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 filt_ufsread (struct knote *kn, long hint);
115 static int filt_ufswrite (struct knote *kn, long hint);
116 static int filt_ufsvnode (struct knote *kn, long hint);
117 static void filt_ufsdetach (struct knote *kn);
118 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
124 #define SETHIGH(q, h) { \
127 tmp.val[_QUAD_HIGHWORD] = (h); \
130 #define SETLOW(q, l) { \
133 tmp.val[_QUAD_LOWWORD] = (l); \
136 #define VN_KNOTE(vp, b) \
137 KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, (b))
139 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
142 * A virgin directory (no blushing please).
144 static struct dirtemplate mastertemplate = {
145 0, 12, DT_DIR, 1, ".",
146 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
148 static struct odirtemplate omastertemplate = {
150 0, DIRBLKSIZ - 12, 2, ".."
154 ufs_itimes(struct vnode *vp)
160 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
162 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
163 ip->i_flag |= IN_LAZYMOD;
165 ip->i_flag |= IN_MODIFIED;
166 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
168 if (ip->i_flag & IN_ACCESS) {
169 ip->i_atime = ts.tv_sec;
170 ip->i_atimensec = ts.tv_nsec;
172 if (ip->i_flag & IN_UPDATE) {
173 ip->i_mtime = ts.tv_sec;
174 ip->i_mtimensec = ts.tv_nsec;
177 if (ip->i_flag & IN_CHANGE) {
178 ip->i_ctime = ts.tv_sec;
179 ip->i_ctimensec = ts.tv_nsec;
182 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
186 * Create a regular file
188 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
189 * struct componentname *a_cnp, struct vattr *a_vap)
193 ufs_create(struct vop_old_create_args *ap)
198 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
199 ap->a_dvp, ap->a_vpp, ap->a_cnp);
202 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
209 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
210 * struct componentname *a_cnp, struct vattr *a_vap)
215 ufs_mknod(struct vop_old_mknod_args *ap)
217 struct vattr *vap = ap->a_vap;
218 struct vnode **vpp = ap->a_vpp;
224 * UFS cannot represent the entire major/minor range supported by
227 if (vap->va_rmajor != VNOVAL &&
228 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
232 /* no special directory support */
233 if (vap->va_type == VDIR)
236 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
237 ap->a_dvp, vpp, ap->a_cnp);
240 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
242 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
243 if (vap->va_rmajor != VNOVAL) {
245 * Want to be able to use this to make badblock
246 * inodes, so don't truncate the dev number.
248 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
251 * Remove inode, then reload it through VFS_VGET so it is
252 * checked to see if it is an alias of an existing entry in
255 (*vpp)->v_type = VNON;
256 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
257 vgone_vxlocked(*vpp);
259 error = VFS_VGET(ap->a_dvp->v_mount, ino, vpp);
270 * Update the times on the inode.
272 * ufs_close(struct vnode *a_vp, int a_fflag)
277 ufs_close(struct vop_close_args *ap)
279 struct vnode *vp = ap->a_vp;
281 if (vp->v_sysref.refcnt > 1)
283 return (vop_stdclose(ap));
287 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
291 ufs_access(struct vop_access_args *ap)
293 struct vnode *vp = ap->a_vp;
294 struct inode *ip = VTOI(vp);
295 struct ucred *cred = ap->a_cred;
296 mode_t mask, mode = ap->a_mode;
304 * Disallow write attempts on read-only filesystems;
305 * unless the file is a socket, fifo, or a block or
306 * character device resident on the filesystem.
309 switch (vp->v_type) {
313 if (vp->v_mount->mnt_flag & MNT_RDONLY)
316 if ((error = ufs_getinoquota(ip)) != 0)
325 #if 0 /* handled by kernel now */
326 /* If immutable bit set, nobody gets to write it. */
327 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
331 /* Otherwise, user id 0 always gets access. */
332 if (cred->cr_uid == 0)
337 /* Otherwise, check the owner. */
338 if (cred->cr_uid == ip->i_uid) {
345 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
348 /* Otherwise, check the groups. */
349 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
350 if (ip->i_gid == *gp) {
357 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
360 /* Otherwise, check everyone else. */
367 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
371 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
376 ufs_getattr(struct vop_getattr_args *ap)
378 struct vnode *vp = ap->a_vp;
379 struct inode *ip = VTOI(vp);
380 struct vattr *vap = ap->a_vap;
383 * This may cause i_fsmid to be updated even if no change (0)
384 * is returned, but we should only write out the inode if non-zero
385 * is returned and if the mount is read-write.
387 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
388 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
390 ip->i_flag |= IN_LAZYMOD;
395 * Copy from inode table
397 vap->va_fsid = dev2udev(ip->i_dev);
398 vap->va_fileid = ip->i_number;
399 vap->va_mode = ip->i_mode & ~IFMT;
400 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
401 ip->i_effnlink : ip->i_nlink;
402 vap->va_uid = ip->i_uid;
403 vap->va_gid = ip->i_gid;
404 vap->va_rmajor = umajor(ip->i_rdev);
405 vap->va_rminor = uminor(ip->i_rdev);
406 vap->va_size = ip->i_din.di_size;
407 vap->va_atime.tv_sec = ip->i_atime;
408 vap->va_atime.tv_nsec = ip->i_atimensec;
409 vap->va_mtime.tv_sec = ip->i_mtime;
410 vap->va_mtime.tv_nsec = ip->i_mtimensec;
411 vap->va_ctime.tv_sec = ip->i_ctime;
412 vap->va_ctime.tv_nsec = ip->i_ctimensec;
413 vap->va_flags = ip->i_flags;
414 vap->va_gen = ip->i_gen;
415 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
416 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
417 vap->va_type = IFTOVT(ip->i_mode);
418 vap->va_filerev = ip->i_modrev;
419 vap->va_fsmid = ip->i_fsmid;
425 ufs_markatime(struct vop_markatime_args *ap)
427 struct vnode *vp = ap->a_vp;
428 struct inode *ip = VTOI(vp);
430 if (vp->v_mount->mnt_flag & MNT_RDONLY)
432 if (vp->v_mount->mnt_flag & MNT_NOATIME)
434 ip->i_flag |= IN_ACCESS;
435 VN_KNOTE(vp, NOTE_ATTRIB);
440 * Set attribute vnode op. called from several syscalls
442 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
443 * struct ucred *a_cred)
447 ufs_setattr(struct vop_setattr_args *ap)
449 struct vattr *vap = ap->a_vap;
450 struct vnode *vp = ap->a_vp;
451 struct inode *ip = VTOI(vp);
452 struct ucred *cred = ap->a_cred;
456 * Check for unsettable attributes.
458 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
459 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
460 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
461 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
464 if (vap->va_flags != VNOVAL) {
465 if (vp->v_mount->mnt_flag & MNT_RDONLY)
467 if (cred->cr_uid != ip->i_uid &&
468 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT)))
471 * Note that a root chflags becomes a user chflags when
472 * we are jailed, unless the jail.chflags_allowed sysctl
475 if (cred->cr_uid == 0 &&
476 (!jailed(cred) || jail_chflags_allowed)) {
478 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
481 ip->i_flags = vap->va_flags;
484 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
485 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
487 ip->i_flags &= SF_SETTABLE;
488 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
490 ip->i_flag |= IN_CHANGE;
491 if (vap->va_flags & (IMMUTABLE | APPEND))
494 if (ip->i_flags & (IMMUTABLE | APPEND))
497 * Go through the fields and update iff not VNOVAL.
499 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
500 if (vp->v_mount->mnt_flag & MNT_RDONLY)
502 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
505 if (vap->va_size != VNOVAL) {
507 * Disallow write attempts on read-only filesystems;
508 * unless the file is a socket, fifo, or a block or
509 * character device resident on the filesystem.
511 switch (vp->v_type) {
516 if (vp->v_mount->mnt_flag & MNT_RDONLY)
522 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
526 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
527 if (vp->v_mount->mnt_flag & MNT_RDONLY)
529 if (cred->cr_uid != ip->i_uid &&
530 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT)) &&
531 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
532 (error = VOP_ACCESS(vp, VWRITE, cred))))
534 if (vap->va_atime.tv_sec != VNOVAL)
535 ip->i_flag |= IN_ACCESS;
536 if (vap->va_mtime.tv_sec != VNOVAL)
537 ip->i_flag |= IN_CHANGE | IN_UPDATE;
539 if (vap->va_atime.tv_sec != VNOVAL) {
540 ip->i_atime = vap->va_atime.tv_sec;
541 ip->i_atimensec = vap->va_atime.tv_nsec;
543 if (vap->va_mtime.tv_sec != VNOVAL) {
544 ip->i_mtime = vap->va_mtime.tv_sec;
545 ip->i_mtimensec = vap->va_mtime.tv_nsec;
547 error = ffs_update(vp, 0);
552 if (vap->va_mode != (mode_t)VNOVAL) {
553 if (vp->v_mount->mnt_flag & MNT_RDONLY)
555 error = ufs_chmod(vp, (int)vap->va_mode, cred);
557 VN_KNOTE(vp, NOTE_ATTRIB);
562 * Change the mode on a file.
563 * Inode must be locked before calling.
566 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
568 struct inode *ip = VTOI(vp);
571 if (cred->cr_uid != ip->i_uid) {
572 error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT);
577 if (vp->v_type != VDIR && (mode & S_ISTXT))
579 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
582 ip->i_mode &= ~ALLPERMS;
583 ip->i_mode |= (mode & ALLPERMS);
584 ip->i_flag |= IN_CHANGE;
589 * Perform chown operation on inode ip;
590 * inode must be locked prior to call.
593 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
595 struct inode *ip = VTOI(vp);
604 if (uid == (uid_t)VNOVAL)
606 if (gid == (gid_t)VNOVAL)
609 * If we don't own the file, are trying to change the owner
610 * of the file, or are not a member of the target group,
611 * the caller must be superuser or the call fails.
613 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
614 (gid != ip->i_gid && !(cred->cr_gid == gid ||
615 groupmember((gid_t)gid, cred)))) &&
616 (error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT)))
621 if ((error = ufs_getinoquota(ip)) != 0)
624 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
625 ip->i_dquot[USRQUOTA] = NODQUOT;
628 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
629 ip->i_dquot[GRPQUOTA] = NODQUOT;
631 change = ip->i_blocks;
632 (void) ufs_chkdq(ip, -change, cred, CHOWN);
633 (void) ufs_chkiq(ip, -1, cred, CHOWN);
634 for (i = 0; i < MAXQUOTAS; i++) {
635 ufs_dqrele(vp, ip->i_dquot[i]);
636 ip->i_dquot[i] = NODQUOT;
642 if ((error = ufs_getinoquota(ip)) == 0) {
644 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
645 ip->i_dquot[USRQUOTA] = NODQUOT;
648 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
649 ip->i_dquot[GRPQUOTA] = NODQUOT;
651 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
652 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
655 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
657 for (i = 0; i < MAXQUOTAS; i++) {
658 ufs_dqrele(vp, ip->i_dquot[i]);
659 ip->i_dquot[i] = NODQUOT;
664 if (ufs_getinoquota(ip) == 0) {
666 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
667 ip->i_dquot[USRQUOTA] = NODQUOT;
670 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
671 ip->i_dquot[GRPQUOTA] = NODQUOT;
673 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
674 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
675 (void) ufs_getinoquota(ip);
679 if (ufs_getinoquota(ip))
680 panic("ufs_chown: lost quota");
682 ip->i_flag |= IN_CHANGE;
683 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
684 ip->i_mode &= ~(ISUID | ISGID);
691 * NB Currently unsupported.
693 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
698 ufs_mmap(struct vop_mmap_args *ap)
704 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
705 * struct componentname *a_cnp)
709 ufs_remove(struct vop_old_remove_args *ap)
712 struct vnode *vp = ap->a_vp;
713 struct vnode *dvp = ap->a_dvp;
717 #if 0 /* handled by kernel now */
718 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
719 (VTOI(dvp)->i_flags & APPEND)) {
724 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
725 VN_KNOTE(vp, NOTE_DELETE);
726 VN_KNOTE(dvp, NOTE_WRITE);
736 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
737 * struct componentname *a_cnp)
741 ufs_link(struct vop_old_link_args *ap)
743 struct vnode *vp = ap->a_vp;
744 struct vnode *tdvp = ap->a_tdvp;
745 struct componentname *cnp = ap->a_cnp;
747 struct direct newdir;
750 if (tdvp->v_mount != vp->v_mount) {
754 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
758 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
762 #if 0 /* handled by kernel now, also DragonFly allows this */
763 if (ip->i_flags & (IMMUTABLE | APPEND)) {
770 ip->i_flag |= IN_CHANGE;
771 if (DOINGSOFTDEP(vp))
772 softdep_change_linkcnt(ip);
773 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
775 ufs_makedirentry(ip, cnp, &newdir);
776 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
782 ip->i_flag |= IN_CHANGE;
783 if (DOINGSOFTDEP(vp))
784 softdep_change_linkcnt(ip);
790 VN_KNOTE(vp, NOTE_LINK);
791 VN_KNOTE(tdvp, NOTE_WRITE);
796 * whiteout vnode call
798 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
802 ufs_whiteout(struct vop_old_whiteout_args *ap)
804 struct vnode *dvp = ap->a_dvp;
805 struct componentname *cnp = ap->a_cnp;
806 struct direct newdir;
809 switch (ap->a_flags) {
811 /* 4.4 format directories support whiteout operations */
812 if (dvp->v_mount->mnt_maxsymlinklen > 0)
817 /* create a new directory whiteout */
819 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
820 panic("ufs_whiteout: old format filesystem");
824 newdir.d_namlen = cnp->cn_namelen;
825 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
826 newdir.d_type = DT_WHT;
827 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
831 /* remove an existing directory whiteout */
833 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
834 panic("ufs_whiteout: old format filesystem");
837 cnp->cn_flags &= ~CNP_DOWHITEOUT;
838 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
841 panic("ufs_whiteout: unknown op");
847 * Rename system call.
848 * rename("foo", "bar");
851 * link("foo", "bar");
853 * but ``atomically''. Can't do full commit without saving state in the
854 * inode on disk which isn't feasible at this time. Best we can do is
855 * always guarantee the target exists.
857 * Basic algorithm is:
859 * 1) Bump link count on source while we're linking it to the
860 * target. This also ensure the inode won't be deleted out
861 * from underneath us while we work (it may be truncated by
862 * a concurrent `trunc' or `open' for creation).
863 * 2) Link source to destination. If destination already exists,
865 * 3) Unlink source reference to inode if still around. If a
866 * directory was moved and the parent of the destination
867 * is different from the source, patch the ".." entry in the
870 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
871 * struct componentname *a_fcnp, struct vnode *a_tdvp,
872 * struct vnode *a_tvp, struct componentname *a_tcnp)
876 ufs_rename(struct vop_old_rename_args *ap)
878 struct vnode *tvp = ap->a_tvp;
879 struct vnode *tdvp = ap->a_tdvp;
880 struct vnode *fvp = ap->a_fvp;
881 struct vnode *fdvp = ap->a_fdvp;
882 struct componentname *tcnp = ap->a_tcnp;
883 struct componentname *fcnp = ap->a_fcnp;
884 struct inode *ip, *xp, *dp;
885 struct direct newdir;
886 ino_t oldparent = 0, newparent = 0;
887 int doingdirectory = 0;
888 int error = 0, ioflag;
891 * Check for cross-device rename.
893 if ((fvp->v_mount != tdvp->v_mount) ||
894 (tvp && (fvp->v_mount != tvp->v_mount))) {
908 #if 0 /* handled by kernel now */
909 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
910 (VTOI(tdvp)->i_flags & APPEND))) {
917 * Renaming a file to itself has no effect. The upper layers should
918 * not call us in that case. Temporarily just warn if they do.
921 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
926 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
930 * Note: now that fvp is locked we have to be sure to unlock it before
931 * using the 'abortit' target.
935 if (ip->i_nlink >= LINK_MAX) {
940 #if 0 /* handled by kernel now */
941 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
942 || (dp->i_flags & APPEND)) {
948 if ((ip->i_mode & IFMT) == IFDIR) {
950 * Avoid ".", "..", and aliases of "." for obvious reasons.
952 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
953 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
954 (ip->i_flag & IN_RENAME)) {
959 ip->i_flag |= IN_RENAME;
960 oldparent = dp->i_number;
963 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
966 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
967 * Cleanup fvp requirements so we can unlock it.
969 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
970 * is setup we can use the 'bad' target if we unlock fvp. We cannot
971 * use the abortit target anymore because of IN_RENAME.
980 * 1) Bump link count while we're moving stuff
981 * around. If we crash somewhere before
982 * completing our work, the link count
983 * may be wrong, but correctable.
987 ip->i_flag |= IN_CHANGE;
988 if (DOINGSOFTDEP(fvp))
989 softdep_change_linkcnt(ip);
990 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
991 DOINGASYNC(fvp)))) != 0) {
997 * If ".." must be changed (ie the directory gets a new
998 * parent) then the source directory must not be in the
999 * directory heirarchy above the target, as this would
1000 * orphan everything below the source directory. Also
1001 * the user must have write permission in the source so
1002 * as to be able to change "..". We must repeat the call
1003 * to namei, as the parent directory is unlocked by the
1004 * call to checkpath().
1006 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
1010 * We are now back to where we were in that fvp, fdvp are unlocked
1011 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
1012 * set. Only the bad target or, if we clean up tvp and tdvp, the
1013 * out target, may be used.
1015 if (oldparent != dp->i_number)
1016 newparent = dp->i_number;
1017 if (doingdirectory && newparent) {
1018 if (error) /* write access check above */
1022 * Once we start messing with tvp and tdvp we cannot use the
1023 * 'bad' target, only finish cleaning tdvp and tvp up and
1024 * use the 'out' target.
1026 * This cleans up tvp.
1034 * This is a real mess. ufs_checkpath vput's the target
1035 * directory so retain an extra ref and note that tdvp will
1036 * lose its lock on return. This leaves us with one good
1037 * ref after ufs_checkpath returns.
1040 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
1041 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1048 * relookup no longer messes with tdvp's refs. tdvp must be
1049 * unlocked on entry and will be locked on a successful
1052 error = relookup(tdvp, &tvp, tcnp);
1054 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1060 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1067 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1068 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1069 * be set if doingdirectory. This means we can use the 'bad' target
1074 * 2) If target doesn't exist, link the target
1075 * to the source and unlink the source.
1076 * Otherwise, rewrite the target directory
1077 * entry to reference the source inode and
1078 * expunge the original entry's existence.
1081 if (dp->i_dev != ip->i_dev)
1082 panic("ufs_rename: EXDEV");
1084 * Account for ".." in new directory.
1085 * When source and destination have the same
1086 * parent we don't fool with the link count.
1088 if (doingdirectory && newparent) {
1089 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1095 dp->i_flag |= IN_CHANGE;
1096 if (DOINGSOFTDEP(tdvp))
1097 softdep_change_linkcnt(dp);
1098 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1103 ufs_makedirentry(ip, tcnp, &newdir);
1104 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1106 if (doingdirectory && newparent) {
1109 dp->i_flag |= IN_CHANGE;
1110 if (DOINGSOFTDEP(tdvp))
1111 softdep_change_linkcnt(dp);
1112 (void)ffs_update(tdvp, 1);
1116 VN_KNOTE(tdvp, NOTE_WRITE);
1119 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1120 panic("ufs_rename: EXDEV");
1122 * Short circuit rename(foo, foo).
1124 if (xp->i_number == ip->i_number)
1125 panic("ufs_rename: same file");
1127 * If the parent directory is "sticky", then the user must
1128 * own the parent directory, or the destination of the rename,
1129 * otherwise the destination may not be changed (except by
1130 * root). This implements append-only directories.
1132 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1133 tcnp->cn_cred->cr_uid != dp->i_uid &&
1134 xp->i_uid != tcnp->cn_cred->cr_uid) {
1139 * Target must be empty if a directory and have no links
1140 * to it. Also, ensure source and target are compatible
1141 * (both directories, or both not directories).
1143 * Purge the file or directory being replaced from the
1146 if ((xp->i_mode&IFMT) == IFDIR) {
1147 if ((xp->i_effnlink > 2) ||
1148 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1152 if (!doingdirectory) {
1156 /* cache_purge removed - handled by VFS compat layer */
1157 } else if (doingdirectory == 0) {
1158 /* cache_purge removed - handled by VFS compat layer */
1164 * note: inode passed to ufs_dirrewrite() is 0 for a
1165 * non-directory file rename, 1 for a directory rename
1166 * in the same directory, and > 1 for an inode representing
1167 * the new directory.
1169 error = ufs_dirrewrite(dp, xp, ip->i_number,
1171 (doingdirectory && newparent) ?
1172 newparent : (ino_t)doingdirectory);
1175 if (doingdirectory) {
1178 if (DOINGSOFTDEP(tdvp))
1179 softdep_change_linkcnt(dp);
1182 if (DOINGSOFTDEP(tvp))
1183 softdep_change_linkcnt(xp);
1185 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1187 * Truncate inode. The only stuff left in the directory
1188 * is "." and "..". The "." reference is inconsequential
1189 * since we are quashing it. We have removed the "."
1190 * reference and the reference in the parent directory,
1191 * but there may be other hard links. The soft
1192 * dependency code will arrange to do these operations
1193 * after the parent directory entry has been deleted on
1194 * disk, so when running with that code we avoid doing
1199 dp->i_flag |= IN_CHANGE;
1202 xp->i_flag |= IN_CHANGE;
1203 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1204 error = ffs_truncate(tvp, (off_t)0, ioflag,
1209 VN_KNOTE(tdvp, NOTE_WRITE);
1211 VN_KNOTE(tvp, NOTE_DELETE);
1217 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1218 * unlocked) remain. We are about to overwrite fvp but we have to
1219 * keep 'ip' intact so we cannot release the old fvp, which is still
1220 * refd and accessible via ap->a_fvp.
1222 * This means we cannot use either 'bad' or 'out' to cleanup any
1227 * 3) Unlink the source.
1229 fcnp->cn_flags &= ~CNP_MODMASK;
1230 fcnp->cn_flags |= CNP_LOCKPARENT;
1231 error = relookup(fdvp, &fvp, fcnp);
1232 if (error || fvp == NULL) {
1234 * From name has disappeared. IN_RENAME will not be set if
1235 * we get past the panic so we don't have to clean it up.
1238 panic("ufs_rename: lost dir entry");
1240 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1246 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1249 * fdvp and fvp are locked.
1255 * Ensure that the directory entry still exists and has not
1256 * changed while the new name has been entered. If the source is
1257 * a file then the entry may have been unlinked or renamed. In
1258 * either case there is no further work to be done. If the source
1259 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1260 * flag ensures that it cannot be moved by another rename or removed
1261 * by a rmdir. Cleanup IN_RENAME.
1265 panic("ufs_rename: lost dir entry");
1268 * If the source is a directory with a
1269 * new parent, the link count of the old
1270 * parent directory must be decremented
1271 * and ".." set to point to the new parent.
1273 if (doingdirectory && newparent) {
1274 xp->i_offset = mastertemplate.dot_reclen;
1275 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1276 /* cache_purge removed - handled by VFS compat layer */
1278 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1279 xp->i_flag &= ~IN_RENAME;
1282 VN_KNOTE(fvp, NOTE_RENAME);
1294 ip->i_flag &= ~IN_RENAME;
1295 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1298 ip->i_flag |= IN_CHANGE;
1299 ip->i_flag &= ~IN_RENAME;
1300 if (DOINGSOFTDEP(fvp))
1301 softdep_change_linkcnt(ip);
1312 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1313 * struct componentname *a_cnp, struct vattr *a_vap)
1317 ufs_mkdir(struct vop_old_mkdir_args *ap)
1319 struct vnode *dvp = ap->a_dvp;
1320 struct vattr *vap = ap->a_vap;
1321 struct componentname *cnp = ap->a_cnp;
1322 struct inode *ip, *dp;
1325 struct dirtemplate dirtemplate, *dtp;
1326 struct direct newdir;
1331 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1335 dmode = vap->va_mode & 0777;
1338 * Must simulate part of ufs_makeinode here to acquire the inode,
1339 * but not have it entered in the parent directory. The entry is
1340 * made later after writing "." and ".." entries.
1342 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1346 ip->i_gid = dp->i_gid;
1350 struct ucred ucred, *ucp;
1354 * If we are hacking owners here, (only do this where told to)
1355 * and we are not giving it TO root, (would subvert quotas)
1356 * then go ahead and give it to the other user.
1357 * The new directory also inherits the SUID bit.
1358 * If user's UID and dir UID are the same,
1359 * 'give it away' so that the SUID is still forced on.
1361 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1362 (dp->i_mode & ISUID) && dp->i_uid) {
1364 ip->i_uid = dp->i_uid;
1366 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1368 * Make sure the correct user gets charged
1370 * Make a dummy credential for the victim.
1371 * XXX This seems to never be accessed out of
1372 * our context so a stack variable is ok.
1375 ucred.cr_uid = ip->i_uid;
1376 ucred.cr_ngroups = 1;
1377 ucred.cr_groups[0] = dp->i_gid;
1382 ip->i_uid = cnp->cn_cred->cr_uid;
1384 if ((error = ufs_getinoquota(ip)) ||
1385 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1386 ffs_vfree(tvp, ip->i_number, dmode);
1392 #else /* !SUIDDIR */
1393 ip->i_uid = cnp->cn_cred->cr_uid;
1395 if ((error = ufs_getinoquota(ip)) ||
1396 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1397 ffs_vfree(tvp, ip->i_number, dmode);
1402 #endif /* !SUIDDIR */
1403 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1405 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1408 if (DOINGSOFTDEP(tvp))
1409 softdep_change_linkcnt(ip);
1410 if (cnp->cn_flags & CNP_ISWHITEOUT)
1411 ip->i_flags |= UF_OPAQUE;
1414 * Bump link count in parent directory to reflect work done below.
1415 * Should be done before reference is created so cleanup is
1416 * possible if we crash.
1420 dp->i_flag |= IN_CHANGE;
1421 if (DOINGSOFTDEP(dvp))
1422 softdep_change_linkcnt(dp);
1423 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1428 * The vnode must have a VM object in order to issue buffer cache
1431 vinitvmio(tvp, DIRBLKSIZ);
1434 * Initialize directory with "." and ".." from static template.
1436 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1437 dtp = &mastertemplate;
1439 dtp = (struct dirtemplate *)&omastertemplate;
1441 dirtemplate.dot_ino = ip->i_number;
1442 dirtemplate.dotdot_ino = dp->i_number;
1443 vnode_pager_setsize(tvp, DIRBLKSIZ);
1444 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1447 ip->i_size = DIRBLKSIZ;
1448 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1449 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1450 if (DOINGSOFTDEP(tvp)) {
1452 * Ensure that the entire newly allocated block is a
1453 * valid directory so that future growth within the
1454 * block does not have to ensure that the block is
1455 * written before the inode.
1458 while (blkoff < bp->b_bcount) {
1460 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1461 blkoff += DIRBLKSIZ;
1464 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1465 DOINGASYNC(tvp)))) != 0) {
1470 * Directory set up, now install its entry in the parent directory.
1472 * If we are not doing soft dependencies, then we must write out the
1473 * buffer containing the new directory body before entering the new
1474 * name in the parent. If we are doing soft dependencies, then the
1475 * buffer containing the new directory body will be passed to and
1476 * released in the soft dependency code after the code has attached
1477 * an appropriate ordering dependency to the buffer which ensures that
1478 * the buffer is written before the new name is written in the parent.
1480 if (DOINGASYNC(dvp))
1482 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1484 ufs_makedirentry(ip, cnp, &newdir);
1485 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1489 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1494 dp->i_flag |= IN_CHANGE;
1495 if (DOINGSOFTDEP(dvp))
1496 softdep_change_linkcnt(dp);
1498 * No need to do an explicit VOP_TRUNCATE here, vrele will
1499 * do this for us because we set the link count to 0.
1503 ip->i_flag |= IN_CHANGE;
1504 if (DOINGSOFTDEP(tvp))
1505 softdep_change_linkcnt(ip);
1513 * Rmdir system call.
1515 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1516 * struct componentname *a_cnp)
1520 ufs_rmdir(struct vop_old_rmdir_args *ap)
1522 struct vnode *vp = ap->a_vp;
1523 struct vnode *dvp = ap->a_dvp;
1524 struct componentname *cnp = ap->a_cnp;
1525 struct inode *ip, *dp;
1532 * Do not remove a directory that is in the process of being renamed.
1533 * Verify the directory is empty (and valid). Rmdir ".." will not be
1534 * valid since ".." will contain a reference to the current directory
1535 * and thus be non-empty. Do not allow the removal of mounted on
1536 * directories (this can happen when an NFS exported filesystem
1537 * tries to remove a locally mounted on directory).
1540 if (ip->i_flag & IN_RENAME) {
1544 if (ip->i_effnlink != 2 ||
1545 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1549 #if 0 /* handled by kernel now */
1550 if ((dp->i_flags & APPEND)
1551 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1557 * Delete reference to directory before purging
1558 * inode. If we crash in between, the directory
1559 * will be reattached to lost+found,
1563 if (DOINGSOFTDEP(vp)) {
1564 softdep_change_linkcnt(dp);
1565 softdep_change_linkcnt(ip);
1567 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1571 if (DOINGSOFTDEP(vp)) {
1572 softdep_change_linkcnt(dp);
1573 softdep_change_linkcnt(ip);
1577 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1579 * Truncate inode. The only stuff left in the directory is "." and
1580 * "..". The "." reference is inconsequential since we are quashing
1581 * it. The soft dependency code will arrange to do these operations
1582 * after the parent directory entry has been deleted on disk, so
1583 * when running with that code we avoid doing them now.
1585 if (!DOINGSOFTDEP(vp)) {
1587 dp->i_flag |= IN_CHANGE;
1589 ip->i_flag |= IN_CHANGE;
1590 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1591 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1593 /* cache_purge removed - handled by VFS compat layer */
1595 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1596 if (ip->i_dirhash != NULL)
1597 ufsdirhash_free(ip);
1600 VN_KNOTE(vp, NOTE_DELETE);
1605 * symlink -- make a symbolic link
1607 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1608 * struct componentname *a_cnp, struct vattr *a_vap,
1613 ufs_symlink(struct vop_old_symlink_args *ap)
1615 struct vnode *vp, **vpp = ap->a_vpp;
1619 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1623 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1625 len = strlen(ap->a_target);
1626 if (len < vp->v_mount->mnt_maxsymlinklen) {
1628 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1630 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1633 * Make sure we have a VM object in order to use
1636 if (vp->v_object == NULL)
1638 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1639 UIO_SYSSPACE, IO_NODELOCKED,
1640 ap->a_cnp->cn_cred, NULL);
1648 * Vnode op for reading directories.
1650 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1651 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1655 ufs_readdir(struct vop_readdir_args *ap)
1657 struct uio *uio = ap->a_uio;
1658 struct vnode *vp = ap->a_vp;
1663 int offset; /* offset into buffer cache buffer */
1664 int eoffset; /* end of buffer clipped to file EOF */
1665 int pickup; /* pickup point */
1670 if (uio->uio_offset < 0)
1673 * Guess the number of cookies needed. Make sure we compute at
1674 * least 1, and no more then a reasonable limit.
1676 if (ap->a_ncookies) {
1677 ncookies = uio->uio_resid / 16 + 1;
1678 if (ncookies > 1024)
1680 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1682 ncookies = -1; /* force conditionals below */
1687 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1693 if (uio->uio_offset >= VTOI(vp)->i_size) {
1696 if (ap->a_ncookies) {
1697 *ap->a_ncookies = cookie_index;
1698 *ap->a_cookies = cookies;
1704 * Loop until we run out of cookies, we run out of user buffer,
1705 * or we hit the directory EOF.
1707 * Always start scans at the beginning of the buffer, don't trust
1708 * the offset supplied by userland.
1710 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1711 pickup = (int)(uio->uio_offset - bp->b_loffset);
1714 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1715 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1717 eoffset = bp->b_bcount;
1719 while (offset < eoffset) {
1720 dp = (struct direct *)(bp->b_data + offset);
1721 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1722 offset + dp->d_reclen > bp->b_bcount) {
1726 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1730 if (offset < pickup) {
1731 offset += dp->d_reclen;
1734 #if BYTE_ORDER == LITTLE_ENDIAN
1736 retval = vop_write_dirent(&error, uio,
1737 dp->d_ino, dp->d_namlen, dp->d_type,
1742 retval = vop_write_dirent(&error, uio,
1743 dp->d_ino, dp->d_type, dp->d_namlen,
1749 cookies[cookie_index] = bp->b_loffset + offset;
1751 offset += dp->d_reclen;
1752 if (cookie_index == ncookies)
1757 * This will align the next loop to the beginning of the
1758 * next block, and pickup will calculate to 0.
1760 uio->uio_offset = bp->b_loffset + offset;
1763 if (retval || error || cookie_index == ncookies ||
1764 uio->uio_offset >= VTOI(vp)->i_size) {
1769 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1772 * Report errors only if we didn't manage to read anything
1774 if (error && cookie_index == 0) {
1776 kfree(cookies, M_TEMP);
1777 *ap->a_ncookies = 0;
1778 *ap->a_cookies = NULL;
1783 *ap->a_ncookies = cookie_index;
1784 *ap->a_cookies = cookies;
1793 * Return target name of a symbolic link
1795 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1799 ufs_readlink(struct vop_readlink_args *ap)
1801 struct vnode *vp = ap->a_vp;
1802 struct inode *ip = VTOI(vp);
1806 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1807 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1808 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1813 * Perform the equivalent of an OPEN on vp so we can issue a
1816 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1820 * Calculate the logical to physical mapping if not done already,
1821 * then call the device strategy routine.
1823 * In order to be able to swap to a file, the VOP_BMAP operation may not
1824 * deadlock on memory. See ufs_bmap() for details.
1826 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1830 ufs_strategy(struct vop_strategy_args *ap)
1832 struct bio *bio = ap->a_bio;
1834 struct buf *bp = bio->bio_buf;
1835 struct vnode *vp = ap->a_vp;
1840 if (vp->v_type == VBLK || vp->v_type == VCHR)
1841 panic("ufs_strategy: spec");
1842 nbio = push_bio(bio);
1843 if (nbio->bio_offset == NOOFFSET) {
1844 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1845 NULL, NULL, bp->b_cmd);
1847 bp->b_error = error;
1848 bp->b_flags |= B_ERROR;
1849 /* I/O was never started on nbio, must biodone(bio) */
1853 if (nbio->bio_offset == NOOFFSET)
1856 if (nbio->bio_offset == NOOFFSET) {
1858 * We hit a hole in the file. The buffer has been zero-filled
1859 * so just biodone() it.
1863 vn_strategy(ip->i_devvp, nbio);
1869 * Print out the contents of an inode.
1871 * ufs_print(struct vnode *a_vp)
1875 ufs_print(struct vop_print_args *ap)
1877 struct vnode *vp = ap->a_vp;
1878 struct inode *ip = VTOI(vp);
1880 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1881 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1883 if (vp->v_type == VFIFO)
1885 lockmgr_printinfo(&vp->v_lock);
1891 * Read wrapper for special devices.
1893 * ufsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1894 * struct ucred *a_cred)
1898 ufsspec_read(struct vop_read_args *ap)
1905 resid = uio->uio_resid;
1906 error = VOCALL(&spec_vnode_vops, &ap->a_head);
1908 * The inode may have been revoked during the call, so it must not
1909 * be accessed blindly here or in the other wrapper functions.
1911 ip = VTOI(ap->a_vp);
1912 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1913 ip->i_flag |= IN_ACCESS;
1918 * Write wrapper for special devices.
1920 * ufsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1921 * struct ucred *a_cred)
1925 ufsspec_write(struct vop_write_args *ap)
1932 resid = uio->uio_resid;
1933 error = VOCALL(&spec_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 special devices.
1943 * Update the times on the inode then do device close.
1945 * ufsspec_close(struct vnode *a_vp, int a_fflag)
1949 ufsspec_close(struct vop_close_args *ap)
1951 struct vnode *vp = ap->a_vp;
1953 if (vp->v_sysref.refcnt > 1)
1955 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1959 * Read wrapper for fifos.
1961 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1962 * struct ucred *a_cred)
1966 ufsfifo_read(struct vop_read_args *ap)
1973 resid = uio->uio_resid;
1974 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1975 ip = VTOI(ap->a_vp);
1976 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1977 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1978 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1983 * Write wrapper for fifos.
1985 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1986 * struct ucred *a_cred)
1990 ufsfifo_write(struct vop_write_args *ap)
1997 resid = uio->uio_resid;
1998 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1999 ip = VTOI(ap->a_vp);
2000 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
2001 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
2006 * Close wrapper for fifos.
2008 * Update the times on the inode then do device close.
2010 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
2014 ufsfifo_close(struct vop_close_args *ap)
2016 struct vnode *vp = ap->a_vp;
2018 if (vp->v_sysref.refcnt > 1)
2020 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2024 * Kqfilter wrapper for fifos.
2026 * Fall through to ufs kqfilter routines if needed
2030 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
2034 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2036 error = ufs_kqfilter(ap);
2041 * Advisory record locking support
2043 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
2048 ufs_advlock(struct vop_advlock_args *ap)
2050 struct inode *ip = VTOI(ap->a_vp);
2052 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
2056 * Initialize the vnode associated with a new inode, handle aliased
2059 * Make sure directories have their VM object now rather then later,
2060 * saving us from having to check on all the myrid directory VOPs
2061 * that might be executed without a VOP_OPEN being performed.
2064 ufs_vinit(struct mount *mntp, struct vnode **vpp)
2073 vp->v_type = IFTOVT(ip->i_mode);
2075 switch(vp->v_type) {
2078 vp->v_ops = &mntp->mnt_vn_spec_ops;
2079 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
2082 vp->v_ops = &mntp->mnt_vn_fifo_ops;
2086 vinitvmio(vp, ip->i_size);
2089 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
2090 vinitvmio(vp, ip->i_size);
2097 if (ip->i_number == ROOTINO)
2098 vp->v_flag |= VROOT;
2100 * Initialize modrev times
2102 getmicrouptime(&tv);
2103 SETHIGH(ip->i_modrev, tv.tv_sec);
2104 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
2110 * Allocate a new inode.
2114 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2115 struct componentname *cnp)
2117 struct inode *ip, *pdir;
2118 struct direct newdir;
2124 if ((mode & IFMT) == 0)
2127 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2131 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
2132 ip->i_gid = pdir->i_gid;
2136 struct ucred ucred, *ucp;
2140 * If we are not the owner of the directory,
2141 * and we are hacking owners here, (only do this where told to)
2142 * and we are not giving it TO root, (would subvert quotas)
2143 * then go ahead and give it to the other user.
2144 * Note that this drops off the execute bits for security.
2146 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2147 (pdir->i_mode & ISUID) &&
2148 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2149 ip->i_uid = pdir->i_uid;
2153 * Make sure the correct user gets charged
2155 * Quickly knock up a dummy credential for the victim.
2156 * XXX This seems to never be accessed out of our
2157 * context so a stack variable is ok.
2160 ucred.cr_uid = ip->i_uid;
2161 ucred.cr_ngroups = 1;
2162 ucred.cr_groups[0] = pdir->i_gid;
2166 ip->i_uid = cnp->cn_cred->cr_uid;
2169 if ((error = ufs_getinoquota(ip)) ||
2170 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2171 ffs_vfree(tvp, ip->i_number, mode);
2177 #else /* !SUIDDIR */
2178 ip->i_uid = cnp->cn_cred->cr_uid;
2180 if ((error = ufs_getinoquota(ip)) ||
2181 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2182 ffs_vfree(tvp, ip->i_number, mode);
2187 #endif /* !SUIDDIR */
2188 ip->i_fsmid = cache_getnewfsmid();
2189 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2191 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2194 if (DOINGSOFTDEP(tvp))
2195 softdep_change_linkcnt(ip);
2196 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2197 priv_check_cred(cnp->cn_cred, PRIV_ROOT, 0)) {
2198 ip->i_mode &= ~ISGID;
2201 if (cnp->cn_flags & CNP_ISWHITEOUT)
2202 ip->i_flags |= UF_OPAQUE;
2205 * Regular files and directories need VM objects. Softlinks do
2206 * not (not immediately anyway).
2208 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2212 * Make sure inode goes to disk before directory entry.
2214 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2217 ufs_makedirentry(ip, cnp, &newdir);
2218 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2226 * Write error occurred trying to update the inode
2227 * or the directory so must deallocate the inode.
2231 ip->i_flag |= IN_CHANGE;
2232 if (DOINGSOFTDEP(tvp))
2233 softdep_change_linkcnt(ip);
2239 ufs_missingop(struct vop_generic_args *ap)
2241 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2242 return (EOPNOTSUPP);
2245 static struct filterops ufsread_filtops =
2246 { 1, NULL, filt_ufsdetach, filt_ufsread };
2247 static struct filterops ufswrite_filtops =
2248 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2249 static struct filterops ufsvnode_filtops =
2250 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2253 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2256 ufs_kqfilter(struct vop_kqfilter_args *ap)
2258 struct vnode *vp = ap->a_vp;
2259 struct knote *kn = ap->a_kn;
2262 switch (kn->kn_filter) {
2264 kn->kn_fop = &ufsread_filtops;
2267 kn->kn_fop = &ufswrite_filtops;
2270 kn->kn_fop = &ufsvnode_filtops;
2276 kn->kn_hook = (caddr_t)vp;
2278 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2279 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2280 lwkt_reltoken(&ilock);
2286 filt_ufsdetach(struct knote *kn)
2288 struct vnode *vp = (struct vnode *)kn->kn_hook;
2291 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2292 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2293 kn, knote, kn_selnext);
2294 lwkt_reltoken(&ilock);
2299 filt_ufsread(struct knote *kn, long hint)
2301 struct vnode *vp = (struct vnode *)kn->kn_hook;
2302 struct inode *ip = VTOI(vp);
2305 * filesystem is gone, so set the EOF flag and schedule
2306 * the knote for deletion.
2308 if (hint == NOTE_REVOKE) {
2309 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2313 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2314 return (kn->kn_data != 0);
2319 filt_ufswrite(struct knote *kn, long hint)
2322 * filesystem is gone, so set the EOF flag and schedule
2323 * the knote for deletion.
2325 if (hint == NOTE_REVOKE)
2326 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2333 filt_ufsvnode(struct knote *kn, long hint)
2335 if (kn->kn_sfflags & hint)
2336 kn->kn_fflags |= hint;
2337 if (hint == NOTE_REVOKE) {
2338 kn->kn_flags |= EV_EOF;
2341 return (kn->kn_fflags != 0);
2344 /* Global vfs data structures for ufs. */
2345 static struct vop_ops ufs_vnode_vops = {
2346 .vop_default = vop_defaultop,
2347 .vop_fsync = (void *)ufs_missingop,
2348 .vop_read = (void *)ufs_missingop,
2349 .vop_reallocblks = (void *)ufs_missingop,
2350 .vop_write = (void *)ufs_missingop,
2351 .vop_access = ufs_access,
2352 .vop_advlock = ufs_advlock,
2353 .vop_bmap = ufs_bmap,
2354 .vop_old_lookup = ufs_lookup,
2355 .vop_close = ufs_close,
2356 .vop_old_create = ufs_create,
2357 .vop_getattr = ufs_getattr,
2358 .vop_inactive = ufs_inactive,
2359 .vop_old_link = ufs_link,
2360 .vop_old_mkdir = ufs_mkdir,
2361 .vop_old_mknod = ufs_mknod,
2362 .vop_mmap = ufs_mmap,
2363 .vop_open = vop_stdopen,
2364 .vop_pathconf = vop_stdpathconf,
2365 .vop_poll = vop_stdpoll,
2366 .vop_kqfilter = ufs_kqfilter,
2367 .vop_print = ufs_print,
2368 .vop_readdir = ufs_readdir,
2369 .vop_readlink = ufs_readlink,
2370 .vop_reclaim = ufs_reclaim,
2371 .vop_old_remove = ufs_remove,
2372 .vop_old_rename = ufs_rename,
2373 .vop_old_rmdir = ufs_rmdir,
2374 .vop_setattr = ufs_setattr,
2375 .vop_markatime = ufs_markatime,
2376 .vop_strategy = ufs_strategy,
2377 .vop_old_symlink = ufs_symlink,
2378 .vop_old_whiteout = ufs_whiteout
2381 static struct vop_ops ufs_spec_vops = {
2382 .vop_default = spec_vnoperate,
2383 .vop_fsync = (void *)ufs_missingop,
2384 .vop_access = ufs_access,
2385 .vop_close = ufsspec_close,
2386 .vop_getattr = ufs_getattr,
2387 .vop_inactive = ufs_inactive,
2388 .vop_print = ufs_print,
2389 .vop_read = ufsspec_read,
2390 .vop_reclaim = ufs_reclaim,
2391 .vop_setattr = ufs_setattr,
2392 .vop_markatime = ufs_markatime,
2393 .vop_write = ufsspec_write
2396 static struct vop_ops ufs_fifo_vops = {
2397 .vop_default = fifo_vnoperate,
2398 .vop_fsync = (void *)ufs_missingop,
2399 .vop_access = ufs_access,
2400 .vop_close = ufsfifo_close,
2401 .vop_getattr = ufs_getattr,
2402 .vop_inactive = ufs_inactive,
2403 .vop_kqfilter = ufsfifo_kqfilter,
2404 .vop_print = ufs_print,
2405 .vop_read = ufsfifo_read,
2406 .vop_reclaim = ufs_reclaim,
2407 .vop_setattr = ufs_setattr,
2408 .vop_markatime = ufs_markatime,
2409 .vop_write = ufsfifo_write
2412 VNODEOP_SET(ufs_vnode_vops);
2413 VNODEOP_SET(ufs_spec_vops);
2414 VNODEOP_SET(ufs_fifo_vops);
2420 ufs_vnoperate(struct vop_generic_args *ap)
2422 return (VOCALL(&ufs_vnode_vops, ap));
2426 * ufs_vnoperatefifo()
2429 ufs_vnoperatefifo(struct vop_generic_args *ap)
2431 return (VOCALL(&ufs_fifo_vops, ap));
2435 * ufs_vnoperatespec()
2438 ufs_vnoperatespec(struct vop_generic_args *ap)
2440 return (VOCALL(&ufs_spec_vops, ap));