ops = &mntp->mnt_vn_fifo_ops;
2021 vinitvmio(vp, ip->i_size);
2024 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
2025 vinitvmio(vp, ip->i_size);
2032 if (ip->i_number == ROOTINO)
2033 vp->v_flag |= VROOT;
2035 * Initialize modrev times
2037 getmicrouptime(&tv);
2038 SETHIGH(ip->i_modrev, tv.tv_sec);
2039 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
2045 * Allocate a new inode.
2049 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2050 struct componentname *cnp)
2052 struct inode *ip, *pdir;
2053 struct direct newdir;
2059 if ((mode & IFMT) == 0)
2062 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2066 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
2067 ip->i_gid = pdir->i_gid;
2071 struct ucred ucred, *ucp;
2075 * If we are not the owner of the directory,
2076 * and we are hacking owners here, (only do this where told to)
2077 * and we are not giving it TO root, (would subvert quotas)
2078 * then go ahead and give it to the other user.
2079 * Note that this drops off the execute bits for security.
2081 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2082 (pdir->i_mode & ISUID) &&
2083 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2084 ip->i_uid = pdir->i_uid;
2088 * Make sure the correct user gets charged
2090 * Quickly knock up a dummy credential for the victim.
2091 * XXX This seems to never be accessed out of our
2092 * context so a stack variable is ok.
2095 ucred.cr_uid = ip->i_uid;
2096 ucred.cr_ngroups = 1;
2097 ucred.cr_groups[0] = pdir->i_gid;
2101 ip->i_uid = cnp->cn_cred->cr_uid;
2104 if ((error = ufs_getinoquota(ip)) ||
2105 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2106 ffs_vfree(tvp, ip->i_number, mode);
2112 #else /* !SUIDDIR */
2113 ip->i_uid = cnp->cn_cred->cr_uid;
2115 if ((error = ufs_getinoquota(ip)) ||
2116 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2117 ffs_vfree(tvp, ip->i_number, mode);
2122 #endif /* !SUIDDIR */
2123 ip->i_fsmid = cache_getnewfsmid();
2124 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2126 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2129 if (DOINGSOFTDEP(tvp))
2130 softdep_change_linkcnt(ip);
2131 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2132 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2133 ip->i_mode &= ~ISGID;
2136 if (cnp->cn_flags & CNP_ISWHITEOUT)
2137 ip->i_flags |= UF_OPAQUE;
2140 * Regular files and directories need VM objects. Softlinks do
2141 * not (not immediately anyway).
2143 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2147 * Make sure inode goes to disk before directory entry.
2149 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2152 ufs_makedirentry(ip, cnp, &newdir);
2153 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2161 * Write error occurred trying to update the inode
2162 * or the directory so must deallocate the inode.
2166 ip->i_flag |= IN_CHANGE;
2167 if (DOINGSOFTDEP(tvp))
2168 softdep_change_linkcnt(ip);
2174 ufs_missingop(struct vop_generic_args *ap)
2176 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2177 return (EOPNOTSUPP);
2180 static struct filterops ufsread_filtops =
2181 { 1, NULL, filt_ufsdetach, filt_ufsread };
2182 static struct filterops ufswrite_filtops =
2183 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2184 static struct filterops ufsvnode_filtops =
2185 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2188 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2191 ufs_kqfilter(struct vop_kqfilter_args *ap)
2193 struct vnode *vp = ap->a_vp;
2194 struct knote *kn = ap->a_kn;
2197 switch (kn->kn_filter) {
2199 kn->kn_fop = &ufsread_filtops;
2202 kn->kn_fop = &ufswrite_filtops;
2205 kn->kn_fop = &ufsvnode_filtops;
2211 kn->kn_hook = (caddr_t)vp;
2213 lwkt_gettoken(&vlock, &vp->v_token);
2214 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2215 lwkt_reltoken(&vlock);
2221 filt_ufsdetach(struct knote *kn)
2223 struct vnode *vp = (struct vnode *)kn->kn_hook;
2226 lwkt_gettoken(&vlock, &vp->v_token);
2227 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2228 kn, knote, kn_selnext);
2229 lwkt_reltoken(&vlock);
2234 filt_ufsread(struct knote *kn, long hint)
2236 struct vnode *vp = (struct vnode *)kn->kn_hook;
2237 struct inode *ip = VTOI(vp);
2240 * filesystem is gone, so set the EOF flag and schedule
2241 * the knote for deletion.
2243 if (hint == NOTE_REVOKE) {
2244 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2248 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2249 return (kn->kn_data != 0);
2254 filt_ufswrite(struct knote *kn, long hint)
2257 * filesystem is gone, so set the EOF flag and schedule
2258 * the knote for deletion.
2260 if (hint == NOTE_REVOKE)
2261 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2268 filt_ufsvnode(struct knote *kn, long hint)
2270 if (kn->kn_sfflags & hint)
2271 kn->kn_fflags |= hint;
2272 if (hint == NOTE_REVOKE) {
2273 kn->kn_flags |= EV_EOF;
2276 return (kn->kn_fflags != 0);
2279 /* Global vfs data structures for ufs. */
2280 static struct vop_ops ufs_vnode_vops = {
2281 .vop_default = vop_defaultop,
2282 .vop_fsync = (void *)ufs_missingop,
2283 .vop_read = (void *)ufs_missingop,
2284 .vop_reallocblks = (void *)ufs_missingop,
2285 .vop_write = (void *)ufs_missingop,
2286 .vop_access = ufs_access,
2287 .vop_advlock = ufs_advlock,
2288 .vop_bmap = ufs_bmap,
2289 .vop_old_lookup = ufs_lookup,
2290 .vop_close = ufs_close,
2291 .vop_old_create = ufs_create,
2292 .vop_getattr = ufs_getattr,
2293 .vop_inactive = ufs_inactive,
2294 .vop_old_link = ufs_link,
2295 .vop_old_mkdir = ufs_mkdir,
2296 .vop_old_mknod = ufs_mknod,
2297 .vop_mmap = ufs_mmap,
2298 .vop_open = vop_stdopen,
2299 .vop_pathconf = vop_stdpathconf,
2300 .vop_poll = vop_stdpoll,
2301 .vop_kqfilter = ufs_kqfilter,
2302 .vop_print = ufs_print,
2303 .vop_readdir = ufs_readdir,
2304 .vop_readlink = ufs_readlink,
2305 .vop_reclaim = ufs_reclaim,
2306 .vop_old_remove = ufs_remove,
2307 .vop_old_rename = ufs_rename,
2308 .vop_old_rmdir = ufs_rmdir,
2309 .vop_setattr = ufs_setattr,
2310 .vop_markatime = ufs_markatime,
2311 .vop_strategy = ufs_strategy,
2312 .vop_old_symlink = ufs_symlink,
2313 .vop_old_whiteout = ufs_whiteout
2316 static struct vop_ops ufs_spec_vops = {
2317 .vop_default = vop_defaultop,
2318 .vop_fsync = (void *)ufs_missingop,
2319 .vop_access = ufs_access,
2320 .vop_close = ufs_close,
2321 .vop_getattr = ufs_getattr,
2322 .vop_inactive = ufs_inactive,
2323 .vop_print = ufs_print,
2324 .vop_read = vop_stdnoread,
2325 .vop_reclaim = ufs_reclaim,
2326 .vop_setattr = ufs_setattr,
2327 .vop_markatime = ufs_markatime,
2328 .vop_write = vop_stdnowrite
2331 static struct vop_ops ufs_fifo_vops = {
2332 .vop_default = fifo_vnoperate,
2333 .vop_fsync = (void *)ufs_missingop,
2334 .vop_access = ufs_access,
2335 .vop_close = ufsfifo_close,
2336 .vop_getattr = ufs_getattr,
2337 .vop_inactive = ufs_inactive,
2338 .vop_kqfilter = ufsfifo_kqfilter,
2339 .vop_print = ufs_print,
2340 .vop_read = ufsfifo_read,
2341 .vop_reclaim = ufs_reclaim,
2342 .vop_setattr = ufs_setattr,
2343 .vop_markatime = ufs_markatime,
2344 .vop_write = ufsfifo_write
2347 VNODEOP_SET(ufs_vnode_vops);
2348 VNODEOP_SET(ufs_spec_vops);
2349 VNODEOP_SET(ufs_fifo_vops);
2355 ufs_vnoperate(struct vop_generic_args *ap)
2357 return (VOCALL(&ufs_vnode_vops, ap));
2361 * ufs_vnoperatefifo()
2364 ufs_vnoperatefifo(struct vop_generic_args *ap)
2366 return (VOCALL(&ufs_fifo_vops, ap));
2370 * ufs_vnoperatespec()
2373 ufs_vnoperatespec(struct vop_generic_args *ap)
2375 return (VOCALL(&ufs_spec_vops, ap));