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 1.63 2007/11/20 21:03:51 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;
224 * UFS cannot represent the entire major/minor range supported by
227 if (vap->va_rmajor != VNOVAL &&
228 makeudev(vap->va_rmajor, vap->va_rminor) == NOUDEV) {
232 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
233 ap->a_dvp, vpp, ap->a_cnp);
236 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
238 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
239 if (vap->va_rmajor != VNOVAL) {
241 * Want to be able to use this to make badblock
242 * inodes, so don't truncate the dev number.
244 ip->i_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
247 * Remove inode, then reload it through VFS_VGET so it is
248 * checked to see if it is an alias of an existing entry in
251 (*vpp)->v_type = VNON;
252 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
253 vgone_vxlocked(*vpp);
255 error = VFS_VGET(ap->a_dvp->v_mount, ino, vpp);
268 * ufs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
274 ufs_open(struct vop_open_args *ap)
276 struct vnode *vp = ap->a_vp;
279 * Files marked append-only must be opened for appending.
281 if ((VTOI(vp)->i_flags & APPEND) &&
282 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) {
286 return (vop_stdopen(ap));
292 * Update the times on the inode.
294 * ufs_close(struct vnode *a_vp, int a_fflag)
299 ufs_close(struct vop_close_args *ap)
301 struct vnode *vp = ap->a_vp;
303 if (vp->v_sysref.refcnt > 1)
305 return (vop_stdclose(ap));
309 * ufs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
313 ufs_access(struct vop_access_args *ap)
315 struct vnode *vp = ap->a_vp;
316 struct inode *ip = VTOI(vp);
317 struct ucred *cred = ap->a_cred;
318 mode_t mask, mode = ap->a_mode;
326 * Disallow write attempts on read-only filesystems;
327 * unless the file is a socket, fifo, or a block or
328 * character device resident on the filesystem.
331 switch (vp->v_type) {
335 if (vp->v_mount->mnt_flag & MNT_RDONLY)
338 if ((error = ufs_getinoquota(ip)) != 0)
347 /* If immutable bit set, nobody gets to write it. */
348 if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
351 /* Otherwise, user id 0 always gets access. */
352 if (cred->cr_uid == 0)
357 /* Otherwise, check the owner. */
358 if (cred->cr_uid == ip->i_uid) {
365 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
368 /* Otherwise, check the groups. */
369 for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
370 if (ip->i_gid == *gp) {
377 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
380 /* Otherwise, check everyone else. */
387 return ((ip->i_mode & mask) == mask ? 0 : EACCES);
391 * ufs_getattr(struct vnode *a_vp, struct vattr *a_vap)
396 ufs_getattr(struct vop_getattr_args *ap)
398 struct vnode *vp = ap->a_vp;
399 struct inode *ip = VTOI(vp);
400 struct vattr *vap = ap->a_vap;
403 * This may cause i_fsmid to be updated even if no change (0)
404 * is returned, but we should only write out the inode if non-zero
405 * is returned and if the mount is read-write.
407 if (cache_check_fsmid_vp(vp, &ip->i_fsmid) &&
408 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0
410 ip->i_flag |= IN_LAZYMOD;
415 * Copy from inode table
417 vap->va_fsid = dev2udev(ip->i_dev);
418 vap->va_fileid = ip->i_number;
419 vap->va_mode = ip->i_mode & ~IFMT;
420 vap->va_nlink = VFSTOUFS(vp->v_mount)->um_i_effnlink_valid ?
421 ip->i_effnlink : ip->i_nlink;
422 vap->va_uid = ip->i_uid;
423 vap->va_gid = ip->i_gid;
424 vap->va_rmajor = umajor(ip->i_rdev);
425 vap->va_rminor = uminor(ip->i_rdev);
426 vap->va_size = ip->i_din.di_size;
427 vap->va_atime.tv_sec = ip->i_atime;
428 vap->va_atime.tv_nsec = ip->i_atimensec;
429 vap->va_mtime.tv_sec = ip->i_mtime;
430 vap->va_mtime.tv_nsec = ip->i_mtimensec;
431 vap->va_ctime.tv_sec = ip->i_ctime;
432 vap->va_ctime.tv_nsec = ip->i_ctimensec;
433 vap->va_flags = ip->i_flags;
434 vap->va_gen = ip->i_gen;
435 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
436 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
437 vap->va_type = IFTOVT(ip->i_mode);
438 vap->va_filerev = ip->i_modrev;
439 vap->va_fsmid = ip->i_fsmid;
444 * Set attribute vnode op. called from several syscalls
446 * ufs_setattr(struct vnode *a_vp, struct vattr *a_vap,
447 * struct ucred *a_cred)
451 ufs_setattr(struct vop_setattr_args *ap)
453 struct vattr *vap = ap->a_vap;
454 struct vnode *vp = ap->a_vp;
455 struct inode *ip = VTOI(vp);
456 struct ucred *cred = ap->a_cred;
460 * Check for unsettable attributes.
462 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
463 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
464 (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
465 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
468 if (vap->va_flags != VNOVAL) {
469 if (vp->v_mount->mnt_flag & MNT_RDONLY)
471 if (cred->cr_uid != ip->i_uid &&
472 (error = suser_cred(cred, PRISON_ROOT)))
475 * Note that a root chflags becomes a user chflags when
476 * we are jailed, unless the jail.chflags_allowed sysctl
479 if (cred->cr_uid == 0 &&
480 (!jailed(cred) || jail_chflags_allowed)) {
482 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) &&
485 ip->i_flags = vap->va_flags;
488 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
489 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
491 ip->i_flags &= SF_SETTABLE;
492 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
494 ip->i_flag |= IN_CHANGE;
495 if (vap->va_flags & (IMMUTABLE | APPEND))
498 if (ip->i_flags & (IMMUTABLE | APPEND))
501 * Go through the fields and update iff not VNOVAL.
503 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
504 if (vp->v_mount->mnt_flag & MNT_RDONLY)
506 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred)) != 0)
509 if (vap->va_size != VNOVAL) {
511 * Disallow write attempts on read-only filesystems;
512 * unless the file is a socket, fifo, or a block or
513 * character device resident on the filesystem.
515 switch (vp->v_type) {
520 if (vp->v_mount->mnt_flag & MNT_RDONLY)
526 if ((error = ffs_truncate(vp, vap->va_size, 0, cred)) != 0)
530 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
531 if (vp->v_mount->mnt_flag & MNT_RDONLY)
533 if (cred->cr_uid != ip->i_uid &&
534 (error = suser_cred(cred, PRISON_ROOT)) &&
535 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
536 (error = VOP_ACCESS(vp, VWRITE, cred))))
538 if (vap->va_atime.tv_sec != VNOVAL)
539 ip->i_flag |= IN_ACCESS;
540 if (vap->va_mtime.tv_sec != VNOVAL)
541 ip->i_flag |= IN_CHANGE | IN_UPDATE;
543 if (vap->va_atime.tv_sec != VNOVAL) {
544 ip->i_atime = vap->va_atime.tv_sec;
545 ip->i_atimensec = vap->va_atime.tv_nsec;
547 if (vap->va_mtime.tv_sec != VNOVAL) {
548 ip->i_mtime = vap->va_mtime.tv_sec;
549 ip->i_mtimensec = vap->va_mtime.tv_nsec;
551 error = ffs_update(vp, 0);
556 if (vap->va_mode != (mode_t)VNOVAL) {
557 if (vp->v_mount->mnt_flag & MNT_RDONLY)
559 error = ufs_chmod(vp, (int)vap->va_mode, cred);
561 VN_KNOTE(vp, NOTE_ATTRIB);
566 * Change the mode on a file.
567 * Inode must be locked before calling.
570 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred)
572 struct inode *ip = VTOI(vp);
575 if (cred->cr_uid != ip->i_uid) {
576 error = suser_cred(cred, PRISON_ROOT);
581 if (vp->v_type != VDIR && (mode & S_ISTXT))
583 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
586 ip->i_mode &= ~ALLPERMS;
587 ip->i_mode |= (mode & ALLPERMS);
588 ip->i_flag |= IN_CHANGE;
593 * Perform chown operation on inode ip;
594 * inode must be locked prior to call.
597 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
599 struct inode *ip = VTOI(vp);
608 if (uid == (uid_t)VNOVAL)
610 if (gid == (gid_t)VNOVAL)
613 * If we don't own the file, are trying to change the owner
614 * of the file, or are not a member of the target group,
615 * the caller must be superuser or the call fails.
617 if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
618 (gid != ip->i_gid && !(cred->cr_gid == gid ||
619 groupmember((gid_t)gid, cred)))) &&
620 (error = suser_cred(cred, PRISON_ROOT)))
625 if ((error = ufs_getinoquota(ip)) != 0)
628 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
629 ip->i_dquot[USRQUOTA] = NODQUOT;
632 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
633 ip->i_dquot[GRPQUOTA] = NODQUOT;
635 change = ip->i_blocks;
636 (void) ufs_chkdq(ip, -change, cred, CHOWN);
637 (void) ufs_chkiq(ip, -1, cred, CHOWN);
638 for (i = 0; i < MAXQUOTAS; i++) {
639 ufs_dqrele(vp, ip->i_dquot[i]);
640 ip->i_dquot[i] = NODQUOT;
646 if ((error = ufs_getinoquota(ip)) == 0) {
648 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
649 ip->i_dquot[USRQUOTA] = NODQUOT;
652 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
653 ip->i_dquot[GRPQUOTA] = NODQUOT;
655 if ((error = ufs_chkdq(ip, change, cred, CHOWN)) == 0) {
656 if ((error = ufs_chkiq(ip, 1, cred, CHOWN)) == 0)
659 (void)ufs_chkdq(ip, -change, cred, CHOWN|FORCE);
661 for (i = 0; i < MAXQUOTAS; i++) {
662 ufs_dqrele(vp, ip->i_dquot[i]);
663 ip->i_dquot[i] = NODQUOT;
668 if (ufs_getinoquota(ip) == 0) {
670 ufs_dqrele(vp, ip->i_dquot[USRQUOTA]);
671 ip->i_dquot[USRQUOTA] = NODQUOT;
674 ufs_dqrele(vp, ip->i_dquot[GRPQUOTA]);
675 ip->i_dquot[GRPQUOTA] = NODQUOT;
677 (void) ufs_chkdq(ip, change, cred, FORCE|CHOWN);
678 (void) ufs_chkiq(ip, 1, cred, FORCE|CHOWN);
679 (void) ufs_getinoquota(ip);
683 if (ufs_getinoquota(ip))
684 panic("ufs_chown: lost quota");
686 ip->i_flag |= IN_CHANGE;
687 if (cred->cr_uid != 0 && (ouid != uid || ogid != gid))
688 ip->i_mode &= ~(ISUID | ISGID);
695 * NB Currently unsupported.
697 * ufs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
702 ufs_mmap(struct vop_mmap_args *ap)
708 * ufs_remove(struct vnode *a_dvp, struct vnode *a_vp,
709 * struct componentname *a_cnp)
713 ufs_remove(struct vop_old_remove_args *ap)
716 struct vnode *vp = ap->a_vp;
717 struct vnode *dvp = ap->a_dvp;
721 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
722 (VTOI(dvp)->i_flags & APPEND)) {
726 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
727 VN_KNOTE(vp, NOTE_DELETE);
728 VN_KNOTE(dvp, NOTE_WRITE);
736 * ufs_link(struct vnode *a_tdvp, struct vnode *a_vp,
737 * struct componentname *a_cnp)
741 ufs_link(struct vop_old_link_args *ap)
743 struct vnode *vp = ap->a_vp;
744 struct vnode *tdvp = ap->a_tdvp;
745 struct componentname *cnp = ap->a_cnp;
747 struct direct newdir;
750 if (tdvp->v_mount != vp->v_mount) {
754 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE))) {
758 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
762 if (ip->i_flags & (IMMUTABLE | APPEND)) {
768 ip->i_flag |= IN_CHANGE;
769 if (DOINGSOFTDEP(vp))
770 softdep_change_linkcnt(ip);
771 error = ffs_update(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
773 ufs_makedirentry(ip, cnp, &newdir);
774 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
780 ip->i_flag |= IN_CHANGE;
781 if (DOINGSOFTDEP(vp))
782 softdep_change_linkcnt(ip);
788 VN_KNOTE(vp, NOTE_LINK);
789 VN_KNOTE(tdvp, NOTE_WRITE);
794 * whiteout vnode call
796 * ufs_whiteout(struct vnode *a_dvp, struct componentname *a_cnp, int a_flags)
800 ufs_whiteout(struct vop_old_whiteout_args *ap)
802 struct vnode *dvp = ap->a_dvp;
803 struct componentname *cnp = ap->a_cnp;
804 struct direct newdir;
807 switch (ap->a_flags) {
809 /* 4.4 format directories support whiteout operations */
810 if (dvp->v_mount->mnt_maxsymlinklen > 0)
815 /* create a new directory whiteout */
817 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
818 panic("ufs_whiteout: old format filesystem");
822 newdir.d_namlen = cnp->cn_namelen;
823 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
824 newdir.d_type = DT_WHT;
825 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
829 /* remove an existing directory whiteout */
831 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
832 panic("ufs_whiteout: old format filesystem");
835 cnp->cn_flags &= ~CNP_DOWHITEOUT;
836 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
839 panic("ufs_whiteout: unknown op");
845 * Rename system call.
846 * rename("foo", "bar");
849 * link("foo", "bar");
851 * but ``atomically''. Can't do full commit without saving state in the
852 * inode on disk which isn't feasible at this time. Best we can do is
853 * always guarantee the target exists.
855 * Basic algorithm is:
857 * 1) Bump link count on source while we're linking it to the
858 * target. This also ensure the inode won't be deleted out
859 * from underneath us while we work (it may be truncated by
860 * a concurrent `trunc' or `open' for creation).
861 * 2) Link source to destination. If destination already exists,
863 * 3) Unlink source reference to inode if still around. If a
864 * directory was moved and the parent of the destination
865 * is different from the source, patch the ".." entry in the
868 * ufs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
869 * struct componentname *a_fcnp, struct vnode *a_tdvp,
870 * struct vnode *a_tvp, struct componentname *a_tcnp)
874 ufs_rename(struct vop_old_rename_args *ap)
876 struct vnode *tvp = ap->a_tvp;
877 struct vnode *tdvp = ap->a_tdvp;
878 struct vnode *fvp = ap->a_fvp;
879 struct vnode *fdvp = ap->a_fdvp;
880 struct componentname *tcnp = ap->a_tcnp;
881 struct componentname *fcnp = ap->a_fcnp;
882 struct inode *ip, *xp, *dp;
883 struct direct newdir;
884 ino_t oldparent = 0, newparent = 0;
885 int doingdirectory = 0;
886 int error = 0, ioflag;
889 * Check for cross-device rename.
891 if ((fvp->v_mount != tdvp->v_mount) ||
892 (tvp && (fvp->v_mount != tvp->v_mount))) {
906 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
907 (VTOI(tdvp)->i_flags & APPEND))) {
913 * Renaming a file to itself has no effect. The upper layers should
914 * not call us in that case. Temporarily just warn if they do.
917 kprintf("ufs_rename: fvp == tvp (can't happen)\n");
922 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
926 * Note: now that fvp is locked we have to be sure to unlock it before
927 * using the 'abortit' target.
931 if (ip->i_nlink >= LINK_MAX) {
936 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
937 || (dp->i_flags & APPEND)) {
942 if ((ip->i_mode & IFMT) == IFDIR) {
944 * Avoid ".", "..", and aliases of "." for obvious reasons.
946 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
947 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & CNP_ISDOTDOT ||
948 (ip->i_flag & IN_RENAME)) {
953 ip->i_flag |= IN_RENAME;
954 oldparent = dp->i_number;
957 VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
960 * fvp still locked. ip->i_flag has IN_RENAME set if doingdirectory.
961 * Cleanup fvp requirements so we can unlock it.
963 * tvp and tdvp are locked. tvp may be NULL. Now that dp and xp
964 * is setup we can use the 'bad' target if we unlock fvp. We cannot
965 * use the abortit target anymore because of IN_RENAME.
974 * 1) Bump link count while we're moving stuff
975 * around. If we crash somewhere before
976 * completing our work, the link count
977 * may be wrong, but correctable.
981 ip->i_flag |= IN_CHANGE;
982 if (DOINGSOFTDEP(fvp))
983 softdep_change_linkcnt(ip);
984 if ((error = ffs_update(fvp, !(DOINGSOFTDEP(fvp) |
985 DOINGASYNC(fvp)))) != 0) {
991 * If ".." must be changed (ie the directory gets a new
992 * parent) then the source directory must not be in the
993 * directory heirarchy above the target, as this would
994 * orphan everything below the source directory. Also
995 * the user must have write permission in the source so
996 * as to be able to change "..". We must repeat the call
997 * to namei, as the parent directory is unlocked by the
998 * call to checkpath().
1000 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
1004 * We are now back to where we were in that fvp, fdvp are unlocked
1005 * and tvp, tdvp are locked. tvp may be NULL. IN_RENAME may be
1006 * set. Only the bad target or, if we clean up tvp and tdvp, the
1007 * out target, may be used.
1009 if (oldparent != dp->i_number)
1010 newparent = dp->i_number;
1011 if (doingdirectory && newparent) {
1012 if (error) /* write access check above */
1016 * Once we start messing with tvp and tdvp we cannot use the
1017 * 'bad' target, only finish cleaning tdvp and tvp up and
1018 * use the 'out' target.
1020 * This cleans up tvp.
1028 * This is a real mess. ufs_checkpath vput's the target
1029 * directory so retain an extra ref and note that tdvp will
1030 * lose its lock on return. This leaves us with one good
1031 * ref after ufs_checkpath returns.
1034 error = ufs_checkpath(ip, dp, tcnp->cn_cred);
1035 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1042 * relookup no longer messes with tdvp's refs. tdvp must be
1043 * unlocked on entry and will be locked on a successful
1046 error = relookup(tdvp, &tvp, tcnp);
1048 if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1054 KKASSERT((tcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1061 * We are back to fvp, fdvp unlocked, tvp, tdvp locked. tvp may
1062 * be NULL (xp will also be NULL in that case), and IN_RENAME will
1063 * be set if doingdirectory. This means we can use the 'bad' target
1068 * 2) If target doesn't exist, link the target
1069 * to the source and unlink the source.
1070 * Otherwise, rewrite the target directory
1071 * entry to reference the source inode and
1072 * expunge the original entry's existence.
1075 if (dp->i_dev != ip->i_dev)
1076 panic("ufs_rename: EXDEV");
1078 * Account for ".." in new directory.
1079 * When source and destination have the same
1080 * parent we don't fool with the link count.
1082 if (doingdirectory && newparent) {
1083 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1089 dp->i_flag |= IN_CHANGE;
1090 if (DOINGSOFTDEP(tdvp))
1091 softdep_change_linkcnt(dp);
1092 error = ffs_update(tdvp, !(DOINGSOFTDEP(tdvp) |
1097 ufs_makedirentry(ip, tcnp, &newdir);
1098 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1100 if (doingdirectory && newparent) {
1103 dp->i_flag |= IN_CHANGE;
1104 if (DOINGSOFTDEP(tdvp))
1105 softdep_change_linkcnt(dp);
1106 (void)ffs_update(tdvp, 1);
1110 VN_KNOTE(tdvp, NOTE_WRITE);
1113 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1114 panic("ufs_rename: EXDEV");
1116 * Short circuit rename(foo, foo).
1118 if (xp->i_number == ip->i_number)
1119 panic("ufs_rename: same file");
1121 * If the parent directory is "sticky", then the user must
1122 * own the parent directory, or the destination of the rename,
1123 * otherwise the destination may not be changed (except by
1124 * root). This implements append-only directories.
1126 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1127 tcnp->cn_cred->cr_uid != dp->i_uid &&
1128 xp->i_uid != tcnp->cn_cred->cr_uid) {
1133 * Target must be empty if a directory and have no links
1134 * to it. Also, ensure source and target are compatible
1135 * (both directories, or both not directories).
1137 * Purge the file or directory being replaced from the
1140 if ((xp->i_mode&IFMT) == IFDIR) {
1141 if ((xp->i_effnlink > 2) ||
1142 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1146 if (!doingdirectory) {
1150 /* cache_purge removed - handled by VFS compat layer */
1151 } else if (doingdirectory == 0) {
1152 /* cache_purge removed - handled by VFS compat layer */
1158 * note: inode passed to ufs_dirrewrite() is 0 for a
1159 * non-directory file rename, 1 for a directory rename
1160 * in the same directory, and > 1 for an inode representing
1161 * the new directory.
1163 error = ufs_dirrewrite(dp, xp, ip->i_number,
1165 (doingdirectory && newparent) ?
1166 newparent : (ino_t)doingdirectory);
1169 if (doingdirectory) {
1172 if (DOINGSOFTDEP(tdvp))
1173 softdep_change_linkcnt(dp);
1176 if (DOINGSOFTDEP(tvp))
1177 softdep_change_linkcnt(xp);
1179 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1181 * Truncate inode. The only stuff left in the directory
1182 * is "." and "..". The "." reference is inconsequential
1183 * since we are quashing it. We have removed the "."
1184 * reference and the reference in the parent directory,
1185 * but there may be other hard links. The soft
1186 * dependency code will arrange to do these operations
1187 * after the parent directory entry has been deleted on
1188 * disk, so when running with that code we avoid doing
1193 dp->i_flag |= IN_CHANGE;
1196 xp->i_flag |= IN_CHANGE;
1197 ioflag = DOINGASYNC(tvp) ? 0 : IO_SYNC;
1198 error = ffs_truncate(tvp, (off_t)0, ioflag,
1203 VN_KNOTE(tdvp, NOTE_WRITE);
1205 VN_KNOTE(tvp, NOTE_DELETE);
1211 * tvp and tdvp have been cleaned up. only fvp and fdvp (both
1212 * unlocked) remain. We are about to overwrite fvp but we have to
1213 * keep 'ip' intact so we cannot release the old fvp, which is still
1214 * refd and accessible via ap->a_fvp.
1216 * This means we cannot use either 'bad' or 'out' to cleanup any
1221 * 3) Unlink the source.
1223 fcnp->cn_flags &= ~CNP_MODMASK;
1224 fcnp->cn_flags |= CNP_LOCKPARENT;
1225 error = relookup(fdvp, &fvp, fcnp);
1226 if (error || fvp == NULL) {
1228 * From name has disappeared. IN_RENAME will not be set if
1229 * we get past the panic so we don't have to clean it up.
1232 panic("ufs_rename: lost dir entry");
1234 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1240 KKASSERT((fcnp->cn_flags & CNP_PDIRUNLOCK) == 0);
1243 * fdvp and fvp are locked.
1249 * Ensure that the directory entry still exists and has not
1250 * changed while the new name has been entered. If the source is
1251 * a file then the entry may have been unlinked or renamed. In
1252 * either case there is no further work to be done. If the source
1253 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1254 * flag ensures that it cannot be moved by another rename or removed
1255 * by a rmdir. Cleanup IN_RENAME.
1259 panic("ufs_rename: lost dir entry");
1262 * If the source is a directory with a
1263 * new parent, the link count of the old
1264 * parent directory must be decremented
1265 * and ".." set to point to the new parent.
1267 if (doingdirectory && newparent) {
1268 xp->i_offset = mastertemplate.dot_reclen;
1269 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1270 /* cache_purge removed - handled by VFS compat layer */
1272 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1273 xp->i_flag &= ~IN_RENAME;
1276 VN_KNOTE(fvp, NOTE_RENAME);
1288 ip->i_flag &= ~IN_RENAME;
1289 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1292 ip->i_flag |= IN_CHANGE;
1293 ip->i_flag &= ~IN_RENAME;
1294 if (DOINGSOFTDEP(fvp))
1295 softdep_change_linkcnt(ip);
1306 * ufs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1307 * struct componentname *a_cnp, struct vattr *a_vap)
1311 ufs_mkdir(struct vop_old_mkdir_args *ap)
1313 struct vnode *dvp = ap->a_dvp;
1314 struct vattr *vap = ap->a_vap;
1315 struct componentname *cnp = ap->a_cnp;
1316 struct inode *ip, *dp;
1319 struct dirtemplate dirtemplate, *dtp;
1320 struct direct newdir;
1325 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1329 dmode = vap->va_mode & 0777;
1332 * Must simulate part of ufs_makeinode here to acquire the inode,
1333 * but not have it entered in the parent directory. The entry is
1334 * made later after writing "." and ".." entries.
1336 error = ffs_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1340 ip->i_gid = dp->i_gid;
1344 struct ucred ucred, *ucp;
1348 * If we are hacking owners here, (only do this where told to)
1349 * and we are not giving it TO root, (would subvert quotas)
1350 * then go ahead and give it to the other user.
1351 * The new directory also inherits the SUID bit.
1352 * If user's UID and dir UID are the same,
1353 * 'give it away' so that the SUID is still forced on.
1355 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1356 (dp->i_mode & ISUID) && dp->i_uid) {
1358 ip->i_uid = dp->i_uid;
1360 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1362 * Make sure the correct user gets charged
1364 * Make a dummy credential for the victim.
1365 * XXX This seems to never be accessed out of
1366 * our context so a stack variable is ok.
1369 ucred.cr_uid = ip->i_uid;
1370 ucred.cr_ngroups = 1;
1371 ucred.cr_groups[0] = dp->i_gid;
1376 ip->i_uid = cnp->cn_cred->cr_uid;
1378 if ((error = ufs_getinoquota(ip)) ||
1379 (error = ufs_chkiq(ip, 1, ucp, 0))) {
1380 ffs_vfree(tvp, ip->i_number, dmode);
1386 #else /* !SUIDDIR */
1387 ip->i_uid = cnp->cn_cred->cr_uid;
1389 if ((error = ufs_getinoquota(ip)) ||
1390 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
1391 ffs_vfree(tvp, ip->i_number, dmode);
1396 #endif /* !SUIDDIR */
1397 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1399 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1402 if (DOINGSOFTDEP(tvp))
1403 softdep_change_linkcnt(ip);
1404 if (cnp->cn_flags & CNP_ISWHITEOUT)
1405 ip->i_flags |= UF_OPAQUE;
1408 * Bump link count in parent directory to reflect work done below.
1409 * Should be done before reference is created so cleanup is
1410 * possible if we crash.
1414 dp->i_flag |= IN_CHANGE;
1415 if (DOINGSOFTDEP(dvp))
1416 softdep_change_linkcnt(dp);
1417 error = ffs_update(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1422 * The vnode must have a VM object in order to issue buffer cache
1425 vinitvmio(tvp, DIRBLKSIZ);
1428 * Initialize directory with "." and ".." from static template.
1430 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1431 dtp = &mastertemplate;
1433 dtp = (struct dirtemplate *)&omastertemplate;
1435 dirtemplate.dot_ino = ip->i_number;
1436 dirtemplate.dotdot_ino = dp->i_number;
1437 vnode_pager_setsize(tvp, DIRBLKSIZ);
1438 error = VOP_BALLOC(tvp, 0LL, DIRBLKSIZ, cnp->cn_cred, B_CLRBUF, &bp);
1441 ip->i_size = DIRBLKSIZ;
1442 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1443 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1444 if (DOINGSOFTDEP(tvp)) {
1446 * Ensure that the entire newly allocated block is a
1447 * valid directory so that future growth within the
1448 * block does not have to ensure that the block is
1449 * written before the inode.
1452 while (blkoff < bp->b_bcount) {
1454 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1455 blkoff += DIRBLKSIZ;
1458 if ((error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) |
1459 DOINGASYNC(tvp)))) != 0) {
1464 * Directory set up, now install its entry in the parent directory.
1466 * If we are not doing soft dependencies, then we must write out the
1467 * buffer containing the new directory body before entering the new
1468 * name in the parent. If we are doing soft dependencies, then the
1469 * buffer containing the new directory body will be passed to and
1470 * released in the soft dependency code after the code has attached
1471 * an appropriate ordering dependency to the buffer which ensures that
1472 * the buffer is written before the new name is written in the parent.
1474 if (DOINGASYNC(dvp))
1476 else if (!DOINGSOFTDEP(dvp) && (error = bwrite(bp)) != 0)
1478 ufs_makedirentry(ip, cnp, &newdir);
1479 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1483 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1488 dp->i_flag |= IN_CHANGE;
1489 if (DOINGSOFTDEP(dvp))
1490 softdep_change_linkcnt(dp);
1492 * No need to do an explicit VOP_TRUNCATE here, vrele will
1493 * do this for us because we set the link count to 0.
1497 ip->i_flag |= IN_CHANGE;
1498 if (DOINGSOFTDEP(tvp))
1499 softdep_change_linkcnt(ip);
1507 * Rmdir system call.
1509 * ufs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1510 * struct componentname *a_cnp)
1514 ufs_rmdir(struct vop_old_rmdir_args *ap)
1516 struct vnode *vp = ap->a_vp;
1517 struct vnode *dvp = ap->a_dvp;
1518 struct componentname *cnp = ap->a_cnp;
1519 struct inode *ip, *dp;
1526 * Do not remove a directory that is in the process of being renamed.
1527 * Verify the directory is empty (and valid). Rmdir ".." will not be
1528 * valid since ".." will contain a reference to the current directory
1529 * and thus be non-empty. Do not allow the removal of mounted on
1530 * directories (this can happen when an NFS exported filesystem
1531 * tries to remove a locally mounted on directory).
1534 if (ip->i_flag & IN_RENAME) {
1538 if (ip->i_effnlink != 2 ||
1539 !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1543 if ((dp->i_flags & APPEND)
1544 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1549 * Delete reference to directory before purging
1550 * inode. If we crash in between, the directory
1551 * will be reattached to lost+found,
1555 if (DOINGSOFTDEP(vp)) {
1556 softdep_change_linkcnt(dp);
1557 softdep_change_linkcnt(ip);
1559 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1563 if (DOINGSOFTDEP(vp)) {
1564 softdep_change_linkcnt(dp);
1565 softdep_change_linkcnt(ip);
1569 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1571 * Truncate inode. The only stuff left in the directory is "." and
1572 * "..". The "." reference is inconsequential since we are quashing
1573 * it. The soft dependency code will arrange to do these operations
1574 * after the parent directory entry has been deleted on disk, so
1575 * when running with that code we avoid doing them now.
1577 if (!DOINGSOFTDEP(vp)) {
1579 dp->i_flag |= IN_CHANGE;
1581 ip->i_flag |= IN_CHANGE;
1582 ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
1583 error = ffs_truncate(vp, (off_t)0, ioflag, cnp->cn_cred);
1585 /* cache_purge removed - handled by VFS compat layer */
1587 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1588 if (ip->i_dirhash != NULL)
1589 ufsdirhash_free(ip);
1592 VN_KNOTE(vp, NOTE_DELETE);
1597 * symlink -- make a symbolic link
1599 * ufs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1600 * struct componentname *a_cnp, struct vattr *a_vap,
1605 ufs_symlink(struct vop_old_symlink_args *ap)
1607 struct vnode *vp, **vpp = ap->a_vpp;
1611 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1615 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1617 len = strlen(ap->a_target);
1618 if (len < vp->v_mount->mnt_maxsymlinklen) {
1620 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1622 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1625 * Make sure we have a VM object in order to use
1628 if (vp->v_object == NULL)
1630 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1631 UIO_SYSSPACE, IO_NODELOCKED,
1632 ap->a_cnp->cn_cred, NULL);
1640 * Vnode op for reading directories.
1642 * ufs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
1643 * int *a_eofflag, int *ncookies, off_t **a_cookies)
1647 ufs_readdir(struct vop_readdir_args *ap)
1649 struct uio *uio = ap->a_uio;
1650 struct vnode *vp = ap->a_vp;
1655 int offset; /* offset into buffer cache buffer */
1656 int eoffset; /* end of buffer clipped to file EOF */
1657 int pickup; /* pickup point */
1662 if (uio->uio_offset < 0)
1665 * Guess the number of cookies needed. Make sure we compute at
1666 * least 1, and no more then a reasonable limit.
1668 if (ap->a_ncookies) {
1669 ncookies = uio->uio_resid / 16 + 1;
1670 if (ncookies > 1024)
1672 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1674 ncookies = -1; /* force conditionals below */
1679 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1685 if (uio->uio_offset >= VTOI(vp)->i_size) {
1688 if (ap->a_ncookies) {
1689 *ap->a_ncookies = cookie_index;
1690 *ap->a_cookies = cookies;
1696 * Loop until we run out of cookies, we run out of user buffer,
1697 * or we hit the directory EOF.
1699 * Always start scans at the beginning of the buffer, don't trust
1700 * the offset supplied by userland.
1702 while ((error = ffs_blkatoff_ra(vp, uio->uio_offset, NULL, &bp, 2)) == 0) {
1703 pickup = (int)(uio->uio_offset - bp->b_loffset);
1706 if (bp->b_loffset + bp->b_bcount > VTOI(vp)->i_size)
1707 eoffset = (int)(VTOI(vp)->i_size - bp->b_loffset);
1709 eoffset = bp->b_bcount;
1711 while (offset < eoffset) {
1712 dp = (struct direct *)(bp->b_data + offset);
1713 if (dp->d_reclen <= 0 || (dp->d_reclen & 3) ||
1714 offset + dp->d_reclen > bp->b_bcount) {
1718 if (offsetof(struct direct, d_name[dp->d_namlen]) > dp->d_reclen) {
1722 if (offset < pickup) {
1723 offset += dp->d_reclen;
1726 #if BYTE_ORDER == LITTLE_ENDIAN
1728 retval = vop_write_dirent(&error, uio,
1729 dp->d_ino, dp->d_namlen, dp->d_type,
1734 retval = vop_write_dirent(&error, uio,
1735 dp->d_ino, dp->d_type, dp->d_namlen,
1741 cookies[cookie_index] = 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, &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 kprintf("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)
1941 ufsspec_close(struct vop_close_args *ap)
1943 struct vnode *vp = ap->a_vp;
1945 if (vp->v_sysref.refcnt > 1)
1947 return (VOCALL(&spec_vnode_vops, &ap->a_head));
1951 * Read wrapper for fifos.
1953 * ufsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1954 * struct ucred *a_cred)
1958 ufsfifo_read(struct vop_read_args *ap)
1965 resid = uio->uio_resid;
1966 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1967 ip = VTOI(ap->a_vp);
1968 if ((ap->a_vp->v_mount->mnt_flag & MNT_NOATIME) == 0 && ip != NULL &&
1969 (uio->uio_resid != resid || (error == 0 && resid != 0)))
1970 VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1975 * Write wrapper for fifos.
1977 * ufsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1978 * struct ucred *a_cred)
1982 ufsfifo_write(struct vop_write_args *ap)
1989 resid = uio->uio_resid;
1990 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1991 ip = VTOI(ap->a_vp);
1992 if (ip != NULL && (uio->uio_resid != resid || (error == 0 && resid != 0)))
1993 VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1998 * Close wrapper for fifos.
2000 * Update the times on the inode then do device close.
2002 * ufsfifo_close(struct vnode *a_vp, int a_fflag)
2006 ufsfifo_close(struct vop_close_args *ap)
2008 struct vnode *vp = ap->a_vp;
2010 if (vp->v_sysref.refcnt > 1)
2012 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2016 * Kqfilter wrapper for fifos.
2018 * Fall through to ufs kqfilter routines if needed
2022 ufsfifo_kqfilter(struct vop_kqfilter_args *ap)
2026 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2028 error = ufs_kqfilter(ap);
2033 * Return POSIX pathconf information applicable to ufs filesystems.
2035 * ufs_pathconf(struct vnode *a_vp, int a_name, int *a_retval)
2039 ufs_pathconf(struct vop_pathconf_args *ap)
2041 switch (ap->a_name) {
2043 *ap->a_retval = LINK_MAX;
2046 *ap->a_retval = NAME_MAX;
2049 *ap->a_retval = PATH_MAX;
2052 *ap->a_retval = PIPE_BUF;
2054 case _PC_CHOWN_RESTRICTED:
2067 * Advisory record locking support
2069 * ufs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
2074 ufs_advlock(struct vop_advlock_args *ap)
2076 struct inode *ip = VTOI(ap->a_vp);
2078 return (lf_advlock(ap, &(ip->i_lockf), ip->i_size));
2082 * Initialize the vnode associated with a new inode, handle aliased
2085 * Make sure directories have their VM object now rather then later,
2086 * saving us from having to check on all the myrid directory VOPs
2087 * that might be executed without a VOP_OPEN being performed.
2090 ufs_vinit(struct mount *mntp, struct vnode **vpp)
2099 vp->v_type = IFTOVT(ip->i_mode);
2101 switch(vp->v_type) {
2104 vp->v_ops = &mntp->mnt_vn_spec_ops;
2105 addaliasu(vp, umajor(ip->i_rdev), uminor(ip->i_rdev));
2108 vp->v_ops = &mntp->mnt_vn_fifo_ops;
2112 vinitvmio(vp, ip->i_size);
2115 if (ip->i_size >= vp->v_mount->mnt_maxsymlinklen)
2116 vinitvmio(vp, ip->i_size);
2123 if (ip->i_number == ROOTINO)
2124 vp->v_flag |= VROOT;
2126 * Initialize modrev times
2128 getmicrouptime(&tv);
2129 SETHIGH(ip->i_modrev, tv.tv_sec);
2130 SETLOW(ip->i_modrev, tv.tv_usec * 4294);
2136 * Allocate a new inode.
2140 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2141 struct componentname *cnp)
2143 struct inode *ip, *pdir;
2144 struct direct newdir;
2150 if ((mode & IFMT) == 0)
2153 error = ffs_valloc(dvp, mode, cnp->cn_cred, &tvp);
2157 ip->i_gid = pdir->i_gid;
2161 struct ucred ucred, *ucp;
2165 * If we are not the owner of the directory,
2166 * and we are hacking owners here, (only do this where told to)
2167 * and we are not giving it TO root, (would subvert quotas)
2168 * then go ahead and give it to the other user.
2169 * Note that this drops off the execute bits for security.
2171 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2172 (pdir->i_mode & ISUID) &&
2173 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2174 ip->i_uid = pdir->i_uid;
2178 * Make sure the correct user gets charged
2180 * Quickly knock up a dummy credential for the victim.
2181 * XXX This seems to never be accessed out of our
2182 * context so a stack variable is ok.
2185 ucred.cr_uid = ip->i_uid;
2186 ucred.cr_ngroups = 1;
2187 ucred.cr_groups[0] = pdir->i_gid;
2191 ip->i_uid = cnp->cn_cred->cr_uid;
2194 if ((error = ufs_getinoquota(ip)) ||
2195 (error = ufs_chkiq(ip, 1, ucp, 0))) {
2196 ffs_vfree(tvp, ip->i_number, mode);
2202 #else /* !SUIDDIR */
2203 ip->i_uid = cnp->cn_cred->cr_uid;
2205 if ((error = ufs_getinoquota(ip)) ||
2206 (error = ufs_chkiq(ip, 1, cnp->cn_cred, 0))) {
2207 ffs_vfree(tvp, ip->i_number, mode);
2212 #endif /* !SUIDDIR */
2213 ip->i_fsmid = cache_getnewfsmid();
2214 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2216 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2219 if (DOINGSOFTDEP(tvp))
2220 softdep_change_linkcnt(ip);
2221 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2222 suser_cred(cnp->cn_cred, 0)) {
2223 ip->i_mode &= ~ISGID;
2226 if (cnp->cn_flags & CNP_ISWHITEOUT)
2227 ip->i_flags |= UF_OPAQUE;
2230 * Regular files and directories need VM objects. Softlinks do
2231 * not (not immediately anyway).
2233 if (tvp->v_type == VREG || tvp->v_type == VDIR)
2237 * Make sure inode goes to disk before directory entry.
2239 error = ffs_update(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2242 ufs_makedirentry(ip, cnp, &newdir);
2243 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2251 * Write error occurred trying to update the inode
2252 * or the directory so must deallocate the inode.
2256 ip->i_flag |= IN_CHANGE;
2257 if (DOINGSOFTDEP(tvp))
2258 softdep_change_linkcnt(ip);
2264 ufs_missingop(struct vop_generic_args *ap)
2266 panic("no vop function for %s in ufs child", ap->a_desc->sd_name);
2267 return (EOPNOTSUPP);
2270 static struct filterops ufsread_filtops =
2271 { 1, NULL, filt_ufsdetach, filt_ufsread };
2272 static struct filterops ufswrite_filtops =
2273 { 1, NULL, filt_ufsdetach, filt_ufswrite };
2274 static struct filterops ufsvnode_filtops =
2275 { 1, NULL, filt_ufsdetach, filt_ufsvnode };
2278 * ufs_kqfilter(struct vnode *a_vp, struct knote *a_kn)
2281 ufs_kqfilter(struct vop_kqfilter_args *ap)
2283 struct vnode *vp = ap->a_vp;
2284 struct knote *kn = ap->a_kn;
2287 switch (kn->kn_filter) {
2289 kn->kn_fop = &ufsread_filtops;
2292 kn->kn_fop = &ufswrite_filtops;
2295 kn->kn_fop = &ufsvnode_filtops;
2301 kn->kn_hook = (caddr_t)vp;
2303 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2304 SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
2305 lwkt_reltoken(&ilock);
2311 filt_ufsdetach(struct knote *kn)
2313 struct vnode *vp = (struct vnode *)kn->kn_hook;
2316 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2317 SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
2318 kn, knote, kn_selnext);
2319 lwkt_reltoken(&ilock);
2324 filt_ufsread(struct knote *kn, long hint)
2326 struct vnode *vp = (struct vnode *)kn->kn_hook;
2327 struct inode *ip = VTOI(vp);
2330 * filesystem is gone, so set the EOF flag and schedule
2331 * the knote for deletion.
2333 if (hint == NOTE_REVOKE) {
2334 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2338 kn->kn_data = ip->i_size - kn->kn_fp->f_offset;
2339 return (kn->kn_data != 0);
2344 filt_ufswrite(struct knote *kn, long hint)
2347 * filesystem is gone, so set the EOF flag and schedule
2348 * the knote for deletion.
2350 if (hint == NOTE_REVOKE)
2351 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2358 filt_ufsvnode(struct knote *kn, long hint)
2360 if (kn->kn_sfflags & hint)
2361 kn->kn_fflags |= hint;
2362 if (hint == NOTE_REVOKE) {
2363 kn->kn_flags |= EV_EOF;
2366 return (kn->kn_fflags != 0);
2369 /* Global vfs data structures for ufs. */
2370 static struct vop_ops ufs_vnode_vops = {
2371 .vop_default = vop_defaultop,
2372 .vop_fsync = (void *)ufs_missingop,
2373 .vop_read = (void *)ufs_missingop,
2374 .vop_reallocblks = (void *)ufs_missingop,
2375 .vop_write = (void *)ufs_missingop,
2376 .vop_access = ufs_access,
2377 .vop_advlock = ufs_advlock,
2378 .vop_bmap = ufs_bmap,
2379 .vop_old_lookup = ufs_lookup,
2380 .vop_close = ufs_close,
2381 .vop_old_create = ufs_create,
2382 .vop_getattr = ufs_getattr,
2383 .vop_inactive = ufs_inactive,
2384 .vop_old_link = ufs_link,
2385 .vop_old_mkdir = ufs_mkdir,
2386 .vop_old_mknod = ufs_mknod,
2387 .vop_mmap = ufs_mmap,
2388 .vop_open = ufs_open,
2389 .vop_pathconf = ufs_pathconf,
2390 .vop_poll = vop_stdpoll,
2391 .vop_kqfilter = ufs_kqfilter,
2392 .vop_print = ufs_print,
2393 .vop_readdir = ufs_readdir,
2394 .vop_readlink = ufs_readlink,
2395 .vop_reclaim = ufs_reclaim,
2396 .vop_old_remove = ufs_remove,
2397 .vop_old_rename = ufs_rename,
2398 .vop_old_rmdir = ufs_rmdir,
2399 .vop_setattr = ufs_setattr,
2400 .vop_strategy = ufs_strategy,
2401 .vop_old_symlink = ufs_symlink,
2402 .vop_old_whiteout = ufs_whiteout
2405 static struct vop_ops ufs_spec_vops = {
2406 .vop_default = spec_vnoperate,
2407 .vop_fsync = (void *)ufs_missingop,
2408 .vop_access = ufs_access,
2409 .vop_close = ufsspec_close,
2410 .vop_getattr = ufs_getattr,
2411 .vop_inactive = ufs_inactive,
2412 .vop_print = ufs_print,
2413 .vop_read = ufsspec_read,
2414 .vop_reclaim = ufs_reclaim,
2415 .vop_setattr = ufs_setattr,
2416 .vop_write = ufsspec_write
2419 static struct vop_ops ufs_fifo_vops = {
2420 .vop_default = fifo_vnoperate,
2421 .vop_fsync = (void *)ufs_missingop,
2422 .vop_access = ufs_access,
2423 .vop_close = ufsfifo_close,
2424 .vop_getattr = ufs_getattr,
2425 .vop_inactive = ufs_inactive,
2426 .vop_kqfilter = ufsfifo_kqfilter,
2427 .vop_print = ufs_print,
2428 .vop_read = ufsfifo_read,
2429 .vop_reclaim = ufs_reclaim,
2430 .vop_setattr = ufs_setattr,
2431 .vop_write = ufsfifo_write
2434 VNODEOP_SET(ufs_vnode_vops);
2435 VNODEOP_SET(ufs_spec_vops);
2436 VNODEOP_SET(ufs_fifo_vops);
2442 ufs_vnoperate(struct vop_generic_args *ap)
2444 return (VOCALL(&ufs_vnode_vops, ap));
2448 * ufs_vnoperatefifo()
2451 ufs_vnoperatefifo(struct vop_generic_args *ap)
2453 return (VOCALL(&ufs_fifo_vops, ap));
2457 * ufs_vnoperatespec()
2460 ufs_vnoperatespec(struct vop_generic_args *ap)
2462 return (VOCALL(&ufs_spec_vops, ap));