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 1.56 2006/09/05 00:55:51 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>
54 #include <sys/namei.h>
55 #include <sys/mount.h>
56 #include <sys/unistd.h>
57 #include <sys/vnode.h>
58 #include <sys/malloc.h>
59 #include <sys/dirent.h>
60 #include <sys/lockf.h>
61 #include <sys/event.h>
64 #include <sys/file.h> /* XXX */
68 #include <vm/vm_extern.h>
70 #include <vfs/fifofs/fifo.h>
76 #include "ufs_extern.h"
77 #include "ffs_extern.h"
82 static int ufs_access (struct vop_access_args *);
83 static int ufs_advlock (struct vop_advlock_args *);
84 static int ufs_chmod (struct vnode *, int, struct ucred *);
85 static int ufs_chown (struct vnode *, uid_t, gid_t, struct ucred *);
86 static int ufs_close (struct vop_close_args *);
87 static int ufs_create (struct vop_old_create_args *);
88 static int ufs_getattr (struct vop_getattr_args *);
89 static int ufs_link (struct vop_old_link_args *);
90 static int ufs_makeinode (int mode, struct vnode *, struct vnode **, struct componentname *);
91 static int ufs_missingop (struct vop_generic_args *ap);
92 static int ufs_mkdir (struct vop_old_mkdir_args *);
93 static int ufs_mknod (struct vop_old_mknod_args *);
94 static int ufs_mmap (struct vop_mmap_args *);
95 static int ufs_open (struct vop_open_args *);
96 static int ufs_pathconf (struct vop_pathconf_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;
223 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
224 ap->a_dvp, vpp, ap->a_cnp);
227 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
229 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
230 if (vap->va_rdev != VNOVAL) {
232 * Want to be able to use this to make badblock
233 * inodes, so don't truncate the dev number.
235 ip->i_rdev = vap->va_rdev;
238 * Remove inode, then reload it through VFS_VGET so it is
239 * checked to see if it is an alias of an existing entry in
242 (*vpp)->v_type = VNON;
243 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
246 error = VFS_VGET(ap->a_dvp->v_mount, ino, vpp);
259 * ufs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
260 * struct thread *a_td)
265 ufs_open(struct vop_open_args *ap)
267 struct vnode *vp = ap->a_vp;
270 * Files marked append-only must be opened for appending.
272 if ((VTOI(vp)->i_flags & APPEND) &&
273 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) {
277 return (vop_stdopen(ap));
283 * Update the times on the inode.
285 * ufs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
286 * struct thread *a_td)
291 ufs_close(struct vop_close_args *ap)
293 struct vnode *vp = ap->a_vp;
295 if (vp->v_usecount > 1)
297 return (vop_stdclose(ap));
301 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
302 * struct thread *a_td)
306 ufs_access(struct vop_access_args *ap)
308 struct vnode *vp = ap->a_vp;
309 struct inode *ip = VTOI(vp);
310 struct ucred *cred = ap->a_cred;
311 mode_t mask, mode = ap->a_mode;
319 * Disallow write attempts on read-only filesystems;
320 * unless the file is a socket, fifo, or a block or
321 * character device resident on the filesystem.
324 switch (vp->v_type) {
328 if (vp->v_mount->mnt_flag & MNT_RDONLY)
331 if ((error = ufs_getinoquota(ip)) != 0)
340 /* If immutable bit set, nobody gets to write it. */
341 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
344 /* Otherwise, user id 0 always gets access. */
345 if (cred->cr_uid == 0)
350 /* Otherwise, check the owner. */
351 if (cred->cr_uid == ip->i_uid) {
358 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
361 /* Otherwise, check the groups. */
362 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
363 if (ip->i_gid == *gp) {
370 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
373 /* Otherwise, check everyone else. */
380 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
384 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap,
385 * struct thread *a_td)
390 ufs_getattr(struct vop_getattr_args *ap)
392 struct vnode *vp = ap->a_vp;
393 struct inode *ip = VTOI(vp);
394 struct vattr *vap = ap->a_vap;
397 * This may cause i_fsmid to be updated even if no change (0)
398 * is returned, but we should only write out the inode if non-zero
399 * is returned and if the mount is read-write.
401 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
402 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
404 ip->i_flag |= IN_LAZYMOD;
409 * Copy from inode table
411 vap->va_fsid = dev2udev(ip->i_dev);
412 vap->va_fileid = ip->i_number;
413 vap->va_mode = ip->i_mode & ~IFMT;
414 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
415 ip->i_effnlink : ip->i_nlink;
416 vap->va_uid = ip->i_uid;
417 vap->va_gid = ip->i_gid;
418 vap->va_rdev = ip->i_rdev;
419 vap->va_size = ip->i_din.di_size;
420 vap->va_atime.tv_sec = ip->i_atime;
421 vap->va_atime.tv_nsec = ip->i_atimensec;
422 vap->va_mtime.tv_sec = ip->i_mtime;
423 vap->va_mtime.tv_nsec = ip->i_mtimensec;
424 vap->va_ctime.tv_sec = ip->i_ctime;
425 vap->va_ctime.tv_nsec = ip->i_ctimensec;
426 vap->va_flags = ip->i_flags;
427 vap->va_gen = ip->i_gen;
428 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
429 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
430 vap->va_type = IFTOVT(ip->i_mode);
431 vap->va_filerev = ip->i_modrev;
432 vap->va_fsmid = ip->i_fsmid;
437 * Set attribute vnode op. called from several syscalls
439 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
440 * struct ucred *a_cred, struct thread *a_td)
444 ufs_setattr(struct vop_setattr_args *ap)
446 struct vattr *vap = ap->a_vap;
447 struct vnode *vp = ap->a_vp;
448 struct inode *ip = VTOI(vp);
449 struct ucred *cred = ap->a_cred;
453 * Check for unsettable attributes.
455 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
456 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
457 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
458 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
461 if (vap->va_flags != VNOVAL) {
462 if (vp->v_mount->mnt_flag & MNT_RDONLY)
464 if (cred->cr_uid != ip->i_uid &&
465 (error = suser_cred(cred, PRISON_ROOT)))
468 * Note that a root chflags becomes a user chflags when
469 * we are jailed, unless the jail.chflags_allowed sysctl
472 if (cred->cr_uid == 0 &&
473 (!jailed(cred) || jail_chflags_allowed)) {
475 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
478 ip->i_flags = vap->va_flags;
481 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
482 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
484 ip->i_flags &= SF_SETTABLE;
485 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
487 ip->i_flag |= IN_CHANGE;
488 if (vap->va_flags & (IMMUTABLE | APPEND))
491 if (ip->i_flags & (IMMUTABLE | APPEND))
494 * Go through the fields and update iff not VNOVAL.
496 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
497 if (vp->v_mount->mnt_flag & MNT_RDONLY)
499 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
502 if (vap->va_size != VNOVAL) {
504 * Disallow write attempts on read-only filesystems;
505 * unless the file is a socket, fifo, or a block or
506 * character device resident on the filesystem.
508 switch (vp->v_type) {
513 if (vp->v_mount->mnt_flag & MNT_RDONLY)
519 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
523 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
524 if (vp->v_mount->mnt_flag & MNT_RDONLY)
526 if (cred->cr_uid != ip->i_uid &&
527 (error = suser_cred(cred, PRISON_ROOT)) &&
528 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
529 (error = VOP_ACCESS(vp, VWRITE, cred))))
531 if (vap->va_atime.tv_sec != VNOVAL)
532 ip->i_flag |= IN_ACCESS;
533 if (vap->va_mtime.tv_sec != VNOVAL)
534 ip->i_flag |= IN_CHANGE | IN_UPDATE;
536 if (vap->va_atime.tv_sec != VNOVAL) {
537 ip->i_atime = vap->va_atime.tv_sec;
538 ip->i_atimensec = vap->va_atime.tv_nsec;
540 if (vap->va_mtime.tv_sec != VNOVAL) {
541 ip->i_mtime = vap->va_mtime.tv_sec;
542 ip->i_mtimensec = vap->va_mtime.tv_nsec;
544 error = ffs_update(vp, 0);
549 if (vap->va_mode != (mode_t)VNOVAL) {
550 if (vp->v_mount->mnt_flag & MNT_RDONLY)
552 error = ufs_chmod(vp, (int)vap->va_mode, cred);
554 VN_KNOTE(vp, NOTE_ATTRIB);
559 * Change the mode on a file.
560 * Inode must be locked before calling.
563 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
565 struct inode *ip = VTOI(vp);
568 if (cred->cr_uid != ip->i_uid) {
569 error = suser_cred(cred, PRISON_ROOT);
574 if (vp->v_type != VDIR && (mode & S_ISTXT))
576 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
579 ip->i_mode &= ~ALLPERMS;
580 ip->i_mode |= (mode & ALLPERMS);
581 ip->i_flag |= IN_CHANGE;
586 * Perform chown operation on inode ip;
587 * inode must be locked prior to call.
590 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
592 struct inode *ip = VTOI(vp);
601 if (uid == (uid_t)VNOVAL)
603 if (gid == (gid_t)VNOVAL)
606 * If we don't own the file, are trying to change the owner
607 * of the file, or are not a member of the target group,
608 * the caller must be superuser or the call fails.
610 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
611 (gid != ip->i_gid && !(cred->cr_gid == gid ||
612 groupmember((gid_t)gid, cred)))) &&
613 (error = suser_cred(cred, PRISON_ROOT)))
618 if ((error = ufs_getinoquota(ip)) != 0)
621 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
622 ip->i_dquot[USRQUOTA] = NODQUOT;
625 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
626 ip->i_dquot[GRPQUOTA] = NODQUOT;
628 change = ip->i_blocks;
629 (void) ufs_chkdq(ip, -change, cred, CHOWN);
630 (void) ufs_chkiq(ip, -1, cred, CHOWN);
631 for (i = 0; i < MAXQUOTAS; i++) {
632 ufs_dqrele(vp, ip->i_dquot[i]);
633 ip->i_dquot[i] = NODQUOT;
639 if ((error = ufs_getinoquota(ip)) == 0) {
641 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
642 ip->i_dquot[USRQUOTA] = NODQUOT;
645 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
646 ip->i_dquot[GRPQUOTA] = NODQUOT;
648 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
649 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
652 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
654 for (i = 0; i < MAXQUOTAS; i++) {
655 ufs_dqrele(vp, ip->i_dquot[i]);
656 ip->i_dquot[i] = NODQUOT;
661 if (ufs_getinoquota(ip) == 0) {
663 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
664 ip->i_dquot[USRQUOTA] = NODQUOT;
667 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
668 ip->i_dquot[GRPQUOTA] = NODQUOT;
670 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
671 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
672 (void) ufs_getinoquota(ip);
676 if (ufs_getinoquota(ip))
677 panic("ufs_chown: lost quota");
679 ip->i_flag |= IN_CHANGE;
680 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
681 ip->i_mode &= ~(ISUID | ISGID);
688 * NB Currently unsupported.
690 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
691 * struct thread *a_td)
696 ufs_mmap(struct vop_mmap_args *ap)
702 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
703 * struct componentname *a_cnp)
707 ufs_remove(struct vop_old_remove_args *ap)
710 struct vnode *vp = ap->a_vp;
711 struct vnode *dvp = ap->a_dvp;
715 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
716 (VTOI(dvp)->i_flags & APPEND)) {
720 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
721 VN_KNOTE(vp, NOTE_DELETE);
722 VN_KNOTE(dvp, NOTE_WRITE);
730 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
731 * struct componentname *a_cnp)
735 ufs_link(struct vop_old_link_args *ap)
737 struct vnode *vp = ap->a_vp;
738 struct vnode *tdvp = ap->a_tdvp;
739 struct componentname *cnp = ap->a_cnp;
741 struct direct newdir;
744 if (tdvp->v_mount != vp->v_mount) {
748 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
752 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
756 if (ip->i_flags & (IMMUTABLE | APPEND)) {
762 ip->i_flag |= IN_CHANGE;
763 if (DOINGSOFTDEP(vp))
764 softdep_change_linkcnt(ip);
765 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
767 ufs_makedirentry(ip, cnp, &newdir);
768 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
774 ip->i_flag |= IN_CHANGE;
775 if (DOINGSOFTDEP(vp))
776 softdep_change_linkcnt(ip);
782 VN_KNOTE(vp, NOTE_LINK);
783 VN_KNOTE(tdvp, NOTE_WRITE);
788 * whiteout vnode call
790 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
794 ufs_whiteout(struct vop_old_whiteout_args *ap)
796 struct vnode *dvp = ap->a_dvp;
797 struct componentname *cnp = ap->a_cnp;
798 struct direct newdir;
801 switch (ap->a_flags) {
803 /* 4.4 format directories support whiteout operations */
804 if (dvp->v_mount->mnt_maxsymlinklen > 0)
809 /* create a new directory whiteout */
811 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
812 panic("ufs_whiteout: old format filesystem");
816 newdir.d_namlen = cnp->cn_namelen;
817 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
818 newdir.d_type = DT_WHT;
819 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
823 /* remove an existing directory whiteout */
825 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
826 panic("ufs_whiteout: old format filesystem");
829 cnp->cn_flags &= ~CNP_DOWHITEOUT;
830 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
833 panic("ufs_whiteout: unknown op");
839 * Rename system call.
840 * rename("foo", "bar");
843 * link("foo", "bar");
845 * but ``atomically''. Can't do full commit without saving state in the
846 * inode on disk which isn't feasible at this time. Best we can do is
847 * always guarantee the target exists.
849 * Basic algorithm is:
851 * 1) Bump link count on source while we're linking it to the
852 * target. This also ensure the inode won't be deleted out
853 * from underneath us while we work (it may be truncated by
854 * a concurrent `trunc' or `open' for creation).
855 * 2) Link source to destination. If destination already exists,
857 * 3) Unlink source reference to inode if still around. If a
858 * directory was moved and the parent of the destination
859 * is different from the source, patch the ".." entry in the
862 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
863 * struct componentname *a_fcnp, struct vnode *a_tdvp,
864 * struct vnode *a_tvp, struct componentname *a_tcnp)
868 ufs_rename(struct vop_old_rename_args *ap)
870 struct vnode *tvp = ap->a_tvp;
871 struct vnode *tdvp = ap->a_tdvp;
872 struct vnode *fvp = ap->a_fvp;
873 struct vnode *fdvp = ap->a_fdvp;
874 struct componentname *tcnp = ap->a_tcnp;
875 struct componentname *fcnp = ap->a_fcnp;
876 struct inode *ip, *xp, *dp;
877 struct direct newdir;
878 ino_t oldparent = 0, newparent = 0;
879 int doingdirectory = 0;
880 int error = 0, ioflag;
883 * Check for cross-device rename.
885 if ((fvp->v_mount != tdvp->v_mount) ||
886 (tvp && (fvp->v_mount != tvp->v_mount))) {
900 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
901 (VTOI(tdvp)->i_flags & APPEND))) {
907 * Renaming a file to itself has no effect. The upper layers should
908 * not call us in that case. Temporarily just warn if they do.
911 printf("ufs_rename: fvp == tvp (can't happen)\n");
916 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
920 * Note: now that fvp is locked we have to be sure to unlock it before
921 * using the 'abortit' target.
925 if (ip->i_nlink >= LINK_MAX) {
930 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
931 || (dp->i_flags & APPEND)) {
936 if ((ip->i_mode & IFMT) == IFDIR) {
938 * Avoid ".", "..", and aliases of "." for obvious reasons.
940 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
941 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
942 (ip->i_flag & IN_RENAME)) {
947 ip->i_flag |= IN_RENAME;
948 oldparent = dp->i_number;
951 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
954 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
955 * Cleanup fvp requirements so we can unlock it.
957 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
958 * is setup we can use the 'bad' target if we unlock fvp. We cannot
959 * use the abortit target anymore because of IN_RENAME.
968 * 1) Bump link count while we're moving stuff
969 * around. If we crash somewhere before
970 * completing our work, the link count
971 * may be wrong, but correctable.
975 ip->i_flag |= IN_CHANGE;
976 if (DOINGSOFTDEP(fvp))
977 softdep_change_linkcnt(ip);
978 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
979 DOINGASYNC(fvp)))) != 0) {
985 * If ".." must be changed (ie the directory gets a new
986 * parent) then the source directory must not be in the
987 * directory heirarchy above the target, as this would
988 * orphan everything below the source directory. Also
989 * the user must have write permission in the source so
990 * as to be able to change "..". We must repeat the call
991 * to namei, as the parent directory is unlocked by the
992 * call to checkpath().
994 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
998 * We are now back to where we were in that fvp, fdvp are unlocked
999 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
1000 * set. Only the bad target or, if we clean up tvp and tdvp, the
1001 * out target, may be used.
1003 if (oldparent != dp->i_number)
1004 newparent = dp->i_number;
1005 if (doingdirectory && newparent) {
1006 if (error) /* write access check above */
1010 * Once we start messing with tvp and tdvp we cannot use the
1011 * 'bad' target, only finish cleaning tdvp and tvp up and
1012 * use the 'out' target.
1014 * This cleans up tvp.
1022 * This is a real mess. ufs_checkpath vput's the target
1023 * directory so retain an extra ref and note that tdvp will
1024 * lose its lock on return. This leaves us with one good
1025 * ref after ufs_checkpath returns.
1028 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
1029 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1036 * relookup no longer messes with tdvp's refs. tdvp must be
1037 * unlocked on entry and will be locked on a successful
1040 error = relookup(tdvp, &tvp, tcnp);
1042 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1048 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1055 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1056 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1057 * be set if doingdirectory. This means we can use the 'bad' target
1062 * 2) If target doesn't exist, link the target
1063 * to the source and unlink the source.
1064 * Otherwise, rewrite the target directory
1065 * entry to reference the source inode and
1066 * expunge the original entry's existence.
1069 if (dp->i_dev != ip->i_dev)
1070 panic("ufs_rename: EXDEV");
1072 * Account for ".." in new directory.
1073 * When source and destination have the same
1074 * parent we don't fool with the link count.
1076 if (doingdirectory && newparent) {
1077 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1083 dp->i_flag |= IN_CHANGE;
1084 if (DOINGSOFTDEP(tdvp))
1085 softdep_change_linkcnt(dp);
1086 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1091 ufs_makedirentry(ip, tcnp, &newdir);
1092 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1094 if (doingdirectory && newparent) {
1097 dp->i_flag |= IN_CHANGE;
1098 if (DOINGSOFTDEP(tdvp))
1099 softdep_change_linkcnt(dp);
1100 (void)ffs_update(tdvp, 1);
1104 VN_KNOTE(tdvp, NOTE_WRITE);
1107 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1108 panic("ufs_rename: EXDEV");
1110 * Short circuit rename(foo, foo).
1112 if (xp->i_number == ip->i_number)
1113 panic("ufs_rename: same file");
1115 * If the parent directory is "sticky", then the user must
1116 * own the parent directory, or the destination of the rename,
1117 * otherwise the destination may not be changed (except by
1118 * root). This implements append-only directories.
1120 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1121 tcnp->cn_cred->cr_uid != dp->i_uid &&
1122 xp->i_uid != tcnp->cn_cred->cr_uid) {
1127 * Target must be empty if a directory and have no links
1128 * to it. Also, ensure source and target are compatible
1129 * (both directories, or both not directories).
1131 * Purge the file or directory being replaced from the
1134 if ((xp->i_mode&IFMT) == IFDIR) {
1135 if ((xp->i_effnlink > 2) ||
1136 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1140 if (!doingdirectory) {
1144 /* cache_purge removed - handled by VFS compat layer */
1145 } else if (doingdirectory == 0) {
1146 /* cache_purge removed - handled by VFS compat layer */
1152 * note: inode passed to ufs_dirrewrite() is 0 for a
1153 * non-directory file rename, 1 for a directory rename
1154 * in the same directory, and > 1 for an inode representing
1155 * the new directory.
1157 error = ufs_dirrewrite(dp, xp, ip->i_number,
1159 (doingdirectory && newparent) ?
1160 newparent : (ino_t)doingdirectory);
1163 if (doingdirectory) {
1166 if (DOINGSOFTDEP(tdvp))
1167 softdep_change_linkcnt(dp);
1170 if (DOINGSOFTDEP(tvp))
1171 softdep_change_linkcnt(xp);
1173 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1175 * Truncate inode. The only stuff left in the directory
1176 * is "." and "..". The "." reference is inconsequential
1177 * since we are quashing it. We have removed the "."
1178 * reference and the reference in the parent directory,
1179 * but there may be other hard links. The soft
1180 * dependency code will arrange to do these operations
1181 * after the parent directory entry has been deleted on
1182 * disk, so when running with that code we avoid doing
1187 dp->i_flag |= IN_CHANGE;
1190 xp->i_flag |= IN_CHANGE;
1191 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1192 error = ffs_truncate(tvp, (off_t)0, ioflag,
1197 VN_KNOTE(tdvp, NOTE_WRITE);
1199 VN_KNOTE(tvp, NOTE_DELETE);
1205 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1206 * unlocked) remain. We are about to overwrite fvp but we have to
1207 * keep 'ip' intact so we cannot release the old fvp, which is still
1208 * refd and accessible via ap->a_fvp.
1210 * This means we cannot use either 'bad' or 'out' to cleanup any
1215 * 3) Unlink the source.
1217 fcnp->cn_flags &= ~CNP_MODMASK;
1218 fcnp->cn_flags |= CNP_LOCKPARENT;
1219 error = relookup(fdvp, &fvp, fcnp);
1220 if (error || fvp == NULL) {
1222 * From name has disappeared. IN_RENAME will not be set if
1223 * we get past the panic so we don't have to clean it up.
1226 panic("ufs_rename: lost dir entry");
1228 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1234 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1237 * fdvp and fvp are locked.
1243 * Ensure that the directory entry still exists and has not
1244 * changed while the new name has been entered. If the source is
1245 * a file then the entry may have been unlinked or renamed. In
1246 * either case there is no further work to be done. If the source
1247 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1248 * flag ensures that it cannot be moved by another rename or removed
1249 * by a rmdir. Cleanup IN_RENAME.
1253 panic("ufs_rename: lost dir entry");
1256 * If the source is a directory with a
1257 * new parent, the link count of the old
1258 * parent directory must be decremented
1259 * and ".." set to point to the new parent.
1261 if (doingdirectory && newparent) {
1262 xp->i_offset = mastertemplate.dot_reclen;
1263 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1264 /* cache_purge removed - handled by VFS compat layer */
1266 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1267 xp->i_flag &= ~IN_RENAME;
1270 VN_KNOTE(fvp, NOTE_RENAME);
1282 ip->i_flag &= ~IN_RENAME;
1283 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1286 ip->i_flag |= IN_CHANGE;
1287 ip->i_flag &= ~IN_RENAME;
1288 if (DOINGSOFTDEP(fvp))
1289 softdep_change_linkcnt(ip);
1300 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1301 * struct componentname *a_cnp, struct vattr *a_vap)
1305 ufs_mkdir(struct vop_old_mkdir_args *ap)
1307 struct vnode *dvp = ap->a_dvp;
1308 struct vattr *vap = ap->a_vap;
1309 struct componentname *cnp = ap->a_cnp;
1310 struct inode *ip, *dp;
1313 struct dirtemplate dirtemplate, *dtp;
1314 struct direct newdir;
1319 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1323 dmode = vap->va_mode & 0777;
1326 * Must simulate part of ufs_makeinode here to acquire the inode,
1327 * but not have it entered in the parent directory. The entry is
1328 * made later after writing "." and ".." entries.
1330 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1334 ip->i_gid = dp->i_gid;
1338 struct ucred ucred, *ucp;
1342 * If we are hacking owners here, (only do this where told to)
1343 * and we are not giving it TO root, (would subvert quotas)
1344 * then go ahead and give it to the other user.
1345 * The new directory also inherits the SUID bit.
1346 * If user's UID and dir UID are the same,
1347 * 'give it away' so that the SUID is still forced on.
1349 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1350 (dp->i_mode & ISUID) && dp->i_uid) {
1352 ip->i_uid = dp->i_uid;
1354 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1356 * Make sure the correct user gets charged
1358 * Make a dummy credential for the victim.
1359 * XXX This seems to never be accessed out of
1360 * our context so a stack variable is ok.
1363 ucred.cr_uid = ip->i_uid;
1364 ucred.cr_ngroups = 1;
1365 ucred.cr_groups[0] = dp->i_gid;
1370 ip->i_uid = cnp->cn_cred->cr_uid;
1372 if ((error = ufs_getinoquota(ip)) ||
1373 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1374 ffs_vfree(tvp, ip->i_number, dmode);
1380 #else /* !SUIDDIR */
1381 ip->i_uid = cnp->cn_cred->cr_uid;
1383 if ((error = ufs_getinoquota(ip)) ||
1384 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1385 ffs_vfree(tvp, ip->i_number, dmode);
1390 #endif /* !SUIDDIR */
1391 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1393 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1396 if (DOINGSOFTDEP(tvp))
1397 softdep_change_linkcnt(ip);
1398 if (cnp->cn_flags & CNP_ISWHITEOUT)
1399 ip->i_flags |= UF_OPAQUE;
1402 * Bump link count in parent directory to reflect work done below.
1403 * Should be done before reference is created so cleanup is
1404 * possible if we crash.
1408 dp->i_flag |= IN_CHANGE;
1409 if (DOINGSOFTDEP(dvp))
1410 softdep_change_linkcnt(dp);
1411 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1416 * The vnode must have a VM object in order to issue buffer cache
1419 vinitvmio(tvp, DIRBLKSIZ);
1422 * Initialize directory with "." and ".." from static template.
1424 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1425 dtp = &mastertemplate;
1427 dtp = (struct dirtemplate *)&omastertemplate;
1429 dirtemplate.dot_ino = ip->i_number;
1430 dirtemplate.dotdot_ino = dp->i_number;
1431 vnode_pager_setsize(tvp, DIRBLKSIZ);
1432 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1435 ip->i_size = DIRBLKSIZ;
1436 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1437 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1438 if (DOINGSOFTDEP(tvp)) {
1440 * Ensure that the entire newly allocated block is a
1441 * valid directory so that future growth within the
1442 * block does not have to ensure that the block is
1443 * written before the inode.
1446 while (blkoff < bp->b_bcount) {
1448 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1449 blkoff += DIRBLKSIZ;
1452 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1453 DOINGASYNC(tvp)))) != 0) {
1458 * Directory set up, now install its entry in the parent directory.
1460 * If we are not doing soft dependencies, then we must write out the
1461 * buffer containing the new directory body before entering the new
1462 * name in the parent. If we are doing soft dependencies, then the
1463 * buffer containing the new directory body will be passed to and
1464 * released in the soft dependency code after the code has attached
1465 * an appropriate ordering dependency to the buffer which ensures that
1466 * the buffer is written before the new name is written in the parent.
1468 if (DOINGASYNC(dvp))
1470 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1472 ufs_makedirentry(ip, cnp, &newdir);
1473 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1477 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1482 dp->i_flag |= IN_CHANGE;
1483 if (DOINGSOFTDEP(dvp))
1484 softdep_change_linkcnt(dp);
1486 * No need to do an explicit VOP_TRUNCATE here, vrele will
1487 * do this for us because we set the link count to 0.
1491 ip->i_flag |= IN_CHANGE;
1492 if (DOINGSOFTDEP(tvp))
1493 softdep_change_linkcnt(ip);
1501 * Rmdir system call.
1503 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1504 * struct componentname *a_cnp)
1508 ufs_rmdir(struct vop_old_rmdir_args *ap)
1510 struct vnode *vp = ap->a_vp;
1511 struct vnode *dvp = ap->a_dvp;
1512 struct componentname *cnp = ap->a_cnp;
1513 struct inode *ip, *dp;
1520 * Do not remove a directory that is in the process of being renamed.
1521 * Verify the directory is empty (and valid). Rmdir ".." will not be
1522 * valid since ".." will contain a reference to the current directory
1523 * and thus be non-empty. Do not allow the removal of mounted on
1524 * directories (this can happen when an NFS exported filesystem
1525 * tries to remove a locally mounted on directory).
1528 if (ip->i_flag & IN_RENAME) {
1532 if (ip->i_effnlink != 2 ||
1533 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1537 if ((dp->i_flags & APPEND)
1538 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1542 if (vp->v_mountedhere != 0) {
1547 * Delete reference to directory before purging
1548 * inode. If we crash in between, the directory
1549 * will be reattached to lost+found,
1553 if (DOINGSOFTDEP(vp)) {
1554 softdep_change_linkcnt(dp);
1555 softdep_change_linkcnt(ip);
1557 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1561 if (DOINGSOFTDEP(vp)) {
1562 softdep_change_linkcnt(dp);
1563 softdep_change_linkcnt(ip);
1567 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1569 * Truncate inode. The only stuff left in the directory is "." and
1570 * "..". The "." reference is inconsequential since we are quashing
1571 * it. The soft dependency code will arrange to do these operations
1572 * after the parent directory entry has been deleted on disk, so
1573 * when running with that code we avoid doing them now.
1575 if (!DOINGSOFTDEP(vp)) {
1577 dp->i_flag |= IN_CHANGE;
1579 ip->i_flag |= IN_CHANGE;
1580 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1581 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1583 /* cache_purge removed - handled by VFS compat layer */
1585 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1586 if (ip->i_dirhash != NULL)
1587 ufsdirhash_free(ip);
1590 VN_KNOTE(vp, NOTE_DELETE);
1595 * symlink -- make a symbolic link
1597 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1598 * struct componentname *a_cnp, struct vattr *a_vap,
1603 ufs_symlink(struct vop_old_symlink_args *ap)
1605 struct vnode *vp, **vpp = ap->a_vpp;
1609 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1613 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1615 len = strlen(ap->a_target);
1616 if (len < vp->v_mount->mnt_maxsymlinklen) {
1618 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1620 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1623 * Make sure we have a VM object in order to use
1626 if (vp->v_object == NULL)
1628 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1629 UIO_SYSSPACE, IO_NODELOCKED,
1630 ap->a_cnp->cn_cred, NULL);
1638 * Vnode op for reading directories.
1640 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1641 * int *a_eofflag, int *ncookies, u_long **a_cookies)
1645 ufs_readdir(struct vop_readdir_args *ap)
1647 struct uio *uio = ap->a_uio;
1648 struct vnode *vp = ap->a_vp;
1653 int offset; /* offset into buffer cache buffer */
1654 int eoffset; /* end of buffer clipped to file EOF */
1655 int pickup; /* pickup point */
1660 if (uio->uio_offset < 0)
1663 * Guess the number of cookies needed. Make sure we compute at
1664 * least 1, and no more then a reasonable limit.
1666 if (ap->a_ncookies) {
1667 ncookies = uio->uio_resid / 16 + 1;
1668 if (ncookies > 1024)
1670 cookies = kmalloc(ncookies * sizeof(u_long), M_TEMP, M_WAITOK);
1672 ncookies = -1; /* force conditionals below */
1677 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1683 if (uio->uio_offset >= VTOI(vp)->i_size) {
1686 if (ap->a_ncookies) {
1687 *ap->a_ncookies = cookie_index;
1688 *ap->a_cookies = cookies;
1694 * Loop until we run out of cookies, we run out of user buffer,
1695 * or we hit the directory EOF.
1697 * Always start scans at the beginning of the buffer, don't trust
1698 * the offset supplied by userland.
1700 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1701 pickup = (int)(uio->uio_offset - bp->b_loffset);
1704 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1705 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1707 eoffset = bp->b_bcount;
1709 while (offset < eoffset) {
1710 dp = (struct direct *)(bp->b_data + offset);
1711 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1712 offset + dp->d_reclen > bp->b_bcount) {
1716 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1720 if (offset < pickup) {
1721 offset += dp->d_reclen;
1724 #if BYTE_ORDER == LITTLE_ENDIAN
1726 retval = vop_write_dirent(&error, uio,
1727 dp->d_ino, dp->d_namlen, dp->d_type,
1732 retval = vop_write_dirent(&error, uio,
1733 dp->d_ino, dp->d_type, dp->d_namlen,
1739 cookies[cookie_index] =
1740 (u_long)bp->b_loffset + offset;
1743 offset += dp->d_reclen;
1744 if (cookie_index == ncookies)
1749 * This will align the next loop to the beginning of the
1750 * next block, and pickup will calculate to 0.
1752 uio->uio_offset = bp->b_loffset + offset;
1755 if (retval || error || cookie_index == ncookies ||
1756 uio->uio_offset >= VTOI(vp)->i_size) {
1761 *ap->a_eofflag = VTOI(vp)->i_size <= uio->uio_offset;
1764 * Report errors only if we didn't manage to read anything
1766 if (error && cookie_index == 0) {
1768 kfree(cookies, M_TEMP);
1769 *ap->a_ncookies = 0;
1770 *ap->a_cookies = NULL;
1775 *ap->a_ncookies = cookie_index;
1776 *ap->a_cookies = cookies;
1785 * Return target name of a symbolic link
1787 * ufs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1791 ufs_readlink(struct vop_readlink_args *ap)
1793 struct vnode *vp = ap->a_vp;
1794 struct inode *ip = VTOI(vp);
1798 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1799 (ip->i_din.di_blocks == 0)) { /* XXX - for old fastlink support */
1800 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1805 * Perform the equivalent of an OPEN on vp so we can issue a
1808 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1812 * Calculate the logical to physical mapping if not done already,
1813 * then call the device strategy routine.
1815 * In order to be able to swap to a file, the VOP_BMAP operation may not
1816 * deadlock on memory. See ufs_bmap() for details.
1818 * ufs_strategy(struct vnode *a_vp, struct bio *a_bio)
1822 ufs_strategy(struct vop_strategy_args *ap)
1824 struct bio *bio = ap->a_bio;
1826 struct buf *bp = bio->bio_buf;
1827 struct vnode *vp = ap->a_vp;
1832 if (vp->v_type == VBLK || vp->v_type == VCHR)
1833 panic("ufs_strategy: spec");
1834 nbio = push_bio(bio);
1835 if (nbio->bio_offset == NOOFFSET) {
1836 error = VOP_BMAP(vp, bio->bio_offset, NULL, &nbio->bio_offset,
1839 bp->b_error = error;
1840 bp->b_flags |= B_ERROR;
1841 /* I/O was never started on nbio, must biodone(bio) */
1845 if (nbio->bio_offset == NOOFFSET)
1848 if (nbio->bio_offset == NOOFFSET) {
1850 * We hit a hole in the file. The buffer has been zero-filled
1851 * so just biodone() it.
1855 vn_strategy(ip->i_devvp, nbio);
1861 * Print out the contents of an inode.
1863 * ufs_print(struct vnode *a_vp)
1867 ufs_print(struct vop_print_args *ap)
1869 struct vnode *vp = ap->a_vp;
1870 struct inode *ip = VTOI(vp);
1872 printf("tag VT_UFS, ino %lu, on dev %s (%d, %d)",
1873 (u_long)ip->i_number, devtoname(ip->i_dev), major(ip->i_dev),
1875 if (vp->v_type == VFIFO)
1877 lockmgr_printinfo(&vp->v_lock);
1883 * Read wrapper for special devices.
1885 * ufsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1886 * struct ucred *a_cred)
1890 ufsspec_read(struct vop_read_args *ap)
1897 resid = uio->uio_resid;
1898 error = VOCALL(&spec_vnode_vops, &ap->a_head);
1900 * The inode may have been revoked during the call, so it must not
1901 * be accessed blindly here or in the other wrapper functions.
1903 ip = VTOI(ap->a_vp);
1904 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1905 ip->i_flag |= IN_ACCESS;
1910 * Write wrapper for special devices.
1912 * ufsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1913 * struct ucred *a_cred)
1917 ufsspec_write(struct vop_write_args *ap)
1924 resid = uio->uio_resid;
1925 error = VOCALL(&spec_vnode_vops, &ap->a_head);
1926 ip = VTOI(ap->a_vp);
1927 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1928 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1933 * Close wrapper for special devices.
1935 * Update the times on the inode then do device close.
1937 * ufsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
1938 * struct thread *a_td)
1942 ufsspec_close(struct vop_close_args *ap)
1944 struct vnode *vp = ap->a_vp;
1946 if (vp->v_usecount > 1)
1948 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1952 * Read wrapper for fifos.
1954 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1955 * struct ucred *a_cred)
1959 ufsfifo_read(struct vop_read_args *ap)
1966 resid = uio->uio_resid;
1967 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1968 ip = VTOI(ap->a_vp);
1969 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1970 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1971 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1976 * Write wrapper for fifos.
1978 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1979 * struct ucred *a_cred)
1983 ufsfifo_write(struct vop_write_args *ap)
1990 resid = uio->uio_resid;
1991 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1992 ip = VTOI(ap->a_vp);
1993 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1994 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1999 * Close wrapper for fifos.
2001 * Update the times on the inode then do device close.
2003 * ufsfifo_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
2004 * struct thread *a_td)
2008 ufsfifo_close(struct vop_close_args *ap)
2010 struct vnode *vp = ap->a_vp;
2012 if (vp->v_usecount > 1)
2014 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2018 * Kqfilter wrapper for fifos.
2020 * Fall through to ufs kqfilter routines if needed
2024 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
2028 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2030 error = ufs_kqfilter(ap);
2035 * Return POSIX pathconf information applicable to ufs filesystems.
2037 * ufs_pathconf(struct vnode *a_vp, int a_name, int *a_retval)
2041 ufs_pathconf(struct vop_pathconf_args *ap)
2043 switch (ap->a_name) {
2045 *ap->a_retval = LINK_MAX;
2048 *ap->a_retval = NAME_MAX;
2051 *ap->a_retval = PATH_MAX;
2054 *ap->a_retval = PIPE_BUF;
2056 case _PC_CHOWN_RESTRICTED:
2069 * Advisory record locking support
2071 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
2076 ufs_advlock(struct vop_advlock_args *ap)
2078 struct inode *ip = VTOI(ap->a_vp);
2080 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
2084 * Initialize the vnode associated with a new inode, handle aliased
2087 * Make sure directories have their VM object now rather then later,
2088 * saving us from having to check on all the myrid directory VOPs
2089 * that might be executed without a VOP_OPEN being performed.
2092 ufs_vinit(struct mount *mntp, struct vnode **vpp)
2101 vp->v_type = IFTOVT(ip->i_mode);
2103 switch(vp->v_type) {
2106 vp->v_ops = &mntp->mnt_vn_spec_ops;
2107 addaliasu(vp, ip->i_rdev);
2110 vp->v_ops = &mntp->mnt_vn_fifo_ops;
2114 vinitvmio(vp, ip->i_size);
2117 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
2118 vinitvmio(vp, ip->i_size);
2125 if (ip->i_number == ROOTINO)
2126 vp->v_flag |= VROOT;
2128 * Initialize modrev times
2130 getmicrouptime(&tv);
2131 SETHIGH(ip->i_modrev, tv.tv_sec);
2132 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
2138 * Allocate a new inode.
2142 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2143 struct componentname *cnp)
2145 struct inode *ip, *pdir;
2146 struct direct newdir;
2152 if ((mode & IFMT) == 0)
2155 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2159 ip->i_gid = pdir->i_gid;
2163 struct ucred ucred, *ucp;
2167 * If we are not the owner of the directory,
2168 * and we are hacking owners here, (only do this where told to)
2169 * and we are not giving it TO root, (would subvert quotas)
2170 * then go ahead and give it to the other user.
2171 * Note that this drops off the execute bits for security.
2173 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2174 (pdir->i_mode & ISUID) &&
2175 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2176 ip->i_uid = pdir->i_uid;
2180 * Make sure the correct user gets charged
2182 * Quickly knock up a dummy credential for the victim.
2183 * XXX This seems to never be accessed out of our
2184 * context so a stack variable is ok.
2187 ucred.cr_uid = ip->i_uid;
2188 ucred.cr_ngroups = 1;
2189 ucred.cr_groups[0] = pdir->i_gid;
2193 ip->i_uid = cnp->cn_cred->cr_uid;
2196 if ((error = ufs_getinoquota(ip)) ||
2197 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2198 ffs_vfree(tvp, ip->i_number, mode);
2204 #else /* !SUIDDIR */
2205 ip->i_uid = cnp->cn_cred->cr_uid;
2207 if ((error = ufs_getinoquota(ip)) ||
2208 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2209 ffs_vfree(tvp, ip->i_number, mode);
2214 #endif /* !SUIDDIR */
2215 ip->i_fsmid = cache_getnewfsmid();
2216 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2218 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2221 if (DOINGSOFTDEP(tvp))
2222 softdep_change_linkcnt(ip);
2223 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2224 suser_cred(cnp->cn_cred, 0)) {
2225 ip->i_mode &= ~ISGID;
2228 if (cnp->cn_flags & CNP_ISWHITEOUT)
2229 ip->i_flags |= UF_OPAQUE;
2232 * Regular files and directories need VM objects. Softlinks do
2233 * not (not immediately anyway).
2235 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2239 * Make sure inode goes to disk before directory entry.
2241 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2244 ufs_makedirentry(ip, cnp, &newdir);
2245 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2253 * Write error occurred trying to update the inode
2254 * or the directory so must deallocate the inode.
2258 ip->i_flag |= IN_CHANGE;
2259 if (DOINGSOFTDEP(tvp))
2260 softdep_change_linkcnt(ip);
2266 ufs_missingop(struct vop_generic_args *ap)
2268 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2269 return (EOPNOTSUPP);
2272 static struct filterops ufsread_filtops =
2273 { 1, NULL, filt_ufsdetach, filt_ufsread };
2274 static struct filterops ufswrite_filtops =
2275 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2276 static struct filterops ufsvnode_filtops =
2277 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2280 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2283 ufs_kqfilter(struct vop_kqfilter_args *ap)
2285 struct vnode *vp = ap->a_vp;
2286 struct knote *kn = ap->a_kn;
2289 switch (kn->kn_filter) {
2291 kn->kn_fop = &ufsread_filtops;
2294 kn->kn_fop = &ufswrite_filtops;
2297 kn->kn_fop = &ufsvnode_filtops;
2303 kn->kn_hook = (caddr_t)vp;
2305 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2306 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2307 lwkt_reltoken(&ilock);
2313 filt_ufsdetach(struct knote *kn)
2315 struct vnode *vp = (struct vnode *)kn->kn_hook;
2318 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2319 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2320 kn, knote, kn_selnext);
2321 lwkt_reltoken(&ilock);
2326 filt_ufsread(struct knote *kn, long hint)
2328 struct vnode *vp = (struct vnode *)kn->kn_hook;
2329 struct inode *ip = VTOI(vp);
2332 * filesystem is gone, so set the EOF flag and schedule
2333 * the knote for deletion.
2335 if (hint == NOTE_REVOKE) {
2336 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2340 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2341 return (kn->kn_data != 0);
2346 filt_ufswrite(struct knote *kn, long hint)
2349 * filesystem is gone, so set the EOF flag and schedule
2350 * the knote for deletion.
2352 if (hint == NOTE_REVOKE)
2353 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2360 filt_ufsvnode(struct knote *kn, long hint)
2362 if (kn->kn_sfflags & hint)
2363 kn->kn_fflags |= hint;
2364 if (hint == NOTE_REVOKE) {
2365 kn->kn_flags |= EV_EOF;
2368 return (kn->kn_fflags != 0);
2371 /* Global vfs data structures for ufs. */
2372 static struct vop_ops ufs_vnode_vops = {
2373 .vop_default = vop_defaultop,
2374 .vop_fsync = (void *)ufs_missingop,
2375 .vop_read = (void *)ufs_missingop,
2376 .vop_reallocblks = (void *)ufs_missingop,
2377 .vop_write = (void *)ufs_missingop,
2378 .vop_access = ufs_access,
2379 .vop_advlock = ufs_advlock,
2380 .vop_bmap = ufs_bmap,
2381 .vop_old_lookup = ufs_lookup,
2382 .vop_close = ufs_close,
2383 .vop_old_create = ufs_create,
2384 .vop_getattr = ufs_getattr,
2385 .vop_inactive = ufs_inactive,
2386 .vop_old_link = ufs_link,
2387 .vop_old_mkdir = ufs_mkdir,
2388 .vop_old_mknod = ufs_mknod,
2389 .vop_mmap = ufs_mmap,
2390 .vop_open = ufs_open,
2391 .vop_pathconf = ufs_pathconf,
2392 .vop_poll = vop_stdpoll,
2393 .vop_kqfilter = ufs_kqfilter,
2394 .vop_print = ufs_print,
2395 .vop_readdir = ufs_readdir,
2396 .vop_readlink = ufs_readlink,
2397 .vop_reclaim = ufs_reclaim,
2398 .vop_old_remove = ufs_remove,
2399 .vop_old_rename = ufs_rename,
2400 .vop_old_rmdir = ufs_rmdir,
2401 .vop_setattr = ufs_setattr,
2402 .vop_strategy = ufs_strategy,
2403 .vop_old_symlink = ufs_symlink,
2404 .vop_old_whiteout = ufs_whiteout
2407 static struct vop_ops ufs_spec_vops = {
2408 .vop_default = spec_vnoperate,
2409 .vop_fsync = (void *)ufs_missingop,
2410 .vop_access = ufs_access,
2411 .vop_close = ufsspec_close,
2412 .vop_getattr = ufs_getattr,
2413 .vop_inactive = ufs_inactive,
2414 .vop_print = ufs_print,
2415 .vop_read = ufsspec_read,
2416 .vop_reclaim = ufs_reclaim,
2417 .vop_setattr = ufs_setattr,
2418 .vop_write = ufsspec_write
2421 static struct vop_ops ufs_fifo_vops = {
2422 .vop_default = fifo_vnoperate,
2423 .vop_fsync = (void *)ufs_missingop,
2424 .vop_access = ufs_access,
2425 .vop_close = ufsfifo_close,
2426 .vop_getattr = ufs_getattr,
2427 .vop_inactive = ufs_inactive,
2428 .vop_kqfilter = ufsfifo_kqfilter,
2429 .vop_print = ufs_print,
2430 .vop_read = ufsfifo_read,
2431 .vop_reclaim = ufs_reclaim,
2432 .vop_setattr = ufs_setattr,
2433 .vop_write = ufsfifo_write
2436 VNODEOP_SET(ufs_vnode_vops);
2437 VNODEOP_SET(ufs_spec_vops);
2438 VNODEOP_SET(ufs_fifo_vops);
2444 ufs_vnoperate(struct vop_generic_args *ap)
2446 return (VOCALL(&ufs_vnode_vops, ap));
2450 * ufs_vnoperatefifo()
2453 ufs_vnoperatefifo(struct vop_generic_args *ap)
2455 return (VOCALL(&ufs_fifo_vops, ap));
2459 * ufs_vnoperatespec()
2462 ufs_vnoperatespec(struct vop_generic_args *ap)
2464 return (VOCALL(&ufs_spec_vops, ap));