2 * Copyright (c) 1982, 1986, 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
35 * $FreeBSD: src/sys/ufs/ufs/ufs_vnops.c,v 1.131.2.8 2003/01/02 17:26:19 bde Exp $
38 #include "opt_quota.h"
39 #include "opt_suiddir.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
51 #include <sys/namei.h>
52 #include <sys/mount.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <sys/malloc.h>
56 #include <sys/dirent.h>
57 #include <sys/lockf.h>
58 #include <sys/event.h>
61 #include <sys/file.h> /* XXX */
65 #include <vm/vm_extern.h>
67 #include <vfs/fifofs/fifo.h>
73 #include "ufs_extern.h"
74 #include "ffs_extern.h"
80 static int ufs_access (struct vop_access_args *);
81 static int ufs_advlock (struct vop_advlock_args *);
82 static int ufs_chmod (struct vnode *, int, struct ucred *);
83 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *);
84 static int ufs_close (struct vop_close_args *);
85 static int ufs_create (struct vop_old_create_args *);
86 static int ufs_getattr (struct vop_getattr_args *);
87 static int ufs_link (struct vop_old_link_args *);
88 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *);
89 static int ufs_markatime (struct vop_markatime_args *);
90 static int ufs_missingop (struct vop_generic_args *ap);
91 static int ufs_mkdir (struct vop_old_mkdir_args *);
92 static int ufs_mknod (struct vop_old_mknod_args *);
93 static int ufs_print (struct vop_print_args *);
94 static int ufs_readdir (struct vop_readdir_args *);
95 static int ufs_readlink (struct vop_readlink_args *);
96 static int ufs_remove (struct vop_old_remove_args *);
97 static int ufs_rename (struct vop_old_rename_args *);
98 static int ufs_rmdir (struct vop_old_rmdir_args *);
99 static int ufs_setattr (struct vop_setattr_args *);
100 static int ufs_strategy (struct vop_strategy_args *);
101 static int ufs_symlink (struct vop_old_symlink_args *);
102 static int ufs_whiteout (struct vop_old_whiteout_args *);
103 static int ufsfifo_close (struct vop_close_args *);
104 static int ufsfifo_kqfilter (struct vop_kqfilter_args *);
105 static int ufsfifo_read (struct vop_read_args *);
106 static int ufsfifo_write (struct vop_write_args *);
107 static int filt_ufsread (struct knote *kn, long hint);
108 static int filt_ufswrite (struct knote *kn, long hint);
109 static int filt_ufsvnode (struct knote *kn, long hint);
110 static void filt_ufsdetach (struct knote *kn);
111 static int ufs_kqfilter (struct vop_kqfilter_args *ap);
113 #define VN_KNOTE(vp, b) \
114 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, (b))
116 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
119 * A virgin directory (no blushing please).
121 static struct dirtemplate mastertemplate = {
122 0, 12, DT_DIR, 1, ".",
123 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
125 static struct odirtemplate omastertemplate = {
127 0, DIRBLKSIZ - 12, 2, ".."
131 ufs_itimes(struct vnode *vp)
137 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
139 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
140 ip->i_flag |= IN_LAZYMOD;
142 ip->i_flag |= IN_MODIFIED;
144 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
146 if (ip->i_flag & IN_ACCESS) {
147 ip->i_atime = (uint32_t)ts.tv_sec;
148 ip->i_atime_ext = ts.tv_sec >> 32;
149 ip->i_atimensec = ts.tv_nsec;
151 if (ip->i_flag & IN_CHANGE) {
152 ip->i_ctime = (uint32_t)ts.tv_sec;
153 ip->i_ctime_ext = ts.tv_sec >> 32;
154 ip->i_ctimensec = ts.tv_nsec;
156 if (ip->i_flag & IN_UPDATE) {
157 if (ip->i_flag & IN_NOCOPYWRITE) {
158 if (vp->v_flag & VLASTWRITETS) {
159 ip->i_mtime = (uint32_t)
160 vp->v_lastwrite_ts.tv_sec;
162 vp->v_lastwrite_ts.tv_sec >> 32;
164 vp->v_lastwrite_ts.tv_nsec;
167 ip->i_mtime = (uint32_t)ts.tv_sec;
168 ip->i_mtime_ext = ts.tv_sec >> 32;
169 ip->i_mtimensec = ts.tv_nsec;
174 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
178 * Create a regular file
180 * ufs_create(struct vnode *a_dvp, struct vnode **a_vpp,
181 * struct componentname *a_cnp, struct vattr *a_vap)
185 ufs_create(struct vop_old_create_args *ap)
190 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
191 ap->a_dvp, ap->a_vpp, ap->a_cnp);
194 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
201 * ufs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
202 * struct componentname *a_cnp, struct vattr *a_vap)
207 ufs_mknod(struct vop_old_mknod_args *ap)
209 struct vattr *vap = ap->a_vap;
210 struct vnode **vpp = ap->a_vpp;
216 * UFS cannot represent the entire major/minor range supported by
219 if (vap->va_rmajor != VNOVAL &&
220 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
224 /* no special directory support */
225 if (vap->va_type == VDIR)
228 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
229 ap->a_dvp, vpp, ap->a_cnp);
232 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
234 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
235 if (vap->va_rmajor != VNOVAL) {
237 * Want to be able to use this to make badblock
238 * inodes, so don't truncate the dev number.
240 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
243 * Remove inode, then reload it through VFS_VGET so it is
244 * checked to see if it is an alias of an existing entry in
247 (*vpp)->v_type = VNON;
248 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
249 vgone_vxlocked(*vpp);
251 error = VFS_VGET(ap->a_dvp->v_mount, NULL, ino, vpp);
262 * Update the times on the inode.
264 * ufs_close(struct vnode *a_vp, int a_fflag)
269 ufs_close(struct vop_close_args *ap)
271 struct vnode *vp = ap->a_vp;
275 return (vop_stdclose(ap));
279 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
283 ufs_access(struct vop_access_args *ap)
285 struct vnode *vp = ap->a_vp;
286 struct inode *ip = VTOI(vp);
290 if (ap->a_mode & VWRITE) {
291 switch (vp->v_type) {
295 if ((error = ufs_getinoquota(ip)) != 0)
304 error = vop_helper_access(ap, ip->i_uid, ip->i_gid, ip->i_mode, 0);
309 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
314 ufs_getattr(struct vop_getattr_args *ap)
316 struct vnode *vp = ap->a_vp;
317 struct inode *ip = VTOI(vp);
318 struct vattr *vap = ap->a_vap;
322 * Copy from inode table
324 vap->va_fsid = devid_from_dev(ip->i_dev);
325 vap->va_fileid = ip->i_number;
326 vap->va_mode = ip->i_mode & ~IFMT;
327 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
328 ip->i_effnlink : ip->i_nlink;
329 vap->va_uid = ip->i_uid;
330 vap->va_gid = ip->i_gid;
331 vap->va_rmajor = umajor(ip->i_rdev);
332 vap->va_rminor = uminor(ip->i_rdev);
333 vap->va_size = ip->i_din.di_size;
334 vap->va_atime.tv_sec =
335 (time_t)(ip->i_atime | ((uint64_t)ip->i_atime_ext << 32));
336 vap->va_atime.tv_nsec = ip->i_atimensec;
337 vap->va_mtime.tv_sec =
338 (time_t)(ip->i_mtime | ((uint64_t)ip->i_mtime_ext << 32));
339 vap->va_mtime.tv_nsec = ip->i_mtimensec;
340 vap->va_ctime.tv_sec =
341 (time_t)(ip->i_ctime | ((uint64_t)ip->i_ctime_ext << 32));
342 vap->va_ctime.tv_nsec = ip->i_ctimensec;
343 vap->va_flags = ip->i_flags;
344 vap->va_gen = ip->i_gen;
345 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
346 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
347 vap->va_type = IFTOVT(ip->i_mode);
348 vap->va_filerev = ip->i_modrev;
354 ufs_markatime(struct vop_markatime_args *ap)
356 struct vnode *vp = ap->a_vp;
357 struct inode *ip = VTOI(vp);
359 if (vp->v_mount->mnt_flag & MNT_RDONLY)
361 if (vp->v_mount->mnt_flag & MNT_NOATIME)
363 ip->i_flag |= IN_ACCESS;
364 VN_KNOTE(vp, NOTE_ATTRIB);
369 * Set attribute vnode op. called from several syscalls
371 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
372 * struct ucred *a_cred)
376 ufs_setattr(struct vop_setattr_args *ap)
378 struct vattr *vap = ap->a_vap;
379 struct vnode *vp = ap->a_vp;
380 struct inode *ip = VTOI(vp);
381 struct ucred *cred = ap->a_cred;
385 * Check for unsettable attributes.
387 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
388 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
389 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
390 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
393 if (vap->va_flags != VNOVAL) {
394 if (vp->v_mount->mnt_flag & MNT_RDONLY)
396 if (cred->cr_uid != ip->i_uid &&
397 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
400 * Note that a root chflags becomes a user chflags when
401 * we are jailed, unless the jail vfs_chflags sysctl
404 if (cred->cr_uid == 0 &&
406 PRISON_CAP_ISSET(cred->cr_prison->pr_caps,
407 PRISON_CAP_VFS_CHFLAGS))) {
409 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
412 ip->i_flags = vap->va_flags;
415 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
416 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
418 ip->i_flags &= SF_SETTABLE;
419 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
421 ip->i_flag |= IN_CHANGE;
422 if (vap->va_flags & (IMMUTABLE | APPEND))
425 if (ip->i_flags & (IMMUTABLE | APPEND))
428 * Go through the fields and update iff not VNOVAL.
430 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
431 if (vp->v_mount->mnt_flag & MNT_RDONLY)
433 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
436 if (vap->va_size != VNOVAL) {
438 * Disallow write attempts on read-only filesystems;
439 * unless the file is a socket, fifo, or a block or
440 * character device resident on the filesystem.
442 switch (vp->v_type) {
447 if (vp->v_mount->mnt_flag & MNT_RDONLY)
453 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
457 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
458 if (vp->v_mount->mnt_flag & MNT_RDONLY)
460 if (cred->cr_uid != ip->i_uid &&
461 (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) &&
462 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
463 (error = VOP_EACCESS(vp, VWRITE, cred))))
465 if (vap->va_atime.tv_sec != VNOVAL)
466 ip->i_flag |= IN_ACCESS;
467 if (vap->va_mtime.tv_sec != VNOVAL)
468 ip->i_flag |= IN_CHANGE | IN_UPDATE;
470 if (vap->va_atime.tv_sec != VNOVAL) {
471 ip->i_atime = (uint32_t)vap->va_atime.tv_sec;
472 ip->i_atime_ext = vap->va_atime.tv_sec >> 32;
473 ip->i_atimensec = vap->va_atime.tv_nsec;
475 if (vap->va_mtime.tv_sec != VNOVAL) {
476 ip->i_mtime = (uint32_t)vap->va_mtime.tv_sec;
477 ip->i_mtime_ext = vap->va_mtime.tv_sec >> 32;
478 ip->i_mtimensec = vap->va_mtime.tv_nsec;
479 vclrflags(vp, VLASTWRITETS);
481 error = ffs_update(vp, 0);
486 if (vap->va_mode != (mode_t)VNOVAL) {
487 if (vp->v_mount->mnt_flag & MNT_RDONLY)
489 error = ufs_chmod(vp, (int)vap->va_mode, cred);
491 VN_KNOTE(vp, NOTE_ATTRIB);
496 * Change the mode on a file.
497 * Inode must be locked before calling.
500 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
502 struct inode *ip = VTOI(vp);
504 mode_t cur_mode = ip->i_mode;
506 error = vop_helper_chmod(vp, mode, cred, ip->i_uid, ip->i_gid,
511 if (cred->cr_uid != ip->i_uid) {
512 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
517 if (vp->v_type != VDIR && (mode & S_ISTXT))
519 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
523 ip->i_mode = cur_mode;
524 ip->i_flag |= IN_CHANGE;
529 * Perform chown operation on inode ip;
530 * inode must be locked prior to call.
533 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
535 struct inode *ip = VTOI(vp);
544 if (uid == (uid_t)VNOVAL)
546 if (gid == (gid_t)VNOVAL)
549 * If we don't own the file, are trying to change the owner
550 * of the file, or are not a member of the target group,
551 * the caller must be superuser or the call fails.
553 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
554 (gid != ip->i_gid && !(cred->cr_gid == gid ||
555 groupmember(gid, cred)))) &&
556 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
561 if ((error = ufs_getinoquota(ip)) != 0)
564 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
565 ip->i_dquot[USRQUOTA] = NODQUOT;
568 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
569 ip->i_dquot[GRPQUOTA] = NODQUOT;
571 change = ip->i_blocks;
572 (void) ufs_chkdq(ip, -change, cred, CHOWN);
573 (void) ufs_chkiq(ip, -1, cred, CHOWN);
574 for (i = 0; i < MAXQUOTAS; i++) {
575 ufs_dqrele(vp, ip->i_dquot[i]);
576 ip->i_dquot[i] = NODQUOT;
582 if ((error = ufs_getinoquota(ip)) == 0) {
584 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
585 ip->i_dquot[USRQUOTA] = NODQUOT;
588 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
589 ip->i_dquot[GRPQUOTA] = NODQUOT;
591 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
592 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
595 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
597 for (i = 0; i < MAXQUOTAS; i++) {
598 ufs_dqrele(vp, ip->i_dquot[i]);
599 ip->i_dquot[i] = NODQUOT;
604 if (ufs_getinoquota(ip) == 0) {
606 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
607 ip->i_dquot[USRQUOTA] = NODQUOT;
610 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
611 ip->i_dquot[GRPQUOTA] = NODQUOT;
613 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
614 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
615 (void) ufs_getinoquota(ip);
619 if (ufs_getinoquota(ip))
620 panic("ufs_chown: lost quota");
622 ip->i_flag |= IN_CHANGE;
623 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
624 ip->i_mode &= ~(ISUID | ISGID);
629 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
630 * struct componentname *a_cnp)
634 ufs_remove(struct vop_old_remove_args *ap)
637 struct vnode *vp = ap->a_vp;
638 struct vnode *dvp = ap->a_dvp;
642 #if 0 /* handled by kernel now */
643 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
644 (VTOI(dvp)->i_flags & APPEND)) {
649 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
650 VN_KNOTE(vp, NOTE_DELETE);
651 VN_KNOTE(dvp, NOTE_WRITE);
661 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
662 * struct componentname *a_cnp)
666 ufs_link(struct vop_old_link_args *ap)
668 struct vnode *vp = ap->a_vp;
669 struct vnode *tdvp = ap->a_tdvp;
670 struct componentname *cnp = ap->a_cnp;
672 struct direct newdir;
675 if (tdvp->v_mount != vp->v_mount) {
680 error = vn_lock(vp, LK_EXCLUSIVE | LK_FAILRECLAIM);
685 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
689 #if 0 /* handled by kernel now, also DragonFly allows this */
690 if (ip->i_flags & (IMMUTABLE | APPEND)) {
697 ip->i_flag |= IN_CHANGE;
698 if (DOINGSOFTDEP(vp))
699 softdep_change_linkcnt(ip);
700 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
702 ufs_makedirentry(ip, cnp, &newdir);
703 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
709 ip->i_flag |= IN_CHANGE;
710 if (DOINGSOFTDEP(vp))
711 softdep_change_linkcnt(ip);
717 VN_KNOTE(vp, NOTE_LINK);
718 VN_KNOTE(tdvp, NOTE_WRITE);
723 * whiteout vnode call
725 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
729 ufs_whiteout(struct vop_old_whiteout_args *ap)
731 struct vnode *dvp = ap->a_dvp;
732 struct componentname *cnp = ap->a_cnp;
733 struct direct newdir;
736 switch (ap->a_flags) {
738 /* 4.4 format directories support whiteout operations */
739 if (dvp->v_mount->mnt_maxsymlinklen > 0)
744 /* create a new directory whiteout */
746 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
747 panic("ufs_whiteout: old format filesystem");
750 newdir.d_ino = UFS_WINO;
751 newdir.d_namlen = cnp->cn_namelen;
752 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
753 newdir.d_type = DT_WHT;
754 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
758 /* remove an existing directory whiteout */
760 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
761 panic("ufs_whiteout: old format filesystem");
764 cnp->cn_flags &= ~CNP_DOWHITEOUT;
765 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
768 panic("ufs_whiteout: unknown op");
774 * Rename system call.
775 * rename("foo", "bar");
778 * link("foo", "bar");
780 * but ``atomically''. Can't do full commit without saving state in the
781 * inode on disk which isn't feasible at this time. Best we can do is
782 * always guarantee the target exists.
784 * Basic algorithm is:
786 * 1) Bump link count on source while we're linking it to the
787 * target. This also ensure the inode won't be deleted out
788 * from underneath us while we work (it may be truncated by
789 * a concurrent `trunc' or `open' for creation).
790 * 2) Link source to destination. If destination already exists,
792 * 3) Unlink source reference to inode if still around. If a
793 * directory was moved and the parent of the destination
794 * is different from the source, patch the ".." entry in the
797 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
798 * struct componentname *a_fcnp, struct vnode *a_tdvp,
799 * struct vnode *a_tvp, struct componentname *a_tcnp)
803 ufs_rename(struct vop_old_rename_args *ap)
805 struct vnode *tvp = ap->a_tvp;
806 struct vnode *tdvp = ap->a_tdvp;
807 struct vnode *fvp = ap->a_fvp;
808 struct vnode *fdvp = ap->a_fdvp;
809 struct componentname *tcnp = ap->a_tcnp;
810 struct componentname *fcnp = ap->a_fcnp;
811 struct inode *ip, *xp, *dp;
812 struct direct newdir;
813 ino_t oldparent = 0, newparent = 0;
814 int doingdirectory = 0;
815 int error = 0, ioflag;
818 * Check for cross-device rename.
820 if ((fvp->v_mount != tdvp->v_mount) ||
821 (tvp && (fvp->v_mount != tvp->v_mount))) {
835 #if 0 /* handled by kernel now */
836 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
837 (VTOI(tdvp)->i_flags & APPEND))) {
844 * Renaming a file to itself has no effect. The upper layers should
845 * not call us in that case. Temporarily just warn if they do.
848 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
853 error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
858 * Note: now that fvp is locked we have to be sure to unlock it before
859 * using the 'abortit' target.
863 if (ip->i_nlink >= LINK_MAX) {
868 #if 0 /* handled by kernel now */
869 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
870 || (dp->i_flags & APPEND)) {
876 if ((ip->i_mode & IFMT) == IFDIR) {
878 * Avoid ".", "..", and aliases of "." for obvious reasons.
880 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
881 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
882 (ip->i_flag & IN_RENAME)) {
887 ip->i_flag |= IN_RENAME;
888 oldparent = dp->i_number;
891 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
894 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
895 * Cleanup fvp requirements so we can unlock it.
897 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
898 * is setup we can use the 'bad' target if we unlock fvp. We cannot
899 * use the abortit target anymore because of IN_RENAME.
908 * 1) Bump link count while we're moving stuff
909 * around. If we crash somewhere before
910 * completing our work, the link count
911 * may be wrong, but correctable.
915 ip->i_flag |= IN_CHANGE;
916 if (DOINGSOFTDEP(fvp))
917 softdep_change_linkcnt(ip);
918 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
919 DOINGASYNC(fvp)))) != 0) {
925 * If ".." must be changed (ie the directory gets a new
926 * parent) then the source directory must not be in the
927 * directory heirarchy above the target, as this would
928 * orphan everything below the source directory. Also
929 * the user must have write permission in the source so
930 * as to be able to change "..". We must repeat the call
931 * to namei, as the parent directory is unlocked by the
932 * call to checkpath().
934 error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred);
938 * We are now back to where we were in that fvp, fdvp are unlocked
939 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
940 * set. Only the bad target or, if we clean up tvp and tdvp, the
941 * out target, may be used.
943 if (oldparent != dp->i_number)
944 newparent = dp->i_number;
945 if (doingdirectory && newparent) {
946 if (error) /* write access check above */
950 * Once we start messing with tvp and tdvp we cannot use the
951 * 'bad' target, only finish cleaning tdvp and tvp up and
952 * use the 'out' target.
954 * This cleans up tvp.
962 * This is a real mess. ufs_checkpath vput's the target
963 * directory so retain an extra ref and note that tdvp will
964 * lose its lock on return. This leaves us with one good
965 * ref after ufs_checkpath returns.
968 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
969 tcnp->cn_flags |= CNP_PDIRUNLOCK;
976 * relookup no longer messes with tdvp's refs. tdvp must be
977 * unlocked on entry and will be locked on a successful
980 error = relookup(tdvp, &tvp, tcnp);
982 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
988 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
995 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
996 * be NULL (xp will also be NULL in that case), and IN_RENAME will
997 * be set if doingdirectory. This means we can use the 'bad' target
1002 * 2) If target doesn't exist, link the target
1003 * to the source and unlink the source.
1004 * Otherwise, rewrite the target directory
1005 * entry to reference the source inode and
1006 * expunge the original entry's existence.
1009 if (dp->i_dev != ip->i_dev)
1010 panic("ufs_rename: EXDEV");
1012 * Account for ".." in new directory.
1013 * When source and destination have the same
1014 * parent we don't fool with the link count.
1016 if (doingdirectory && newparent) {
1017 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1023 dp->i_flag |= IN_CHANGE;
1024 if (DOINGSOFTDEP(tdvp))
1025 softdep_change_linkcnt(dp);
1026 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1031 ufs_makedirentry(ip, tcnp, &newdir);
1032 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1034 if (doingdirectory && newparent) {
1037 dp->i_flag |= IN_CHANGE;
1038 if (DOINGSOFTDEP(tdvp))
1039 softdep_change_linkcnt(dp);
1040 (void)ffs_update(tdvp, 1);
1044 VN_KNOTE(tdvp, NOTE_WRITE);
1047 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1048 panic("ufs_rename: EXDEV");
1050 * Short circuit rename(foo, foo).
1052 if (xp->i_number == ip->i_number)
1053 panic("ufs_rename: same file");
1055 * If the parent directory is "sticky", then the user must
1056 * own the parent directory, or the destination of the rename,
1057 * otherwise the destination may not be changed (except by
1058 * root). This implements append-only directories.
1060 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1061 tcnp->cn_cred->cr_uid != dp->i_uid &&
1062 xp->i_uid != tcnp->cn_cred->cr_uid) {
1067 * Target must be empty if a directory and have no links
1068 * to it. Also, ensure source and target are compatible
1069 * (both directories, or both not directories).
1071 * Purge the file or directory being replaced from the
1074 if ((xp->i_mode&IFMT) == IFDIR) {
1075 if ((xp->i_effnlink > 2) ||
1076 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1080 if (!doingdirectory) {
1084 /* cache_purge removed - handled by VFS compat layer */
1085 } else if (doingdirectory == 0) {
1086 /* cache_purge removed - handled by VFS compat layer */
1092 * note: inode passed to ufs_dirrewrite() is 0 for a
1093 * non-directory file rename, 1 for a directory rename
1094 * in the same directory, and > 1 for an inode representing
1095 * the new directory.
1097 error = ufs_dirrewrite(dp, xp, ip->i_number,
1099 (doingdirectory && newparent) ?
1100 newparent : (ino_t)doingdirectory);
1103 if (doingdirectory) {
1106 if (DOINGSOFTDEP(tdvp))
1107 softdep_change_linkcnt(dp);
1110 if (DOINGSOFTDEP(tvp))
1111 softdep_change_linkcnt(xp);
1113 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1115 * Truncate inode. The only stuff left in the directory
1116 * is "." and "..". The "." reference is inconsequential
1117 * since we are quashing it. We have removed the "."
1118 * reference and the reference in the parent directory,
1119 * but there may be other hard links. The soft
1120 * dependency code will arrange to do these operations
1121 * after the parent directory entry has been deleted on
1122 * disk, so when running with that code we avoid doing
1127 dp->i_flag |= IN_CHANGE;
1130 xp->i_flag |= IN_CHANGE;
1131 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1132 error = ffs_truncate(tvp, (off_t)0, ioflag,
1137 VN_KNOTE(tdvp, NOTE_WRITE);
1139 VN_KNOTE(tvp, NOTE_DELETE);
1145 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1146 * unlocked) remain. We are about to overwrite fvp but we have to
1147 * keep 'ip' intact so we cannot release the old fvp, which is still
1148 * refd and accessible via ap->a_fvp.
1150 * This means we cannot use either 'bad' or 'out' to cleanup any
1155 * 3) Unlink the source.
1157 fcnp->cn_flags &= ~CNP_MODMASK;
1158 fcnp->cn_flags |= CNP_LOCKPARENT;
1159 error = relookup(fdvp, &fvp, fcnp);
1160 if (error || fvp == NULL) {
1162 * From name has disappeared. IN_RENAME will not be set if
1163 * we get past the panic so we don't have to clean it up.
1166 panic("ufs_rename: lost dir entry");
1168 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1174 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1177 * fdvp and fvp are locked.
1183 * Ensure that the directory entry still exists and has not
1184 * changed while the new name has been entered. If the source is
1185 * a file then the entry may have been unlinked or renamed. In
1186 * either case there is no further work to be done. If the source
1187 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1188 * flag ensures that it cannot be moved by another rename or removed
1189 * by a rmdir. Cleanup IN_RENAME.
1193 panic("ufs_rename: lost dir entry");
1196 * If the source is a directory with a
1197 * new parent, the link count of the old
1198 * parent directory must be decremented
1199 * and ".." set to point to the new parent.
1201 if (doingdirectory && newparent) {
1202 xp->i_offset = mastertemplate.dot_reclen;
1203 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1204 /* cache_purge removed - handled by VFS compat layer */
1206 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1207 xp->i_flag &= ~IN_RENAME;
1210 VN_KNOTE(fvp, NOTE_RENAME);
1222 ip->i_flag &= ~IN_RENAME;
1223 if (vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM) == 0) {
1226 ip->i_flag |= IN_CHANGE;
1227 ip->i_flag &= ~IN_RENAME;
1228 if (DOINGSOFTDEP(fvp))
1229 softdep_change_linkcnt(ip);
1240 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1241 * struct componentname *a_cnp, struct vattr *a_vap)
1245 ufs_mkdir(struct vop_old_mkdir_args *ap)
1247 struct vnode *dvp = ap->a_dvp;
1248 struct vattr *vap = ap->a_vap;
1249 struct componentname *cnp = ap->a_cnp;
1250 struct inode *ip, *dp;
1253 struct dirtemplate dirtemplate, *dtp;
1254 struct direct newdir;
1259 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1263 dmode = vap->va_mode & 0777;
1266 * Must simulate part of ufs_makeinode here to acquire the inode,
1267 * but not have it entered in the parent directory. The entry is
1268 * made later after writing "." and ".." entries.
1270 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1274 ip->i_gid = dp->i_gid;
1278 struct ucred ucred, *ucp;
1282 * If we are hacking owners here, (only do this where told to)
1283 * and we are not giving it TO root, (would subvert quotas)
1284 * then go ahead and give it to the other user.
1285 * The new directory also inherits the SUID bit.
1286 * If user's UID and dir UID are the same,
1287 * 'give it away' so that the SUID is still forced on.
1289 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1290 (dp->i_mode & ISUID) && dp->i_uid) {
1292 ip->i_uid = dp->i_uid;
1294 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1296 * Make sure the correct user gets charged
1298 * Make a dummy credential for the victim.
1299 * XXX This seems to never be accessed out of
1300 * our context so a stack variable is ok.
1303 ucred.cr_uid = ip->i_uid;
1304 ucred.cr_ngroups = 1;
1305 ucred.cr_groups[0] = dp->i_gid;
1310 ip->i_uid = cnp->cn_cred->cr_uid;
1312 if ((error = ufs_getinoquota(ip)) ||
1313 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1314 ffs_vfree(tvp, ip->i_number, dmode);
1320 #else /* !SUIDDIR */
1321 ip->i_uid = cnp->cn_cred->cr_uid;
1323 if ((error = ufs_getinoquota(ip)) ||
1324 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1325 ffs_vfree(tvp, ip->i_number, dmode);
1330 #endif /* !SUIDDIR */
1331 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1333 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1336 if (DOINGSOFTDEP(tvp))
1337 softdep_change_linkcnt(ip);
1338 if (cnp->cn_flags & CNP_ISWHITEOUT)
1339 ip->i_flags |= UF_OPAQUE;
1342 * Bump link count in parent directory to reflect work done below.
1343 * Should be done before reference is created so cleanup is
1344 * possible if we crash.
1348 dp->i_flag |= IN_CHANGE;
1349 if (DOINGSOFTDEP(dvp))
1350 softdep_change_linkcnt(dp);
1351 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1356 * The vnode must have a VM object in order to issue buffer cache
1359 vinitvmio(tvp, DIRBLKSIZ, DIRBLKSIZ, -1);
1362 * Initialize directory with "." and ".." from static template.
1364 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1365 dtp = &mastertemplate;
1367 dtp = (struct dirtemplate *)&omastertemplate;
1369 dirtemplate.dot_ino = ip->i_number;
1370 dirtemplate.dotdot_ino = dp->i_number;
1371 nvnode_pager_setsize(tvp, DIRBLKSIZ, DIRBLKSIZ, -1);
1372 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1375 ip->i_size = DIRBLKSIZ;
1376 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1377 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1378 if (DOINGSOFTDEP(tvp)) {
1380 * Ensure that the entire newly allocated block is a
1381 * valid directory so that future growth within the
1382 * block does not have to ensure that the block is
1383 * written before the inode.
1386 while (blkoff < bp->b_bcount) {
1388 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1389 blkoff += DIRBLKSIZ;
1392 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1393 DOINGASYNC(tvp)))) != 0) {
1398 * Directory set up, now install its entry in the parent directory.
1400 * If we are not doing soft dependencies, then we must write out the
1401 * buffer containing the new directory body before entering the new
1402 * name in the parent. If we are doing soft dependencies, then the
1403 * buffer containing the new directory body will be passed to and
1404 * released in the soft dependency code after the code has attached
1405 * an appropriate ordering dependency to the buffer which ensures that
1406 * the buffer is written before the new name is written in the parent.
1408 if (DOINGASYNC(dvp))
1410 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1412 ufs_makedirentry(ip, cnp, &newdir);
1413 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1417 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1422 dp->i_flag |= IN_CHANGE;
1423 if (DOINGSOFTDEP(dvp))
1424 softdep_change_linkcnt(dp);
1426 * No need to do an explicit VOP_TRUNCATE here, vrele will
1427 * do this for us because we set the link count to 0.
1431 ip->i_flag |= IN_CHANGE;
1432 if (DOINGSOFTDEP(tvp))
1433 softdep_change_linkcnt(ip);
1441 * Rmdir system call.
1443 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1444 * struct componentname *a_cnp)
1448 ufs_rmdir(struct vop_old_rmdir_args *ap)
1450 struct vnode *vp = ap->a_vp;
1451 struct vnode *dvp = ap->a_dvp;
1452 struct componentname *cnp = ap->a_cnp;
1453 struct inode *ip, *dp;
1460 * Do not remove a directory that is in the process of being renamed.
1461 * Verify the directory is empty (and valid). Rmdir ".." will not be
1462 * valid since ".." will contain a reference to the current directory
1463 * and thus be non-empty. Do not allow the removal of mounted on
1464 * directories (this can happen when an NFS exported filesystem
1465 * tries to remove a locally mounted on directory).
1468 if (ip->i_flag & IN_RENAME) {
1472 if (ip->i_effnlink != 2 ||
1473 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1477 #if 0 /* handled by kernel now */
1478 if ((dp->i_flags & APPEND)
1479 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1485 * Delete reference to directory before purging
1486 * inode. If we crash in between, the directory
1487 * will be reattached to lost+found,
1491 if (DOINGSOFTDEP(vp)) {
1492 softdep_change_linkcnt(dp);
1493 softdep_change_linkcnt(ip);
1495 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1499 if (DOINGSOFTDEP(vp)) {
1500 softdep_change_linkcnt(dp);
1501 softdep_change_linkcnt(ip);
1505 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1507 * Truncate inode. The only stuff left in the directory is "." and
1508 * "..". The "." reference is inconsequential since we are quashing
1509 * it. The soft dependency code will arrange to do these operations
1510 * after the parent directory entry has been deleted on disk, so
1511 * when running with that code we avoid doing them now.
1513 if (!DOINGSOFTDEP(vp)) {
1515 dp->i_flag |= IN_CHANGE;
1517 ip->i_flag |= IN_CHANGE;
1518 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1519 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1521 /* cache_purge removed - handled by VFS compat layer */
1523 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1524 if (ip->i_dirhash != NULL)
1525 ufsdirhash_free(ip);
1528 VN_KNOTE(vp, NOTE_DELETE);
1533 * symlink -- make a symbolic link
1535 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1536 * struct componentname *a_cnp, struct vattr *a_vap,
1541 ufs_symlink(struct vop_old_symlink_args *ap)
1543 struct vnode *vp, **vpp = ap->a_vpp;
1547 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1551 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1553 len = strlen(ap->a_target);
1554 if (len < vp->v_mount->mnt_maxsymlinklen) {
1556 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1558 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1561 * Make sure we have a VM object in order to use
1564 if (vp->v_object == NULL)
1565 vinitvmio(vp, 0, PAGE_SIZE, -1);
1566 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1567 UIO_SYSSPACE, IO_NODELOCKED,
1568 ap->a_cnp->cn_cred, NULL);
1576 * Vnode op for reading directories.
1578 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1579 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1583 ufs_readdir(struct vop_readdir_args *ap)
1585 struct uio *uio = ap->a_uio;
1586 struct vnode *vp = ap->a_vp;
1591 int offset; /* offset into buffer cache buffer */
1592 int eoffset; /* end of buffer clipped to file EOF */
1593 int pickup; /* pickup point */
1598 if (uio->uio_offset < 0)
1601 * Guess the number of cookies needed. Make sure we compute at
1602 * least 1, and no more then a reasonable limit.
1604 if (ap->a_ncookies) {
1605 ncookies = uio->uio_resid / 16 + 1;
1606 if (ncookies > 1024)
1608 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1610 ncookies = -1; /* force conditionals below */
1615 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
1622 if (uio->uio_offset >= VTOI(vp)->i_size) {
1625 if (ap->a_ncookies) {
1626 *ap->a_ncookies = cookie_index;
1627 *ap->a_cookies = cookies;
1633 * Loop until we run out of cookies, we run out of user buffer,
1634 * or we hit the directory EOF.
1636 * Always start scans at the beginning of the buffer, don't trust
1637 * the offset supplied by userland.
1639 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1640 pickup = (int)(uio->uio_offset - bp->b_loffset);
1643 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1644 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1646 eoffset = bp->b_bcount;
1648 while (offset < eoffset) {
1649 dp = (struct direct *)(bp->b_data + offset);
1650 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1651 offset + dp->d_reclen > bp->b_bcount) {
1655 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1659 if (offset < pickup) {
1660 offset += dp->d_reclen;
1663 #if BYTE_ORDER == LITTLE_ENDIAN
1665 retval = vop_write_dirent(&error, uio,
1666 dp->d_ino, dp->d_namlen, dp->d_type,
1671 retval = vop_write_dirent(&error, uio,
1672 dp->d_ino, dp->d_type, dp->d_namlen,
1678 cookies[cookie_index] = bp->b_loffset + offset;
1680 offset += dp->d_reclen;
1681 if (cookie_index == ncookies)
1686 * This will align the next loop to the beginning of the
1687 * next block, and pickup will calculate to 0.
1689 uio->uio_offset = bp->b_loffset + offset;
1692 if (retval || error || cookie_index == ncookies ||
1693 uio->uio_offset >= VTOI(vp)->i_size) {
1698 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1701 * Report errors only if we didn't manage to read anything
1703 if (error && cookie_index == 0) {
1705 kfree(cookies, M_TEMP);
1706 *ap->a_ncookies = 0;
1707 *ap->a_cookies = NULL;
1712 *ap->a_ncookies = cookie_index;
1713 *ap->a_cookies = cookies;
1722 * Return target name of a symbolic link
1724 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1728 ufs_readlink(struct vop_readlink_args *ap)
1730 struct vnode *vp = ap->a_vp;
1731 struct inode *ip = VTOI(vp);
1735 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1736 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1737 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1742 * Perform the equivalent of an OPEN on vp so we can issue a
1745 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1749 * Calculate the logical to physical mapping if not done already,
1750 * then call the device strategy routine.
1752 * In order to be able to swap to a file, the VOP_BMAP operation may not
1753 * deadlock on memory. See ufs_bmap() for details.
1755 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1759 ufs_strategy(struct vop_strategy_args *ap)
1761 struct bio *bio = ap->a_bio;
1763 struct buf *bp = bio->bio_buf;
1764 struct vnode *vp = ap->a_vp;
1769 if (vp->v_type == VBLK || vp->v_type == VCHR)
1770 panic("ufs_strategy: spec");
1771 nbio = push_bio(bio);
1772 if (nbio->bio_offset == NOOFFSET) {
1773 error = VOP_BMAP(vp, bio->bio_offset, &nbio->bio_offset,
1774 NULL, NULL, bp->b_cmd);
1776 bp->b_error = error;
1777 bp->b_flags |= B_ERROR;
1778 /* I/O was never started on nbio, must biodone(bio) */
1782 if (nbio->bio_offset == NOOFFSET)
1785 if (nbio->bio_offset == NOOFFSET) {
1787 * We hit a hole in the file. The buffer has been zero-filled
1788 * so just biodone() it.
1792 vn_strategy(ip->i_devvp, nbio);
1798 * Print out the contents of an inode.
1800 * ufs_print(struct vnode *a_vp)
1804 ufs_print(struct vop_print_args *ap)
1806 struct vnode *vp = ap->a_vp;
1807 struct inode *ip = VTOI(vp);
1809 kprintf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1810 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1812 if (vp->v_type == VFIFO)
1814 lockmgr_printinfo(&vp->v_lock);
1820 * Read wrapper for fifos.
1822 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1823 * struct ucred *a_cred)
1827 ufsfifo_read(struct vop_read_args *ap)
1834 resid = uio->uio_resid;
1835 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1836 ip = VTOI(ap->a_vp);
1837 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1838 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1839 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1844 * Write wrapper for fifos.
1846 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1847 * struct ucred *a_cred)
1851 ufsfifo_write(struct vop_write_args *ap)
1858 resid = uio->uio_resid;
1859 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1860 ip = VTOI(ap->a_vp);
1861 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1862 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1867 * Close wrapper for fifos.
1869 * Update the times on the inode then do device close.
1871 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
1875 ufsfifo_close(struct vop_close_args *ap)
1877 struct vnode *vp = ap->a_vp;
1879 if (VREFCNT(vp) > 1)
1881 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1885 * Kqfilter wrapper for fifos.
1887 * Fall through to ufs kqfilter routines if needed
1891 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
1895 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1897 error = ufs_kqfilter(ap);
1902 * Advisory record locking support
1904 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
1909 ufs_advlock(struct vop_advlock_args *ap)
1911 struct inode *ip = VTOI(ap->a_vp);
1913 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
1917 * Initialize the vnode associated with a new inode, handle aliased
1920 * Make sure directories have their VM object now rather then later,
1921 * saving us from having to check on all the myrid directory VOPs
1922 * that might be executed without a VOP_OPEN being performed.
1925 ufs_vinit(struct mount *mntp, struct vnode **vpp)
1933 vp->v_type = IFTOVT(ip->i_mode);
1935 switch(vp->v_type) {
1938 vp->v_ops = &mntp->mnt_vn_spec_ops;
1939 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
1942 vp->v_ops = &mntp->mnt_vn_fifo_ops;
1946 vinitvmio(vp, ip->i_size,
1947 blkoffsize(ip->i_fs, ip, ip->i_size),
1948 blkoff(ip->i_fs, ip->i_size));
1951 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen) {
1952 vinitvmio(vp, ip->i_size,
1953 blkoffsize(ip->i_fs, ip, ip->i_size),
1954 blkoff(ip->i_fs, ip->i_size));
1962 if (ip->i_number == UFS_ROOTINO)
1963 vsetflags(vp, VROOT);
1965 * Initialize modrev times
1967 ip->i_modrev = init_va_filerev();
1973 * Allocate a new inode.
1977 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1978 struct componentname *cnp)
1980 struct inode *ip, *pdir;
1981 struct direct newdir;
1987 if ((mode & IFMT) == 0)
1990 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
1994 ip->i_flags = pdir->i_flags & (SF_NOHISTORY|UF_NOHISTORY|UF_NODUMP);
1995 ip->i_gid = pdir->i_gid;
1999 struct ucred ucred, *ucp;
2003 * If we are not the owner of the directory,
2004 * and we are hacking owners here, (only do this where told to)
2005 * and we are not giving it TO root, (would subvert quotas)
2006 * then go ahead and give it to the other user.
2007 * Note that this drops off the execute bits for security.
2009 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2010 (pdir->i_mode & ISUID) &&
2011 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2012 ip->i_uid = pdir->i_uid;
2016 * Make sure the correct user gets charged
2018 * Quickly knock up a dummy credential for the victim.
2019 * XXX This seems to never be accessed out of our
2020 * context so a stack variable is ok.
2023 ucred.cr_uid = ip->i_uid;
2024 ucred.cr_ngroups = 1;
2025 ucred.cr_groups[0] = pdir->i_gid;
2029 ip->i_uid = cnp->cn_cred->cr_uid;
2032 if ((error = ufs_getinoquota(ip)) ||
2033 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2034 ffs_vfree(tvp, ip->i_number, mode);
2040 #else /* !SUIDDIR */
2041 ip->i_uid = cnp->cn_cred->cr_uid;
2043 if ((error = ufs_getinoquota(ip)) ||
2044 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2045 ffs_vfree(tvp, ip->i_number, mode);
2050 #endif /* !SUIDDIR */
2051 ip->i_din.di_spare7E = 0;
2052 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2054 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2057 if (DOINGSOFTDEP(tvp))
2058 softdep_change_linkcnt(ip);
2059 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2060 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2061 ip->i_mode &= ~ISGID;
2064 if (cnp->cn_flags & CNP_ISWHITEOUT)
2065 ip->i_flags |= UF_OPAQUE;
2068 * Regular files and directories need VM objects. Softlinks do
2069 * not (not immediately anyway).
2071 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2072 vinitvmio(tvp, 0, PAGE_SIZE, -1);
2075 * Make sure inode goes to disk before directory entry.
2077 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2080 ufs_makedirentry(ip, cnp, &newdir);
2081 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2089 * Write error occurred trying to update the inode
2090 * or the directory so must deallocate the inode.
2094 ip->i_flag |= IN_CHANGE;
2095 if (DOINGSOFTDEP(tvp))
2096 softdep_change_linkcnt(ip);
2102 ufs_missingop(struct vop_generic_args *ap)
2104 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2105 return (EOPNOTSUPP);
2108 static struct filterops ufsread_filtops =
2109 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsread };
2110 static struct filterops ufswrite_filtops =
2111 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufswrite };
2112 static struct filterops ufsvnode_filtops =
2113 { FILTEROP_ISFD, NULL, filt_ufsdetach, filt_ufsvnode };
2116 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2119 ufs_kqfilter(struct vop_kqfilter_args *ap)
2121 struct vnode *vp = ap->a_vp;
2122 struct knote *kn = ap->a_kn;
2124 switch (kn->kn_filter) {
2126 kn->kn_fop = &ufsread_filtops;
2129 kn->kn_fop = &ufswrite_filtops;
2132 kn->kn_fop = &ufsvnode_filtops;
2135 return (EOPNOTSUPP);
2138 kn->kn_hook = (caddr_t)vp;
2140 /* XXX: kq token actually protects the list */
2141 lwkt_gettoken(&vp->v_token);
2142 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2143 lwkt_reltoken(&vp->v_token);
2149 filt_ufsdetach(struct knote *kn)
2151 struct vnode *vp = (struct vnode *)kn->kn_hook;
2153 lwkt_gettoken(&vp->v_token);
2154 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2155 lwkt_reltoken(&vp->v_token);
2160 filt_ufsread(struct knote *kn, long hint)
2162 struct vnode *vp = (struct vnode *)kn->kn_hook;
2163 struct inode *ip = VTOI(vp);
2167 * filesystem is gone, so set the EOF flag and schedule
2168 * the knote for deletion.
2170 if (hint == NOTE_REVOKE) {
2171 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2175 off = ip->i_size - kn->kn_fp->f_offset;
2176 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2177 if (kn->kn_sfflags & NOTE_OLDAPI)
2179 return (kn->kn_data != 0);
2184 filt_ufswrite(struct knote *kn, long hint)
2187 * filesystem is gone, so set the EOF flag and schedule
2188 * the knote for deletion.
2190 if (hint == NOTE_REVOKE)
2191 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2198 filt_ufsvnode(struct knote *kn, long hint)
2200 if (kn->kn_sfflags & hint)
2201 kn->kn_fflags |= hint;
2202 if (hint == NOTE_REVOKE) {
2203 kn->kn_flags |= (EV_EOF | EV_NODATA);
2206 return (kn->kn_fflags != 0);
2209 /* Global vfs data structures for ufs. */
2210 static struct vop_ops ufs_vnode_vops = {
2211 .vop_default = vop_defaultop,
2212 .vop_fsync = (void *)ufs_missingop,
2213 .vop_read = (void *)ufs_missingop,
2214 .vop_reallocblks = (void *)ufs_missingop,
2215 .vop_write = (void *)ufs_missingop,
2216 .vop_access = ufs_access,
2217 .vop_advlock = ufs_advlock,
2218 .vop_bmap = ufs_bmap,
2219 .vop_old_lookup = ufs_lookup,
2220 .vop_close = ufs_close,
2221 .vop_old_create = ufs_create,
2222 .vop_getattr = ufs_getattr,
2223 .vop_inactive = ufs_inactive,
2224 .vop_old_link = ufs_link,
2225 .vop_old_mkdir = ufs_mkdir,
2226 .vop_old_mknod = ufs_mknod,
2227 .vop_open = vop_stdopen,
2228 .vop_pathconf = vop_stdpathconf,
2229 .vop_kqfilter = ufs_kqfilter,
2230 .vop_print = ufs_print,
2231 .vop_readdir = ufs_readdir,
2232 .vop_readlink = ufs_readlink,
2233 .vop_reclaim = ufs_reclaim,
2234 .vop_old_remove = ufs_remove,
2235 .vop_old_rename = ufs_rename,
2236 .vop_old_rmdir = ufs_rmdir,
2237 .vop_setattr = ufs_setattr,
2238 .vop_markatime = ufs_markatime,
2239 .vop_strategy = ufs_strategy,
2240 .vop_old_symlink = ufs_symlink,
2241 .vop_old_whiteout = ufs_whiteout
2244 static struct vop_ops ufs_spec_vops = {
2245 .vop_default = vop_defaultop,
2246 .vop_fsync = (void *)ufs_missingop,
2247 .vop_access = ufs_access,
2248 .vop_close = ufs_close,
2249 .vop_getattr = ufs_getattr,
2250 .vop_inactive = ufs_inactive,
2251 .vop_print = ufs_print,
2252 .vop_read = vop_stdnoread,
2253 .vop_reclaim = ufs_reclaim,
2254 .vop_setattr = ufs_setattr,
2255 .vop_markatime = ufs_markatime,
2256 .vop_write = vop_stdnowrite
2259 static struct vop_ops ufs_fifo_vops = {
2260 .vop_default = fifo_vnoperate,
2261 .vop_fsync = (void *)ufs_missingop,
2262 .vop_access = ufs_access,
2263 .vop_close = ufsfifo_close,
2264 .vop_getattr = ufs_getattr,
2265 .vop_inactive = ufs_inactive,
2266 .vop_kqfilter = ufsfifo_kqfilter,
2267 .vop_print = ufs_print,
2268 .vop_read = ufsfifo_read,
2269 .vop_reclaim = ufs_reclaim,
2270 .vop_setattr = ufs_setattr,
2271 .vop_markatime = ufs_markatime,
2272 .vop_write = ufsfifo_write
2275 VNODEOP_SET(ufs_vnode_vops);
2276 VNODEOP_SET(ufs_spec_vops);
2277 VNODEOP_SET(ufs_fifo_vops);
2283 ufs_vnoperate(struct vop_generic_args *ap)
2285 return (VOCALL(&ufs_vnode_vops, ap));
2289 * ufs_vnoperatefifo()
2292 ufs_vnoperatefifo(struct vop_generic_args *ap)
2294 return (VOCALL(&ufs_fifo_vops, ap));
2298 * ufs_vnoperatespec()
2301 ufs_vnoperatespec(struct vop_generic_args *ap)
2303 return (VOCALL(&ufs_spec_vops, ap));